Genetics and the Epidemiology of Chronic Diseases Edited by James V. NEeL Margery W. SHAW Wirriam J. ScHuLn Department of Human Genetics University of Michigan Medical School Ann Arbor, Michigan A Symposium June 17-19, 1963, Ann Arbor, Michigan, Sponsored by U.S. Public Health Service, University of Michigan Medical School, University of Michigan School of Public Health U.S. DEPARTMENT OF HEALTH, EDUCATION, AND WELFARE Public Health Service Division of Chronic Diseases Washington, D.C. 20201 PUBLIC HEALTHY Public Health Service Publication No. 1163 February 1965 For sale by the Superintendent of Documents, U.S. Government Printing Office Washington, D.C., 20402 - Price $1.50 cents PUBLIC HEALTH P refa ce fa The papers which follow were presented at a Symposium on Con- tributions of Genetics to Epidemiologic Studies of Chronic Diseases, held at Ann Arbor, Michigan, June 17-19, 1963, and co-sponsored by the Division of Chronic Diseases, U.S. Public Health Service, and the Schools of Medicine and Public Health of the University of Michigan. The purpose of the symposium was to consider the role and limita- tions of the genetic approach in studies concerned with the epidemi- ology of a variety of chronic diseases. The papers presented have been organized under five headings, as follows: 1) Basic genetic and epidemiologic principles; 2) Critical reviews of selected, illustrative problems; 3) Problems in experimental design; 4) Papers describing current genetico-epidemiologic studies; and 5) Genetic counseling. The first three groups are self-explanatory, but the fourth and fifth require a word of comment. The papers comprising the fourth group were delivered in a series of workshops in which the emphasis was as much on methodology and the limitations inherent in certain approaches as on actual results. An effort was made to ensure that these contributions mirrored the types of problems with which the epidemiologist and his confreres of the 1960’s are occupied. The paper on genetic counseling was included in response to an expressed need for consideration of a topic which is increasingly coming to the attention of state health organizations. The arrangers of the program and the editors are most grateful for the assistance in the planning rendered by the Symposium Planning Committee, the valuable services rendered by the Department of Post- graduate Medicine of the University of Michigan Medical School in the conduct of the program, and gratefully acknowledge financial assistance in the publication of the symposium proceedings by the Division of Chronic Diseases, the National Cancer Institute, and the National Institute of Arthritis and Metabolic Diseases of the Public Health Service. 296 Im SYMPOSIUM PLANNING COMMITTEE DR. MARGERY W. SHAW, DR. ALicE CHENOWETH, University of Michigan Children’s Bureau, Welfare Adminis- Dr. MERTON HONEYMAN, tration, U.S. Department of Health, Connecticut State Health Department Education, and Welfare Dr. LEE ScHACHT, Minnesota State Health Department PUBLIC HEALTH SERVICE U.S. DEPARTMENT OF HEALTH, EDUCATION, AND WELFARE DR. SIMON ABRAHAMS, Dr. GEORGE K. TOKUHATA, Division of Radiological Health Division of Chronic Diseases (Chair- Dr. WILFRED DaAviD, man) Division of Chronic Diseases Dr. KATHERINE S. WILSON, Dr. SAMUEL Fox, National Institutes of Health Heart Disease Control Branch, Divi- Dr. CarL WITKOP, JR., sion of Chronic Diseases National Institutes of Health Dr. GLEN McDoNALD, Diabetes and Arthritis Branch, Divi- ston of Chronic Diseases Iv Table of Contents PART I PREFACE... .....ccmmmm smn m mmm mm mmm mmm mm mmm mmm BASIC GENETIC AND EPIDEMIOLOGICAL PRINCI- PLES Genetics and Epidemiology, by Thomas Francis, Jr______ Genetic Studies of Family Units, by Bertram L. Hanna __ Problems of Ascertainment in the Analysis of Family Data, by James F. Crow_ ________________________ Estimation of Genetic Parameters in Population Studies, by William J. Schull _____________________ _______ Hereditary Factors Elucidated by Twin Studies, by Bent Harvald and M. Hauge _________________________ » ABO Blood Groups, Secretor Status, and Susceptibility to Chronic Diseases: An Example of a Genetic Basis for Family Predispositions, by J. A. Fraser Roberts__ ~ Epidemiological Concepts Applied to Studies of Chronic Diseases, by Abraham M. Lilienfeld. ______________ PART II THE EVALUATION OF GENETIC FACTORS IN SE- LECTED ILLUSTRATIVE DISEASES Diabetes Mellitus, by James V. Neel, Stefan S. Fajans, Jerome W. Conn, and Ruth T. Davidson___________ Coronary Artery Disease, by Victor A. McKusick ~ ____ “The Geographic and Genetic Characteristics of Amyo- trophic Lateral Sclerosis and Other Selected Chronic Neurological Diseases, by Leonard T. Kurland ______ PART III PROBLEMS IN THE EXPERIMENTAL DESIGN OF EPIDEMIOLOGICAL STUDIES Selection Techniques for Rare Traits, by Leslie Kish____ The Reliability of Survey Data, by Charles F. Cannell ___ A Trial of Methods for Collecting Household Morbidity Data, by Kenneth R. Wilcox, Jr__________________ Page III 23 45 61 77 87 105 133 145 165 177 PART IV CURRENT EPIDEMIOLOGICAL PROBLEMS Intrafamilial Transmission of Rheumatoid Arthritis, by Sidney Cobb________________ Family and Genetic Studies of Rheumatoid Arthritis and Rheumatoid Factor in Blackfeet Indians, by Thomas A. Burch, William M. O’Brien, and Joseph J. Bunim_ LONGEVITY AND CORONARY ARTERY DISEASE Family Patterns of Longevity and Mortality, by Bernice H.Cohen_____________________________________ Coronary Heart Disease in Relation to Blood Pressure and Cholesterol Levels in Population Studies, by Fredrick H. Epstein and Marcus O. Kjelsberg_ _ ____ Inherited Antigenic Differences in Serum Beta Lipopro- teins in Relation to Coronary Artery Disease and Diabetes, by Baruch S. Blumberg, Barbara Mishell and Sam Visnich________________________________ REGIONAL VARIATIONS IN MORBIDITY AND MOR- TALITY Regional Variations in the Incidence of Spina Bifida, by David Hewitt___________________________________ Vital Record Incidence of Congenital Malformations in New York State, by A. M. Gittelsohn and S. Milham, Jr_ Methodological Problems in the Use of Standard Vital Statistical Data in the Study of Neonatal Mortality and Birth Weight, by Douglas Grahn______________ MALIGNANT DISEASES Familial Factors in Lung Cancer and Smoking, by George K. Tokuhata___________________ _______________ Do Malignancies Result from Diagnostic and Therapeutic Radiation? by T. Sterling, E. Saenger, and R. A. BOUUTBT. cnx wis msin 0 misono sms vio moms i am smn Congenital Malformations and Cancer in the Sibships of Leukemic Children, by Robert W. Miller PART V GENETIC COUNSELING Some Aspects of Genetic Counseling, by Franz J. Kall- vi Page 209 227 237 265 279 295 305 321 339 365 373 PART I Basic Genetic and Epidemiological Principles Genetics and Epidemiology Opening Comments By Tuomas Francis, Jr., M.D., Henry Sewall University Professor of Epidemiology, Chairman of the Department of Epidemiology, School of Public Health, and Professor of Epidemiology, Department of Pediatrics ana Communicable Diseases, Medical School, University of Michigan, Ann Arbor, Mich. I would like to join Dr. Neel in welcoming you to Michigan. The excellent program he and his associates have arranged with you and for you gives promise of stimulation and effect. We are fortunate in our environment here that the close relationship between the Departments of Epidemiology and Human Genetics provides a constant opportunity for interchange in study and ideas. We follow the current trend toward desegregation in these fields and hope that the present pro- gram will be of general aid in this direction. The type of opening comments with which I am most familiar are phrases such as pass, or I can open, or I'll open for a couple. This relates to a very complicated intellectual exercise which I can explain in a form of genetic vulgate. It requires a certain supply of un- restricted funds. A group of skilled persons compete in trying to develop a formula which most successfully explains the configuration and structure of a hand. There is a pool of genes from which one can draw or they are distributed, and the contestant attempts to fit them into sequences or combinations. They have different degrees of penetrance. An effort is made to establish linkages, sometimes known as pairs; they may even be recognized by the theorist in peculiar positions—for example there is a pairing known as back to back which has a special power. But in addition, the effort may be to find the correct gene to fit a given locus. In some sequences the dominant characteristic results in a hand with a flush; in others the proper arrangement emphasizes the dominant feature of the hand as straight. In these situations an effort is made to match the items as homozygotes, and this adds further power whereas the appearance of a heterozygote may not provide the essential combination and the experimenter, recognizing that his hypothesis of hand formation is not further tenable, will announce that he drops out and forfeits his chance and will shortly undertake a new formulation. Another type of combination is seen when the same code number may reappear in successions and pairs. This may result in an effort toward explanation known as Queens full of genes. There is a distinct statistical probablility which applies to these formulations, but at times the concept of wild genes is introduced, meaning that certain of them can be accepted in any locus and with increased dominance, so that strong combinations may rapidly be formed and their value supported, but the statistical significance is often difficult to calculate. The investigator who completes and defends his formula of hand formation most effectively is rewarded by a modified Mendelian prize—not of colored beans, but a collection of discs. Instead of black and white, however, they are commonly blue, red, and white. The successful contestant then arranges them in piles and calculates their distribution until the next exercise is begun. As a result of his suc- cessful manipulations of the genes, the winner earns the title, Genius. This brief and incomplete discussion can only mention that the exercise has different sets of guidelines. One form is called penny ante—named for the character in a child’s book, ‘Henny Penny.” It means the contestants are chicken. Another level which is under- taken with grave demeanor and full appreciation of the values involved commonly requires limited light, a haze of tobacco incense and some- times other stimuli, so that the persons will poke about in the gene pool with better appreciation of its content. I could describe this better had I a better knowledge of the genetic code. This is to welcome you to the game and hope that you will defend your formulations with judgment and enthusiasm, recognizing, how- ever, that a sequence of four does not gain acceptance at full value. Other contestants may demand full evidence or offer competing infor- mation of greater power. Ladies and gentlemen, I shall now proceed to open. We have previously described epidemiology as the study of the laws and factors governing the occurrence and distribution of disease and disorder in a population. These factors include the characteristics of the population, the causative agencies, and the biological, social, and physical environment. Of all the factors which may be involved in the occurrence or absence of disease, the genetic influence is probably the most pervasive. Our very presence here to discuss the effect of 2 our past upon our future is the consequence of our heritage. But even casual reflection leads to the realization that this survival cannot be attributed to any one item in our genetic code; rather the influences are multiple or polygenic. It is the combination which counts. This adds to the complexity of an already complicated problem for, as Mayr (1961) points out, there is as yet no adequate analysis of a single polygenic trait in man, although they include characteristics that are particularly important to mankind. Similarly, there are few, if any, disorders whose occurrence is not the resultant of multiple convergent factors. So when the human geneticist turns to disease and disorder in the population as his basis of genetic analysis, he is promptly in epidemiology. And when he asserts the concept of multiple factors to produce an effect, he is in full cry epidemiologically. Conversely, where the epidemiologist seeks explanation for familial or other group aggregations of health or disease, he is immediately involved in genetic problems. In each case the investigator is concerned with “popula- tion thinking” (Mayr, 1961), and the beginning of formation of an epidemosis (Francis, 1959). An inherited characteristic may be brought to view only by appro- priate challenge. It has been repeatedly emphasized that whether a gene is harmful or useful or indifferent is commonly related to the environment in which the bearer is exposed. The genetic epidemiol- ogist then becomes ecologist seeking significant correlations between a group disorder and one or other stress from the great array of en- vironmental influences. The success of these efforts will obviously be related to the uniqueness of the stress and the directness of its effect, to the frequency of the basic defect and to the ease of detection of the disorder under consideration. Geographic variations may be valu- able but also confusing. And even if one finds a correlation, perhaps all he can say is, “Redheads are more sensitive to heat.” It is clear that this type of extensive and intensive investigation is not entered into readily, unless the problem is of major importance. Neverthe- less, it is the starting point for mapping the numerous genetic factors which go into determining resistance or susceptibility of a population to pathogenic stresses in the environment. In a heterogeneous popu- lation and a rapidly changing environment the possibilities seem infinite. The difficulty in establishing genetic correlations for certain dis- orders, as distinct from environmental effects, has been well recognized. Nevertheless, it is one of the basic hopes of the Community Health Study which we are conducting in Tecumseh, Michigan. Taking atherosclerosis and coronary heart disease as an example, the hypothe- sis is briefly this: Atherosclerosis and coronary heart disease are the result of a basic metabolic disturbance. Despite much discussion and inferential data, it is not established that coronary heart disease has a significant familial distribution nor has the weight of genetic factors 3 versus environmental influences in causation been adequately deter- mined. Nevertheless, when a case of coronary heart disease occurs, the probability is that additional cases are more likely to occur in the family of the index case than in similar families in which there is no identified case. This should be true whether the determining factor is genetic or environmental, since the household is more likely to be sharing a common environment and way of life; the blood kindred is also more likely to be sharing the same genetic determinants. The kindred thus constitutes a pool of susceptibles for continued study in comparison with other unaffected kindreds. The total community was identified as individuals, as households, and as genetic kindreds. In 1959-60 extensive histories were obtained and careful physical examinations were made by physicians from the clinical departments of the medical school in the clinic established at Tecumseh. A variety of laboratory tests included X-ray, EKG, de- terminations of blood cholesterol, sugar, uric acid, proteins, and urine examinations. In this community of 10,000 people there are 4,500 kindreds and something over 2,000 households. Nearly 90 percent of all ages participated in the study, and those who didn’t were fully identified. Diagnostic criteria were carefully established and utilized after complete review of the findings. The entire procedure has re- quired great effort and much time, so that many of the fascinating data are but now becoming available for proper analyses. One feature of value has been the division of the population into ten representative samples, so that the rotation through the clinic always constitutes a representative segment of the population. Special studies can then be conducted without having to draw new samples and without involv- ing the entire population. The second round of interviews and examinations began a year ago. They have been enlarged in certain details of the history and in some phases of the physical and laboratory examinations. The partici- pation continues high. A continued surveillance of the community and the people is also being established. There is need to emphasize that the components of special interest come out of this natural community in a natural unselected manner. They exist in a natural ecologic setting. Their position as individuals, as households or kindreds in the study derives from the data obtained by methods of interview and examination uniformly applied to the total population, not by prior selection. The various data can be viewed independent of any clinical diagnosis, as indices of suspicion, by observing the subsequent course of those with and without the particular characteristic. It is here that the oncoming members, the children and younger segments, of the hypothetical susceptibles can be further compared with matched groups or kindreds of presumably nonsusceptibles. In this manner the natural risk of developing a disorder can be measured against a general ecological background 4 which will hopefully bring out the individual or group variation. It is expected that the role of the genetic factor in aggregation of disorder can be established not only for coronary heart disease, but for diabetes, hypertension, chronic pulmonary disease, rheumatoid disease, and others. They have numerous overlapping features, so that etiologic relationships of various abnormalities and their combinations may be considered in a common host population. For example, are coronary heart disease and diabetes different manifestations of the same basic disorder, of the same polymorphism? A still further objective is to find new ways of determining the nature of susceptibility through research in basic science—physiologic, biochemical, physical, etc., applied to the community; and as already indicated, efforts will be made to detect the beginnings or precursory stages. Here the careful continued study of the younger age groups and the respective cohorts of the kindreds are expected to be the most rewarding in the search for, or application of, preventive measures. Continued ecologic observation should also aid in revealing natural exposures or changes which would constitute environmental patho- genic agencies. In support of that objective a mapping of the biological, social, and physical environment is being made. The role of physical activity, as well as the social and psychological reactions of the individuals are to be incorporated with other data of possible importance. I think it is safe to say that comprehensive epidemio- logical information of high clinical excellence is for the first time becoming available from a total natural community. The question of numbers is always before us, but our purpose is not merely one of measuring incidence. It is fair to say that there are enough cases of our target disorders to provide a good collection of potential susceptibles for detailed and continued study. Attention to these and their comparison groups will be enhanced in the sense of biological and clinical investigation, while that to the total population can be reduced in intensity. The prospects for the ongoing study are that the genetic information from this epidemiological base will contribute new, valuable data for comparison with environmental influences at the same time. It is apparent, however, that the pathologic index of coronary occlusion or even probable coronary heart disease may be unsuitable for analysis of oligogenic determinants. The case of acute coronary occlusion may be, and I suspect this is so, only the accident in the course of a basic aberration present in the population generally. In that event it would require that the genetic determinant apply to the accident rather than to the underlying disturbance, which would seem more difficult of achievement. This emphasizes also that for our purposes it is necessary to project inquiry beyond a demonstration of a correlation between some factor and a result, in order to determine 5 the pathogenic mechanism which may then be approached prophylac- tically or therapeutically. The influence of environment upon genetic expression is generally considered in terms of the external environment. The genetic makeup and potential, on the other hand, is thought of as part of the internal environment, intrinsic, inherent, and immutable. This ignores the possibility that other components of the internal environment may play a role in altering the genetic effect; that mutations may not be quite as chancy as they seem. It is not within my competence to discuss this in detail except to point out that a variety of presumably external influences may have an intimate relation to the internal genetic mechanisms. Increasing knowledge of viral infections and cellular biology demonstrate the manner in which viruses may par- ticipate in and alter genetic mechanisms of cells, particularly when they persist in a temperate or latent stage as part of the internal or intracellular environment. This may result in new karyotypes, in new antigens, in escape from general regulatory mechanisms. Viruses may exert genic repressions. At what stage do these somatic effects influence organismal heredity? The dominant influence genetics had earlier exerted upon the interpretation of spontaneous tumors or leukemia in different strains of mice may in reality relate more to susceptibility to a virus than to tumor production per se. Moreover, viruses may be constant accompaniments of the reproductive process. Can a virus which has wide distribution through the blood stream induce sufficient variation in the germ cells to result in transmissible changes in the code? I retain a rather lurking suspicion that the malarial parasite has a role in inducing the sickle-cell change rather than only as a selective phenomenon. Chemical and physical agents raise the same questions. But this is only a brief inquiry into problems of genetic epidemiology where the possibility of induced mutation under natural conditions rather than evolutionary selection of the sponta- neous mutant is considered. It emphasizes, however, some of the additional considerations which arise when one attempts to study all the factors which may govern the occurrence and distribution of a genetic disorder in a population. BIBLIOGRAPHY 1. Francis, T., Jr., 1959, The Epidemiological Approach to Human Ecology, Am. J. Med. Sci., 237: 677-684. 2. Mayr, E., 1961, Evaluation and Man’s Progress, Daedalus, Proc. Am. Acad. Arts and Sci., 90: 460-461. Genetic Studies of Family Units By Bertram L. HANNA, * Ph. D., Departments of Medicine and Neurology, Indiana University, Indianapolis, Ind. The common interest of the participants in this symposium is the etiology of chronic disease. The factors responsible for the produc- tion of an observed state in the individual may be broadly classified as (1) those which are intrinsic, transmitted to the zygote from which he developed through a marvelous series of steps which, though thoroughly studied during the past century are not yet fully com- prehended, and (2) the myriad of extrinsic determiners, known and unknown, which have acted from conception to the time of examina- tion to modify the organization of the individual. We wish today to devote our discussion to the intrinsic determiners of disease, realizing full well that it is impossible in many cases to ignore all of those environmental variables and stresses continually operating upon the human organism. GENERAL CONSIDERATIONS The epidemiologic approach to the study of etiology consists essentially of comparing the distributions of disease in groups charac- terized by similarities of age, sex, race, location, occupation, or ethnic background in an attempt to characterize diseased and nondiseased groups and to discern those factors which may be correlated with the occurrence of disease in a population. The occurrence of family *Dr. Hanna died in February 1964 of complications of diabetes. This is one of his final scientific contri- butions. His death is keenly felt by all of his colleagues, but perhaps particularly so by the participants in this symposium, who will recall vividly the quiet courage with which he carried on at this meeting despite an attack of acute glaucoma. 7 aggregations in the population provides information, generally over- looked in epidemiological studies, which will permit the examination of inherited intrinsic determiners which are of significance in the etiology of disease. The binomial nature of the genetic system in sexually reproducing organisms permits the construction of simple mathematical models for the genotypic expectation among the off- spring of matings or in a population. The method of genetic analysis thus consists of a comparison of sample statistics with parameters specified by genetic theory. In the strictest sense the genetic study of family units includes only studies of segregation and linkage. In practice, however, the geneticist frequently utilizes the armamentar- ium of the epidemiologist to obtain ancillary information both from families and from the population which enables him, by assuming specific genetic models, to examine various problems not directly amenable to the more precise methodologies. Several authors have listed criteria suggestive of genetic factors in traits of unknown etiology (David and Snyder, 1952; Neel and Schull, 1954; Steinberg, 1959; Morton, 1962). Among these are: (1) A greater frequency of occurrence in relatives than in the general population. (2) Greater concordance in identical than in fraternal twins. (3) Increased frequency of defects at the same site or of the same functional system in relatives as compared with the general population. (4) An excess of consanguineous parentage for a presumed reces- sive trait. (5) Onset at a characteristic age without a known precipitating event. It is obvious that no one of these relationships provides, either singly or in any combination, the critical evidence required to estab- lish a genetic etiology for disease. Despite every effort to rule out environmental determination there will always remain in the mind of the cautious investigator some doubt of the validity of any con- clusion based upon such criteria. These “epidemiologic impressions” should, however, suggest more detailed examination of family patterns. Evidence for a genetic etiology may be obtained by demonstrating that (1) the observed distribution of offspring in families is compatible with the expectation under a specific genetic model, or (2) the dis- tribution of phenotypes in the population is compatible with the distribution expected under the same genetic model. A good fit of the observed data with a genetic model does not conclusively demon- strate a genetic etiology. The genetic approach is, however, quite powerful since the nonrandom genetic variation is incorporated in the structure of the model, with environmental variation being considered random among family members. In unsophisticated epidemiologic studies, genetic variation is assumed to be random; 8 the power of such studies could be greatly increased by incorporating into their design knowledge of family relationships derived from genetic theory (Fisher, 1918; Kempthorne, 1957). SEGREGATION MODELS AND MODES OF INHERITANCE The segregation models employed for the analysis of traits pro- duced by genes not located on the sex chromosome are summarized in table 1, in which two alternative types of genes (alleles) at a single locus are designated A and a. Matings in the population fall into six different classes as listed under the column headed “parental genotypes” in table 1. The expectation for offspring differs for each of these mating types. If it is possible to identify the genotypes (AA, Aa, and aa), of parents and of all individuals within sibships, then there is little difficulty in testing the hypothesis that traits A and a result from two alleles at the same locus which segregate ac- cording to genetic expectation within families. An investigator simply examines his families, assigns them to the proper mating type groups and then does a simple statistical test of goodness of fit to determine whether the frequencies observed among the offspring are compatible with expectation under the model. TABLE 1.—Models for the genetic analysis of autosomal traits with family data Offspring expectation Parental genotypes AA Aa aa AA X AA 1 — — AA X Aa 1/2 1/2 — Aa X Aa 1/4 1/2 1/4 AA X aa —_ 1 — Aa X aa i 1/2 2 aa X aa . — In practice, the use of segregation models is rarely so simple and straightforward. The complicating factors may be classified broadly as (1) dominance, (2) ascertainment, (3) interactions, and (4) multiple etiologies. While each demonstrable allele at a segre- gating locus must have some effect upon the structure or function of the organism, the activity differential within an allelic series need not be straightforward. Suppose, for example, that an allele A results in the production of two units of demonstrable cellular pro- tein, its allele @ producing one. If the dominance relationship at this locus is additive, then assays ranging from two (genotype aa) through four (genotype AA) are expected in the population. A deviation from additive dominance might occur, however, such that genotype Aa has a value of four, suggesting that the allele a acts differently in the presence of A than when paired with another a. 9 735-712 0—65——2 In addition to true “gene dominance,” we may speak of trait domi- nance. This will appear for traits for which the proper tools for heterozygote recognition are not available to the investigator. For our discussion today we will speak of dominant and recessive traits rather than dominant and recessive genes since the trait is the ob- served event from which the investigation of etiology proceeds. The problems of ascertainment will be discussed in detail by Dr. Crow later in this program. We shall only note here that family sampling is complicated by dominance relationships, by the small size of human families, which permit some parents potentially capa- ble of producing affected offspring to be omitted from a sample, and by differing probabilities of including in a sample families having more than one affected offspring. Family samples collected under various types of ascertainment must thus be treated statistically in different manners to obtain comparable reliable results. The investigator is generally unable to measure directly the action of a single gene locus. He is required to work with end-products, produced through multiple developmental and physiologic inter- actions during the lifetimes of the individuals. Environmental factors or other genes segregating in a family may produce variation in the expression of a genetic trait or may prevent the appearance of symp- toms in a genetically susceptible individual (incomplete penetrance). Variable expression presents no problem for genetic analysis if the different manifestations are recognizable, but incomplete penetrance, since it leads to a modification of the observed ratio among the offspring, is a serious problem. Genetic analysis is impossible if penetrance is low. If a disease having delayed onset is rare, the penetrance may be calculated from the distribution of age of onset in family members, without any genetic assumptions (Morton, 1957). If incomplete penetrance is entirely due to delayed onset, the average penetrance lies between ff(2)@(z)dz and Sf fi(2)G(z)dz where f(z) is the frequency of age z at death or last examination among normal and affected siblings, f1(z) is the same frequency with the index cases excluded, and G(2) is the cumulative frequency of onset at age z among affected cases. Heterogeneity in age of onset among families will cause such a penetrance estimate to be approximate only (Morton, 1957). For a dominant trait which occasionally skips a generation, pene- trance may be estimated from the number of unaffected carriers in the direct line of descent from an affected ancestor to the first affected descendant in that line, if the trait is sufficiently rare that the chance of a “normal” carrier mating into the family is negligible. If the penetrance is low, however, long skips may fail to be recorded so that the penetrance will be overestimated. A proportion of the cases observed in a body of data may have arisen through nonrandom nonrecurrent events such as mutation, 10 chance expression of a heterozygote, or purely environmental effects such as prenatal anoxia, maternal-fetal incompatibility, toxic drug therapy, or trauma. With respect to family analysis these are all “true phenocopies” and should, where possible, be excluded from the analysis. Families examined for a test of genetic etiology may be divided broadly into two groups. Some sibships will include two or more affected children and have been called multiplex families by Morton (1959). Other families will contain only a single affected child (simplex families) which may occur as a result of a number of different precipitating events. A certain proportion of them will have the same genetic etiology as do those cases in the high risk families, but others may occur as a result of environmental factors, illegitimacy, errors in diagnosis, or from genetic causes other than those found in the high risk families. Morton distinguishes these two groups as chance isolated and sporadic families. Methods are available for estimating the proportion of sporadic cases in the data, making use of the distribution of affected children in multiplex and simplex families (Morton, 1959). These methods will be discussed by Dr. Crow in the following paper. For many diseases of clinical interest the proper diagnostic tools are not yet available to distinguish between two independently inherited diseases leading to the same observable abnormality. If one of these is autosomal and the other sex linked, genetic separation is not difficult, but two diseases inherited in the same manner will defy genetic analysis unless both are rare, in which case the chance of both occurring in one sibship is remote. AUTOSOMAL DOMINANT INHERITANCE The inability of the investigator to recognize a heterozygote requires modification of the model in order to test genetic hypotheses. If we now recognize A in table 1 as a dominant gene, it will be seen that for an autosomal dominant trait, affected children will occur only when at least one parent is affected. Thus for an autosomal dominant trait, transmission is directly from one generation to the next with the two sexes equally affected and with each affected child having an affected parent except for the rare case in which mutation occurs. Because of the limitation on human family size, not all families with an affected parent will be expected to have an affected child. If the trait is sufficiently rare in the population it may be further assumed that every affected individual is heterozygous. If this is the case the majority of families observed will be of parental genotypes Dd X dd and will have a segregation frequency of one half. This assumption appears reasonable except for dominant traits 11 which are quite widespread in the population. It may be shown, for example, that if the incidence of an autosomal dominant condition in the population is 1/5,000 individuals, the probability of including in a sample of one hundred families any marriages other than those between heterozygous affected individuals and homozygous normals is less than one in twenty. For conditions having a frequency of less than 1/50,000 the risk for an irregular sample of any reasonable size is negligible. The model for the test of genetic segregation for a rare dominant trait is a true genetic model, since the offspring expectations are derived directly from genetic theory. A simple statistical test of significance is sufficient to demonstrate whether or not one may accept the null hypothesis that the trait in question is inherited in this manner. If a dominant trait is common in the population so that the frequency of affected-by-affected matings is elevated, it is necessary to obtain an estimate of the frequency of the condition in the population of marriageable age in order to make tests of genetic hypotheses. In the absence of evidence of assortative mating the Hardy-Weinberg equilibrium model may be employed to estimate allelic frequencies for the locus in question. From these the expecta- tion in the sample of families may be obtained as shown in table 2. Because of limited family size the distribution of offspring in a family will not necessarily identify the parental genotype, so that in a sample of families only two mating types may be distinguished; affected X affected and affected X normal. This method lacks the power of the one previously discussed since an estimate of a population pa- rameter, which itself has sampling variance, is taken as the basis for model construction. An analysis of this type should be supported by a population analysis to demonstrate that the observed frequencies are distributed as expected under the proposed genetic model. The early acceptance of the genetics of the ABO blood groups was based in large measure on parallel family and population analyses of this type (Bernstein, 1925; Strandskov, 1931). AUTOSOMAL RECESSIVE INHERITANCE A condition which is recognized as an autosomal recessive trait will occur among children from three types of matings in the population (table 1). If the recessive genes are allelic, all children of affected parents will be affected, regardless of the sex of the children. If both parents are normal heterozygotes, the proportion of affected offspring is 1/4; if one parent is affected, it is 1/2. If the trait is rare, most of the parents and relatives of an affected sibship will be normal. For rare recessive diseases the rate of parental consanguinity is elevated. A carrier of a rare gene is unlikely to marry a carrier of 12 TABLE 2.— Models for the genetic analysis of common autosomal dominant traits with family data, where q*=the frequency of unaffected individuals in the population, and p+q=1 Parental matings Offspring expectation Frequency of matings Phenotypes Genotypes AA Aa aa Dominant X Dominant. .___________ AA X AA pt PB AA X Aa 4piq pig 2piq Aa X"Aa 4p’q? pq 2p’q? pq? p? p*q(1+p) pq? Dominant X Recessive..___.________ AA X aa 2p2q? 2plq? Aa X aa 4pq? 2pq’ 2pq? 2pq? 2pq? Recessive X Recessive....___._______ aa X aa qt qt EXPECTATION IN FAMILIES Affected Normal (a) One parent affected. _______________________________________ 1/(14-q) q/(1+q) (b) Both parents affected_.._.___________________________________ (142q)/(14q)? q¥/(14q)? the same gene if he mates at random; the likelihood is greatly increased if he marries a relative. It can be shown that the proportion of cases of a recessive disease in the population which result from con- sanguineous marriage is approximately «/(¢+a), where ¢ is the frequency of the gene in the population and « is the population in- breeding coefficient. Since lim «a/(g+a)=1 as ¢ approaches zero, the proportion of cases derived from consanguineous marriage provides evidence for the relative incidence of the condition in the population. It will be seen from table 1 that children with a recessive disorder can result from only three types of parental matings. These matings may be separated for analysis by examination of the parental geno- types. The mating of an affected by a homozygous normal parent will not be included in a body of data ascertained through the children, but will constitute the larger proportion of affected by normal matings obtained in data ascertained through the parents, particularly if the condition is rare in the population. For a rare recessive trait, the majority of matings producing abnormal children will be between phenotypically normal heterozygotes. If ascertainment is through the children, families with only one child must be excluded since the fact that every such family has an affected child will lead to a serious overestimate of the segregation ratio. If, on the other hand, ascer- tainment is through the parents, such families need not be excluded. Methods for the family analysis of recessive traits are derived entirely from genetic theory and are therefore quite powerful. These methodologies have been reviewed by Neel and Schull (1954), Li (1961), Steinberg (1959), and Morton (1959, 1962) and will not be discussed further here. \ 13 SEX-LINKED INHERITANCE Models employed for the analysis of traits produced by genes located on the X chromosome are summarized in table 3; segregating alleles may be designated X* and X° to indicate dominant and recessive alleles located on the X chromosome. It will be seen from this table that a male offspring always receives his Y chromosome from his father and his X chromosome from his mother. Each female offspring carries the X chromosome of her father and has a 1/2 chance of carrying either X chromosome of her mother. In a population, the distribution of X chromosomes in males is thus a random sample of the X chromosomes of their mothers, presumably randomly selected from among the females of the previous genera- tion. If the trait is recessive and sex linked, most cases will occur in the male offspring of normal parents, and no relatives will be affected except males on the maternal side of the family. Affected males will not transmit the trait to either sex but all of their daughters will be carriers. The frequency of a sex-linked recessive trait in females will be approximately the square of the frequency in males. Excluding XO females (Turner's syndrome), affected females will arise only from matings of affected fathers and carrier mothers and, if the trait is rare, there will be elevated consanguinity in families with affected females. If a condition is inherited as a sex-linked dominant trait, every affected offspring will have an affected parent, barring mutation and misconception. If the condition is rare, the frequency in females will be approximately twice the frequency i in males in the population. If a mother is affected, the expectation is that 1/2 of her children of both sexes will be affected, but affected fathers will transmit the condition to all of their daughters and none of their sons. TABLE 3.— Models for the genetic analysis of sex-linked traits with family data Parental genotypes Offspring expectation Mother | Father Female Male XAXA | XA(Y) XAXA XA(Y) XAXs | XA(Y) 1/2XAX A [ZA Ke 1/2XA(Y) 1/2Xs(Y) XAXA | Xa(Y) XAX XA(Y XAXs Xs (Y) 1/2XAXs Jj2Xexs 1/2XA(Y) ix XaXs XA(Y) XaX Xa(Y) XaXs Xs (Y) XoX» Xa(Y) Other types of sex-linked inheritance have been suggested to explain the patterns observed for various conditions. In the early literature, several pedigrees appeared which suggested the occurrence of a condition resulting from a gene located on the Y chromosome (Y linkage). Such a gene would be transmitted directly from a father to all of his sons and could never occur in a female. Stern (1957) 14 reviewed this early work and concluded that there is no evidence to support the claim for ¥ linkage for most of the conditions examined. Only one condition, the occurrence of hairy pinnae has been reexamined in recent years, but critical evidence in support of ¥ linkage has not yet been forthcoming. Haldane (1936) suggested that there might be occasional crossing over between homologous portions of the X and Y chromosome in man, producing partial sex-linked inheritance patterns. Critical reevaluation of earlier studies by Morton (1957a) casts serious doubt upon the validity of this hypothesis. SEX LIMITED INHERITANCE Some autosomal traits are limited in their expression to one sex. These are generally associated in some manner with sexual differentia- tion or reproductive physiology. If the trait is autosomal recessive, all the criteria of this mode of inheritance apply except that only one sex is at risk. If the trait is autosomal dominant and is limited to males, one half of the sons of unaffected carrier females will be affected so that a pedigree may, on cursory examination, suggest sex-linked inheritance. One half of the sons of an affected father would, however, be affected for an autosomal dominant trait, whereas all children would be normal were the trait sex-linked recessive. In addition, an affected XO daughter could occur through nondisjunction in a carrier for a sex-linked recessive but could not occur if the trait is sex limited. LINKAGE MODELS The association of inherited traits in members of a family affords an opportunity to investigate genetic linkage, which is the occurrence of genes responsible for the traits at loci near each other on the same chromosome. It is not specific alleles which are linked, however, but genic loci. The association of a disease condition with a specific blood type in a family provides no evidence for linkage since other blood types resulting from alleles will also segregate with the disease if the responsible loci are on the same chromosome. A test of the null hypothesis of genetic independence may, however, be made with families in which at least one parent segregates both traits. The expectation of offspring from the cross of double heterozygote by homozygote parents for linked loci having a crossover frequency ¢, is given in table 4. From data of this type an estimate of ¢ may be obtained directly. The test of linkage in humans is complicated by the fact that it is only rarely known whether a parent is of genotype AB/ab (coupling 15 TABLE 4.— Models for the analysis of linkage with backcross matings of (a) coupling and (b) repulsion phase, where c is the frequency of recombination between two loci on the same chromosome Offspring expectation Parental genotypes AB/ab Ab/ab | aB/ab ab/ab (a) AB/ab X ab/ab.___.._._____ (1-c¢)/2 c/2 c/2 (1—-c)/2 (b) Ab/aB Xab/aB_.._......... c/2 (1—c)/2 (1-c)/2 c/2 phase) or Ab/aB (repulsion phase). In addition, human families are rarely large enough to permit the total genetic potentiality of the parents to be expressed. Bernstein (1931) devised the earliest method of scoring families independent of phase to provide a test of the null hypothesis. The historical development of linkage analysis in humans has been reviewed by Morton (1955) who developed a sequential probability ratio test requiring a much smaller sample size for a given risk of error than do earlier methods (Morton 1955, 1956, 1957a). Smith (1959) has applied Bayes’ theorem to obtain, with Morton’s tables of likelihood scores, an estimate of probability for the occurrence of linkage. Only two autosomal linkages have been substantiated to date. Renwick and Lawler (1955) demonstrated linkage between the loci for nail-patella syndrome and the ABO blood groups, and Morton (1956) showed linkage between the loci for elliptocytosis and the Rh blood groups. Elliptocytosis was shown to be very closely linked to Rh in four large pedigrees but to be independent in three others, suggesting that at least two different genic loci may be responsible for this condition. INCIDENCE OF GENETIC DISEASE The incidence of disease has always been of major interest in the study of epidemiology. The use of family models for the study of diseases of known genetic etiology may greatly increase the precision of such estimates. Studies of the frequency of cystic fibrosis, a chronic and usually fatal disease of children have produced prevalence estimates of from 1/10,000 (Selander, 1962) to 1/100 individuals (Goodman and Reed, 1952). One excellent epidemiologic study suggests an incidence of approximately 1/3,700 live births (Steinberg and Brown, 1960). These authors summarize succinctly the deficiencies of their method, noting the absence of confirmed diagnoses, the incompleteness of sampling, the problems of migration, and the inability to determine exactly the initial population represented by their affected group. Our studies of cystic fibrosis further point up the problems faced by the epidemiologist. In a carefully controlled study of cystic fibrosis 16 sibships carried on during the past 2 years, approximately 44 percent of 170 probands referred to us by local physicians, hospitals, and National Cystic Fibrosis Research Foundation chapters have been found, on careful reexamination with sweat tests confirmed by inde- pendent laboratories, to have something other than cystic fibrosis despite clinical findings which on the surface might suggest this diagnosis (Merritt et al., 1963). It is apparent that this diagnosis has become popular in the past few years, although it may not have been so at the time of earlier studies. Cystic fibrosis appears to segregate in families as an autosomal recessive trait (Anderson and Hodges, 1946; Lowe et al., 1949). The absence of a significant increase in parental consanguinity suggests that the gene may be relatively common in the population. There is general agreement among the workers in this country that despite improvements in clinical management in recent years the likelihood is quite small that an affected individual has survived to reproduce. The evidence from our data suggests that the selection against affected individuals prior to the time of diagnosis is about 1.6 times as great as that of the normal individual. The mutation rate of 1X1073 estimated by Goodman and Reed (1952) is inordinately high and we believe, together with these authors and with Steinberg and Brown (1960), that the concept of heterozygote advantage is perhaps more realistic at the present time. If mutation at this locus is of the magnitude reliably reported for other loci in man, selection of less than 2 percent against the normal homozygote is sufficient to maintain cystic fibrosis in the population. The inherited nature of cystic fibrosis is not widely known to the public and no evidence of assortative mating is seen in the present generation. Each parent of a cystic proband must be a carrier of the responsible gene and, barring mutation, must have received this from a carrier grandparent. Each of the siblings of this parent will therefore have a precise probability of being a carrier which is specified by genetic theory. Examination of the frequency of cystic fibrosis among the offspring of these siblings therefore gives an estimate of the frequency of cystic fibrosis carriers in the population and of the frequency of the mutant allele producing this disease. Two different methods are available for this analysis. These may be called (1) the population or single selection method, and (2) the segregation or complete selection method. These differ in the assump- tions required for model construction as given in table 5. Let P be the proportion of homozygous normal and @ the proportion of heterozy- gous carrier individuals in the population of marriageable age. For the population model, P and Q are constant in successive generations. The probability that a carrier individual mates with a normal homo- zygote is P; with an Aa carrier, . The probability of being an Aa ” carrier is then =p for each of the siblings of a known carrier parent. The risk that a sibling will marry a carrier in the population is @, so that the probability of a first cousin of the proband being affected is 2/(4—Q) X@xi=5e Sor Since Q/2=q and (2P+Q)/2=1—¢q where q is the frequency for the mutant allele, the risk for a first cousin is a From this we estimate ¢g=4R/(T+2R) where R is the number affected among 7 cousins in a random sample of cousin sibships. TABLE 5.— Assumptions regarding the population and the families required for the estimation of allelic frequency from first cousin data by alternative methods Assumption about the— Population method Segregation method A. Population___________ 1. Random mating with respect to trait___| 1. Rondon mating with respect to rait. 2. Genetic equilibrium . Negligible mutation____________________ 1. Negligible mutation. . Uniform parental fertility . _____________ 2. Uniform parental fertility. . Random survival of parents to repro- ductive age. COD This model assumes genetic equilibrium in the population, but no such assumption is made for families; the information required to permit correction for mutation or differential selection within families is not available at present. In addition, the required random sample of offspring sibships from the parental generation is difficult and ex- pensive to obtain. The reliability of estimates obtained by this method may thus be questioned, even if extremely large samples are examined. For the segregation model, a sibling of a known carrier parent is taken as proband for the sibship consisting of his children, which is thus ascertained independent of the occurrence of cystic fibrosis. If it is assumed that all grandparental matings are Aa X AA, an estimate of the maximum possible allelic frequency in the population is ob- tained. If it is assumed that all grandparental matings are Aa X Aa, a minimum estimate is obtained. These estimates and their standard errors are given in table 6. Tt is clear that the values derived from the segregation model provide unbiased estimates of the maximum and minimum allelic frequencies possible in the parental generation of marriageable age. The “correct’’ allelic frequency (¢*) lies between these two; as ¢* decreases in magnitude the difference ¢*—@max approaches zero. For frequencies in the range found for cystic fibrosis, the difference between ¢m,. and the frequency estimated from data meeting all conditions specified for the population model is approximately 5X 107%. 18 TABLE 6.—The estimation of maximum (q) and minimum (q) frequencies of the allele for cystic fibrosis from cousin data. R is the number of affected among T cousins examined Assumed grandparental matings Allelic Variance frequency AA X Aa (maximum estimate) ____. __ q = 4R/T V(Q@ = q(4—q)/T Aa X Aa (minimum estimate) _._____ q = 3R/T V(@ = qa@—-q)/T The major advantage of the segregation method is that if the oc- currence or nonoccurrence of affected cousins is not a criterion for the inclusion of sibships in the body of data, random sampling of the parental generation is not required. No assumptions regarding the population equilibrium or selection within families are required unless one attempts to interpolate a “corrected” frequency from the esti- mated maximum and minimum values. The incidence among conceptions in the offspring generation is then /=g¢* with 95 percent confidence limits of (¢+0¢)2. Estimates obtained from a sample of 982 cousins of 74 cystic fibrosis proband sibships are given in table 7. These data indicate an incidence of 15/100,000; the probability that the incidence of cystic fibrosis in the population represented by this sample is greater than 69/100,000 is less than 0.025. TABLE 7.—The maximum incidence of cystic fibrosis estimated from the frequency of occurrence among first cousins of affected probands Sibships Total A flected children children Proband.......ccesessas 74 233 95 1st cousin... - where ¢=1—p=3/4. Thus, 27/64 of all three-child families have none affected, and among the 37/64 that have one or more affected, 27/37 have an isolated case. This offers little opportunity to distinguish between familial and sporadic cases, and poor information for testing Mendelian ratios. In addition to the difficulty with sporadic cases, another trouble- some methodological problem arises when a Mendelian ratio is to be compared with the observed ratios in children, where the genotypes of the parents were determined by observation of the children. If sibships of children segregating for a recessive trait are examined, only those families will be included that have one or more affected children, since there is ordinarily no way of knowing in the other cases that the parents were heterozygous. Thus the observed ratio of affected to normal children will be biased. THE A PRIORI CORRECTION Table 1 gives a set of data on the family distribution of 124 cases of cystic fibrosis with normal parents, arranged according to sibship size. As can be seen, there is a total of 124 affected among 269 children and the affected proportion (46.1 percent) is far in excess of the 25 percent expected on simple recessive inheritance. Clearly, one reason for the excess is the bias previously referred to—the non- inclusion of families where the parents were heterozygous, but failed to have a homozygous recessive child. These would add to the number of normal children and thereby reduce the proportion affected. This suggests a way of correcting the data. If all families where both parents are heterozygous are included, the expected number of affected children in a sibship of size sis sp. In a family of size s, the probability of having no affected children and thus being omitted from the data is ¢°, where ¢g=3/4. Thus, in a family of size s, the corrected number of affected children is sp/(1—¢*). The expected numbers for each family size can be worked out in this way and their total compared with the observed number. This is called the a priori method and was introduced by Weinberg (1912). It is described in several textbooks (e.g., Stern, 1960; Neel and Schull, 1954; Li, 1961). This simple procedure contains a hidden assumption. The method is appropriate only if all families have an equal probability of being included in the sample. In most studies a family with several affected persons has a greater chance of being included than a family with only one affected. As first emphasized in a classical paper by R. A. 24 Fisher (1934) the manner in which the data are gathered influences the method of analysis. We therefore must look for more generally applicable methods. For general reviews, see Bailey (1961, Chapter 14) and Morton (1962). CLASSIFICATION OF ASCERTAINMENT METHODS I shall use a system of classification and vocabulary introduced by Morton (1959), which follows in some ways an earlier system of Bailey (1951). If the families are obtained by random sampling through the parents without regard to the children (usually not pos- sible for recessive conditions) this is complete selection. Incomplete selection is ascertainment through the affected children, which means that families with none affected are omitted. With incomplete selection there are three possibilities: A. Truncate selection. Every affected individual is independently ascertained, as if the population were exhaustively examined. A family with a large number of affected children is no more likely to be included than one with a small number. This might happen if ascertainment were by family units. The distribution of affected children among selected families is approximately binomial except for omission of the class with none affected, and thus is truncated. B. Single selection. The method of ascertainment is such that no family is discovered more than once, as would be the case when the probability of ascertainment approaches zero. Such a situation might arise, for example, if the sample were ob- tained by taking children from a certain grade in school. The probability of a family being included in the study is propor- tional to the number affected, whereas in truncate selection it is independent of the number affected. C. Multiple selection. Most actual examples are somewhere between these two extremes. If the data are gathered in such a way that the individual ascertainments are identified, or if the ascertainment probability is known, there are appropriate methods of analysis. Multiple selection includes truncate and single selection as special cases. DEFINITIONS AND SYMBOLS Any individual in a sibship who is independently ascertained will be called a proband. (The word propositus is sometimes used, but since it also has other meanings I shall use proband.) 735-712 0—65——3 25 Quantities to be estimated: p=the probability of an affected individual, e.g., p=1/4 for a completely penetrant recessive trait with heterozygous parents. g=1—p=the probability of not being affected. w=probability of an affected person being a proband, i.e.; being independently ascertained. Estimators will be indicated by a circumflex or “hat,” e.g. p. Observed data: s=the sibship size. r=the number of affected persons in the sibship. a=the number of probands in the sibship. N=total number of families. T=total number of children. R=total number affected. WEINBERG PROBAND METHOD The Proband Method was also invented by Weinberg (1912). It has been discussed by Fisher (1934) who provided the error formula. Because of its ease of computation and applicability to all data where the ascertainments have been recorded, it is well suited for an initial analysis. In many cases this will be sufficient, but if a more refined treatment is wanted the Maximum Likelihood Method to be described later can be used. The argument for the Proband Method is a simple, intuitive one. Each ascertained affected individual (proband) is regarded as providing the information that his parents are capable of producing affected children; then the remaining members of the sibship provide an un- biased estimate of the ratio of affected to normal. If there is more than one proband in the sibship, the family will be counted repeatedly. A demonstration that the Weinberg Proband Method gives a consistent estimate of the true proportion of recessives is given in appendix 1. I shall consider multiple selection first, and then regard single selection and truncate selection as special cases. Multiple Selection In this procedure we regard a sibship of size s with r affected as a family of s—1 with r—1 affected. Then each family is counted as many times as there are probands. For example, family 4 in table 1, with 7 children and 3 affected, of which 2 were independently ascer- 26 tained, would be regarded as a sibship of 6 with 2 affected and would be counted twice. The formula for estimating the true proportion affected, p, is > atr—1) a(r—1 PS ab=D 4 where the summation is over all families. The data in table 1 have been treated in this manner. As can be seen from the table 59 TABLE 1.— Distribution by sibships of 124 individuals with cystic fibrosis (data from Dr. Charles C. Lobeck) Number Number Number Sibship number in sibship affected of a(s—1) a(r—1) a(a—1) probands 8 ¥ a — bt BD bt ed pk ROOD SRW BW CORO BD — Pt DO BD BD bt bt pd dt fk fk dt fk fd fd fd DD bt fd fd ed dt fd pd DD G0 bet bk pd ft RRR mm mt BOR) RONNIE mmm DROW WOE RNNDNG GN OW es RDN WW WWW WWW Wwww admin Bab bade ARO Grrr DI uuxeS CNNOOC COCOCOCO COOCOC ONNNS COOOO COOOCOO CONOCOO COOoON COCO oNnoceo HRN NDRNNN RRND RNAaRR WWWWW WWWwWw Wome d ade HRNNOO COCOO0C0O0 COOCOMM HNNNDG COOOC iii) NBROOM Iii DOO ORD bt bt bt ft Do 3 TaBLE 1.— Distribution by sibships of 124 individuals with cystic fibrosis (data from Dr. Charles C. Lobeck)—Continued Number Number | Number Sibship number in sibship affected of a(s—1) a(r-1) a(a—1) probands 8 r a tt bt = BD RORORORON RDNNNDN NRE RNRNNN Tk bk bk pk od fk kd bd od pk ok bd fk bd od kd pk dd bd BO BO DD COOOC COON rim ht bt bt bt bh ok kd pt bk et pt et BB mmm mm tt 2 COOOCO OOOO COCOOCO COCOOC OCOCCOO0 OOM Imm 8 COOOO COCOOOD OCOOOO OOOO OC OCOOoOoO oooeo 2 124 ~N — © This is the procedure originally suggested by Weinberg. However it assigns too much weight to large families and not enough to small families. In Maximum Likelihood Theory it is shown that optimum weighting is attained by assigning to each group a weight that is inversely proportional to its variance. In practice, this makes only a trivial change in the estimate; but it is only a slight increase in amount of computation and leads at the same time to an estimate of the vari- ance of the estimate. Fisher (1934) has shown that the variance of the estimate from a single sibship is _pql1+7+pr(s—3)] V- a(s—1) 2 where = is estimated by - _Soufa—1) Sar) 5 Table 2 shows the data from table 1 rearranged so that the numbers for each sibship are pooled. If each family is to be weighted in inverse proportion to its variance, this can be accomplished by multi- plying each family size estimate by C, where 1/C=1+4 w+ pym(s—3). 28 TABLE 2.—Data from table 1 grouped by sibship size 8 a(s—1) a(r—1) a(a—1) yc Ca(s—1) Ca(r—1) 9 2 0 2.08 4.33 .963 8 2 0 1.98 4.04 1.010 7 3 0 1.88 3.73 1. 507 30 7 2 1.78 16. 88 3.940 10 1 0 1.68 5.97 L507 48 15 8 1.58 30. 40 9. 530 42 12 2 1.47 28. 45 8.130 38 9 6 1.37 27.65 6. 550 26 8 4 127 20. 40 6.290 218 59 22 | 141.85 38.607 = _ 271 (first estimate) P=oR™ 2 f=pg=378 L=13+poi (2-3) C a 1DPox (8 A 38.607 Pils 272 (improved estimate) PAD) _ Vii = 001306 o%=+/.001396=. 037 . . A . . The improved estimate, p, is given by a > Ca(r—1) @) P=SSCa(s—1) where a, r, and s now stand for the pooled numbers in all sibships of the same size. The variance of is estimated by _ p(1—p) Vices Ca(s—1) (5) where v 1 CF rip (—3) The data of table 2 are analyzed in this manner. As can be seen, the improved estimate of p is hardly different from the first estimate. From these data, the best estimate of the true proportion of affected is .272 with a standard error of .037. The absence of any difference in the first and the improved estimates shows that there is no consistent association between proportion affected and sibship size, although this could of course be tested more directly. Single Selection (7—0) As the probability of ascertainment gets small it becomes less and less likely that a family will be ascertained more than once. In the 29 limit a=1 and C=1 for any ascertained family, and the proportion affected is estimated by eas 1) R—N 6) SSE) NN The variance becomes __ Pe _ Pq Vi=SsSG6—1 T—N @ This estimate is a maximum likelihood estimate (Fisher, 1934; Neel and Schull, 1954), hence is fully efficient. Thus the proband method increases in efficiency as = gets smaller. Notice that if this is applied to the data of table 1, Pao .233. That this underestimates the true proportion is expected since it is applied to data where some families were ascertained more than once. The variance of this estimate, which would be appropriate if the ascertainment probability were very small, is (.233)(.767)/189= 000945. The standard error is .031, so in this example the two estimates, one based on an ascertainment probability of 0 and other with #=.37, do not differ significantly from each other. Truncate Selection (r=1) In this case every affected individual is ascertained independently, soa=r. The estimate of p is prea 1) i 3), » or if the data are grouped 2.r(s—1) by family size as in table 2, a The variance formula is the same as that given for multiple selection—Equations (2) and (5)—when r=1 and a=r. Applied to the data of table 1, we obtain p,=112/363=.309. This is a substantial overestimate because the formula for truncate selection was applied to data where the ascertainment probability was in fact considerably less than 1. The three estimates, .233 when = is assumed to be zero, .272 when = is .37, and .309 when is assumed to be one, illustrate the dependence of p on the ascertainment probability. Clearly, the estimates differ; but perhaps the most important point is that the estimate of p, though somewhat dependent on =, is not very sensitive to errors in the esti- mation of =. In data where the ascertainments are not recorded and when there is no other basis for estimating =, about the only procedure available is to analyze the data on the two extreme assumptions of #=0 and 30 n=1, and take the estimate of p to be at some unknown point between the two extreme estimates. The ascertainment probability, =, though needed for the estimation of p, is of no genetic interest. However, the number of ascertain- ments divided by = gives an estimate of the prevalence of the condition (see Morton, 1962). THE HALDANE METHOD If #=1, that is with truncate selection, Fisher's Maximum Likeli- hood Method is available, although the computations are more difficult. The procedure was first used by Haldane (1932). The proportion affected, p, is estimated from the votal number affected, R, by », in the equation (8) where n, is the number of sibships of size s, and g=1—7. In this case, I is the total number affected in families of size two or more; families of one are not informative and are therefore omitted. The variance of p is given by 1 oan, [1—spg*'—¢ vial em 5 Equation (8) is most easily solved by trial and error, choosing trial values of p and estimating p by interpolation. The calculations are greatly simplified by the use of tables. Li (1961, p. 66) gives tables of sp/(1—¢*) and s(1—spg*'—¢°)/pq(1—¢*)* for various values of p and s. Using Li’s tables and the data in table 1, the expected total number is 101.0 when p=.25, 108.6 when p»p=.30 and 115.8 when p=.35. The observed value, R, is 115. Interpolating, p=.355. That this overestimates the value of p is expected since the assumption of truncate selection is inappropriate for these data. If = is other than 1, but known, the Maximum Likelihood Method can be extended to include this (Haldane, 1938). The computation is troublesome, but the procedure is discussed by Neel and Schull (1954, p. 225). 1 shall not dissuss it because it is very much like the methods of the next section. IMPROVED PROCEDURES For routine purposes the Weinberg Proband Method is easy to use and generally satisfactory. It is, I believe, the method of choice for a first analysis of the data. There is the further advantage that the 31 more elaborate Maximum Likelihood Method usually requires an iterative solution based on a trial value of p; for this purpose the value estimated by the Proband Method can be used. There are two principal criticisms of the methods discussed thus far. One is that, although these provide estimates of p based on certain assumptions, there are no tests for the validity of the assumptions. The second objection is that these provide no basis for separation of familial and sporadic cases. For more extensive tests Fisher's Maximum Likelihood Method is the standard procedure. This permits more elaborate hypotheses to be tested and allows for internal checks. The difficulty is that the equations can almost never be solved directly, so that complex iter- ative methods are needed. The calculations are greatly facilitated by the use of scoring procedures developed by Fisher. These have been adapted for segregation analysis by Finney (1949), Bailey (1951, 1961), and Morton (1959). The adaptation of maximum likelihood scores for separation of familial and sporadic cases and for tests of the assumptions are mainly due to Morton (1959). I shall illustrate some of the procedures by a further analysis of the cystic fibrosis data already discussed. The first problem is the separation of sporadic cases (mu tations, phenocopies, etc.) from familial cases. This can be accomplished by distinguishing between multiplex sibships (two or more affected) and simplex (only one affected). If both sporadic and familial cases are rare and uncorrelated in occurrence, and if the recurrence rate is very small for sporadic cases, then multiplex sibships will consist almost exclusively of familial cases. So the operational procedure is this: We assume that any sibship with two or more affected is familial, whereas simplex sibships are a mixture of sporadic cases and familial sibships which happen to have only one affected. We therefore analyze multiplex and simplex families separately and attribute any excess of affected in the latter to sporadic cases. Letting z be the proportion of sporadic cases among all cases, we can estimate z and p from the distribution of simplex »s. multiplex families. We can get an independent estimate of p from the distri- bution of the number affected in multiplex families. The parameters, such as p, z, and =, are estimated by the Method of Maximum Likelihood. In this method the probability of the observed data is expressed as a function of these parameters, and the values of the parameters that jointly maximize the probability of the observed results are taken as the best estimates. This is accom- plished by equating the derivative of the probability to zero and solving the resulting equations. One improvement is to use the logarithm of the probability of the observed results rather than the probability; this simplifies the algebra and doesn’t change the result, for the same values of the parameters that maximize the probability also maximize 32 the log of the probability. The log-probability expression also lends itself to a procedure for finding the solutions by iteration. We start with a trial value of the quantity to be estimated, say p. The Weinberg Sib Method provides a good starting value. Calling the trial value po, we obtain an improved estimate, p;, by the equation Us 10 P=mtg ( ) where Up, and Kop, are quantities that will be defined later. From this improved estimate a still better estimate, p., is given by + Un, (11) p= K,.,, and so on, until the desired accuracy is attained. Each family is assigned a score, u,, and contributes an amount of information, k,,, relative to the quantity p to be estimated. These quantities are additive, so the total score for all families is U,=Zu, and K,,=Zk,,. The score, ur, is the derivative of the logarithm of the probability with respect to p, evaluated at the trial value p,. Likewise, the amount of information is the expected value of the second derivative, also evaluated at the trial value. The general theory of maximum likelihood scores is discussed in statistics textbooks (e.g., Rao, 1952). For a general discussion of maximum likelihood scoring methods as applied to problems in segre- gation analysis see Bailey (1961, appendix 1). I shall illustrate some of Morton's simpler methods by continuing the analysis of the data in table 1. To avoid any contribution from sporadic cases, I shall consider first only multiplex sibships (»>1). Within such sibships the probability of » affected in a sibship of size s, and of the sibship being ascertained at least once is (rena 1—(1—pr)'—wspg’" The estimate of = can be obtained conveniently in either of two ways. The distribution of probands is clearly equivalent to truncate selection, since (1) there are no families represented unless there was at least one proband, and (2) there is no ascertainment bias in proband selection— by the mere fact of their being probands they are independently ascertained. Thus = can be estimated (as it was earlier in the Weinberg method) a ey 1 or by the Haldane Method (replacing R by the total num- ber of a s by the number affected, », and n, by the number of (12) 33 families with r affected in equation (8)). These two procedures give almost identical results, .37 and .36, as estimates of =. The score to be assigned to each sibship is 7 Up=ry0 = (13) where values of ¢, are given for various family sizes in table 3 for the trial value of p,=.25. Likewise, table 3 gives the information values, k,,, to be assigned to a sibship of size s. The values given in table 3 are based on m=.35. Each multiplex family in table 1 is given a « and k score. For example, in sibship number 1 where s=10 and r=3 u,=3(5.3333)—17.192=1.192 k,,=34.873. Adding up the values for the 27 informative sibships (those with at least three sibs and with two or more affected) gives U,=>u,=—0.350 K,,=>k,,=298.393. The improved estimate of p is p=.25—.350/298.39=.249. A better estimate could be obtained by repeating the procedure with a trial value of .249. This would necessitate a new table 3 based on this value. However, the first estimate was so close that no further iteration is necessary. The variance of the estimate is the reciprocal of K,,, hence V,,=1/298.39=.00335 and the standard error of p is 0p, =V Vy, =.058 Now, as an independent estimate of p and as a test for sporadic cases, we compare the simplex and multiplex families. Among the 71 informative sibships (those with at least 2), there are 38 simplex and 33 multiplex. Each family is assigned a score depending on only two things, the family size and whether it is simplex or multiplex. Such scores are given in table 4, based on the values p=.25, ==.35, and z=0. For example, family 1 is multiplex (r>>1) with 10 sibs; therefore its scores are: u,=1.299, u,=—1.459, k,,=13.72, k,,=17.30, and k,,= —15.41. Family 7, being simplex with 7 sibs, is scored as: u,=—6.989, u,—4.324, k,,—14.62, k;;=5.60, and k,,—=—9.05. 34 Proceeding in this manner through all 71 families and adding all the scores together, we obtain for the total scores: U,=16.026 K,,=642.90 U=—17848 K,,=12595 K,.=—260.49. When two or more parameters, such as p and z in this case, are being estimated simultaneously, it is conventional to write the K’s in the form of an Information Matrix: K,, K, 642.90 —260.49 Bn Ku —260.49 125.95 To obtain the variances and the improved estimates, the matrix is inverted. This is easily done for small matrices, such as the 2x2 matrix of this example. We designate by K?? the element in the inverted matrix corresponding to the position of the element K,, in the original matrix, and so on. Then the elements of the inverted matrix are Kr?=K,,/A, K**=K,,/A, and K"*=—K,,/A, where A=K,,K,.— (K,;)®>. Thus, the inverted matrix (called the Variance- Covariance matrix) is Kr Kr .00960 .01986 Kr K== .01986 .04901 The improved estimates for the two parameters are P1="Po+ Up KPoPo+ U.K?" (14) 21=To+ U, K*%+ U, K% and the variances are Vy, =K?%%, V, =K%%. In this example, p,=.25 and 2,=0. Hence pr=.25+ (16.026) (.00960) + (— 7.848) (.01986) = .248 V,,=.00960, 0, =.098 2,=0.00+ (—7.848)(.04901) + (16.026) (.01986) = —.066 V.,=.04901, 0, =.221. Again, the trial value of p was close enough so that a second cycle of iteration is not needed. In these data there is excellent agreement with the .25 proportion of recessive homozygotes expected on Men- delian theory. Furthermore, there is no evidence for sporadic cases, 35 since z does not deviate significantly from zero. On the other hand, the standard error of the estimate of z is so large that even a substantial number of sporadic cases cannot be ruled out. One of the most convenient properties of the uv and k scores is that they are additive for different sets of data. We can combine the results of the two independent analyses by summing the scores. Thus, the combined scores are: U,=—.350+16.026=15.676 K,,—298.393+642.90=941.293. These lead to the new information matrix 941.29 —260.49 —260.49 125.95 which inverts to the variance-covariance matrix .00248 .00514 .00514 .01857 |. The combined estimates from both analyses are H=25+ (15.676) (.00248) + (—7.848) (.00514) = 2485 V;=.00248, ¢;=.050 %—0.004 (—7.848)(.01857) + (15.676) (.00514) = — .065 V;=.0186, 0; =.136. Clearly, in these data there is excellent agreement with the 3:1 Mendelian expectation and no evidence for any appreciable contri- bution from sporadic cases. In fact, taken at face value, the number of isolated cases is slightly less than expected, as indicated by the negative sign of U,. However, the occurrence of a substantial fraction of sporadic cases cannot be ruled out because of the large variance of the estimate of z. Although sporadic cases do not appear to be an important part of the total incidence in cystic fibrosis, this is not true of many other conditions. In deafness there is a large component caused by recessive genes, but also a substantial sporadic fraction, as revealed by this kind of analysis. For the details see Chung, Robison and Morton (1959). It may appear puzzling that the Maximum Likelihood estimate of p has a larger variance (.00248) than the variance from the Proband Method (.001396). This is because I assumed in the Proband analysis that there were no isolated cases, whereas in the Maximum Likelihood this was not assumed. If we make this assumption, then the variance of p is simply the reciprocal of K,,, 1/941.29=.00106. The estimate of p is p=.25+ (15.676) (.00106) = .267. 36 We can compare these results with those obtained by the Proband Method: Proband: »=.272, V;=.00140, ¢;=.037 Maximum Likelihood: p=.267, V;=.00106, sj=.033. The greater efficiency of the Maximum Likelihood Method is shown by its smaller variance. The relative efficiency of the Proband Method may be expressed as the ratio of the variances, 106/140=76 percent. In other words 76 observations analyzed by Maximum Likelihood provides as much statistical precision as 100 observations analyzed by the Proband Method. In all the foregoing discussion I have not taken into consideration errors in estimating the ascertainment probability, =. We estimated x from the data, but since = enters directly into the computation of p, errors in the éstimation of = will affect the estimate of p. This can be taken into account in the Maximum Likelihood Method by includ- ing Ur, ker, kpx, and k.. with the other scores (Morton and Chung, 1959). The effect will be to increase Vj, so the procedures that I have used have both overestimated the precision of the estimate, 2. However, the effect of this is not great, for as pointed out earlier the estimate of p is relatively insensitive to small errors in the estima- tion of mr. This kind of analysis can be carried considerably further. Addi- tional tests for internal consistency can be included. The distribution of the number of times that an individual proband is ascertained provides a test for the assumption of the independence of separate ascertainments (Morton, 1959). The same kind of analysis can also be extended to other kinds of sibships to test dominant and sex- linked inheritance. For other examples and more details see Morton (1959), Morton and Chung (1959), and Morton (1962). Tables of u and k scores, of which tables 3 and 4 are examples, are available for various values of the parameters. Also the procedures have been adapted to computer analysis; the program is called SEGRAN (Morton, 1962). TABLE 3.—Scores for distribution of number affected (r) in sibships with more than one affected (r>1) x 1 r P= n=.35 oa Le 8 €p kpp 3 11.320 3.057 4 12. 022 6. 540 5 12.773 10. 431 6 13.571 14.704 7 14.414 19.324 8 15. 300 24.254 9 16. 227 29. 451 10 17.192 34.873 37 TABLE 4.—Scores for simplex (r=1) vs. multiplex (r’>>1) sibships 1 p=3 z=0 x=35 r=1 r>1 8 kep kez kpz Up Us Up Us —1.150 .288 4.183 —1.046 4.81 .30 -120 —2.306 . 685 3.680 —1.003 8.49 .75 —-2.52 —3.468 1.229 3.220 -1.141 ny 1.40 —3. 96 —4.636 1.970 2.803 -1.191 13.00 2.35 —b5.52 —5.810 2.972 2.428 —1.242 14.11 3.69 -7.22 —6. 989 4.324 2.092 —1.204 14.62 5.60 —9.05 —8.174 6.144 1.793 —1.348 14. 66 8.28 —11.02 —9.364 8.586 1. 530 —1.403 14.33 12.04 —13.14 30. un men mavens —10. 560 11.859 1.299 —1. 459 13.72 17.30 —15.41 ADDITIONAL INFORMATION ON THE FREQUENCY OF ISOLATED CASES ’ If the trait being analyzed is caused by one or more rare recessive genes there will be an appreciable contribution to familial cases from consanguineous marriages. However the sporadic cases are probably not increased appreciably by inbreeding. This provides another basis for separation of sporadic from familial cases. Following Morton (1962 and earlier papers) we let y =proportion of sporadic among isolated cases a =average inbreeding coefficient in the population F,=inbreeding coefficient of isolated cases Fy;=inbreeding coefficient of familial cases; then, since simplex families are a mixture of sporadic and chance isolated cases, Fi=yat(1—y)F, from which _F—F, = 15) and a=cy (16) where ¢ is the proportion of simplex families. In obtaining ¢ and F, the families are weighted by the number of probands. The data on cystic fibrosis contain so few instances of parental con- sanguinity that this method is of no use. It was used very effectively by Chung, Robison, and Morton (1959) in the analysis of deaf mutism. The proportion of sporadic cases, x, as estimated by equation (14) agreed very well with the completely independent estimate obtained from equations (15) and (16). 38 ESTIMATION OF THE NUMBER OF LOCI INVOLVED It is often difficult to distinguish between a situation where all cases of a disease are caused by the same gene and the alternative where any one of several genes can cause the condition. Sometimes this distinction can be made clinically; in other instances there may be different modes of inheritance. In rare cases linkage can be used. For example, Morton (1956) showed that linkage of elliptocytosis with the Rh locus is significantly heterogeneous; the data are most satis- factorily explained by two clinically indistinguishable loci, one closely linked to the Rh locus and the other independent of it. In practice this method is almost completely restricted to dominant or X-linked conditions. For recessive traits it is desirable to know whether all cases are caused by the same gene or whether any one of several genes, when homozygous, can produce the condition. One approach is through consanguinity data. If the trait is completely recessive its frequency in a population with inbreeding coefficient Fis (1—F)Z¢+ FZq:, where ¢; is the frequency of the individual recessive allele. The larger the number of genes, the smaller will be the frequency of each, and Z¢; will be corre- spondingly larger than Z¢?. Since, with fully penetrant recessives 2qi is the trait frequency in a randomly mating population, the increase with inbreeding relative to the normal incidence can be used to measure the number of genes. The regression of the incidence (after sporadic cases are removed) on the inbreeding coefficient ¥' may be written as I=1—e¢-4+8" x A BF a7 where A+ B=2q,=ngand A=2¢}>(ng)? from which (A+ B)?/A gives a minimum estimate of the number of genes (Morton, 1962 and earlier). This is a correct estimate only in the unlikely circumstance that all alleles are equally frequent, otherwise it is too small. This procedure, applied by Chung, Robison, and Morton (1959) to data on deaf mutism led to the estimate of at least 36 alleles. It should be mentioned that this method does not tell whether these alleles are at different loci. Formally, they could all be at the same locus provided all heterozygous combinations had normal hearing. The hypothesis of many separate loci is more consistent with evidence from experimental organisms. The second procedure follows a suggestion of A. G. Steinberg (1962). We inquire into the frequency of the trait in relatives (usually cousins) of index cases. If the gene is rare the frequency of the trait in cousins will be nearly proportional to the gene fre- quency in the population, the proportionality constant being the 39 probability that the cousin carries the same allele as the homozygous propositus. If the trait can be caused by any of several recessive genes their individual frequency is less and therefore the incidence in cousins of probands will be less. We can derive a quantitative expression for cousins by utilizing figure 1. Figure 1.—A pedigree showing two cousins. Large letters designate persons; small letters, gametes. Gametes z and w come from the unrelated spouses not shown in the pedigree. We desire the probability that Fis affected, given that EF is affected. If Eis affected either A or B must be a carrier, unless there has been a mutation, which possibility we ignore. The probability that y carries the same allele as z, having inherited it from the same grand- parent, is 1/4. There is also a chance that both grandparents, A and B, were carriers, but if the recessive gene frequency is small this probability is negligible. The chance that gamete z carries the recessive gene is simply p, the population gene frequency. Thus the probability of an affected cousin is pt approximately; (18) the population incidence, 7, is I=)2 (19) If there are k genes, of equal frequency, p then approximately _P P= 4 (20) I=Zp?=kPt (21) and k can be estimated by eliminating 7 from (20) and (21) I . k=16p? approximately. (22) 40 A more accurate formula is derived in Appendix 2, but is not needed for any data available at present. It is important to em- phasize that any variation in the frequency of p from locus to locus will decrease the value of k, so equation (22) will tend to under- estimate the number of loci if k is greater than 1. Dr. A. G. Steinberg (personal communication) reports some data from Ohio giving 7=.000267 and P=.00425. Taking these figures at face value gives k= (.000267)/16(.00425)>=.92. Thus there is no evidence for more than one locus. However, there is considerable uncertainty, especially in the incidence figure. Until better data are available, this should be regarded only as illustrative of the method. ACKNOWLEDGMENTS I should like to thank Dr. Charles C. Lobeck for supplying the unpublished cystic fibrosis data used throughout the article. I am indebted to Dr. George Hutchinson for pointing out a better weight- ing system in the Proband Method than I had previously used. Dr. Arthur Steinberg has been most helpful in discussions of the problem of estimating the number of genes. Mr. Takeo Maruyama prepared the computer program for the Maximum Likelihood scores in tables 3 and 4, which are based on some earlier tables of Dr. Newton Morton. SUMMARY When the segregation ratio for a recessive gene is estimated from sibships where the heterozygosity of the parents has been ascertained through their having produced one or more affected children, the observed proportion of homozygous recessive individuals generally overestimates the true proportion. Several methods for eliminating this bias are described. The Weinberg Proband Method is simple to use and leads to a consistent estimate of the segregation ratio, but it is not fully efficient (in the statistical sense). It is useful when the greatest of precision is not needed, and as a first estimate for iterative approaches to more precise estimates. The Maximum Likelihood Method is asymptotically efficient and has the advantage of permit- ting tests of some of the assumptions and separation of familial from sporadic cases (mutations, phenocopies, etc.) in sibships where there is only one affected person. The methods are described and illus- trated with data from sibships where at least one child has cystic fibrosis of the pancreas. The fraction of sporadic cases can also be estimated when the gene is rare by the difference in parental consanguinity rate between sibships with only one affected (simplex) and those with more than one (multiplex). 41 735-712 0—65—4 Two ways are described for determining whether all cases of a disease are caused by the same recessive gene, or any of several genes may cause the disease. One method depends on the relative incidence in children of related and unrelated parents. The other depends on the incidence in the cousins of index cases. APPENDIX 1 DEMONSTRATION THAT THE PROBAND METHOD GIVES A CONSISTENT ESTIMATE OF THE SEGREGATION RATIO Consider sibships of size s. Let z be the probability that a family with r affected will be ascertained at least once. Then z=1—(1—m)". Let a be the number of ascertainments in a family of size r. The estimated number of ascertainments among families ascertained at least once is nr nr I= a-m =z The value yielded by the sib method, ?, is A _r=1 P=" > a(s—1)z (3) pg r=1 = a(r—1)z (3) org where p is the true probability of an affected individual. Replacing a by its estimated value, d==r/z, "3 r(r—1) (+) pg r=1 w(s—1)>r (+) PQ" 2 s—2)! s(s—1)p? 2 = Ha) P ! (s—1)! — Br s(s—Dp 2, =m? A = agar _s(s—=Dp*(p+g*? s(s—1)p(p+g)! =p since p+¢=1. When #—0, z— =r, and a— 1, then Siwr(r—1) (+) pr Soar(s—1) (4) Pre which is the same expression as before. A = 42 APPENDIX 2 A MORE EXACT FORMULA FOR THE INCIDENCE OF A RECESSIVE DISEASE IN THE COUSINS OF PROBANDS Referring to Figure 1, if E.is affected then at least one of the grandparents, A and B, must be a carrier (unless there has been a mutation, which possibility we ignore). The a priori probability of both A and B being carriers is (2pq)?, where p is the frequency of the abnormal gene and ¢g=1—p is the frequency of the normal gene. The probability that one or the other is a carrier, but not both, is 2X 2pq X@*=4pg. Thus the ratio of the two probabilities is 4p2¢?/4pgd, or p:q. We then use Bayes’ principle as follows: Aor Bisa| Both A and B carriers carrier A priori probability ratio_._____________ q PD. Conditional probability that C and D | 1 (3/4)2. are both normal. Conditional probability that z carries | 1/4 1/3 (1/3 rather than 1/2 an abnormal gene. because C is normal). A posteriori probability ratio_._________ 7/4=X 3p/16=Y. Therefore the probability that gamete y carries the recessive gene is 1 x 1 y 1 1.» sX+YT3X+Y a(i—pd ~at16 The probability that gamete z carries the ‘abnormal gene is p, the population gene frequency. (This is p rather than p/(1+p) because the relevant gene frequency is the frequency at birth, not in adulthood.) Putting all this together, the probability that individual F is affected is 2 p=i+F, or approximately P= p/4 (This ignores the information provided by possible additional normal sibs of C and D, which would reduce the p?/16 component slightly.) If there are k genes at different loci, any one of which can cause the disease when homozygous, the probability of an affected cousin is P=3zrp, (1+2)+ 2(1—z)p? (1 =P z= =p! 2 I=73p} (3) where 7 is the incidence of the disease and z; is the proportion of cases caused by the #t® gene. The right term in equation (1) comes from cases in F due to a dif- ferent gene than the one causing the disease in E. If the k genes are equally frequent, neglecting p?/16, we have P=1p+(k—1)p! (4) I=kp? 5) 43 from which k can be estimated from P and I by eliminating p. This leads to I k=F—en (6) where ¢c— (k-1)/k. This can be most easily solved by using equation (22) to give a trial value for k. This value is used to calculate ¢, which is substituted into equation (6) to give an improved estimate of k. The process may be repeated if necessary for greater accuracy. The improved estimate for Steinberg’s data is 0.93, hardly different from the value 0.92 obtained by equation (22). If the genes are of unequal frequency this will generally be an underestimate of k. 10. 11. 12. 13. 14. 15. 16. 17. 18. REFERENCES . Bailey, N. T. J., 1951. A classification of methods of ascertainment and analysis in estimating the frequencies of recessives in man. Ann. Eugen. 16: 223-225. . Bailey, N. T. J., 1961. Mathematical Theory of Genetic Linkage. Oxford: The Clarendon Press. Chung, C. S., Robison, O. W., and Morton, N. E., 1959. A note on deaf mutism. Ann. Human Genet. 23: 357-366. Finney, D. J., 1949. The truncated binomial distribution. Ann. Eugen. 14: 319-328. Fisher, R. A., 1934. The effects of methods of ascertainment upon the estimation of frequencies. Ann. Eugen. 6: 13-25. Haldane, J. B. S., 1932. A method for investigating recessive characters in man. J. Genet. 28: 251-255. . Haldane, J. B. S., 1938. The estimation of the frequencies of recessive conditions in man. Ann. Eugen. 8: 255-262. . Li, C. C., 1961. Human Genetics: Principles and Methods. New York: McGraw-Hill Book Co. . Morton, N. E., 1956. The detection and estimation of linkage between the genes for elliptocytosis and the RH blood type. Amer. J. Hum. Genet. 8: 80-96. Morton, N. E., 1959. Genetic tests under incomplete ascertainment. Amer. J. Hum. Genet. 11: 1-16. Morton, N. E., 1962. Segregation and linkage. In Methodology in Human Genetics, Ed. by W. Burdette. San Francisco: Holden-Day, Inc., pp. 17-52. Morton, N. E., and Chung, C. 8. 1959. Formal genetics of muscular dystrophy. Amer. J. Human Genet. 11: 360-379. Neel, J. V., and Shull, W. J., 1954. Human Heredity. University of Chicago Press. Rao, C. R., 1952. Advanced Statistical Methods in Biometric Research. New York: John Wiley and Sons. Steinberg, A. G., 1962. Population genetics: Special cases. Methodology in Human Genetics, Ed. by W. Burdette. San Francisco: Holden-Day, Inc., pp. 76-91. Steinberg, A. G. Personal communication. Stern, C., 1960. Principles of Human Genetics. San Francisco: W. H. Freeman and Co. Weinberg, W., 1912. Methode und Fehlerquellen der Untersuchung auf Mendleschen Zahlen beim Menschen. Arch. Rass. u. Ges. Biol. 9: 165-174. Estimation of Genetic Parameters in Population Studies By WirLiam J. ScuuL, Ph. D., Department of Human Genetics, School of Medicine, University of Michigan, Ann Arbor, Mich. Interwoven into the fabric of many, if not all, epidemiologic studies are a variety of parameters of genetic interest. Among these are or may be: 1) the frequencies of the genes associated with the characteristics under study assuming, of course, that these characteristics are either genetic in origin or are significantly altered in their manifestation by one’s genetic constitution; 2) the relative fitnesses to be assigned to each of the genotypes; that is, the extent to which individuals of a given genotype, for all genotypes, are represented by progeny in subsequent generations as contrasted with some standard; and 3) the rates of spontaneous mutation. On occasion, one may be interested in 4) the frequency of nonpaternity, i.e., the frequency with which the mother of a child incorrectly identifies the child’s father; 5) the correlation between uniting gametes either as a consequence of inbreeding or of phenotypic assortative mating; and 6) the existence of maternal and/or paternal effects. Needless to say, other important parameters exist and it should be patent that, brief as this enumeration is, human genetics does not want for values to estimate. It should be equally obvious that a presentation such as this can hardly be exhaustive. Accordingly, we shall direct our attention toward only two problems, and even then the treatment cannot be complete. Specifically, we shall consider the estimation of gene frequencies in panmictic, that is, in randomly breeding populations and the estimation of the average coefficient of inbreeding. We have chosen the first of these because of the com- monness of its occurrence; the second has been chosen largely because 45 of the current interest in inbreeding as a means of evaluating genetic loads—the burdens imposed upon a population by virtue of the existence of individuals not of the optimal type (see Crow, 1963; Levene, 1963; Li, 1963; Sanghvi, 1963). Before we turn to these problems, however, certain general remarks on the estimation of genetic parameters seem in order. Some general remarks on the estimation of genetic parameters: Rarely, if ever, are populations of human beings amenable to genetically meaningful manipulation by an investigator. As a comsequence, in human genetics one is often called upon to deal with collections of observations which are largely unintelligible unless they can be reduced to at most a few numbers, or facts, which summarize the salient fea- tures of the data. It is frequently desirable, therefore, to be able to calculate from a set of data a quantity, termed an estimate, which may be taken as the value of a numerically unknown quantity, called a parameter, characteristic of the population from which these data are presumably drawn. To estimate these parameters we assume that the population is distributed in some form, and that one could com- pletely specify the distribution of individuals within this population if the value(s) of the parameter(s) upon which the distribution is functionally dependent was known. Available from which to estimate the value is a set of observations, a sample. Intuitively, we recognize that a given sample may be representative or nonrepresentative of the population. That is, we are aware that if one were to sample re- peatedly from the population the samples, and the estimates derived therefrom, would vary one from another. It follows that the only infallible way of ascertaining the value of the parameter involves a study of the entire population, but we presume that this is neither feasible nor possible. It is of importance, then, that the method of estimation which we employ provide estimates whose behavior is predictable. Alternatively stated, there are certain minimum attri- butes which the estimates should possess. Customarily, we seek estimates which have the properties we term consistency, unbiased- ness, and efficiency. When we assert that a statistic is consistent, we imply that as the sample size increases without limit, the estimate tends to the parameter in question. More formally, an estimate which converges in proba- bility to the estimated value, as the sample size tends to infinity, is called a consistent estimate (Cramer, 1946, p. 351). An estimate is said to be unbiased if, for fixed sample size, the average value derived from repeated samples converges in probability to the “true value.” An estimate can be consistent and yet biased. Thus, for example, as an estimate of the variance, the quantity (xi—X)* 25 46 where x, is the observation on the i*" individual (i=1, 2,...,n) in a sample of size n with mean X, is consistent but negatively biased; i.e., the mean of a series of such estimates will be smaller than the true variance by an amount which is a function of the sample size. When the bias of an estimate is known, it can, of course, be removed. We say that an estimate is efficient if, for a given and finite sample size, the estimate has minimum variance. In this sense, efficient estimates exist only under rather restrictive conditions, but asymptoti- cally efficient estimates, estimates which possess minimum variance for samples of infinitely increasing size, frequently occur. This con- cept of efficiency of an estimate is from the late Sir Ronald Fisher who demonstrated many years ago (1922) that a method of estimation, which he termed the method of maximum likelihood, always gives rise to estimates which are asymptotically efficient. Briefly, the method of maximum likelihood is as follows: Suppose that y is a random variable with a distribution dependent solely upon p. We may indicate this symbolically as f(y;p). If we now assume that a sample of size N is drawn from a population of y’s, then the likelihood, say L, of obtaining a given sample will be N L=1{(y; p) f(y2; P) (ys; P) - . - £ (In; p)=1 f(yi;p). This probability, which is functionally dependent solely upon p, is called the “likelihood function.” Now the principle of maximum likelihood states that the best estimate of p is that value which makes the likelihood a maximum. That is, the best estimate of p is afforded by the real positive root, if one exists, of the solution of Sp) ==0, 1) or, since I. must be positive or at least not negative, the estimate of p may be obtained from the solution of dlogL 5 =0. 2 The latter is generally the more convenient form to use. Fisher defines the expected amount of information in the sample with respect to p as 1m=—E (5; 3) or as : 1m=E (LEY) @ 47 In large samples we take as the variance of p the inverse of the infor- mation; hence »=(1(p)~". (5) As an alternative to the latter, one may calculate the variance from the following relationship: PY 1 de, 2! i255) & where e, is the expected number in the i** class. Unfortunately, and all too frequently, particularly when more than one parameter is estimated, the likelihood equation may defy explicit solution. When this occurs, one has recourse to iterative methods of solving the equation, the most common of which has been called maximum likelihood scoring. An especially lucid account of this procedure is to be found in N. T. J. Bailey’s Introduction to the Mathematical Theory of Genetic Linkage. In brief, one chooses a provisional value for p, say p,, which makes S(p) nearly zero. One then calculates I(p;) and obtains an improved estimate p= p1+[S(p1)/1(p1)]. The score, S(p), and the weight, I(p), are then revised on the assump- tion that p=p,. This is repeated until the desired accuracy is obtained. The notions here developed with respect to one parameter can be and have been extended to the case of more than one parameter, but let us now turn to the estimation of gene frequencies. Gene frequency estimation: If one considers a single diploid locus at which two alleles, A and B, may exist, any one of the following mutually exclusive but exhaustive array of genotype-phenotype cor- respondences could occur: 1) AA, AB, and BB are all phenotypically distinct; 2) AA is the same as AB, but distinct from BB; 3) BB is the same as AB, but distinct from AA; 4) AA is the same as BB, but distinct from AB; 5) AA, AB, and BB are indistinguishable. Traditionally, we describe the first of these cases as an instance of “no dominance”; 2 and 3 as instances of “dominance”, and for 4 and 5 no common genetic expressions exist. Cotterman (1953) has suggested, however, that case 5, which is both unknown and un- knowable, be termed the identity system. He further proposed that genotype-phenotype correspondences such as those set out above be called regular phenotype systems if all members of a given genotype have the same phenotype, and if this holds for all genotypes. Now some regular phenotype systems are permutationally equivalent. That 48 is to say, a change of gene symbols is all that distinguishes one from the other as in cases 2 and 3 above. To identify those systems which were permutationally distinct from the array of all possible phenotype systems, Cotterman coined the expression phenogram. It can be readily established, by enumeration, that with 2 alleles, there exist 4 phenograms; whereas with 3 alleles there are 52. If there are more than 3 alleles, enumeration is not feasible, but Bennett (1957) has shown by combinatorial analysis that with 4 alleles there are 5,525 phenograms and with 5, no less than 11,698,156. Each pheno- gram represents, in effect, a unique problem in gene frequency esti- mation. The formidableness of an attempt to consider exhaustively the estimation problems associated with 3 alleles, not to say more, is self-evident. However, we know that the gene frequency estimates will be indeterminant if the number of phenotypes, P, does not equal or exceed the number of genes, G, associated with the phenogram. We further know that if P=G, then only one consistent set of esti- mates will exist; whereas if P>G, more than one consistent set can exist not all of which will be maximally efficient. Thus, in the 2 allele case, at most 3 of the 4 phenograms can give rise to deter- minate gene frequencies, and, in fact, only 2 do (see table 1). In the case of 3 alleles, at most 42 of the 52 phenograms will yield unique estimates, and some of these must also be indeterminant. The precise number of the latter is not known nor can it be safely conjectured on intuitive grounds. We present in table 2 the 4 phenograms for which estimates exist, and to illustrate the method of maximum likelihood, we derive here a fifth. Consider a case of three alleles where the genotypes AA, AB, and BB represent one phenotypically distinct group; AC and CC another; TABLE 1.— Regular two allele phenotype systems and the appropriate gene frequency estimates. The genotypes which are presumed to be phenotypically indistinguish- able are connected by a bar Genotype-Phenotype Maximum likelihood correspondences Phenotypes Observed Expected estimates AB AA b np? AB c 2npq petite =2n " a 2n Pa AA BB BB E nq? AB AB A— a (1-g’n p=1—q or d BB = nq? — +i a AA BB AA BB n q z I i-g AB AB c 2npq AA. BB AA+BB n—c¢ (1-2pq)n p and q are indeterminate AB “identity” AA BB 49 TABLE 2.—Some regular three allele phenotype systems and the appropriate gene The genotypes which are presumed to be phenotypically indistinguishable are connected by a bar frequency estimates. Genotype-Phenot; Maximum likelihood pi Li vi Phenotypes Observed Expected el AB AA a np? _2a+b+f P=" AA BB AB b 2n PY _bt2e+d AC BC BB c nq? =" cc BC a 2ngr f=l-p—q V(p)=RU=P) CC e nr? 2n a(l-q) AC i 2npr Vig)= on AB AB D a 2npq p= (145 AA BB A b (p*+2pr)n o AO BO B c (q*+2qr)n a=a(1+3) 00 00 d nr? D D r=(r+3) (147) p pr ha vo 2 (4-sp +2) —1_ [ed 8n pa+r p'=1 5 (4 _gqp Vio (4 sat) q'=1- rd ap/, pa Cov (p,q) ~a(4 24) P= +2 n D=1-p'—q'—r PE AA+AB+AC (P*+2pq+2pr)n vi r=nf- 43 / us BB+BC (q*+2qr)n n AC BC cc c nr? p=1-4/ 2° .CC p(2—p) — fete _Je Vp) ==" q = 5 ~42—a(1-p)) Via) = _bpa2-p) Cov (p,q) n(—p) V(r)=V(p)+V(q) F2 Cov (p,@ AA b np? b p=/2 BB ¢ nq? a rest a 1—p?—q? = AB - (1-p’-q)n q : a AD r=1-p—q AA / \e2 VOI ="% AC BC Vig=1"2 N | pr. (@ in cc Cov (p,)=—%3 Vir) =V V OVE oa and BC still a third. 50 It should be noted that the gene A is “silent”; it invariably “simulates” B or C. Suppose, now, that in a sample of N, n, individuals were of the type (AA or AB or BB), n, were of the type (AC or CC), and n; were of the type BC. We assume that this sample was drawn from a population randomly mating with no mutation, migration, nor selective differentials. Thus, we assume that ny, ny, and n; are multinomially distributed with parameters N, p, and q. The likelihood equation is, therefore, ! “nnn! (P*+2pq+q*)m (r*+2pr)na(2qr)n,. Since p+q-+r=1, only 2 of the 3 gene frequencies are independent. It follows that if we were to replace r, say, by 1—p—q, differentiate the resulting expression with respect to p and q, set the partials so obtained equal to zero, and solve the 2 equations simultaneously for p and q we would obtain the maximum likelihood estimates. How- ever, in this case, since P=G, the maximum likelihood estimates may be obtained more simply merely by equating the expected relative frequencies of the 3 phenotypes to their observed relative frequencies. Suppose, then, that a=n;/N, b=n,/N, and c=n,/N; we have p*+2pq+qg’=a (7) r’4-2pr=>b (8) 2qr=c. (9) From (7) we note p’+2pq+¢’=(p+q)*=(1—r)* r’—2r+ (1—a)=0 whence and from the quadratic form we have t=1++a but only the root less than one can be a gene frequency. If this value for r is substituted in (9), we obtain AC =v and, finally, p=1—q—r. * Clearly, though A is “silent,” its frequency in the population can be estimated. The variances and covariances of the estimates given above are somewhat more tedious to obtain. However, they may be computed from the information matrix, I, as follows: by ij I= La Tag 51 where Inp=2N[2(1—q)*’+q(r+2p)]/r(r+2p) Ipq=2N[r+pr+2p]/r(r+2p) Io=2N[r’(2—r)+2pq(1+r)+q’r]/qr(4+2p). Now the variances are merely o2=I14/D 7pa="—Ipo/D ga=1Ip,/D where D=I,I,—1I. Though the algebraic expressions for the variances and covariances are quite cumbersome, numeric values can be readily obtained in the manner just described. Once one has generated a set of estimates, what then? Of immediate interest, of course, is the reliability of the estimates; this hinges upon the appropriateness of the genetic model, the randomness of the sample, etc. Let us examine briefly the more important of these. First, is the model appropriate? This can be tested directly from the data from which the estimates are obtained whenever the number of phenotypes exceeds the number of parameters to be estimated by at least one. Under these circumstances one merely compares the observed numbers of individuals in the various phenotypes with those expected, and the significance of the discrepancies which are en- countered can be evaluated by chi-square. If, as in the present case, the number of phenotypes does not satisfy the stricture just cited, the appropriateness of the model cannot be tested in the manner described. However, if data exist on two generations, it is generally possible to test the model indirectly through a test based upon the frequency of segregating families (see, for example, Neel and Schull, 1954, pp. 200-203). In either event, it must be borne in mind that the model involves two major assumptions; namely, that the genotype-phenotype correspondences are as postulated and that the population sampled is in Hardy-Weinberg equilibrium. Thus, if the model is deemed inappropriate because of an improbably large chi-square, one does not necessarily conclude that the genotype-phenotype correspondences are not as postulated. The poor fit of the model to the data may result because (a) the genotype-phenotype correspondences are not as postulated but the population satisfies the equilibrium conditions, or (b) the genotype-phenotype correspondences are as postulated but the population is not in equilibrium, or (c) neither assumption holds. 52 As a general rule, one can distinguish between these alternatives only through further observations. Second, is the sample random? Of principle concern is the occur- rence of correlated observations within the sample. The commonly used estimation equations assume a sample of uncorrelated observa- tions. In much of the published data on gene frequency estimates this restriction is violated. The violations invariably stem from the inclusion of biologically related persons whose genotypes and, there- fore, phenotypes are necessarily correlated. We know that the inclusion of relatives tends in general to have little effect upon the estimate itself, but may have a rather marked effect upon the variance of the estimate. If one ignores the relationships within the data, which is tantamount to treating everyone as unrelated, the variance so obtained underestimates the ‘‘true’’ variance. Thus, in a contrast of two populations, one could assert that a significant difference obtains when, in fact, no difference exists. When one is confronted with a sample which includes relatives, three courses of action are open. First, the problem may be ignored as has generally been the case. The possible consequences of this we have just described. Secondly, one may discard the relatives from the sample in such manner that only unrelated individuals remain. In view of the difficulty often attendant upon the collection of the data, this is an unattractive alternative. Finally, one can allow in the estimation procedure for the correlations which occur. The necessary adjust- ments are not simple. Cotterman (1947) developed a system of weighting appropriate to the two-allele case which gave rise to unbiased, consistent estimates. This method, though simple of use, was not fully efficient. Finney (1948, 1948a) subsequently set out in detail the appropriate maximum likelihood solutions. Most investigators have viewed the computations as too formidable and, as a consequence, relatively little use has been made of the method as judged by the literature. In the case of the ABO blood groups the method proved particularly cumbersome, a deterrrent to its use. Recently, Downs has devised a graphical method which holds promise. Briefly, the procedure is as follows: Suppose one has a sample of size N which is subdivisible into n; unrelated individuals, n; combinations of parent-offspring, n; combinations of siblings, etc. One begins by obtaining an inefficient estimate of p and q, the frequencies of the genes A and B, say. With these estimates one calculates the informa- tion matrix associated with n;, with ng, etc. Now, the contribution of a single observation (individual) to the matrix of nj, of a single observation (parent-child) to n,, etc., can be obtained by interpolation from graphs. Fortunately, the information matrices can be shown to be additive, and to be relatively insensitive to small changes in the values of p and q. Thus, though the estimates of p and q must be obtained iteratively, the information matrices, for practical purposes, 53 need be computed only once. While some sacrifice in precision follows from the use of graphs and the failure to iterate the variance and covariance estimates, the procedure is so much better than current practices that these small elements of approximation seem unimportant. Estimation of the coefficient of inbreeding: One of the more frequent departures from random mating, and one of great interest currently is inbreeding. To assert that a population is inbreeding is to imply that relatives marry more frequently than chance would dictate. Such marriages, that is, the marriage of relatives, we term consan- guineous, and the offspring derived therefrom we term inbred. Con- siderable effort has been directed toward delineating the effect of in- breeding on quantitative as well as qualitative variables. No attempt to appreciate these effects could be fruitful without mention of the coefficient of inbreeding. Sewall Wright, to whom the term as well as the concept is due, has defined this coefficient as the ‘correlation between uniting gametes,” the general formula for which he deduced to be F=3 APF) “where F, is the coefficient of inbreeding of any common ancestor that makes the connecting link between a line of ancestry tracing back from the sire and one tracing back from the dam. The numbers of generations from sire and dam to such a common ancestor are des- ignated n, and ng, respectively. The contribution of a particular tie between the pedigrees of sire and dam is (})%*at!(1+F,) and the total coefficient is simply the sum of all such contributions.” More recently, Malecot (1948) has defined the coefficient as the probability that an individual will possess at a given genetic locus two genes identical in origin. An extension of the concept outlined in the previous paragraph, from an individual to a population gives rise to the notion of the average coefficient of inbreeding. This average may be viewed in a number of different contexts. On the one hand, and as the name might suggest, it may be viewed as the probability that an “average” individual will possess two genes identical in origin at a given genetic locus. On the other hand, this average may be viewed as the rate of mixing of two populations, one randomly mating (F=0), and one completely inbred (F=1). It is the estimation of this average co- efficient of inbreeding which is of primary interest to us, and this average may be estimated either directly or indirectly. Let us consider the latter possibility first. Indirect estimation can proceed in either of two ways. If the pop- ulation is sufficiently well-known that its breeding formula can be specified, one could obtain an estimate of the coefficient of inbreeding by simulation, 7.e., with the aid of digital computers and Monte Carlo 54 methods. Alternatively, one might attempt to estimate this coeffi- cient much as one estimates gene frequencies. That is to say, one might assume a population randomly inbreeding at some fixed rate, F, wherein the array of genotypes were in a stable equilibrium described by the distribution 2a BFA d-20-F 20044, where q, is the frequency of the gene A, and the coefficients of the A’s describe the frequencies of the various genotypes. If an estimate of F derived from this distribution is to be meaningful, clearly the population must be randomly inbreeding at a fixed rate, and a stable equilibrium must obtain. It is doubtful whether in practice either of these conditions are met; however, the error which is introduced may, in many instances, be negligible. Li and Horvitz (1953) have shown that from the above distribution a variety of consistent estimates of F can be generated. Among the methods of estimation which they describe are methods based upon the (1) total proportion of heterozygotes, (2) product moment corre- lation between uniting gametes, (3) determinant of the gametic cor- relation matrix, (4) value of chi-square assuming panmixia in the population, (5) sum of proportions of alleles in homozygous condition among their respective total frequencies, and (6) method of maximum likelihood. When there are only two alleles, the six methods yield identical expressions for F. This is not so when the number of alleles exceeds two. Unfortunately, the sampling variances to be as- sociated with the first five methods of estimation are not known, and thus it is not clear which, if any, is the method of preference. In the general case, Li and Horvitz were unable to obtain explicit solutions for the gene frequencies and F by the method of maximum likelihood. It is possible, however, to obtain explicit expressions for certain special cases, one of which we wish to consider. We propose to examine a situation such as that which presumably obtains with respect to the ABO blood group system. The argument proceeds as follows: Consider a genetic model which assumes three genes, no mutation, migration, nor selection, and a population inbreeding at a constant rate. If we let p, q, and r be the frequencies of the three genes (say, 14, IB, and I°) and F the constant rate of inbreeding, then the ex- pected relative frequencies of the four phenotypes are merely, Phenotype Frequency Oo rF+(1—F)r? Brim ee em pF+(1—F)(p+2r)p Baccano tm wi wwe ssn qF+(1—F)(q+2r)q BB cc cvcmcarren sw memes RA AR RA 2pq(1—F). 55 Now it can be shown that the values of p, q, r, and F which solve the following equations will also solve the maximum likelihood equations: Fr+4r*(1—F) =0 Fp+ (p*+2pr)(1—F)=A Fq+(q’+2qr)(1—F)=B _ 2pq(1—F) =AB. where O, A, B, and AB are the observed relative frequencies of the blood types O, A, B, and AB, respectively. It is not too formidable a task to show from these equations that 2: VZ'—8 (RB+2B) (2A AB+1B) 4(AB+2B) _ AB(1—p) 1 eA—p+1E) r=1—p—q and _ _0—r? Tr—r? where Z=4A.AB+4A.B+4B.AB4-3AB". Clearly, there are 2 solutions to the equations, and while the values of P, q, and r will be real in both instances, the values of F may be real or imaginary, positive or negative. Several circumstances can give rise to imaginary values not the least of which is a system markedly out of equilibrium. Among the real values of F, only those which satisfy 1Z=F=0 are admissible. The variances and covariances of these estimates, which are of a somewhat more complicated form, are to be found in a note published elsewhere (Schull, Ito, and Soni, 1963). For the general case of multiple alleles, it has been argued (see Li and Horvitz, 1953) that since the method of maximum likelihood does not yield the usual gene frequency estimates it may be best to accept the conventional values and estimate F under this set of conditions. The importance of these differing points of view can hardly be overstated. For example, in a sample of 115 individuals of whom 49 are of type A, 35 of type B, 24 of type O, and 7 of type AB, one approach leads to an estimate of F of 0.7781, whereas the other leads to an estimate of approximately 0.00005. A direct estimate of F within this population gives rise to a value of about 0.006 (Schull, Yanase, and Nemoto, 1962). If one accepts this latter estimate as the ‘true’ value, then both methods would appear to be off by a factor of 120 or so, but in different directions. 56 An alternative to the procedures just outlined involves the calcula- tion of the coefficients of inbreeding to be associated with a random sample of individuals drawn from the population of interest. This approach has the added merit of making no assumptions about either fixed rates of inbreeding nor equilibria. There is, however, somewhat more work involved, and in instances of multiple common ancestors or multiple forms of relationship it is an easy matter to fail to enu- merate all of the different pathways as required by the conventional method. Accordingly, we give a method for calculating the coefficient of inbreeding recently published by Kudo (1962) in which errors seem less likely. We begin with several definitions. The pathway CAX in figure 1 Kudo terms a paternal line; whereas the pathway BDX he terms a maternal line. Collectively, CAX and BDX comprise an ancestral line pair. Moreover, since the maternal and paternal lines share the same individual X, they are said to form a loop in contrast, say, with the lines CAX and BDY which do not share an individual or individ- uals. The loop defined by CAX and BDX is said to be of length L where L=n,+nq-+1 and n, and nq are as previously defined. Suppose, for illustrative purposes, we were assigned the task of computing the coefficient of inbreeding associated with the case set out in figure 2. It is convenient to assign odd numbers to males and even numbers to females, and to better illustrate the method we have dotted the lines of descent of lesser importance. We begin by exhaustively enumerat- ing the paternal and maternal lines. We note (see figure 3) that this enumeration defines 16 ancestral line pairs beginning with the pair 1-3-9 and 2-5-11, and terminating with the pair 1-4-8 and 2-6-10. We further note that as set out in figure 3 each ancestral line pair is represented by a single, unique cell, in an array of 4X4 cells. The Roman numerals serve to identify the generations in the pedigree exclusive of the generation in which the individual in question occurs; the numbers are assigned proceeding backwards in time. It should be noted that the number of paternal or maternal lines will always be 2%-1 where R is the maximum generation number when assigned in the manner described. In the present instance, R=3, and the number of maternal and paternal lines is 22. We now search systematically for cells corresponding to an ances- tral line pair in which a specific individual occurs in both the maternal and the paternal line. For example, consider the first paternal line, namely, 1-3-9. We note that this line shares no common individuals with the first maternal line (2-5-11), nor the second (2-5-12), nor the fourth (2-6-10), but does share an individual, 9, with the third ma- ternal line (2-6-9). We hatch in or in some other manner mark the cell corresponding to the intersection of these lines, that is, lines 1-3-9 and 2-6-9. We continue this search, marking the appropriate cells, until all ancestral line pairs have been examined. In the present case, we 57 736-712 0—65——75 I | | I | ol 3 3 4 4 om 9 10 7 8 2511 2512 V4 2610 7 find only one additional cell to that described above; the additional cell corresponds to the paternal line, 1-3-10, and the maternal line, 2-6-10. The coefficient of inbreeding we can show is merely Fp humber of marked cells 2(total number of cells) 58 This formula is appropriate to any pedigree with any number of common ancestors provided none of the latter are themselves inbred. Appropriate adjustments can be made, however, to cover this situation (see Kudo, 1962), or the case of sex-linked genes. Many years ago, Wright and McPhee (1925) proposed an approxi- mate method of calculating coefficients of inbreeding particularly useful in the analysis of unusually complex pedigrees. This method, which consists essentially of a random sampling of ancestral line pairs, would seem to be practicable in those instances, such as the so-called “consanguinity villages” of Japan, where most members of the community are related, and often multiply so. This approach, as well as Kudo’s method, is readily adapted to digital computers. In conclusion: As we stated at the outset, no presentation of the present kind could hope to be exhaustive. Our goals were, therefore, quite limited, namely, to expose some of the methods of population genetics and some of the associated problems. We have attempted to achieve these objectives through a consideration of but two kinds of genetic parameters. Other geneticists would undoubtedly have selected other parameters. Of far greater importance than the choice of the parameter is the methodology, and this would vary little. Most of us subscribe to the notion that a dialogue between epidemi- ologists, on the one hand, and geneticists, on the other, must neces- sarily be fruitful for both our disciplines. A dialogue presupposes, however, that we share some form of communication. Clearly, this is not our specialized jargons which are apt to be mutually incom- prehensible, rather it is our common interest in the population study as a means of solving our problems. REFERENCES 1. Bailey, N. T. J., 1961. Mathematical Theory of Genetic Linkage. Oxford: The Clarendon Press. 2. Bennett, J. H., 1957. The enumeration of genotype-phenotype correspond- ences. Heredity 11: 403-409. 3. Cotterman, C. W., 1947. A weighting system for the estimation of gene frequencies from family records. Contributions of Laboratory of Vertebrate Zoology, University of Michigan, No. 33. pp. 1-21. 4. Cotterman, C. W., 1953. Regular two-allele and three-allele phenotype systems. Amer. J. Hum. Genet. 5: 193-235. 5. Cotterman, C. W., 1954. Estimation of gene frequencies in nonexperimental populations. In: Statistics and Mathematics in Biology (Eds.: Kempthorne, 0.; Bancroft, T. A.; Gowen, J. W.; and Lush, J. L.) Ames: Iowa State University Press. pp. 449-465. 6. Cramer, H., 1946. Mathematical Methods of Statistics. Princeton: Princeton University Press. 7. Crow, J. F., 1963. The concept of genetic load: a reply. Amer. J. Hum. Genet. 15: 310-315. 8. DeGroot, M. H., 1956. Efficiency of gene frequency estimates for the ABO system. Amer. J. Hum. Genet. 8: 39-43. 9. Downs, T. Personal communication. 59 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 60 Finney, D. J., 1948. The estimation of gene frequencies from family records. I. Factors without dominance. Heredity 2: 199-218. Finney, D. J., 1948a. The estimation of gene frequencies from family records. II. Factors with dominance. Heredity 2: 369-390. Fisher, R. A., 1922. On the mathematical foundations of theoretical statis- tics. Phil. Trans. (a) 222: 309-368. Kudo, A., 1962. A method for calculating the inbreeding coefficient. Amer. J. Hum. Genet. 14: 426-432. Levene, H., 1963. Inbred genetic loads and the determination of population structure. Proc. Nat. Acad. Sci. 50: 583-592. Li, C. C., 1963. The way the load ratio works. Amer. J. Hum. Genet. 15: 316-321. Li, C. C. and D. G. Horvitz, 1953. Some methods of estimating the in- breeding coefficient. Amer. J. Hum. Genet. 5: 107-117. Malecot, G., 1948. Les Mathématiques de I’ Hérédité. Paris: Masson et Cie. Neel, J. V.and W. J. Schull, 1954. Human Heredity. Chicago: University of Chicago Press. Sanghvi, L. D., 1963. The concept of genetic load: a critique. Amer. J. Hum. Genet. 15: 298-309. Schull, W. J., Ito, K. P., and A. Soni, 1963. A note on inbreeding and the island of Susak. Jugoslavenska Akademija Znanosti i Umjetnosti (in press). Schull, W. J., Yanase, T., and H. Nemoto, 1962. Kuroshima: The impact of religion on an island’s genetic heritage. Human Biology 34: 271-298. Wright, S. and H. C. McPhee, 1925. An approximate method of calculating coefficients of inbreeding and relationship from livestock pedigrees. J. Agri. Res. 31: 377-383. Hereditary Factors Elucidated by Twin Studies * By B. Harvawp, M.D., ano M. Have, M.D., The Institute for Human Genetics, University of Copenhagen, Copenhagen, Denmark. INTRODUCTION Ever since the days of Galton (1822-1911) twins have been a useful tool in the hands of human geneticists. The contributions of the German psychiatric school have been especially important in further- ing the methodology of twin studies. Thus Siemens (1924) elaborated a polysymptomatic similarity test for the diagnosis of zygosity and Luxenburger (1940) laid down rules for correct sampling techniques. The principle of the classical twin method is simple enough and illustrated in table 1. Monozygotic twins (MZ) carry identical genes while dizygotic (DZ) or fraternal twins are genetically as different as ordinary brothers and sisters. Phenotypical differences between MZ partners must be due to environmental differences (a), whereas DZ partners may also differ because of different genetic endowments (b). On the assumption that the intra-pair difference of environment is the same for MZ and DZ twins and therefore will cause a phenotypical difference (a) of the same magnitude in the two groups, it may be concluded that a trait which shows a substantial difference in the degree of resemblance encountered in pairs in the two groups must, to a certain extent, depend on genetic factors. The comparison may be made on the basis of all-or-none data as well as of measurement data. In the former case, twin pairs where both partners show the same trait are called concordant, whereas pairs where only one of the partners is affected are called discordant. A significantly higher concordance rate in MZ than in DZ pairs, for instance with regard to a special disease, indicates a genetic influence. A concordance rate approaching 1.00 in the MZ twins and at the same *This investigation was supported by research grant C-948 and G M 09418-02 from National Institutes of Health, Public Health Service, U.S. Department of Health, Education, and Welfare, Bethesda, Md. 61 THE PRINCIPLE OF THE CLASSICAL TWIN METHOD. a MZ Loz zzz J concordance discordance b a DZ vz zzzzz4 | concordance discordance difference due to environment Q n Oo 1] difference due to gene effect For all - or - none data: onc. — Conc. H = c MZ DZ 1 Re“ Conc. (1) For quantitative traits: H _ var.,, —vdaryz var. DZ (2) time significantly higher than in DZ pairs will be found in instances where genetic factors are of paramount etiological importance. Various formulas have been elaborated in order to reach a precise expression of the relative influence of genes and environment with regard to the variation of different traits in human populations. One of these is the formula of the H-value (Holzinger, 1929) given in table 1 which attempts to utilize concordance rates in MZ and DZ twin pairs to derive a quantitative estimate of heritability. This formula is subject to many criticisms. First, the value of H depends almost totally on the definition of concordance with regard to the trait in 62 question. For diabetes mellitus, concordance may mean the mani- festation of diabetes in the co-twin before a given age, the mani- festation at any age, and finally the finding of an abnormal glucose tolerance test without manifest diabetes. In epilepsy, concordance may mean chronic epilepsy in the co-twin, rare or a single epileptic seizure during the whole life, or seizures only after such special provo- cation as fever, infection, or head injuries. The concordance rates in these cases will differ substantially resulting in very different values of H. The estimate of heritability on the basis of measurement data as shown in formula (2) is somewhat more meaningful, but here too some reserve with regard to the interpretation is called for. First of all, the assumption that the intra-pair difference of environment is on an average the same in MZ and DZ pairs is dubious. For monochorionic MZ pairs the difference of the intra-uterine environment in some cases may be extreme because of a common placenta, in consequence of which one of the twins may “starve out” the co-twin by virtue of vascular anastomoses. This factor may cause an underestimate of H. During childhood parents are inclined to stress the similarity of MZ twins and therefore treat them more alike than is the case in DZ pairs, where the twins themselves are generally very eager to stress their dissimilarities. Neglect of such biases will attribute too large a share to heredity as a factor in the diversity of non-identical twins. A factor working in the same direction is the tendency of MZ twins to stick together, “hunt together” as the psychologists like to call it. Thus, because of their special twin situation, they subject themselves to the same environmental influences. Of quite another nature is the fact that MZ twins, because of their identical genetic complement, will actively choose environments which may show striking intra-pair similarity, such as similar occupations and similar partners. This is a fine example of the intimate inter- action of genes and environment, and it may be considered a question of taste, if the influence of this similarity of environment after all should not be taken as an expression of a gene effect. Reflections of the same sort may be applied to the relationship between DZ same-sexed twins (DZ) and DZ different-sexed twins (DZ). If the concordance rate in DZ, pairs exceeds that of DZ, pairs this implies a sexual dependance of the manifestation of the trait in question, either because of endogenous factors or because of environmental factors differing between males and females. Generally the DZ, pairs are excluded and the H-value calculated on the basis of the MZ and DZ, pairs only, but whether this procedure is correct is debatable. Special caution is called for when dealing with small samples or unreliable tests. The significance level of a given value of H is difficult to assess, but still the H-value has the merit of simplicity in spite of all limitations. 63 TYPES OF TWIN STUDIES The different ways in which twin studies may be approached are shown in table 2. TABLE 2.—Different types of twin studies Twin case reports... coo... Concordant MZ pairs. Discordant MZ pairs. TWINBerIes. cou conmunmmsmvivemumuion Comparison between MZ and DZ pairs Comparison between MZ twins brought up apart and MZ twins brought up together. “Twin control method’ ..._..___.__ Comparison between MZ twinpartners one of whom has been under specific environmental influence. Case reports, usually of concordant MZ pairs, crowd the literature of human genetics. Even if the value of such single cases is limited, as they can neither prove the presence of a gene effect nor allow any estimate of the relative importance of genetic factors, case reports both on common and rare traits have often been useful in indicating a field for further investigation, either as family studies or based on the collection of twin series. Detailed case reports of discordant MZ pairs are by far more con- clusive and may be utilized in the analysis of environmental differences. Especially in psychiatry, this method has proven useful in elucidating the etiologic importance of different exogenous factors with regard to the trait in question. Comparison of MZ and DZ pairs is the classical twin method dis- cussed above, being informative with regard to the nature/nurture relationship, as well as the question of “false heredity,” so often left open by family studies, which cannot decide whether an intrafamilial accumulation of cases is due to common genes or common environmen t. On the other hand, only on rare occasions can these data be used to test some specific genetic hypothesis. The method provides no answer to the problem whether dominant or recessive genes, single-gene or polygenic combinations, or one or many genotypes are responsible for the appearance of a trait. Further drawbacks of this method are the difficulty of getting a sufficient number of affected twins in the case of uncommon traits, and the difficulty of representative sampling. This means the twin approach will usually be limited to common traits. On the other hand, the method is suitable for the study of quantitative traits and traits with a complicated mode of inheritance, where family studies are often inconclusive. The sampling technique is difficult and there will be a tendency toward overrepresentation of concordant pairs. A detailed discussion of these questions has been given by G. Allen (1955). Comparison of MZ twins brought up apart and MZ twins brought up together suggests whether similarities in MZ twins brought up together may be more or less due to similarities of their environment, 64 in this way serving as a sort of control of the results of the ordinary twin method. Furthermore, twins brought up apart allow an estimate of the extent to which environment may change a given character. The drawbacks of the method are, on the one hand, the extreme difficulty of getting a sufficient amount of material because of the rarity of such twins, and, on the other hand, the fact that the foster homes in which these twins are found generally do not differ as much as we would wish them to. The method has thus far been restricted to the study of different anatomical, physiological, and psychological traits (Newman et al., 1937, Juel-Nielson, Mogensen, 1957, and Shields, 1962). The designation, “the twin control method,” has been used for the study of MZ twin pairs, where one of the partners has been subjected to some specific environmental influence, and where a comparison of the affected and the unaffected partners may be informative. A study of this sort has been carried out with twins who had sustained a head injury (Dencker, 1958) and who later on complained of sequelae of the injury. It could be demonstrated that the symptoms which the head-injured patients ascribed to their injury were found with ap- proximately the same frequency in the co-twins, who had not sustained any head injury. Thus it might be concluded that the symptoms had very little or nothing to do with the head injury, but could be ascribed to the premorbid psychic constitution of the patient. In pharma- cotherapeutic assays, where the drug is given only to one of the partners of MZ pairs, the twin control method makes it possible to draw conclusions from materials of very limited size. Whatever method is used, it is of primary importance to make a correct zygosity diagnosis. Generally the zygosity diagnosis is quite simple. Especially in monochorionic MZ twins, however, there will be a tendency to classify the more dissimilar pairs as DZ. Because the series nearly always are of a limited size, a wrong classification of even a single or a few pairs may highly influence the results. The extensive use of blood and serotyping has, in this connection, proved indispensable. In this way it is possible to obtain a group of twins who have a 100 percent probability of being dizygous, and another group with a high probability of monozygosity. THE TWIN STUDY OF THE INSTITUTE FOR HUMAN GENETICS As mentioned above the problem of representative sampling has been an obstacle in many earlier twin studies. This point is clearly demonstrated by an overrepresentation of MZ pairs compared with DZ pairs and of concordant pairs proportional to discordant pairs. In order to avoid both of these factors, in 1954 we planned an extensive 65 followup study of all twins born in Denmark in a definite period, i.e., the years 1870-1910. This program has now very nearly been ac- complished, and, in the following, some of our results and conclusions will be discussed with the purpose of illustrating the practical and theoretical difficulties met with in this sort of study. In the period 1870-191v, 37,914 twin births (including stillbirths) were recorded in Denmark. Because of the very high infant mortality of earlier years, it may be estimated that between 50 and 60 percent of the pairs born in the period mentioned were broken by the death of one or both partners before the age of 5 years. These early broken pairs have been excluded for several reasons, namely, 1) the difficulty of establishing a reliable zygosity diagnosis in the same-sexed pairs, 2) the very incomplete medical data which it has been possible to procure in these cases, and 3) the fact that the whole project has been focused upon the study of pathological traits appearing in the later years of life. Including the losses by emigration this leaves an estimate of less than 15,000 pairs available for study. Until now satisfactory contact has been established with 6,893 pairs, where full details of the medical history have been obtained. A further 2,000 pairs have been traced but the information is still insufficient. This, however, leaves 56,000 pairs which we have not been able to trace. This may imply an uncontrollable bias, which it has not been possible to avoid. The diagnosis of zygosity was based on a modified similarity test. A questionnaire was sent to all twins of identical sex, containing questions about the degree of similarity between the partners of the pair. Pairs were considered probably monozygous (MZ) when the information obtained either from the twins themselves or from close relatives familiar with both of them indicated that they were conspic- uously similar in appearance and that no difference in eye or hair color was found. If, further, reliable information was presented to the effect that a twin had repeatedly been confused with his partner, this has been taken as a very reliable indication of a high probability of monozygosity. When the answers to the questions about conspicuous similarity between the partners were clearly in the negative, the pair was placed in the group of probably dizygous twins (DZ;). Wherever vague, uncertain, or conflicting answers were obtained and the zygosity diagnosis was therefore doubtful, extensive blood group examinations were used where both partners of a pair were alive at the time of examination. Where this was not the case, the pair was classified as being of uncertain zygosity. In a small sample of the material, comprising 165 same-sexed pairs, we have checked the method used in diagnosing the zygosity by sub- jecting these pairs to a thorough examination of their blood and serum groups including the following systems: A;A,BO, MNS, Rh (with 5 66 antisera), P, Lewis, Duffy, Kell, Lutheran, Gm, Hp, and Ge. It may be calculated that complete identity between the twin partners as regards all these tests gives on the average an MZ probability of about 98 percent. Out of 78 pairs which had been classified as MZ ac- cording to the questionnaire, all except 1 pair presented no difference with regard to blood or serum group. Eighty pairs considered DZ according to the questionnaire showed one or more intrapair blood group differences in 76 cases, whereas the remaining 4 pairs had identical blood groups. Out of the total of 165 pairs, 7 had been classified as being of uncertain zygosity on the basis of the question- naires. Four of these had identical blood and serum groups. Thus it turns out that our diagnostic procedure is sufficiently reliable. Table 3 shows the composition of that part of the material which has been included in the present analysis. It appears that the material agrees with the Weinberg (1901) formulation, according to which the number of the MZ twins is the total number minus 2 times the number of different-sexed pairs. The cases of uncertain zygosity diagnosis are probably mostly MZ pairs, which makes an MZ/DZ ratio of very nearly 1/3, which is the rate of the average Danish population. TABLE 3.— Material included in the present analysis Same sex: Probably MZ. ...ccscrcesnemmummmn 1,528 Provably DZ; ..cccwispupeneinmununsn 2,609 Type of zygosity uncertain. ........ 231 Different sex (DZ3).....ciunminunimmnsns 2, 525 Total. ..occmcssnsmsueeimuamennons 6,893 Most of the pairs for which the diagnosis of zygosity is uncertain are pairs in which both partners have died before the time of exami- nation. This, too, implies a source of bias. The information concerning the medical history of the twins has been procured first by way of questionnaires sent to all living twins, or, in case of death of both partners, to some surviving near-relative of the twins. Information was requested about all admissions to hospitals, whatever the cause had been, and about symptoms of any of a number of specified diseases. If the questionnaires remained unanswered, the twins or their relatives were visited by one of the investigators. All stays in hospitals have been verified by a review of records. Supplementary information has, if necessary, been obtained from general practitioners. All death certificates and autopsy records of deceased twins have been checked irrespective of the suggested cause of death. 67 SPECIFIC DISEASES IN THE TWIN PANEL As might be calculated from the total number of twins, an adequate number of pairs could be obtained only with regard to common diseases. A diagnosis of cancer was made in a total of 1,038 indi- viduals. In a number of these cases, however, the cancer diagnosis had not been verified either histologically, by operation or by autopsy. Tables 4 and 5 show the concordance rate in different zygosity groups. All individuals in whom a diagnosis of cancer was established were considered index cases. Therefore in the calculation of the concordance rate both partners of the concordant pairs are probands, and the pair will be counted twice. k TABLE 4.— Malignant growth in twins (1038 pairs) Concordance rates with regard to cancer of same site MZ DZ; DZ, Verified cancer. SE mr ER TAS mtn mm me 8/160 8/335 4/321 Cancer diagnosis uncertain_______. ___.__.__.. 6/47 4/62 4/49 Total mmm 14/207 12/397 8/370 0.05> P>0.01 ZY 20sity Iagn0sIs UNCOLLAIN, c.ouvsmmsssmmmn sms imi RRs SRS ome mm sm mem mem —-- 4/64 TABLE 5.—Malignant growth in twins (1038 pairs) Concordance rates with regard to cancer of all sites MZ DZ DZs Verified cancer... o.oo eee 17/160 43/335 36/321 Cancer AIagNOoSIS UNCERTAIN. cous cos mans mmm amon mmm wae we 16/47 12/62 8/49 Ot). ov wun smusmmnusvans ssrA RRR 33/207 55/397 44/370 Zygosity diagnosis uncertain A 12/64 It is necessary to distinguish two different degrees of concordance: full concordance with regard to cancer of the same localization as the cancer of the index case, and concordance with regard to cancer of all localizations. If all pairs where either the cancer diagnosis or the zygosity diagnosis is not well established are left out of consideration, no significant differences are found between MZ and DZ pairs. Thus in spite of the considerable size of the material it cannot be proved on this basis that hereditary factors are at all active in cancer. The group with uncertain cancer diagnosis is characterized by a rather high number of concordant pairs, especially with regard to cancer of the same localization. This probably reflects the inclina- tion of physicians to suspect cancer, and especially cancer of the same localization, as a cause of death in a patient known by him to 68 have had a co-twin who had died from cancer. Thus the concordance rates found after inclusion of these doubtful cases are no doubt far too high, and the comparison between the concordance rates of MZ and DZ, pairs must be looked upon with scepticism. When dealing with a disease showing a relatively late age of onset and a prolonged period of manifestation, such as is the case in cancer, some sort of age correction is called for. This may be done in different ways. According to Weinberg (1927) discordant pairs for which the nonaffected co-twin has not yet entered the manifestation period are not counted at all. Pairs where the co-twin has not passed the whole period are counted half. This procedure is not, however, very practical in cancer, where a period of manifestation is difficult to define. The same objection may be raised to the more elaborate procedure of Stromgren (1935), where each pair is to be counted exactly according to the fraction of the manifestation period which has been passed by the nonaffected co-twin. Slater (1953) has suggested calculating the probability of the healthy co-twin having manifested the disease at the time of examination on the basis of the differences between the ages of manifestation in con- cordant twins. This method, too, is rather time consuming and requires a rather considerable number of concordant pairs for a meaningful calculation of probabilities. In our material we have tried a more simple age correction, by leaving out of the analysis all pairs where the healthy co-twin has died before the age of manifestation of malignant growth in the proband. Even after this, however, the concordance rate with regard to cancer of all localizations remains virtually unchanged in all zygosity groups, viz., 0.15 for MZ, 0.18 for DZ,, and 0.14 for DZ,. Our figures support the conclusion that genetic factors determining susceptibility to cancer in general do not play any detectable role in human cancer. The existence of genetic factors determining the development of malignant growth in specific organs cannot be totally ruled out, but on the basis of the rather low concordance rates it may be estimated that their importance, compared with unknown exogenous factors, is very small. Cancer exemplifies the application of the twin method in the case of a rather well-defined all-or-none character, where the diagnosis may be made with reasonable certainty. In many diseases, however, the calculation of concordance rates is not too meaningful. This, for instance, is the case in different forms of arteriosclerotic disease, where it would be more meaningful to try to get a quantitative measure of the degree of arteriosclerosis in the co-twins. An investigation of this sort will be difficult and has not been practicable in our material. Instead, we have chosen to use the occurrence or nonoccurrence of coronary occlusion as an indicator of the degree of arteriosclerosis. As is apparent from table 6 no differences as to the concordance rate 69 can be demonstrated between MZ and DZ, pairs. The difference at the 5 percent significance level between DZ, and DZ; on the other hand indicates the unequal sex-distribution in coronary occlusion, which could be due either to gene effects or different environmental influences in males and females. TABLE 6.—Cardio-vascular diseases in twins MZ ., Dz DZ, COrORATY OPCIUSION. . cc csvmmmmsinm mmm ERE EER a em mmm 20/102 24/155 10/133 Le 0 aN 0.05>P>0.01 Arterial LL —— 4 ER PE SER EERE AAR aE ———————————— 20/80 10/106 4/106 (202 pairs)... .cc.cicomrmrmmmmmmm—— —- 0.01>F>0. 001 Cerebral apoplexy. --| 22/98 16/148 22/179 (425 PAITS) ooo eee eee 0. 05>P>0.01 In arterial hypertension, on the contrary, a difference significant at the one percent level may be demonstrated between MZ and DZ, pairs. The absolute figures are of limited value, as they will only indicate concordance with regard to clinically established hypertensive disease. An estimate of the relative genetic influence in determining the blood pressure necessitates a determination of the intra-pair differences of blood pressure in MZ and DZ pairs in different age groups and at different blood pressure levels. A program of this sort is going on at present, but the material is still too small for any conclusions. Also in cerebral apoplexy a difference between MZ and DZ, pairs can be demonstrated. With regard to psychiatric disorders the material is not very suitable. An intensive psychiatric study is not possible as so many of the probands and their co-twins have died or their original psychiatric pattern has been changed in old age. Only a few comments, therefore, shall be added to the survey given in table 7. The total number of institutionalized oligophrenics is only slightly higher than the number expected based on the total number of institutionalized oligophrenics in Denmark, indicating’ that after the fifth year of life, mental de- ficiency is not more common in twins than in the general population. In epilepsy all types have been included, as the inhomogeneous clinical examinations of the probands do not allow a distinction between genuine and symptomatic epilepsy. The total number of epileptic pairs is about double as high as expected from the general population. This may either be a reflection of the higher number of birth injuries in twins, or indicate that the estimate for the general population is too low. A demonstration of genetic factors influencing resistance to in- Jectious diseases and antibody formation is given in table 8. Whereas no significant difference could be demonstrated between the concord- ance rates in MZ and DZ, pairs with regard to death from acute infections, a highly significant difference is found in tuberculosis. It 70 cannot be taken for granted, however, that this difference is at all due to genetic factors. In the case of infectious diseases the tendency of MZ twins to “hunt together” may influence the results, partly because they will more often be exposed to the same contacts and partly because of their higher possibility of infecting each other. TABLE 7.— Psychiatric disorders in twins MZ DZ, DZ; Mental QeCIoNeY. ...cccumsssneinsvesmmumve - 12/18 0/20 0/29 (B7 DOITS).... ome iss iol st i i wo mm i 0 0 1 ss i 0.001>P Epilepsy --cueeemmmme emma me eee a nm ———— 10/27 6/43 4/57 (IIT DAISY evn sn rmsnnmms mmm ns mmme nme mm ————————————— EE 0.05>P> 0.01 Manic-depressive PEYCNOSIS, . « «cucuivssmonn ve iumwmmmsmmmmmmmmnm——— 10/15 2/23 0/17 (85 DAIS) «ovivin wm mS SES Se SE Ew 0.001>P SOHIZODITOIIR. ov mmm mm mmm mp ws mem wR REE SH RR 4/9 4/33 2/29 (71. PBIB cc nmm mmm mmm mmm mm se wm mm wm mmo mi a Bn i i A 4/21 0/38 0/37 (00 DAITDY cinemas ow ww ese wi a os wm a em TaBLE 8.— Infections and antibody production in twins MZ DZ D2Zs Death from acute infection. meee 10/127 14/221 26/233 (B81 PAIIS) cnn nce mcmn rere canna mmm n enn mmm m——————— TP DIOTCIIOBES sisi mmmmn ws iia ow 50/135 42/267 36/246 COBB DIBITS).....cc mics iii i i oi so 0.001>P BRGUIHALIBTIOVER.. .. «cmmmmmmmmmmmm sine SS SRR A AAA ER CAE A mA 30/148 12/226 14/202 (G78. DBILE) co nome em mmm m mmm em ms mp ERS i 0.001>P Rheumatoid arthritis 16/47 2/71 8/70 Lr — 0.001>P BronCHIBL BRUATNS con wusumm nm ER Se SR ee i 30/64 24/101 22/91 C250 DIT) nm mem mmimm ne Sm SS IR ES An AR BES SAS RE 0.01>P> 0.001 In rheumatic fever and rheumatoid arthritis it may similarly be doubted, if the difference between MZ and DZ, twins indicates an hereditary tendency to these specific forms of antigen-antibody reaction. The higher concordance rate in MZ pairs might equally well be the result of the more intimate contact of MZ partners with common streptococcal infections and common environmental influ- ences. With the same reservations, the rather high concordance rate in MZ twins with bronchial asthma may be interpreted as a reflec- tion of the hereditary character of allergic reactions. Caution is called for in the evaluation of these different concordance rates, based as they are solely on the clinical manifestation of the trait in question. This may be clearly demonstrated in peptic ulcer, where a geographically delimited part of the material with both partners alive has been carefully scrutinized with X-ray examination of both the probands and their co-twins (Gotlieb Jensen, 1964). Table 9 shows that the concordance rates found in this way are con- siderably higher, but still the MZ-DZ, difference is unimpressive, 71 as an indication that the intrafamilial accumulation of ulcer cases must be primarily due to mutual environmental factors of ulcerogenic effect, and not to mutual genes. TABLE 9.— Peptic ulcer in twins MZ DZ, DZ ‘Total material, clinical 38a ONlY....cccocscunssssummnennennns duane 30/144 24/209 18/243 (GOGH IBIE) cnc corms rms emmm mena n mms mem mn ems Sr ma 0.208 0.114 0.074 0.05>P> 0.01 Intensive study: After exclusion of early broken pairs... oo. 8/27 14/62 2/52 0.206 0.226 0.038 0.01>P> 0.001 After X-ray examination _.___________________________________ 10/21 18/40 7/30 0.476 0.450 0.233 Table 10 shows the findings in diabetes mellitus and serves to illus- trate the importance of some sort of age correction in cases where a more exact estimate of the concordance rate is desired. In the total material a highly significant difference is found between the rate of 0.47 in MZ and of 0.10 in DZ,. After exclusion of all discordant pairs where the co-twin has not been observed for at least 70 years, the rates have changed to 0.73 and 0.34, respectively. Calculation of an H-value in the first case gives 0.43, in the second case, 0.59. Age correction (Slater, 1953) gives a concordance rate of 0.71 in MZ and 0.22 in DZ, with an H-value of 0.63. The application of Slater’s method, however, is complicated by considerable uncertainty in establishing the exact time of manifestation of the disease in the instances where only one twin is affected. It therefore was found necessary to make use of the time when the diagnosis was made instead of the time of manifestation. However, several extraneous factors may influence the time at which the diagnosis is made, espe- cially in elderly persons, where the diabetes is often diagnosed by chance in connection with a medical examination for quite other reasons. In the material of Then-Bergh (1939) where pairs, one of whom had clinical diabetes and the other of whom had a diabetic glucose- TABLE 10.— Diabetes mellitus in twins MZ DZ DZ Total material _ -— pp 36/76 12/122 10/116 EE Te — 0.47 0.10 0.09 0.001>P After exclusion of pairs where the healthy co-twin has not reached the age of: CRT ccc om 0.49 0.11 0.09 0) WORTS.. uc mm so om 0.55 0.13 0.09 TO VORIS. coer RRR ER SR HS TRE Se SR 0.73 0.34 0.29 ALS COrraotion 8.15: BIAtOr. . c .ovsssimmnsnninsssnmrnsanssrnsnssmnranren 36/51 12/55 10/55 0.71 0.22 0.18 72 tolerance-test were considered concordant, the concordance rate in MZ twins approached 1.00, also making the H-value approach 1.00. This illustrates the difficulty of deriving a meaningful H-value for clinical traits where the manifestation period is not well-defined and the expressivity of the gene is varying. In several clinical family studies a genetic association between different clinical traits has been suspected. In the families of patients with Graves’ disease an accumulation of cases of other types of thyroid disease has been demonstrated (Bartels, 1941). Similarly, a significantly increased incidence of cancer of the stomach has been found in the families of patients with pernicious anemia (Mosbech, 1953). In the present twin material, with the complete medical histories of more than 13,000 people, there is an opportunity of further elucidation of problems of this sort. Table 11 gives a survey of the occurrence of different diseases in the nondiabetic co-twins of our diabetic probands. A considerable number of these persons is expected to possess the diabetes gene or genes. In diseases occurring with another frequency in this group than in the general population, it must be considered if this may somehow be due to a pleiotropic effect of the diabetic genotype. In most of the examples chosen there is no significant difference between the number expected and the number observed (for example, with respect to gallstones, coronary occlusion, and tuberculosis: disorders very frequent in the diabetic patients themselves). This may indicate that these disorders are a consequence of the diabetic disease, and not a direct gene effect. On the other hand, violent death and death from surgical complications are significantly more common among the nondiabetic co-twins. We have no satis- factory explanation of this finding. Malignant growths are less frequent in the nondiabetic co-twins than expected, a probable sign of some sort of protection from cancer connected with the diabetes genotype. TABLE 11.—Occurrence of different diseases in 256 nondiabetic co-twins of diabetic probands Expected Observed RL EE 11 13 Gallstone ______ 10 10 Tuberculosis. ...... 12 13 Coronary occlusion. 9 11 Rheumatic fever______ 11 9 Arterial hypertension. 6 9 Cerebral apoplexy..... 8 12 Bronchial asthma. ___. —_— _— a 5 9 Malignant growth... - cco cocsneneminsrassmms esse SEES SRS RT RAR eS 23 13 0.05>P>0.01 Death from acute infection. . eee 11 12 WHOIOINE BOUIN in imi hie is os eR 6 0.01>P Death from compleated SUPBLY - ...oovesssssssmssmasmrsn sss mneN ame eae 5 16 0.001>P 73 735-712 O—65—6 The only measurement data which until now have been procured from the material with reasonable certainty are the ages at death. As shown in table 12 the intrapair difference in pairs, where both partners are dead at the time of examination, is significantly smaller in MZ than in DZ, and DZ, pairs. On the basis of the intrapair variances an H-value may be calculated, which is supposed to indicate the fraction of the variance in DZ, twins due to genetic factors. For the total material the value is 0.29. TABLE 12.— Longevity in twins MZ DZ DZ, Average intrapair difference of age at death (years)..._.._ 14. 540. 94 18. 60. 84 20. 42-0. 99 0.01>P>0. 001 Intrapail VAranee. coc sncsoassmanmnensemsnusmssemmmm 209 293 378 2 diff.2 ) 2Xnumber of pairs Var.pz, —var.mz ~ var.paz The lifespan of the individual may be considered the final result of the cooperation of all gene effects and all environmental influences during life. The H-value calculated on the basis of the lifespan variances in the present material has been found to be considerably higher in the urban than in the rural population. This may be due to a more uniform environment in the urban districts, and suggests that the progressive urbanization of society may gradually lead to a state where the variation of longevity throughout the population is predominantly determined by genetic factors. The absolute figures must be considered with reserve, first of all because the pairs broken before the fifth year of life have not been included in the calculation. Moreover, restriction of the calculation to pairs where both partners have died by the time of examination may mean a bias of unpredictable effect. Several other items are being intensively studied in our twin ma- terial. A criminologic research program has been started; the results, however, are not ready for publication as yet. The same is the case in our study of fertility in twins, which seems to show a significantly smaller difference in MZ than in DZ pairs with regard to the number of offspring. This paper has aimed at illustrating the numerous difficulties met within a clinical twin study, among which the difficulty of getting a material of sufficient size should be stressed. Even in a material of the present size only the very common disorders may be elucidated. The analysis is complicated by diagnostic uncertainty, which in combination with sampling biases may lead to fallacies of different kinds. In a way the negative results, as ours with regard to cancer, 74 are more conclusive than the positive ones, where the justification of a comparison between different zygosity groups in nearly all cases may be doubted. SUMMARY The basic principles of twin studies are discussed. So are the limitations of the classical twin method where concordances in MZ and DZ pairs brought up together are compared. Doubt whether environmental differences for identical twin partners are really of the same magnitude as for nonidentical twins is considered the most important objection to the validity of the conclusions drawn from this method. The problems met within clinical twin studies are illustrated from a Danish twin material with followup of nearly 7,000 pairs of twins born in the years 1870-1910. From this material it may be concluded that genetic factors play an unimportant role in the etiology of malignant growths, coronary occlusion, and peptic ulcer. On the other hand, a significantly higher concordance rate in MZ than in DZ pairs could be demonstrated in arterial hypertension, cerebral apoplexy, mental deficiency, epilepsy, manic-depressive psychosis, tuberculosis, rheumatic fever, rheumatoid arthritis, bron- chial asthma, and diabetes. Furthermore it can be demonstrated that the intrapair difference of age at death is significantly smaller in MZ than in DZ twins. REFERENCES 1. Allen, G., 1955. Comments on the analysis of twin samples. Acta Genet. Med. et Gemel. 4: 143-160. . Bartels, E. D., 1941. Heredity in Graves’ Disease. Munksgird, Copenhagen. . Dencker, S. J., 1958. A followup study of 128 closed head injuries in twins using co-twins as controls. Acta psychiat. et neurol. 33: Suppl. 123. 4. Gotlieb Jensen, K., 1964. Peptic ulcer in twins. To be published. 5. Holzinger, K. J., 1929. The relative effect of nature and nurture influences on twin differences. J. Educat. Psychol. 20: 241-248. 6. Juel-Nielsen, N. and Mogensen, A., 1957. Uniovular twins brought up apart. Acta Genet. Med. et Gemel. 7: 430-433. 7. Luxenburger, H., 1940. Zwillingforschung als Methode der Erbforschung des Menschen. In: Handbuch der Erbbiologie des Menschen by G. Just, Volume II, 213-248. Springer, Berlin. 8. Mosbech, J., 1953. Heredity in Pernicious Anaemia. Munksgdrd, Copen- hagen. 9. Newman, H. H., Freeman, F. H., and Holzinger, K. J. 1937. Twins: A Study of Heredity and Environment. University of Chicago Press, Chicago. 10. Shields, J., 1962. Monozygotic Twins, Brought Up Apart and Brought Up Together. Oxford University Press, London. 11. Siemens, H., 1924. Die Zwillingspathologie. Ihre Bedeuting, Ihre Methodik, Ihre Bisherige Ergebnisse. Springer, Berlin. 12. Slater, E., 1953. Psychotic and neurotic illnesses in twins. Medical Research Council Special Report Series. 278: 1-385. 5 13. Strémgren, BE. 1935. Zum Ersatz des Weinbergschen “abgekiinzten Ver- fahrens’””. Zugleich ein Beitrag zur Frage von der Erblichkeit des Frkrankungsalters bei der Schizophrenie. Ztschr. f. d. ges. Neurol. wu. Psychiat. 153: 784-797. 14. Then-Bergh, H., 1939. Zur Frage der psychischen und neurologischen Erscheinungen bei Diabeteskranken und deren Verwandten. Zischr. f. d. ges. Neurol. u. Psychiat. 165: 278-283. 15. Weinberg, W., 1908. Uber den Nachweis der Vererbung beim Menschen. Jahresh. Verein f. vaterl. Naturk. in Wiirttemberg. 64: 368-382. 16. Weinberg, W., 1927. Mathematische Grundlagen der Probandenmethode. Zeitschr. Abstgs. und Vererbgsl. 48: 179-228. 76 ABO Blood Groups, Secretor Status, and Susceptibility to Chronic Diseases: An Example of a Genetic Basis for Family Predispositions By J. A. Fraser Roserts, M.D., D. Sc, F.R.C.P., Clinical Genetics Research Unit, Medi- cal Research Council, Institute of Child Health, The Hospital for Sick Children, Great Ormond Street, London. The search for possible associations between blood groups and disease has a long history, in fact it goes back for rather more than 40 years. The earlier literature was, however, confused, negative and conflicting. There were several reasons for this, of which the most important was that the numbers studied were far too small. In retrospect it can be seen that there were two positive results. Buchanan and Higley in 1921 did produce results which foreshadowed the association between group O and peptic ulceration and between group A and pernicious anaemia. But their own conclusion was that no association existed. Years later Ugelli (1936) in Italy got the same result for ulcer. These isolated suggestive findings made little or no impact, however, for they were swamped in the mass of inconclusive results. It was not until 1953 that the first study to carry more or less general conviction was published (Aird et al., 1953). This study was based on very large numbers and showed with very high probability that cancer of the stomach is about 20 percent more frequent in persons of group A than in those belonging to group O. Soon afterwards an even stronger association was found between group O and peptic ulceration (Aird et al., 1954). This time the numbers were really large and the significance of the association so high that I think most people were convinced that an empirical or ( epidemiological finding had been confirmed with high probability. Since then a very considerable literature has been produced. Some associations seem established beyond reasonable doubt, others seem highly probable, while with others there are no more than suggestive indications, or indeed a plain conflict of evidence. And, of course, it is clear that the great majority of diseases show no blood group associations or at least nothing measurable with practicable numbers. I have said that the great majority of observers and critics seem to be convinced that some associations at least are proved empirically beyond reasonable doubt. One further point may be mentioned in passing. That such associations would be discovered had been predicted, notably by Fisher and Ford. The theoretical importance of the associations is, in fact, considerable, with all the bearings they have on the conflict between those who uphold the neutral, or nearly neutral gene, and those who uphold balanced polymorphism. A minority of critics, however, notably Dr. A. S. Weiner (1962), are not even convinced that the empirical evidence is sound; and though perhaps it is a small minority it is articulate, so I think I ought to present the evidence for just one disease. This is duodenal ulceration, for which the evidence is most overwhelming; it is also the disease ' which I propose to talk about almost exclusively in this paper. Once "we are convinced that at least one single association is reasonably proved, other associations for which the evidence is more problem- atical can be left for the clarification that time will bring. Table 1 shows an up-to-date compilation by Hanley (1963) of the results for duodenal ulceration in persons of group O compared to group A. The third column shows relative incidences. These are obtained by simple cross-multiplication of the figures in the disease and cor- responding control series. A relative incidence of 1.59, as shown in the first line, means that the incidence of ulceration in people of group O is 1.59 as compared with unity in people of group A. The total numbers are now very large, more than 16,000 people with duodenal ulcer. The series come from centres all over the world, from Great Britain, the United States, Denmark, Norway, Austria, France, 1taly, Greece, Japan, India, West Africa, Israel and Australia. Hanley has compared the relative incidence in people of group O as against those of group A. A comparison of group O versus the other three groups, A, B and AB, added together would have given even lower probabilities. It will be seen that in all 22 series the incidence in group O is higher than in group A. At no less than 16 centres individually the difference is highly significant. The total x? can be divided into two parts. The first, with one degree of freedom, measures the divergence of the estimated mean value of 1.38 from the 1 expected if there were no association. It is no less than 297, or 17 times its standard error. This corresponds 8 TABLE 1.—The analysis of the incidence of duodenal ulcer in persons of group O relative to its incidence in persons of group A (from Hanley) Number | Relative Centre in disease | incidence x? series 0:A London. oo ooee ee 946 1.59 38.82 Manchester a 423 1.27 4.89 Newcastle. 482 1.46 14.08 Liverpool 1489 1.33 11.056 Glasgow. . 1642 1.10 2.37 Co BEEN avin rr RENEE TR Cv eam wa me oS Gee em wm mi 680 1.42 17.36 mf Sra mrmemv amma ————————————————— 579 1.49 19. 62 Vienna 1) - - 1160 121 7.74 = A= sre " 449 1.42 10. 89 JOWR CWE OB. cco mimi mri si om i WR 2386 1.47 71.37 PENG. vo sens orsveomivanss ams susp omseses ie sae En RS aS 211 1.59 6. 89 LODAON. oor uvncnsesnspms 564 1.28 7.19 TDORY Ou canons Has AN SESS EE RRR SEE HSH SE Emam mmm —————————— 163 1.67 7.00 IDAAIN , » ri csmrnaninamm smn mma mame 390 1.30 3.86 Jerusalem... o.oo ioiees —— 158 1.46 3.81 IMA AIOBIIPOUGI. oc ccs ww ines sm cms mm min mw mn i 722 1.48 21. 49 Paris. __ 1195 1.40 25. 02 Bombay. ..oucsvumunssnensmns 380 1.16 1.10 ay SER 628 1.52 21.38 Copenh RRR RS A ARREARS AAR hmmm A ———————— 1223 1.33 19.74 Athens. — er 147 1.04 0.02 Sydney. —- 192 2.22 21. 50 Total. . 16,0090 Looceinvisn 337.26 Mean weighted relative incidence=1.38. Ninty-five percent fiducial limits=1.33 to 1.43. Difference from unity (degrees of freedom =1) x?=297.48. Heterogeneity between centres (degrees of freedom =21) x2=39.78. to an infinitesimal probability. It is far smaller than a billionth of a millimeter in the orbit of Neptune, or a billionth of a milligram in the estimated mass of the universe. The fiducial limits of the estimate are quite narrow; there is a 95 percent probability that the true value lies within the bracket 1.33 to 1.43. The remainder of the total x? represents the heterogeneity of the 22 centres; it is 40 for 21 degrees of freedom. There is significant heterogeneity, the probability being about 1 in 2,000. But I think the point to note is how small it is compared to the divergence of the total averaged incidence from unity. Anyway, it would be sur- prising if the relative incidence were to be the same at such varied centres in so many different parts of the world, where the incidence of the disease and the relevant environmental factors are so different. From the beginning of these observations in their modern phase it was pointed out that although the associations might be real enough they might be secondary, and so of little real importance. It might be that there are strains in the populations who are, say, high in group O and at the same time especially prone to duodenal ulceration, just as examination of people in an eastern European city might well show a secondary association between group B and dark hair, which would merely be a reflection of racial origin. The variety of the series, however, from so many different parts of the world, and their general concordance, makes this explanation very unlikely. Never- theless, a direct test is desirable. Such a test was proposed by Pro- 79 fessor Penrose. It can be seen in sibships which are segregating for ABO blood groups whether the index case with the ulcer is unduly often of group O. In other words, we can see whether people of group O are more likely to develop duodenal ulceration than their own non-O sibs. If they are, the hypothesis of stratification is disproved. The labour of carrying out this test is truly Herculean, but it has been done, thanks to the pertinacity of Clarke and his associates in England and Buckwalter and his associates in the United States. Their com- bined results are again quoted by Hanley (1963). His figures include recent data supplied by Clarke et al. and hitherto unpublished. TABLE 2.— Analysis of the change of the propositus being group O in the sibships segregating for blood group O. (Method of analysis that of C.A.B. Smith—see Clarke et al., 19566) (From Hanley, 1963) Group O propositi Centre Total DO prop sibships Variance Observed | Expected Liverpool (Clarke et al., 1963)... ocean 159 86 78.0027 37.3136 Towa (Buckwalterelal,, 1000). ...csxcoscsenoncnmneswns 96 56 46.9797 22. 0082 POLO) vm ns rrr rs vr TR CE EE 255 142 125. 0724 59. 3218 The difference divided by its standard error (y/variance) _ 16.9276 _ 59.3218 The combined results show that in segregating sibships the observed number of index cases who are of group O is 142, as compared with 125 if there were no association. The difference is significant at the 5 percent level, which is all that can be expected with these small numbers. The difference is, in fact, fully consistent with a primary association and significantly against the hypothesis of an association secondary to stratification. Personally I fail to see how anyone can feel doubtful about the reality of the association between blood group O and duodenal ulcera- tion. But, of course, the evidence is empirical, or epidemiological, in the broad sense. It is the same kind of evidence as that for an asso- ciation between cigarette smoking and lung cancer. No causal mechanism has yet been discovered either with duodenal ulceration or with any of the other conditions which show blood group associations. Hazarding a guess, 1 feel inclined to agree with those who suggest that the mechanism is likely to turn out to be immunological. You are, of course, aware of the recent literature on possible asso- ciations between ABO blood groups and infective diseases. None of these can be regarded as proved as yet, but there are the suggestions about group A and smallpox and group B and plague. There is also 80 2.198. p=<0.05. the finding of McDonald and Zuckerman (1962) of a highly significant excess of group O amongst those with upper respiratory illness who show evidence of infection with the Asian influenza virus, A2; this work, however, needs confirmation. Then there is the curious finding resurrected from the literature of the twenties that people of group O became more often, and with high significance, seronegative after anti- syphilitic treatment than did people of group A. It may indeed be that clues as to the meaning of the associations will emerge first from studies on micro-organisms. But I must return to the subject of chronic diseases. Another association, which I think we must accept as proved with very high probability indeed is that between nonsecretion and duo- denal ulceration. Secretion or, alternatively, nonsecretion in the saliva and some other body fluids of substances having ABO specificity depends on a single gene pair, secretion being dominant to nonsecre- tion. The loci for the ABO genes and for the secretor-nonsecretor genes are not genetically linked. Actually the association between nonsecretion and duodenal ulceration is closer than that between group O and ulceration, but as the numbers are so much smaller the probabilities are not so infinitesimal. Nevertheless, the evidence is overwhelming. As Doll et al. (1961) have shown recently, the separate effects of ABO blood group and of secretion must be regarded as multiplicative rather than additive. It should be mentioned again here that persons of groups A, B and AB go together in respect of susceptibility to duodenal ulceration. There is no difference in relative incidence between persons of these three groups; and all differ by the same amount from persons of group O. Hence we can think of two blood group classes only, namely O and not-O. Thus there are four pheno- types to consider: secretor, not-O; secretor, O; nonsecretor, not-O; nonsecretor, O. These can be symbolized SN, SO, sN, sO. The relative incidence of duodenal ulceration in the four phenotypes can be calculated. The results are shown in table 3. TABLE 3.— Relative incidence of duodenal ulceration in subjects of the four phenotypes Relative Incidence Phenotype Doll Clarke Secretor, not-O (SN). 1 1 Secretor, O (80) ARR SE SE NERS 1.4 1.3 NNonSeeretor, DOLD (SN)... ccuvsnussnsnmusssrmssmsmsmusrnm ans ne mea amma 1.8 1.7 NOnSerretor, O (80)... ccvoxsasnnasnnssinsnsnsnmnsmmssnmssnsssnmssniissnnsasnammmm 2.7 2.4 One column shows figures taken from the results of Doll et al. (1961), the other the results from Liverpool (including additional unpublished figures) as given by Hanley (1963). I should mention 81 that with Doll’s results I have combined three series without weighting to give a single series of figures. It will be seen how close these esti- mates are. Taking the lowest incidence, that in persons who are secretors and not-O, as unity, the relative incidence rises steadily, until in the limit, with people who are nonsecretors and group O, it is no less than 2.4 or 2.7. I think that there is little doubt that there is a genetic element in the etiology of duodenal ulceration. The best study I know is that of Doll and Buch (1950). They found that 45 out of 562 brothers of men with ulcer were similarly affected. The expected number, based on a comparable control population and matched age, was 16.67. Thus the incidence amongst brothers was 8.0 percent, against an expectation of 3.0 percent, or 2.7 times as great. Doll and Buch concluded that this resemblance between brothers must be largely genetic. One reason for this was that the resemblance between fathers and sons was much the same, and, of course, brothers share an environment which is more closely similar than do father and son. ABO blood groups and secretor status are wholly determined by heredity, and relatives tend to resemble each other in these respects to a greater degree than do unrelated people. It is possible, there- fore, to calculate how much of the resemblance between brothers in respect of duodenal ulceration can be attributed to their resemblance, for genetic reasons, in ABO blood group and secretor status. In other words, we can estimate how much of the resemblance in sus- ceptibility to ulcer is due to the effect of genes at two known inde- pendent loci. TABLE 4.— Comparison of incidence of ulcer in general population and in brothers 1 2 3 4 5 6 Column 2 Expected Column 5 Population | times Doll's | frequencies Expected | times Doll's Phenotype frequency | relative inci- of those frequencies | relative inci- dence of with D.U. | in brothers dence of table 3 table 3 BN sc cvuuneimusuwinmemmnwuansmenin . 3825 . 3825 .262 . 326 .326 we .3675 . 5145 .353 . 368 .615 .1275 . 2295 .158 .143 . 257 .1225 . 3308 .227 .163 . 441 Total 1.457 1.539 In the calculations of table 4, I have used the figures of Doll et al. (1961), as given in table 3. These are the relative incidences: secretor, not-O, 1; secretor, O, 1.4; nonsecretor, not-O, 1.8; nonsecretor, 0,2.7. We also need figures for the gene frequencies. For simplicity I have taken the frequency of O to be .7; of not-O, .3; of secretion, .5; of nonsecretion, .5. These are in fact close to the real frequencies in Great Britain and in much of North America. They are just about those that would be observed in northern England. 82 Given these estimates, the frequencies in the population of persons belonging to the four phenotypes are given in column 2. We can now multiply by the factors of table 3, which gives us column 3. Reducing the total to unity, we get column 4, which shows the ex- pected frequencies of the four phenotypes amongst those suffering from duodenal ulceration. Actually, these estimates correspond closely to the observed frequencies amongst those with ulcers. Apply- ing the Hardy-Weinberg law, which works so well in blood group calculations, we can estimate the frequencies of the four phenotypes amongst brothers of those with duodenal ulcers. The result is given in column 5. The brothers are naturally intermediate in blood groups and secretor status between the index cases and the general population. Finally, just as before, we multiply these last figures by the factors of table 3. This gives column 6, which gives a measure of the amount of duodenal ulceration in brothers. The ratio of the total of column 3 to the total of column 6 gives the relative incidence of ulcer in brothers to that in the general population, as determined, that is, by the genetic resemblance of brothers in regard to ABO group and secretion-nonsecretion. The answer comes to 1.056. That is, the effect of genes at these two loci makes the incidence of duodenal ulcer in brothers of index cases 5.6 percent higher than in the general population. Taking Doll and Buch’s estimate of an increase in frequency of ulceration in brothers of 170 percent over that in the general population, we see that only 3 percent of the degree of resemblance between brothers is to be ascribed to the known genes at two loci which are known to affect susceptibility, and this in spite of the considerable effect they have on degree of susceptibility in terms of the individual. What are we to make of this finding? It was clear on the face of it that resemblance in susceptibility to a disease and the side effect, if one may use that loose term, of a single major gene must be small, unless indeed the difference in susceptibility of the different pheno- types is very great indeed, and this is a very unlikely thing to find. What it does fit in with is the usual concept of multifactorial inheri- tance as the basis for genetically determined differences in resistance and susceptibility to common diseases. The genes at the two known loci are in this particular respect forming part of a multifactorial system, and the small contribution they make reinforces the conception of multifactorial systems based on the additive effect of many genes. It fits in with Darlington and Mather’s (1949) con- ception of polygenic inheritance, a criterion for which is that the effect of any single gene pair shall be small in relation to the total variation. There is nothing odd about the finding that ancillary effects of major genes play their part in multifactorial systems. This has never been denied by those who have been the foremost advocates of multifactorial determination, though they would think it likely 83 that the bulk of multifactorial determination is due to genes of small effect, genes which do not have a major effect in some other context. The idea is in fact at least 45 years old. I wish I had been able to verify my reference here, but I have not; it is all so long ago. Perhaps Professor Neel, or someone else who has been a real Drosophila geneticist, can help me. [Editorial note: cf. Dobzhansky, T., 1927. Studies on the manifold effect of certain genes in Drosophila melano- gaster. Zirt. f. ind. Abstam.—und Vererbungslehre 43: 330-388.] My recollection is that somewhere about 1916, one of the earlier geneticists—it may have been Sturtevant—measured the size of certain organs, for example, the length of spermathecae, in wild type and white-eyed flies. He found that the gene difference made small differences to these metrical characters. It seems to be generally true that in animals, as contrasted to plants, differences in genetically determined degrees of resistance and susceptibility to infective diseases are multifactorially determined. The same thing is probably equally true of genetic differences in respect of chronic and noninfective diseases. In fact I would go a step further. Evidence seems to be accumulating that even with defects and deformities, whose causation is partly genetic, the genetic basis may also be multifactorial, at least in some instances. One very important finding here is that of my colleague, Dr. Cedric Carter, on congenital pyloric stenosis (Carter 1961a, 1961b). The sex incidence in this condition is very unequal: one boy in 200 male live births; one girl in every 1,000 female births. What Carter has found is that the incidence in sibs and children of affected women is notably higher than in sibs and children of affected men. For example, about 7 or 8 percent of sons or brothers of affected men are affected, but no less than about 20 percent of sons and brothers of affected women. This is quite incompatible with any single gene hypothesis. Women who are affected are affected in spite of the protection conferred by their sex; they are persons of unusually high genetic susceptibility, and this is faithfully mirrored in their relatives. There must be, in fact, degrees of susceptibility —that is, a continuous distribution and, pretty certainly, a multifactorial basis. We are finding indications of the same thing with harelip. (Roberts, Carter and Buck, in preparation.) There are two conflicting pieces of evidence. In the first place the incidence in sibs and in children of affected persons is the same. This points to a dominant gene, of partial penetrance of course. In the second place, when two or more relatives other than sibs are affected, they tend quite often to be on both sides of the family, which equally clearly points to a recessive gene. Our feeling is that the evidence really points to neither. On the other hand, multifactorial determination would fit perfectly well and we have some preliminary indication of the same sex relationship as in congenital pyloric stenosis. I can also mention some results on 84 talipes equino-varus, which go the same way (Wynne-Davies, 1963). But I have strayed too far from chronic diseases, and will only say that the general character of inheritance seems likely to be multifac- torial. There is also the point that the genetic element is usually relatively small, though not invariably so, in relation to the nongenetic factors involved in causation. But, of course, even if small it is nonetheless important in understanding the etiology of the various conditions. A final word might be added on blood group and disease associations in general. The proposition that they do not exist must, I think, be rejected. The next objection, that they might be secondary to racial or some other stratification seems exceedingly unlikely, to say the least. Furthermore, in the one instance in which the hypothesis has been put to the test it has been contradicted by the findings. But some might still say: so what? What do these associations matter in practical terms? Of course the evidence remains epidemiological. No satisfactory explanation, no indication of mechanisms, have yet emerged. But I feel sure they will. Epidemiology is, of course, the beginning and not the end of research. It shows that there is a prob- lem to investigate and may point the way to useful lines to follow. If due note had been taken of the fact that at the Retreat, a mental hospital at York founded by Quakers and mostly taking in Quaker patients, general paralysis was almost unknown at a time when it represented a substantial fraction of admissions to ordinary mental hospitals, then the connection between syphilis and general paralysis might have been discovered far earlier than it actually was. There have been some interesting findings in regard to blood group and disease associations, but nothing very meaningful so far. I can add one new observation, due to Hanley (1963) whose summaries I have already quoted. Using an improved technique for measuring levels of serum pepsinogen, he was able to show that persons of group O have on the average higher levels than persons of group A. This is in normal subjects, not those with ulcer. The difference is very highly significant. There is also a sex difference. Interestingly, there is no difference between secretors and nonsecretors, so presumably a different mechanism may be involved. I do feel sure, however, that in due time the underlying mechanisms will be found and that the result will be a useful contribution to the study of several diseases. REFERENCES 1. Aird, I., Bentall, H. H., and Roberts, J. A. Fraser, 1953. Relationship between cancer of the stomach and ABO blood groups. Brit. Med. J. I: 799-801. - 2. Aird, I., Bentall, H. H., Mehigan, J. A., and Roberts, J. A. Fraser, 1954. Blood groups in relation to peptic ulceration and carcinoma of the colon, rectum, breast, and bronchus; association between ABO groups and peptic ulceration. Brit. Med. J. 2: 315-321. 85 10. 1. 12. 13. 86 . Buchanan, J. A., and Higley, E. T., 1921. The relation of blood groups to disease. Brit. J. Exp. Path. 2: 247-255. . Carter, C. O., 1961a. The inheritance of congenital pyloric stenosis. Brit. Med. Bull. 17: 251-254. Carter, C. O., 1961b. In Clinical Aspects of Genetics, edited by F. Avery Jones. London: Pitman. Darlington, C. D., and Mather, K., 1949. The Elements of Genetics. London: Allan and Unwin. Doll, R., and Buch, J., 1950. Hereditary factors in peptic ulcer. Ann. Eugen. 15: 135-146. . Doll, R., Drane, H., and Newell, A. C., 1961. Secretion of blood group substances in duodenal, gastric and stomal ulcer, gastric carcinoma, and diabetes mellitus. Gut 2: 352-359. Hanley, W. B., 1963. Hereditary aspects of duodenal ulcer. M.D. Thesis, University of Liverpool. McDonald, J. C., and Zuckerman, A. J., 1962. ABO blood groups and acute respiratory virus disease. Brit. Med. J. 2: 89-90. Ugelli, L., 1936. Distribuzione dei gruppi sanguingni negli individui porta- tori di ulcere gastroduodenali. Policlinico (sez prat.) 43: 1591-1594. Weiner, A. 8., 1962. Blood Groups and Disease (A Critical Review). Lancet 1: 813-816. Wynne-Davies, R., 1963. Personal communication. Epidemiological Concepts Applied to Studies of Chronic Diseases* By ABranam M. LiuienreLp, M.D., Department of Chronic Diseases, Johns Hopkins School of Hygiene and Public Health, Baltimore, Md. Students of the chronic diseases have become increasingly interested in the distribution of disease in the population and of those factors that influence this distribution, that is, in the epidemiological aspects of disease. The value of such knowledge has been repeatedly demon- strated in the study of the infectious diseases and of some diseases of current medical and public interest such as heart disease and various forms of cancer. There is an increasing indication that such knowledge is useful to the geneticist who is concerned with the popu- lation aspects of his discipline. This afternoon, time permits only a rather brief summary of this subject. Since many of the aspects have been reviewed recently in various articles and textbooks, we would first like to provide a broad overall view of epidemiological methods in general terms and, then, illustrate the application of some of these methods to two forms of cancer in which we have been interested. In studying the distribution of a disease in a population, attention is paid to variations in disease frequency by such factors as time, place and certain individual characteristics of those affected by the disease. Thus, the epidemiologist studies the changes in the frequency of the disease over a period of time, its distribution by various geo- graphic areas and the frequency of disease by age, race, sex, socio- economic class, individual living habits such as smoking, and exposure to agents such as radiation. He is interested in determining what characterizes the diseased person in contrast to the nondiseased in the population. *Some of the studies, whose results are reported in this paper, were supported in part by a Public Health Service career program award number GM-K6-13,901 from the National Institute of General Medical Sciences, by a grant from the Mary Reynolds Babcock Foundation, Winston-Salem, North Carolina, and by Training Grant No. CT-5085 from the National Cancer Institute. 87 Basically, the epidemiologist attempts to determine whether there are statistical associations between a disease and certain population characteristics or attributes of possible etiological importance. Thus, the first phase of the epidemiological approach to a disease problem is the determination of these statistical associations. The second phase is the derivation of biological inferences from the pattern of statistical associations. It is the totality of the series of statistical associations and these biologic inferences that are termed the “epidemiology of the disease.” The methods of determining statistical associations can be classified as follows: (A) Studies of General Population Characteristics 1. Vital Statistics including mortality and morbidity data obtained on a routine basis. 2. Special Population Surveys. (B) Studies of Individual Characteristics 1. Observational Studies (a) Cross-sectional study (b) Retrospective study (¢) Prospective study 2. Human Experimental Studies In our discussion of specific examples, we will illustrate several of these methods. Their relative advantages and disadvantages have been discussed in previous publications (Lilienfeld, 1959; MacMahon et al., 1960). Once these associations have been determined, the two general types of hypotheses that must be considered to explain these asso- ciations are: (1) a causal hypothesis, which states that the factor associated with the disease is a cause of the disease, or (2) an indirect association hypothesis, which states that the statistical association is indirect and noncausal and that it is a reflection of the existence of a common underlying X factor which predisposes the individual to have the disease as well as to have the characteristic. The problem then consists of determining the relative plausibility of these two hypotheses. This is usually done by integrating data obtained from experimentation, both animal and human, studies of biochemical and pathogenetic mechanisms and various types of further epidemiological studies. Some of the issues involved in these problems have been recently discussed (Yerushalmy and Palmer, 1959). With this structural outline as a background, we would like to re- view some of the epidemiological aspects of breast cancer to illustrate the types of data and the types of reasoning used by the epidemiologist. We have already indicated that the epidemiologist is interested in the age distribution of a disease. Figure 1 contains the average annual age-specific death rates for breast cancer by sex. These rates are plotted on semi-logarithmic paper, which represents the 88 rate of change of mortality with age. The death rates for females suggest that over a lifetime, they can be resolved into two linear components, one consisting of those prior to and one of those after the 40 to 45 year age group and that the rates in the second part of the curve have a lower slope than those in the first. This observa- tion has been interpreted as being consistent with the hypothesis that the proportion of women in the general population who are “susceptible” to breast cancer decreases after 40 to 45 years of age. Since this coincides with the average age, of onset of menopause, these data suggest that entrance into menopause decreases susceptibility to breast cancer. With respect to the males there is also a suggestion that the rate of increase of mortality, as shown by the change in the slope of the line is slightly less after 40 to 50 years of age. Another possible explanation for this change in slope which must be considered is that we are dealing with two separate sets of etio- logical factors. DeWaard et al. (1960) suggested that the underlying etiological factor for breast cancer was excessive estrogenic activity, but that prior to 40 to 45 years of age, ovarian estrogenic activity was of etiological importance, whereas after the menopause it was the estrogenic activity of adrenal origin. They attempted to look at this indirectly and presented some data to suggest that more of the postmenopausal breast cancer patients had obesity and hyper- tension than a control group, whereas this was not observed in the premenopausal group of cancer patients. Their study leaves a great deal to be desired but further studies comparing the characteristics of these two groups of breast cancer patients with controls would appear worthwhile. Differences in these two groups with respect to a history of nursing habits have been reported by Kamoi (1960). He found that female breast cancer patients over 45 years of age gave a history of nursing their children for shorter periods of time than a control group whereas those between 35-44 years of age did not show such a difference. We shall discuss the pertinence of length of nursing to breast cancer later. The most consistently observed factors associated with female breast cancer concern marital and pregnancy habits. Breast cancer is more frequent among those never married than among those who marry and breast cancer patients on the whole marry at a later age than do women in general. The latter association results in their age at first pregnancy being later than usual and possibly also in their having less children, an observation which has been consistently made (Lilienfeld, 1961; Wynder et al., 1960). In figure 2 there are presented the age-specific incidence rates of reported cases of breast cancer in 1949 among single and married women in upstate New York; the term “single” applies to women reported as such at the time of report in contrast to those who were reported as being married, widowed or divorced, which have been 89 735-712 0—65——T7 200 | 100 50 FEMALE ° T 100,000 POPULATION wu o T 1.0 @ a ST » b - < w o | | 08 I ceeess 1939-4) xxxxx 1949 - S| 01 > 1 1 1 1 1 20 30 40 S50 60 70 80 90 AGE (YEARS) FiGure 1.—Average annual age-specific death rates (per 100,000 population) for breast cancer by sex, white population, U.S., 1939-41, 1949-51. grouped together as the ‘“‘married” group. It is apparent that the resolution of the incidence rates into two linear components is present for both the single and married groups. The rates comprising the first component are almost identical for both the single and the married women. The excess incidence of female breast cancer among single women appears to be limited to the latter period of life, after about 40 years of age. Similar patterns have been observed in selected metropolitan areas of the United States, the Netherlands, and England and Wales (Dorn and Cutler, 1955; Stocks, 1955; Versluys, 1955). In accordance with our previous interpretation of the age pattern, one possible explanation for the variation of incidence with marital 90 500 400 8 300- < 3 200- a oO a wl 2 100 = = ¥ 70d S s50- g a S - 2 . [14 Ww a — wn Ww 2 oO a 107 SINGLE .... 2 : MARRIED x x x x o 1 r g 5 x w bs 0 7 . ac w m o = = Z | —— , A Spay " 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 AGE OF REPORT Ficure 2.—Annual age-specific incidence rates of reported cases of cancer of the female breast by marital status, New York State, exclusive of New York City, 1949. status may be that single women as a group enter the menopausal state later, and consequently remain in the more highly susceptible state longer than do married women. In an attempt to test this hypoth- esis, the menopausal histories of patients admitted to the Roswell Park Memorial Institute were studied; the results have already been reported in detail (Lilienfeld, 1956). Briefly, we found that the fre- quency of artificial menopause was 30 percent less among single than among married women and that artificial menopause occurred about 10 years earlier than natural menopause. No difference in time of occurrence of natural menopause in the two groups was noted. Also, patients with breast cancer had a lower frequency of artificial meno- 91 pause than did female patients with cancers of other sites. These results were interpreted as being consistent with the hypothesis that artificial menopause may have some protective effect with regard to breast cancer, and that the excess incidence of breast cancer among single women may be related to their having a lower frequency of artificial menopause. From a biologic viewpoint these results are of interest since they suggest a relationship of differences in frequency of breast cancer by marital status to certain factors associated with the endocrine system. The inverse relationship between breast cancer and artificial menopause has been confirmed by MacMahon and Feinleib (1960) in another study of hospital patients. However, it is important to remember that these data are based on hospital popu- lations and they should be confirmed by studies of the general population. One aspect of breast cancer in which there has been considerable interest concerns nursing habits—an interest that seems quite natural. We have recently reviewed in detail the results of investigations indicating that a larger percentage of breast cancer patients had never nursed their children than other women and/or that breast cancer patients nursed their children for shorter periods of time than women in general (Lilienfeld, 1963). We indicated that the possible relationship of nursing history to breast cancer was far from being firmly established since the previous reports did not have consistent results. Also, these studies have been retrospective studies in which women with breast cancer were queried about their nursing habits, 15 to 25 years prior to the time of interview. The validity of this type of information may be readily questioned. The possible relationship between nursing habits and breast cancer is provocative with respect to the development of a biological hypoth- esis. Present evidence suggests that ovarian function can influence the development of breast cancer. One factor that depresses ovarian function is prolonged nursing. Thus, nursing may not have a direct protective effect on breast cancer but an indirect one through its depressive influence on ovarian function. The hypothesis is attrac- tive since it has been invoked as a possible explanation for the differ- ences in mortality rates of female breast cancer in different countries. To illustrate this, figure 3 contains a comparison of the age- adjusted mortality rates from female breast cancer in several countries for the period 1952-1954 as collected by Segi ef al. (1959). In com- paring these rates one must keep in mind the differences in death certification practices as well as the availability of medical care in these countries. The most striking fact emerging from this comparison is the remarkably low death rate in Japan. Wynder et al. (1960) have suggested that it may be a result of the fact that Japanese mothers nurse their children for a two-year period, which is longer than in the other countries. 92 Rate per 100,000 pop. 0 5 10 15 20 25 T T T T 1 DENMARK NETHERLANDS ENGLAND & WALES CANADA SOUTH AFRICA SCOTLAND NEW ZEALAND US. WHITE SWITZERLAND ISRAEL IRELAND BELGIUM U.S. NONWHITE AUSTRALIA NORTH IRELAND SWEDEN NORWAY GERMANY, F.R. FRANCE AUSTRIA ITALY FINLAND PORTUGAL JAPAN Firoure 3.—Age-adjusted female breast cancer mortality rates in 24 countries, 1952-54, according to Segi (1959). We thought it would be of interest to make a similar comparison with respect to death rates of breast cancer among males; these are presented in table 1, which indicates that for both males and females the death rates are lower in both Finland and Japan. The fact that the low mortality rates in Japan are present for both sexes raises questions as to whether differences in breast feeding habits are a reasonable explanation for differences in the frequency of breast cancer among females. If one assumes that these two diseases are similar (which may be an erroneous assumption), it would appear that there are factors present in Finland and Japan which decrease the risk of both the male and female inhabitants developing and dying from breast cancer. These differences in mortality should be studied further, particularly if one is interested in determining whether they reflect differences in the sociocultural environment, such as nursing 93 habits in these countries or differences of a genetic-constitutional -_, nature. Several types of studies come to mind, e.g., a comparison of “the frequency of breast cancer among Japanese-Americans who were born in Japan with that among American-born women of Japanese descent. Similar studies of those of Finnish birth and descent would be important. TABLE 1.— Average annual age-adjusted death rates (per 100,000 population) for breast cancer among females (1954-1955) and males (1950-1957) for selected foreign countries and the United States* Country | Female Male DERIMIALE. . .oarsnrssmne sss messmsnen mst mR En SEER S nS ee re me mn Si 24.2 0.23 England and Wal 23.2 0.25 Canada 22.2 0.22 UNICO SLALES (WHITEY , ccm mes som wmss mmo wm nm mi mo im om i 0 0 21.7 0.25 TIA BU0S TOT INIYOY sms sires sisi i 19.1 0.35 BIOCONTROL 18.9 0.15 Germany (Fed. Repub.) _..._. 15.1 0.21 ance... 14.0 0.34 Finland 1.7 0.04 Japan. ____ 3.9 0.07 *Adjusted with standard population used by Segi (1959) from the age distribution of the total population of 46 countries, 1950. Stimulated by animal investigations indicating the influence of genetic factors, many familial studies of female breast cancer have been carried out. These studies have consisted of determining the prevalence of breast cancer among the relatives of a group of breast cancer patients and comparing it with an expected number, estimated from some type of control group. In table 2, we have summarized the results of these studies with respect to the relative prevalence or mortality of breast cancer among mothers and sisters of breast cancer patients. Of these 8 studies, 6 showed the presence of familial aggre- gation and 2 did not. In those studies showing familial aggregation, there is a fairly consistent finding of a twofold excess frequency of breast cancer among mothers and sisters. On the whole, these results suggest the influence of genetic factors. However, one must also keep in mind the possibility that familial aggregation may be a manifesta- tion of environmental factors common to members of the same family. TABLE 2.—Summary of results of studies of familial aggregation of breast cancer Prevalence or mortality of breast cancer among: Number of Author, Year cases Mothers of cases Sisters of cases studied Observed| Expected| Ratio: O/E| Observed| Expected| Ratio: O/E Jacobsen, 1946... ....... 200 21 2 3.0 13 5 2.6 Penrose ef al., 1948_ 300 25 1n 2.3 23 7 3.3 Passey et al., 1952. 585 23 20 Ll ncn [esmmmeses]usee ummm Smithers, 19048 ____ 556 29 13.9 21 leer emilee Ee em Se Woolf, 1955. _____________ 200 4 21 1.9 8 3.2 2.5 Anderson et al., 1958 _____ 544 9 7.7 1.2 28 12.4 2.3 Murphy and Abbey, 1959_ 200 7 3 2.3 2 6 0.3 Macklin, 1959... .________ 295 11 5.6 2.0 14 5 2.8 94 The degree of familial aggregation can be evaluated in the light of the other epidemiological data that we have reviewed. These data, viewed totally, are consistent with indicating that endogenous, hormonal factors are of etiologic importance and one would naturally expect these factors to be under genetic control. Therefore, it would seem to us that we should expect a greater degree of familial aggregation than actually has been observed, particularly since similar degrees of familial aggregation have been noted in studies of cancer sites where the epidemiological evidence suggests that environmental factors may play a considerable role, such as in the case of gastric cancer. This relative inconsistency of pattern awaits explanation. In addition to the specific individual factors already discussed, there are several others that have been reported as being associated with the occurrence of breast cancer among females. It has been observed that the incidence and/or mortality rates from breast cancer are higher in the upper socioeconomic groups than in the lower ones (Graham et al., 1960), higher in the urbanized areas of the United States (Levin et al., 1960), higher among the Jewish group in New York City than among other religious groups (Newill, 1961), higher among the native than the foreign born in the United States (Haenszel, 1961), and higher among those with benign breast conditions (Lewison and Lyons, 1953). Further study of these differences is needed. With regard to the suggestive etiologic importance of endocrine factors, it is of interest that Sommers (1955) and Brachetto-Brian and De Srulijes (1954) observed histologic changes of the ovary suggesting the existence of hyperestrinism in women with breast cancer. Their studies were based on autopsy material, limiting the inferences that can bemade. There have been several reports that disordered thyroid function, as shown by a higher frequency of goiter, may be related to breast cancer (Bogardus and Finley, 1961; Liechty et al., 1963). Again, many of these studies are subject to methodological objections, although further, better-controlled observations are warranted. It would appear fruitful to compare a profile of the endocrine status of breast cancer patients with that of controls. Attempts at such studies have been limited by methodological problems which may be resolved in the near future. As Bulbrook and Strong (1958) have indicated recently, whatever studies have been conducted have yielded results which in many respects are disappointing. Perhaps, the solution of the methodological problems will eventually permit answers to some of the questions that have been raised. In considering possible epidemiological approaches to a disease problem which appears principally to be a result of endogenous factors, it occurred to us that it would be worthwhile to study the characteris- tics of male patients with breast cancer. Because of the rarity of this condition among males and because of the seeming importance of female-type hormones, these individuals might be strikingly 95 different. During 1961 and 1962, such a study was conducted by Schottenfeld and Diamond (1963); the details of the study will be reported elsewhere but we will briefly summarize the results here. It was possible to interview 53 male breast cancer patients who had been treated during 1955-1962 in several clinics in many of the large cities in the U.S. For purposes of comparison, 53 patients with gynecomastia and 53 with cancer of the colon, matched by age and race were also interviewed. There was an excess frequency of Jews, single men, and those with later ages at first marriage among the breast cancer patients than among the others; although these were not statistically significant they were consistent with epidemiological data for females with breast cancer. Of greater interest were the findings with respect to a history of orchitis, orchiectomy, the receipt of hormonal therapy, family history, X-ray exposure and antecedent benign breast disease. The results of the comparisons for these specific factors which may be of carcino- genic importance, are presented in table 3. In general, the number of patients with any of these characteristics is rather small, but there is an excess frequency of them among the breast cancer patients as com- pared to the control groups. Since several patients had one or more of these characteristics, the total number of patients who had one or more of these characteristics was computed; 22 percent of the breast cancer patients gave such a history as compared to 8 percent of the gynecomastia and 2 percent of the colon cancer group. TABLE 3.—Summary of history of characteristics among male patients with breast cancer, gynecomastia, and colon cancer Number of patients with history of— Total of | Percent Total Hor- patients | with one Study group pa- mones Family | Expo- | Antece- | with one | or more tients | Orchi- | Orchi- | (thyroid,| history | sure to dent or more | charac- tis |ectomy | cortisone,| of cancer | X-ray | benign | charac- | teristics testoster- | of male to breast | teristics one, stil- | breast | chest | disease besterol) Breast cancer. ____ 53 4 1 3 2 2 4 11 22 Gynecomastia. ___ 53 1 3 0 0 0 4 8 Colon cancer... 53 0 0 1 0 0 0 1 2 Of specific interest from an endocrine viewpoini are the four patients with a history of orchitis and one with an orchiectomy. In view of the other epidemiological evidence the most acceptable interpretation is that orchitis produced a certain degree of testicular atrophy resulting in hormonal changes, which in turn played an etiological role in the development of the breast cancer. However, one cannot exclude the possibility that those with a history of orchitis may have had a concurrent infection of breast tissue which directly produced the neoplastic transformation. 96 The history of familial aggregation in two patients is extraordinarily high in view of the rarity of the disease. However, interpretation of this finding is difficult since it may reflect either genetic factors or environmental factors, such as the familial aggregation of infectious diseases producing orchitis. The exposure to chest X-rays is somewhat unexpected since this is the first instance, insofar as we are aware, of the suggestion that X-rays may be implicated in breast cancer. It should be noted that these represent exposures to therapeutic X-rays, probably of high doses. Any generalization of these results should be made cautiously in view of the limited nature of these data. But they are sufficiently interesting and provocative to serve as a basis for serious consideration of further research along similar lines. The review of some of the salient features of breast cancer illustrates the epidemiological approach to a disease in which endogenous factors appear to be of major etiological importance. Despite the complexi- ties of the relationships, we discussed breast cancer first, since in a conference devoted to genetic contributions to epidemiologic studies, we wanted to indicate that epidemiology is not necessarily limited to a consideration of environmental determinants of disease, but attempts to achieve an integrated, holistic concept. Now, we would like to consider some aspects of the epidemiology of a cancer site where the epidemiological evidence indicates that the environment plays a major etiological role. There is no need to review in detail the evidence that implicates cigarette smoking, ex- posure to certain selected occupations, and an urban factor which may be air pollution, as being of importance in the etiology of lung cancer. We will limit our discussion to a brief and selective review of some of the available data. Many retrospective studies have indicated that a larger proportion of lung cancer patients are cigarette smokers than a comparative group of controls, and of those who smoke, a larger proportion are heavy smokers. For various methodological reasons it was considered desirable to confirm this association by means of a prospective study. Such a study was initiated in 1952 by Hammond and Horn (1958) of the American Cancer Society. Histories of smok- ing habits were obtained on about 190,000 men aged 50 to 69 years; these men were traced for 44 months. Death certificates were obtained on all who died and additional medical information was obtained on those who were reported to have died of cancer. Altogether about 12,000 deaths were reported, of which 2,249 were attributed to cancer. The death rates for verified bronchogenic carcinoma by amount of cigarettes smoked are presented in figure 4. We note the increasing rate of dying from lung cancer with increasing amount of cigarettes smoked. Of additional interest is the observation that the death rate from lung cancer was lower for those who had been regular cigarette smokers in the past but were not smoking at the time of the study. 9 250 » 217.3 x S 200 Ww 20 z < = 8 iso} 143.9 < oc eo ti Ww oof w EX-REGULAR = CIGARET SMOKERS < E sof 44,0 we < w o 3.4 Ob, NEVER NONE ~~ OCCASL LESS THAN 10-19 20-39 40+ SMOKED 10 CIGARETS CIGARETS CIGARETS CIGARETS Ficure 4.—Age-adjusted death rates of well-established cases of bronchogenic carcinoma (exclusive of adenocarcinoma) by number of cigarettes smoked daily. Two additional prospective studies by Doll and Hill (1956) and by Dorn (1959) had similar results. Hammond and Horn were also able to look at the influence of a possible urban factor. In figure 5, there are presented the death rates from lung cancer among cigarette smokers and nonsmokers in rural and urban areas of various sizes. The lung cancer death rates were a great deal higher among regular cigarette smokers than among nonregular smokers, regardless of whether they lived in urban or in rural areas. However, for both cigarette smokers and nonsmokers, the death rates were higher in the more urbanized areas than in the rural ones. But the data do also indicate that the contribution from smoking is greater than that from urbanization. In addition to these two environmental agents—smoking and urbanization—there is evidence that certain occupational exposures increase the lung cancer risk. But the contribution of this to the total amount of lung cancer is relatively small. Recently, it has been shown (Tokuhata and Lilienfeld, 1963), as Tokuhata will report this afternoon, that there is a familial factor influencing the occurrence of lung cancer. This can be most reasonably interpreted as reflecting a genetic component which would influence an individual’s susceptibility to the major environmental agent— cigarette smoking. 98 100 0 « 85.2 Ww > z Sr 70.9 7.7 Z 65.2 o o © o o Sof x w a w - = 25) E 14.7 g 93 oc 47 0 0 ates CITY OF CITY OF SUBURB OR TOWN RURAL 50,000+ 10,000-50,000 Bl NEVER SMOKED REGULARLY [J ciGARET Age standardized death rates due to well-established cases of bronchogenic carcinoma (exclu- sive of adenocarcinoma) by urban-rural classification. Rates for cigarette smokers are com- pared with those for men who never smoked regularly.43) Fieure 5.—Age-adjusted death rates of well-established cases of bronchogenic carcinoma (exclusive of adenocarcinoma) for regular cigarette smokers and nonsmokers by urban-rural classification. To provide some idea of the relative risk associated with cigarette smoking, it has been estimated that a person who smokes two or more packs of cigarettes a day has a lifetime probability of dying from lung cancer of about 1 in 10 as compared to a nonsmoker whose lifetime probability is about 1 in 700. This suggests that there may be a segment in our population who are susceptible to this particular environmental agent. This is not an unreasonable assumption in view of what we know about the biology of disease. It is also rea- sonable to assume that this susceptibility reflects genetic factors, particularly in the light of the results of the familial studies. From an epidemiological viewpoint and also from a genetic view- point it would be of interest to determine the specific factors or indexes of susceptibility. Unfortunately, there is no theoretical basis for “designing a study to determine such factors. The only type of approach that could be adopted is a “shotgun’’ one in which a large group of males in appropriate age groups, perhaps 40,000 or 50,000 such individuals, would be enrolled in a followup study. Initially, information would be obtained with respect to various demographic characteristics, smoking habits and other environmental exposures. 99 Blood samples would be collected, frozen and stored. The group would then be followed for a number of years and the blood samples for those who die of lung cancer, together with a group of controls, would be examined for various genetic markers such as blood groups, enzymes, etc., to determine if there are any relationships with lung cancer. Needless to say, such a study would require a great deal of effort but the results might be very worthwhile. Such information would be of interest biologically, since one of the areas of mutual concern to geneticists and epidemiologists is a specification of the genetic- environmental interaction in this disease entity. From a public health viewpoint, such information would be useful since there is a real question concerning our ability to change the smoking habits of masses of people. However, if we could select those individuals who are susceptible to the environmental agent, they might be more readily influenced. This general approach could be usefully applied to several other chronic diseases such as breast cancer, hypertension, etc. In our discussion of the application of epidemiologic methods to the chronic diseases, we have attempted to orient our thinking to the contributions of genetics to epidemiologic studies. However, we should not lose sight of the other side of the coin—the contributions that epidemiologic methods could make to genetic studies. Time does not permit an expansion of this theme. Briefly, we would like to call attention to the need by geneticists for reliable data on the incidence of disease as a basis for studies of linkage and modes of inheritance. There is also a need for determining the environmental causes of chromosomal aberrations, such as those producing nondis- junction, and of mutations. No doubt there will be some discussion of these issues in the workshops this afternoon. In this rather brief exposition of the use of epidemiologic concepts in the chronic diseases, we have tried to emphasize that the epi- demiologist, who is interested in studying the chronic diseases, depends upon the accumulation of knowledge of various kinds, clinical, genetic, epidemiological and others. He is interested in the totality of the data obtained by different approaches, from which a pattern emerges and serves as a basis for deriving fairly valid inferences concerning the biology of the disease in human populations. REFERENCES 1. Anderson, V. E., Goodman, H. O., and Reed, S. C., 1958. Variables Related to Human Breast Cancer. Minneapolis, University of Minnesota Press. 2. Bogardus, G. M., and Finley, J. W., 1961. Breast Cancer and Thyroid Disease. Surgery 49: 461-468. 100 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. . Brachetto-Brian, D., and De Srulijes, L. K., 1954. Le role de l'ovaire dans I’hyperestrogenisme chez les femmes avec cancer la mamelle. International (6) Cancer Congress, Sao Paulo (Abstr. of Papers): p. 249. Quoted in CA 4: Newsletter XIV, Nov. 1954. . Bulbrook, R. D., and Strong, J. A., 1958. Hormone Studies in Breast Cancer, in Raven, R. W. (Ed.). Cancer, Vol. 6, Chpt. 7: 215. London, Butterworth. . De Waard, R., De Laive, J. W. J., and Baanders-Vanhalewijn, E. A., 1960. On the Bimodal Age Distribution of Mammary Carcinoma. Brit. J. Cancer 14: 437-448. Doll, R., and Hill, A. B,, 1956. Lung Cancer and Other Causes of Death in Relation to Smoking—A Second Report on the Mortality of British Doctors. British Med. J. 1071-1081. . Dorn, H. F., 1959. Tobacco Consumption and Mortality from Cancer and Other Diseases. Pub. Health Reports 74: 581-593. . Dorn, H., and Cutler, S. J., 1955. Morbidity from Cancer in the United States. Part I. Variation in Incidence by Age, Sex, Race, Marital Status and Geographic Region, Public Health Monograph No. 29, Public Health Reports. . Graham, S., Levin, M., and Lilienfeld, A., 1960. The Socioeconomic Distri- bution of Cancer in Various Sites in Buffalo, N.Y., 1948-1952. Cancer 13: 180-191. Haenszel, W., 1961. Cancer Mortality Among the Foreign-born in the United States. J. Nat. Cancer Inst. 26: 37-132. Hammond, E. C., and Horn, D., 1958. Smoking and Death Rates—Report on 44 Months of Followup of 187,783 Men. J.A.M.A. 166: 1294-1308. Jacobsen, O., 1946. Heredity in Breast Cancer: A Genetic and Clinical Study of Two Hundred Probands. London, H. K. Lewis. Kamoi, M., 1960. Statistical Study on Relation between Breast Cancer and Lactation Period. Third Report. Study through Relative Liability. Tohoku J. Exper. Med. 72: 72-77. Levin, M. L., Haenszel, W., Carroll, B. E., Gerhardt, P., Handy, V. H., and Ingraham, S. C., II, 1960. Cancer Incidence in Urban and Rural Areas of New York State. J. Nat. Cancer Inst. 24: 1243-1257. Lewison, E. F., and Lyons, J. G., Jr., 1953. Relationship between Benign Breast Disease and Cancer. A.M.A. Arch. Surg. 66: 94-114. Liechty, R. D., Hodges, R. E., and Burket, J., 1963. Cancer and Thyroid Functions. J.A.M.A. 183: 30-32. Lilienfeld, A. M., 1956. The Relationship of Cancer of the Female Breast to Artificial Menopause and Marital Status. Cancer 9: 927-934. Lilienfeld, A. M., 1959. Epidemiologic Methods and Inferences, in Hilleboe, H. E., and Lammiore, G. W., editors: Preventive Medicine, Philadelphia, W. B. Saunders Company. Lilienfeld, A. M., 1961. Some Epidemiologic Aspects of Cancer of the Breast. Proceedings Fourth National Cancer Conference, Philadelphia, J. B. Lippincott Co., pp. 215-223. Lilienfeld, A. M., 1963. The Epidemiology of Breast Cancer. Presented at Annual Meeting, Amer. Assn. Cancer Research, May 23, 1963. MacMahon, B., and Feinleib, M., 1960. Breast Cancer in Relation to Nursing and Menopausal History. J. Nat. Cancer Inst. 24: 733-753. MacMahon, B., Pugh, T. F., and Ipsen, J., 1960. Epidemiologic Methods. Boston, Little, Brown and Co. Macklin, M. T., 1959. Comparison of the Number of Breast Cancer Deaths Observed in Relatives of Breast Cancer Patients and the Number Expected on the Basis of Mortality Rates. J. Nat. Cancer Inst. 22: 927-951. 101 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 102 Murphy, D. P., and Abbey, H., 1959. Cancer in Families. (A Study of the Relatives of 200 Breast Cancer Probands.) Cambridge, Harvard Univ. Press. Newill, V. A., 1961. Distribution of Cancer Mortality among Ethnic Sub- groups of the White Population of New York City, 1953-58. J. Nat. Cancer Inst. 26: 405-417. Passey, R. D., Waiman, M., Armstrong, E. C., and Rhodes, J., 1952. He- redity in Human Breast Cancer. Acta Unio Internat. Contra Cancrum 8: 184-186. Penrose, L. S., Mackenzie, H. J., and Karn, M. N., 1948. A Genetical Study of Human Mammary Cancer. Eug. 14: 234-266. Schottenfeld, D., Lilienfeld, A., and Diamond, H., 1963. Some Observations on the Epidemiology of Breast Cancer among Males. Am. J. Public Health (in press). Segi, M., Fuzisaku, S., Kurihara, M., Narai, Y., Ito, M., and Ogata, Y., 1959. Age-adjusted Death Rates from Malignant Neoplasms for Selected Sites, by Sex, in 24 Countries, in 1952-53, 1954-55, and 1956-57. Sendi, Japan, Tohoku Univ. School of Medicine, Dept. of Public Health. Smithers, D. W., 1948. Family Histories of 459 Patients with Cancer of the Breast. Brit. J. Cancer 2: 163-167. Sommers, S. C., 1955. Endocrine Abnormalities in Women with Breast Cancer. Lab. Invest. 4: 160-174. Stocks, P., 1955. Social Status in Relation to Carcinoma of the Breast. Schweiz Z. Path. Bakt. 18: 706-717. Tokuhata, G., and Lilienfeld, A., 1963. Familial Aggregation of Lung Cancer in Humans. J. Nat. Cancer Inst. 30: 289-311. Versluys, J. J., 1955. Marriage, Fertility, and Cancer Mortality of the Specifically Female Organs: Mammary Carcinoma. Brit. J. Cancer 9: 239-245. Woolf, C. M., 1955. Investigations on Genetic Aspects of Carcinoma of the Stomach and Breast, in Univ. of California Publications in Public Health. Vol. 2, p. 265. Berkeley, Univ. of California Press. Wynder, E. L., Bross, I. J., and Hirayama, T. A., 1960. A Study of the Epidemiology of Cancer of the Breast. Cancer 13: 559-601. Yerushalmy, J., and Palmer, C. E., 1959. On the Methodology of Investi- gations of Etiologic Factors in Chronic Diseases. J. Chron. Dis. 10: 27-40. PART II The Evaluation of Genetic Factors in Selected, Illustrative Diseases Diabetes Mellitus® By James V. NegL,> M.D., Ph. D., Steran S. Fasans?® M.D., Jerome W. Conn, M.D. and Rura T. Davipson,? B.S., University of Michigan Medical School, Ann Arbor, Mich. INTRODUCTION Diabetes mellitus is in many respects a geneticist’s nightmare. As a disease, it presents almost every impediment to a proper genetic study which can be recognized. Firstly, the nature of the basic defect is unknown, and as long as this is true, the possibility that the disease may be heterogeneous hovers overhead. Thus, in recent years it has been variously postu- lated that the primary defect in the majority of cases is an abnormal insulin molecule (Conn and Fajans, 1961), an excess of insulin antag- onists (Vallance-Owen and Lilley, 1961), an initial overproduction of insulin followed by excessive antagonist activity (Neel, 1962), or an intrinsically higher rate of release of fatty acids and ketone bodies for oxidation, with, as one consequence, an impaired sensitivity to insulin (Randle, Garland, Hales, and Newsholme, 1963). The further possi- bility has been considered that the primary defect has nothing to do with insulin at all, but results from an abnormality in the basement membrane of the small blood vessels (review in Siperstein, in press). Secondly, the frequency of this disorder in the general population is not well defined, a problem if one wishes to test hypotheses which draw on the concept of gene frequency. Thus, from household inter- views of the civilian, noninstitutionalized population, conducted in 1,900 sampling units throughout the United States, the frequency of 1 This investigation’ was supported in part by Grants A-888 and AM 04108-03 GEN from the National Institutes of Health. The participation in this study of Dr. Lee Schacht, Dr. T. Edward Reed, and Dr. Richard Post is gratefully acknowledged. ? Department of Human Genetics. 3 Department of Internal Medicine. 735-712 0—65——S8 105 recognized diabetes mellitus was estimated (rates per 100 persons) to be as follows (U.S. National Health Survey, 1960): Age Males Females 0-24 ____ 0.11 0.07 Doth cv wn mmmmime nam n mmm 0.49 0.38 CL I —— 1.12 1.37 55-64 oo ________ 2.52 3.15 65-74. eee ———————— 3.44 5.03 T+ eee 3.15 3.88 On the other hand, when screening-type glucose tolerance tests, followed by more extensive studies when indicated, have been per- formed, significantly higher figures have been obtained. For example, Chesrow and Bleger (1955), after exclusion of known diabetics, found 64 “definite” and 24 “potential” diabetics among 1,000 residents of an infirmary who were over 60 years of age. Allowance for known and/or deceased diabetics would be expected to bring the total cumulative rate to close to 10 percent. Brandt and Tupper (1960) have reported even higher figures for inmates of county hospitals. Obviously, these figures while useful, cannot be used for normative values. Nonspecific factors which may affect glucose tolerance, such as decreased mobility, increased bed rest, chronic illness, decreased food intake, and other factors, frequently operative in older people, may contribute to the high incidence of abnormal carbohydrate metabolism in these series. More to our needs, in a large scale survey in Georgia, 6.7 percent of persons aged 60-69 and 7.4 percent of persons over 70 were found to have significantly lowered glucose tolerance. Both Negro and Caucasian subjects were included; highest rates were encountered in Negro females and lowest in Negro males. These figures include known diabetics (Petrie, McLoughlin, and Hodgins, 1954). In a health appraisal program at the University of Michigan, among the first 548 subjects evaluated, 90 percent male, with an average age of 56.5 years, there were 10 or 1.8 percent with known diabetes but an additional 59 or 10.8 percent were thought to have previously unsus- pected diabetes (Rush and Tupper, 1960). Criteria for diagnosis in 42 of this latter group were a blood sugar above 160 mg percent at 1 hour, and above 120 mg percent at 2 hours. In 17 of the 59 subjects, a diagnosis of ‘“‘diabetes” was made solely on the basis of the 2-hour postprandial (100 gm glucose) blood sugar test. These examinations were made without the use of a preparatory diet and on the same day that other routine tests such as chest and gastrointestinal roentgeno- grams were carried out. In a comparison of the 2-hour postprandial blood sugar determined in this fashion with the standard oral glucose tolerance tests performed by the same authors in 110 patients, it was noted that the 2-hour value of the glucose tolerance tests was lower (mean 31 mg percent, range 1-117 mg percent) in approximately 75 106 percent of this group. Thus, the figure of 10.8 percent for “persons with unsuspected diabetes’ appears to be high. It should also be noted that the 10.8 percent figure was “derived from the examining physician’s final diagnosis of each patient concerned. No attempt was made in any of these cases to impose an absolutely standardized inter- pretation of the laboratory results.” Whatever the true incidence of diabetes may be, these findings sug- gest that a rather considerable number of persons with what we must at the very least term ‘abnormal carbohydrate tolerance” and at the most mild diabetes, will ordinarily live out their days without this fact becoming evident. The results of twin studies provide further evidence to this effect. Thus, Then Bergh (1938) studied 35 monozygous and 35 dizygous twin pairs age 43 or over, ascertained through the fact of diabetes in one member of the pair. Fifteen monozygous and two dizygous pairs were concordant. He was able to administer glucose tolerance tests to 11 of the apparently normal monozygous co-twins—all the tests were felt to be abnormal. A followup study would be of great value, to see for how many pairs concordance in the clinical sense ultimately developed. Be that as it may, it is apparent that the evolution of diabetes mellitus follows a wide variety of courses, with, even in the advanced years, many persons with the predisposition not having overt disease. Thirdly, and closely related to the foregoing, is the fact that it is by no means clear whether diabetes mellitus should be regarded as a distinct, qualitative departure from normal glucose metabolism, or whether, rather, the clinical entity is the ‘‘tail’’ of a continuous dis- tribution, with some individuals in this tail exhibiting secondary com- plications which contribute to the impression of a discrete disease. There has never been the extensive, detailed population survey which would permit an opinion on this point, although we shall later present some pertinent data. To date the geneticist, taking his clue from the clinician, has usually approached diabetes as if it were a discrete entity. A rather different approach is indicated if there is a con- tinuous, unimodal distribution to the basic defect(s). Fourthly, now, it is clear that the frequency with which the disease is diagnosed appears to be strongly influenced by environmental variables, especially the general nutritional level. For instance, the frequency of diabetes appears to differ among repatriate Jewish groups in Israel (Cohen, 1963). The ultimate strength accorded this finding as evidence for the action of environmental factors will be strongly influenced by information still unfolding regarding intermarriage in the countries of dispersion (discussion in Goldschmidt, 1963). A further uncertainty pertains to death rates from diabetes prior to repatriation. 107 Fifthly, the disease may become apparent at any age from 1 to 100. The geneticist at any given time is thus working with only a fraction of the possible diabetics, and must accordingly, if he wishes to think in terms of ratios, make adjustments for this fact, adjustments the more difficult because the published age-of-onset and prevalence curves are at variance with one another and the sexes are not equally affected by the disease (Spiegelman and Marks, 1946; Joslin ef al., 1959; U.S. National Health Survey, 1960). Sixthly, until recently the so-called juvenile diabetics seldom re- produced. Attempts to determine whether, the genetic basis for juvenile diabetes might differ from that of the adult type thus have been handicapped by a scarcity of certain mating types rather critical to any tests of hypotheses. Seventhly, and finally, there are well-known problems in the reproducibility of certain diagnostic procedures. Thus, when the position of individual points on the GTT is subject to statistical analysis, considerable variability is observed in the results of repeated tests. For instance, Hayner, Waterhouse, and Gordon (1963), in a series of observations on 28 prisoners on standard prison fare, with few exceptions in the fifth decade, have found the following absolute differences and standard deviations for corresponding values on glucose tolerance tests repeated one week apart: 0 hours—4.4+3.8 mg percent; % hour—16.84+15.2 mg percent; 1 hour—20.0 4+ 18.0 mg percent; 1% hours—14.14+12.5 mg percent; 2 hours—11.9+13.5 mg percent; and 3 hours—16.1+14.7 mg percent. In spite of the magni- tude of these differences and standard deviations between replicate observations at the various points of the two curves, it needs, for clinical and diagnostic purposes, to be emphasized that both members of the pair of glucose tolerance tests either fell well within the limits defined as normal (see below), or abnormal (1 case) for 27 of the 28 prisoners. The individual with the mildly abnormal glucose tolerance tests had a family history of diabetes mellitus. One individual in this series had one normal test and one showing an elevation of the 1-hour level of the glucose tolerance test. Why, you may wonder after this dreary recital, would any geneticist ever venture into this obviously genetically unprofitable arena? The answer is very simple. This is a relatively common and potentially fatal disorder in which there is much evidence for a familial pre- disposition. Not only is it a tempting target for the investigator, but its frequency raises some fascinating problems not encountered in the study of such rare metabolic disorders as galactosemia or albinism. Assuming a genetic basis, how did a disease with such an impact on reproduction reach such relatively high frequencies? Why hasn’t natural selection reduced the frequency of the diabetic geno- type(s) to that of most other serious diseases in whose etiology genetic factors seem of importance? There was a further, practical reason 108 for our venturing into this field some ten years ago, namely, the development of the cortisone-conditioned glucose tolerance test (Fajans and Conn, 1954), with the possibility of a very useful handle for the identification of a considerable proportion of individuals with the diabetic genotype long before the onset of clinical disease. REVIEW OF LITERATURE That diabetes mellitus occurs more frequently in some families than might be expected on the basis of chance alone has been recog- nized for centuries, although Allan (1933), Pincus and White (1933; 1934a, b), and Levit and Pessikova (1934) were apparently the first to place this impression on a sound statistical basis. Since these familial concentrations were observed within neighboring socio- economic levels, in the absence of any identifiable environmental factors, a genetic etiology seemed likely. The many studies of the past three decades designed to elucidate the most likely genetic basis for this nonrandom distribution have in recent years been reviewed by Grunnet (1957), Steinberg (1959, 1961), Lamy, Frézal, and Rey (1961), and Simpson (1962). At one time or another, such diverse simple genetic mechanisms have been suggested as irregular dominance, recessiveness, and a heterozygous-homozygous relation- ship, wherein the heterozygote tends to develop the mild diabetes of the later years and the homozygote the more severe diabetes of the earlier years. Each of these theories continues to have its proponents, but recently a number of investigators, impressed by the difficulty of explaining the facts with a single pair of genes, have begun to consider a more complex, multigenic type of inheritance. The fact that able investigators could reach such diverse and/or uncertain conclusions reflects the difficulties in the study of the disease which were enumerated earlier. The findings thought to suggest recessive inheritance fail to indicate whether the presumed recessive genes occur at only one or at several different loci. Now, if all cases of a disease are due to recessive inheritance involving the same locus, all the offspring of two affected parents should themselves be affected. No other reasonably simple genetic hypothesis leads to this expectation. Accordingly, a rather critical test of at least one of the hypotheses regarding diabetes would appear to be afforded by the findings in the offspring of conjugal diabetics. If only one locus is concerned, then all the offspring should eventually develop diabetes. If, however, more than one locus is concerned, or if heterozygotes for this “recessive” gene sometimes de- velop diabetes, then the offspring would not all be expected to develop diabetes. } 109 Post (1962) has recently presented an analysis of such critical material, based in part on the data of West (1960), and in part on pedigrees collected in connection with the study to be described shortly. He reported that in a total material of 161 offspring ranging in age from the first to the seventh decade, 67 from West’s data, and 94 from our own study files, there were 13 diabetics. Expectation for a group with this age distribution, if all were destined ultimately to develop diabetes, was 8.7. The difference between observation and expecta- tion was not statistically significant. In this observation, it is of the utmost importance that ascertain- ment be solely through the parents or that, if otherwise, less adequate allowance be made for this fact. In reviewing both the paper of West (1960) and the data utilized by Post (1962), in preparation for this discussion, we have been impressed that problems of ascertainment are still with us. Thus, in the work by West (1960), “all the subjects tested with a family history of diabetes came to our attention through their diabetic parents or their nondiabetic siblings.” The frequency of ascertainment of the latter type, through nondiabetic siblings, is not stated. To the extent that it occurred, it would tend to produce a bias in the number expected, in the direction of a relative deficiency of affected. Reexamination of the kindreds on which Post (1962) based his report also reveals some probable bias, in that in four sibships it seems very likely that ascertainment was significantly influenced by the fact that the conjugal diabetics had produced an affected offspring, i.e., that ascertainment was at least in part through the offspring. The simplest way to correct for this fact is to omit the offspring through whom ascertainment occurred (diabetic or normal) from consideration. Furthermore, one of the sibships was removed for diagnostic reasons. After these corrections, and with the addition of two new sibships, our own material contains 99 offspring of conjugal diabetics, of whom 6 were already diabetic. Combined with West’s material, with the reservations stated above, the total now becomes 166 offspring, 11 affected. Expectation in a group of this age structure on the hypoth- esis that eventually all will be affected is 6.6; X2(,) is 1.9 and the difference is not significant. I should emphasize that we have scored as diabetic in this accounting only those whose diabetes was known when the family came to attention, because this gives us diagnostic comparability with the figures of the surveys on which U.S. fre- quencies are based. The numbers involved are still quite small. Although the data are consistent with the hypothesis that all the offspring will in time be affected, they do not exclude the hypothesis that some of the children will remain normal. Thus, let us assume for the sake of the argument that these affected parents are actually all heterozygous for a gene 110 which can produce diabetes in either homozygote or heterozygote. * Then three-quarters of their children would be expected to develop diabetes but one-quarter should escape the disease. This would reduce expectation in this sample to 4.98 individuals. The difference between observation and expectation is still not significant (X)=3.5, .10>>P>>.05); we cannot exclude such a different hypothesis on the basis of the available data! Earlier we considered the possibility that a substantial proportion of those with the genetically determined predisposition to diabetes will fail to develop the disease during their lifetimes. If this is so, then even among the offspring of conjugal diabetics we might expect a considerable proportion of the children to remain clinically unaffected. This makes the earlier mentioned excess of affected over expectation among the offspring of conjugal diabetics, even though nonsignificant in these limited numbers, potentially somewhat puzzling. In this connection, we should recognize the possibility that our hospital sample is weighted for severity of diabetes, with a possible effect on the figures. PLAN OF PRESENT STUDY Some 10 years ago we initiated a study on the genetics of diabetes which was designed to take advantage of the possibility that the individual predisposed to diabetes might be recognizable years before the onset of clinical disease by his response to a standardized dose of cortisone. Standard glucose tolerance curves (GTT), the great majority repeated after the administration of cortisone, have now been studied in some 103 controls and 573 apparently healthy, first degree relatives of diabetic patients seen at the University of Michigan Hospital. The controls are for the most part medical students, physicians, technicians, and nurses. Blood sugars were determined by the Nelson modification of the Somogyi method. In this presenta- tion we propose first to utilize a portion of these data to test the hypothesis of simple recessive inheritance for diabetes mellitus, a hypothesis recently found by Steinberg (1959, 1961) to be “the only one consistent with the several large samples of family data which have been published.” Since the data do not fully conform to expecta- tion on this hypothesis, we will then explore alternative explanations of the findings. We are at the outset badly in need of some definitions and standards. Diabetes mellitus, as previously discussed (Jackson and Woolf, 1957; Fajans and Conn, 1959) and as we shall shortly demonstrate again, *The distribution of homozygotes and heterozygotes in populations being what it is, it is of course very unlikely all these parents would be heterozygotes; the assumption is made solely for illustrative purposes. 111 is not a disease of sudden onset, but one in which there have usually been distinct and progressive metabolic abnormalities for many years before the onset of overt disease. There is a wide range of response in the ‘normal’ glucose tolerance curve, and an equally wide range of variation in the impairment of glucose metabolism in the so-called diabetic. What kinds of definitions and boundaries can we set up for purposes of identification? In this presentation, in addition to ‘normal’ people, we propose to recognize four classes of people who may fit into stages in the natural history of diabetes. Frank, florid, overt, or symptomatic diabetes is the most advanced of these stages. This stage is represented by the individual whose impairment of glucose metabolism or “complications” of the disease have brought him to medical attention. The next term to be used is latent, or asymptomatic but clinically detectable, diabetes. A latent diabetic is an individual who has no symptoms referable to the disease but in whom a diagnosis of diabetes can be made either by the finding of an elevated fasting blood sugar level or with the use of the standard glucose tolerance test. To define latent diabetes, we must be able to define the normal. In table 1 are summarized by age-decade and sex the results of the glucose tolerance tests performed in the course of this study on 113 individuals between the ages of 10 and 39 whose family histories were negative for diabetes mellitus. It should be emphasized that these were not the usual cursory medical family histories but that detailed pedigrees were obtained on each individual! All subjects were on a 3-day ‘‘preparatory’’ diet containing 300 gm of carbohydrate per diem and were given a test meal consisting of 1.75 gm of glucose per kilogram of “ideal” body weight, calculated from the subject’s height and age. In the calculation of ideal body weight all ages beyond 25 were equated to 25, since according to life insurance tables “ideal” body weight stays constant for a given height after this age. For the two decades for which significant numbers are available, it will be noted that male values average slightly higher than female, although for the 20-29 age interval (for which the most data are available) only one of the differences, at 2 hours, is significant at the 5 percent level. Incidentally, in this and other similar com- parisons to follow it must be borne in mind that inasmuch as any individual glucose level at one time point is not independent of the level at the next or preceding time point, these ¢ tests are not inde- pendent of one another. There is no clear age trend in this age interval. Since the sex difference is small in relation to the standard deviation, we have felt justified in deriving one composite curve for age 10-29 (average age 24.7 years), sexes combined, to be used in certain comparisons to follow; this curve is given as the last entry in the table. It is to be noted that the curves for ages 30 to 39 are the same or lower. Fajans and Conn (1954) and Conn and Fajans 112 (1961) have previously defined (in part on an experience with this same material) a diabetic glucose tolerance curve as one exhibiting a one-hour value of 160 mgm/100 ml or more, a 1%-hour value of 140 mgm/100 ml or more, and a 2-hour value of 120 mgm/100 ml or more. It will be seen that these values fall two standard deviations above the control curve, so that the definition appears suitably conservative for this age group. By contrast, a normal glucose tolerance curve was defined as one showing a one-hour value of less than 160 mgm/100 ml, a 1%-hour value of less than 135 mgm/100 ml, and a 2-hour value of less than 110 mgm/100 ml. In this connection, it must be emphasized that some of our controls may of course later develop diabetes despite the negative family histories; the control curve is thus an approximation. Be this as it may, we define a “latent” diabetic as an individual whose glucose tolerance curve is abnormal as defined above, and a ‘normal’ individual as one whose curve falls within the defined limits. It is clear that the dividing line between “latent” and “overt” diabetes is arbitrary. TaBLE 1.— Mean glucose tolerance tests by decade and sex, in individuals aged 10-39 with a negative family history for diabetes. The significance of all differences shown has been examined by a t-test, with one, two, and three asterisks indicating significance at the .05, .01, and .001 levels respectively. Variations in the numbers for the various test points in this and subsequent tables result from the fact that complete curves were not obtained on all individuals Hours after test meal Age Sex Value 0 1% 1 14 2 2% 3 10-19 M x 3 111 70 7 76 1 1 1 20-29 M f4or | 82.340.9 (130.5+3.1 [106.44+3.7 | 91.7+3.0 | 90.1+1.9 | 87.2+41.9 | 85.242.2 a 6.2 21.6 26.9 20.3 13.7 13.7 16.0 N 53 49 53 47 50 52 30-39 M %4o03 | 84.5+1.8 [127.14+4.6 [104.34+7.1 | 84.7+5.1 | 88.5+4.9 | 90.443.1 | 79.9+4.7 a 7.0 17.9 27.6 19.8 19.1 11.8 18.1 N 15 15 15 15 15 13 15 10-19 F x 79 116.3 100 88.3 101 85 7.9 N 3 3 3 3 3 3 3 20-29 F %4-03 | 80.340.9 (125.8+4-3.8 | 98.93.90 | 88.14+3.1 | 84.3+2.2 | 83.342.6 | 80.4+2.8 a 5.3 2.2 23.3 18.4 12.8 15.0 16.4 N 35 35 35 35 35 34 34 30-39 F fox | 79.7£2.2 120 2477.0 93.746.9 | 76.8+8.8 | 78.1+5.0 | 83.04+8.4 | 83.74+6.5 a 5.3 7 16.8 19.6 12.1 18.8 16.0 N 6 5 5 6 5 “9 20-29 | M-F Diff. 2.0 4.7 7.5 3.6 *5.8 3.9 4.8 10-29 | M,Ft f+o; | 81.5+0.6 [127.942.4 (103.042.7 | 90.1+2.1 | 88.0+1.4 | 85.6+1.5 | 83.1+1.7 a 5.9 22.8 25.6 19.5 13.7 14.3 15.9 N 92 88 92 85 92 87 90 {Mean age—24.7 years. The next term to be defined is “abnormal cortisone-GTT,” which may indicate the presence of “subclinical diabetes.” In table 2 are summarized by decade and sex the findings of the cortisone-glucose tolerance tests of 101 nondiabetic individuals between the ages of 10 and 39 with negative family histories for diabetes. The test was administered as follows: after completion of the baseline glucose tolerance tests the subjects continued on the standard prep diet for lunch and dinner of that day. A second glucose test was administered 113 the following morning. At 8% and again at 2 hours before that test was scheduled, the subject received cortisone acetate orally. If he weighed less than 160 lbs the dose was 50 mgm both times; if he weighed 160 lbs or more, the dose was 62.5. It should be emphasized that the dose of cortisone is arbitrary; a variety of other dose-time schedules could be employed but of course each would require a different definition of an ‘‘abnormal’’ response. It must also be stressed that the cortisone-glucose tolerance test should still be regarded as an investigative procedure and is not presently recom- mended for routine clinical use. TABLE 2.— Mean cortisone-glucose tolerance tests by decade and sex, in individuals aged 10-39 with a negative family history for diabetes. The significance of all differences shown has been examined by a t-test, with one, two, and three asterisks indicating significance at the .05, .01, and .001 levels respectively. Variations in the numbers for the various test points in this and subsequent tables result from the fact that complete curves were not obtained on all individuals Hours after test meal Age Sex Value 0 1% 1 1% 2 21 3 10-19 M x 87 150 120 79 68 N 1 1 1 1 1 20-29 M X+40x | 91.0+0.8 (152.02. 5 [139.043.6 [115.04:3.1 [107.242.4 | 99.141.9 | 92.942.7 o 5.7 17.4 25.8 21.1 16.9 13.5 19.0 N 50 45 51 46 51 50 50 30-39 M X4or | 94.4+2.9 (151.745.7 (137. 7411.4|114. 1430. 0({105. 1+4.0 | 96.54-4.4 | 90.1+4.7 ao 11.2 21.9 44.3 30.0 15.5 16.9 18.1 N 15 15 15 15 15 15 15 10-19 F Xx 84.3 139.7 110.0 104.7 84.7 80.7 87.3 N 3 3 3 3 3 3 3 20-29 F X4ox | 92.6+1.3 (154.24+4.9 (144.07.3 [119.246.1 | 107.943.8| 99.743.0 | 94.942. 9 o 7.3 26.2 38.8 33.1 20.3 15.8 15.4 N 29 29 29 29 29 28 29 30-39 F X+ox 93.0 173.0 152.5 136.0 132.5 102.5 89.0 N 2 2 2 2 2 2 2 20-29 | M-F Diff. -1.6 -2.2 —=5.0 —4.2 -0.7 —0.6 -2.0 10-29 | M,F X+ox | 91.3+0.7 (152.32. 5 (139. 5+3.4 |116.2+3.0 |106.34+2.0 | 99.541.6 | 93.14+2.0 0 6.5 21.6 31.3 26.0 18.5 14.4 18.0 N 83 78 84 78 84 81 As before, based on the most numerous age group (20-29) there is a suggestion of a sex difference, of small magnitude, and not significant for any of the “point” comparisons; but now it is the females who exhibit the higher values. There are only two age-sex groups in which the numbers are adequate to search for age trends—none are apparent. Accordingly, and again as a basis for comparisons to be described shortly, we have felt justified in deriving a single composite curve for both sexes, ages 10-29; this is the last entry in table 2. The combined curve for ages 30-39 is approximately the same as that given for ages 10-29. Comparison of this entry with the corre- sponding entry of table 1 illustrates the well-known effect of cortisone on glucose tolerance; variability, as measured by the standard devia- tion, is not significantly increased, but yet six of the seven standard deviations are larger after cortisone. Fajans and Conn (1954, 1961) have defined an abnormal response to cortisone as a 2-hour value in 114 excess of 140 mgm/100 ml; table 2 indicates that this is a properly conservative standard. The last term to be defined is prediabetes. The prediabetic is an individual who has a (genetic) predisposition to the disease but in whom no diagnostic test in use today discloses an abnormality. Thus by this definition a prediabetic has a normal glucose tolerance test as well as a normal cortisone-glucose tolerance test. As is apparent from the magnitude of the standard deviations, there is great individual variation in the glucose tolerance curve. How much of this is “random,” of the type discussed in the opening section, and how much real, is not clear. Now, it is a matter of some interest, both for the results obtained in the controls and in the offspring of diabetics to be considered shortly, to determine whether there is a correlation between corresponding points on the standard and cortisone glucose tolerance curves of any given individual. This question is investigated in table 3. The correlations, of the order of .35-.40, are generally significant, but, viewed otherwise, under the conditions of these observations, only approximately 15 percent of the variation in the position of the cortisone tolerance curve at any one time would be removed by knowledge of the corresponding position on the standard test. While it seems quite likely that the true correlation is somewhat higher, being obscured by random noise in the system, the fact is that no investigator has published the type of observations necessary to validate that statement, namely, repetition within a short interval of both types of curves 3-5 times in a series of some 20 individuals. TaBLE 3.— Correlations between corresponding points on the glucose tolerance curve and the cortisone-glucose tolerance curve, in individuals of age-sex composition and genetic background as specified in the table. One asterisk indicates signif- icance at the .06 level and two at the .01 level Correlations between results of Sisndad-GT?e and cortisone-GTT at: Genetic background | Age and sex 0 hours 1 hour 2 hours 3 hours N r+tor N r+or N ror N rtor Controls. ...ccccauus 10-29, M, F___| 83 | **0.316 84 | **0.355 84 | **0.338 81 **0. 489 Offspring of con- 10-20, M, F___ 20 0.389 20 *0. 544 20 *0. 553 18 0.354 jugal diabetic arents. Offspring of one 10-29, M, F___| 107 | **0.373 | 106 | **0.490 | 105 | **0.334 | 101 **0.533 diabetic and one normal parent. The age span covered by these controls is unfortunately limited. However, through the kindness of Dr. C. J. Tupper, we have had access to the recent results of the periodic health appraisal of the senior staff of the University of Michigan. As one aspect of this evaluation, each individual received a screening, two point (0 and 2 hour) glucose tolerance test. In recent years (subsequent to the ex- 115 perience described on page 106), the patients have been asked to adhere to a preparatory diet. After the fasting blood specimen had been drawn, a standard test meal of 100 gm of glucose in lemon-flavored water was administered. The findings regarding the 2-hour blood sugar in those individuals with a negative family history for diabetes mellitus are given in table 4. Individuals thought to have abnor- mally high values at the 2-hour determination subsequently received the standard GTT, preceded by a prep diet. At the time of the evaluation, a medical history was obtained from each person by an internist. TABLE 4.— The blood sugar level 2 hours after 100 gms of glucose in those University of Michigan employees undergoing a routine health appraisal who gave a negative family hastory for diabetes mellitus. Subjects were predominantly of the male sex Age 20-29 30-39 40-49 50-59 60 X+o% 80.5+3.9 86.1+1.3 91.142.1 94.44+4.6 107.245. 6 a 18.0 20.7 22.1 23.9 26.7 N 21 267 107 27 23 No.>120 mgm %, 0 18(6.7) 9(8.4) 3(11.1) 9(39.1) It is apparent, first, that the values for the 20-29 and 30-39 age groups are actually somewhat lower than our own more carefully screened control series, being significantly different at the 1 percent level for the 20-29 group. This is most likely due to the fact that on a per kilogram basis, the glucose load in our procedure was greater. Secondly, there is a very clear tendency for an increase in both the mean and standard deviation with age. The consequence of this is of course that with age an increasing proportion of individuals exceed the value of 120 mgm percent at 2 hours, one of the three cut-off points earlier used for the delineation of diabetes in a sample age 10-29. The number and percent exceeding this level at each age interval are also given in table 4. While the numbers for the older age intervals are unfortunately limited, the fact that in this selected series 24 per- cent of the 50 men age 50 or above had blood sugar levels about 120 mgm percent at 2-hours is a rather impressive finding. While a cer- tain amount of undetected diabetes is not surprising at this age level, the magnitude of this figure, taken in conjunction with the other “control” figures presented earlier, not only raises some questions concerning the diagnostic standards for diabetes in the elderly, but, as we shall see, some especially troublesome issues for the interpreta- tion of family studies. Using a different testing procedure (cortisone glucose tolerance test) we have previously suggested that diagnostic criteria may have to be modified with advancing age (Conn and Fajans, 1961). For clinical purposes, it needs to be pointed out that although the 2-hour post-glucose blood sugar level is an excellent screening procedure, it cannot be employed as a diagnostic procedure 116 for diabetes mellitus if only slightly elevated. To be diagnostic it must be preceded by elevation of the blood sugar level at 1 and 1% hours as previously defined (Fajans and Conn, 1961). Preliminary analysis of data.—In the preliminary analysis of the data to be presented three points of importance to the final form of the analysis were established. Firstly, individuals with an abnormal cortisone-4T'T have a much greater likelihood of developing diabetes than those whose test is normal. Thus, Conn and Fajans (1961) have reported that of 128 nondiabetic relatives of diabetic patients to whom cortisone-GTT’s were administered and who have been followed from 1 to 7 years, 57 were judged abnormal at the time of the initial test. Subsequently diabetes developed in 15 of these 57. Of the 71 subjects with negative or normal responses, diabetes has developed in only 2. It is thus apparent that the test identifies a predisposed group, although, on the one hand, the issue of whether all positive responders will develop diabetes remains open, and, on the other hand, some predisposed may have negative tests only a relatively short time before the onset of clinical disease. A second point already established by these data is that the per- centage of positive responders to the cortisone-GTT increases with age (Conn and Fajans, 1961). Thus, of 78 relatives of diabetics who were themselves nondiabetic and under the age of 18, only 6 gave positive responses to the cortisone test, whereas of 25 such relatives over 48 years of age, 12 were found to manifest a positive response. It is apparent that a positive response to the test cannot possibly be an innate indicator of genetic susceptibility to diabetes, but, to the extent that it identifies the predisposed, an indicator of a phase in the evolution of the disease. A third and final point established by the preliminary analysis is that in the predisposed—notably the offspring of marriages in which both partners had clinical diabetes mellitus—the mean glucose toler- ance curve exhibits a significant deviation from that of the control group from an early age onwards (Neel, in press). Table 5 describes the mean glucose tolerance test results before and after cortisone in 21 offspring of conjugal diabetics (8 males, 13 females), aged 10-29 (average age of 22.7- years), where ascertainment was through an affected parent. Results have been pooled in this fashion because of the small numbers. The actual distribution of the values obtained at 1, 1%, and 2 hours is shown in Fig. 1. In addition to the 21 offspring tested, there were in these families 8 other children in this age group who were not tested, none of whom were reported to have clinical diabetes. As before, correlation coefficients have been computed for four of the seven reference points before and after cortisone. These correlations are given in table 3. Again we see moderate correlations, possibly a little greater than seen in the controls, although the small numbers available make positive conclusions impossible. 117 13° VALUE DxD NO CORTISONE odndhmbll gun a 50 100 150 200 250 154 DxD CORTISONE 0 4 adem elm dx 50 100 150 200 250 15 Dx N, 44 NO CORTISONE 0 iB 50 100 150 200 250 15 Dx N,¢?¢ NO CORTISONE 1° VALUE 15 DxD NO CORTISONE Ne 0 50 100 150 200 250 mgm % GLUCOSE 15 DxD CORTISONE 0 50 100 150 200 250 15 Dx N, 44 NO CORTISONE i | 50 100 150 200 250 15 DxN, #¢ NO CORTISONE 50 100 150 200 250 1° VALUE 15 Dx N, ¢¢ CORTISONE 50 100 150 200 250 mgm % GLUCOSE 15 Dx N, ¢¢% CORTISONE . ulbsa . 50 100 150 200 250 5 Nx N NO CORTISONE 0 ry 50 100 150 200 250 15 Nx N CORTISONE 0 FT 50 100 150 200 50 100 150 200 250 14° VALUE 15 Dx N, #4 CORTISONE 2° VALUE 15 DxD NO CORTISONE — 50 100 150 200 250 15 DxD CORTISONE 0 50 100 150 200 250 15 DxN, gs NO CORTISONE — T 50 100 150 200 250 15 D x N, ¢¢ NO CORTISONE 50 100 50 200 250 2° VALUE 15 DxN, &¢ CORTISONE 50 100 150 200 250 15 DxN, #¢ ] CORTISONE 0 1 dalam - 50 100 150 20! 250 5 Nx N NO CORTISONE 0 3 yre—y 50 100 150 200 250 15 Nx N CORTISONE 0 fA) 50 100 150 200 250 50 100 150 200 250 15 DxN, #¢ CORTISONE ly 50 100 150 200 250 15 Nx N NO CORTISONE 0 ppre——y ————y 50 100 150 200 250 15 Nx N CORTISONE 0 2 50 100 150 200 250 Ficures 1 a and b.—The distribution of the 1-, 1%-, and 2-hour blood glucose values on a standard- and cortisone-glucose tolerance test, in three classes of individuals: Offspring of conjugal diabetics, offspring of one diabetic and one normal parent (ascertained through diabetic parent), and controls. All in- dividuals age 10-29. Further explanation in text. Figs. 2 and 3 contrast the mean curves in this group with those Table 6 gives a point-by-point comparison observed in the controls. of the two curves. 118 The D X D offspring consistently yield higher TABLE 5.— The results of standard and cortisone-GTT’s in 21 offspring of conjugal Mean age of the subjects was 22.7 years diabetics of both sexes and aged 10-29. Hours after test meal Type of | Value & test 0 14 1 14 2 2% 3 Standard.| X+4ox | 86.5+1.6 | 132.0-:5.3 | 132.1£7.6 | 113.048. 2 | 103.245. 9 Z) 24.3 34.7 37.5 27.0 101. 0+6. 2 28.5 03.84-4. 4 20.3 N 21 21 21 21 21 21 21 Cortisone.| Xzox | 99.042.2 | 160.14+5.7 | 164.4-£10.1 | 148.24+10.6 | 128. 68.4 | 120.8+7.3 | 109.246. 2 aT 10.0 25.6 45.0 46.1 37.7 319 26.4 N 20 20 20 19, 19 18 170 —— CONTROLS A —«=e= OFFSPRING of Dx N ” 160 ~~ \ ——=— OFFSPRING of Dx D 150 2 € 140 on € yl oO 130 Oo > = oO S 120 o a o 110 100 90 1 1 1 1 1 — I | all 0 7 1 13 2 23 3 TIME AFTER TEST MEAL Figure 2.—The mean standard glucose tolerance curve in three classes of in- dividuals: controls, offspring of one diabetic and one normal parent (ascertained All individuals through diabetic parent), and offspring of conjugal diabetics. age 10-29. Further explanation in text. 119 140 = CONTROLS i === OFFSPRING of Dx N HA ——— F _ i \ AY OFFSPRING of D x D * 20 E on £ . 110 w wn Oo O 3 100 O oO oO Oo @ 90 80 F 70 x 1 1 1 1 J I I 2r 0 1 | I+ 2 24 3 TIME AFTER TEST MEAL Ficure 3.—The mean cortisone-glucose tolerance curve in three classes of in- dividuals: controls, offspring of one diabetic and one normal parent (ascertained through diabetic parent), and offspring of conjugal diabetics. All individuals age 10-29. Further explanation in text. TABLE 6.— Differences between corresponding points on the GTT for males and females age 10-29 with genetic backgrounds as indicated. A ‘single asterisk indicates significance at the 5 percent level, a double asterisk significance at the I percent level, and a triple asterisk at the 0.1 percent level, on the basis of a t-test Difference at: Type of comparison | Type of test 0 hour 1% 1hour |1}4hours| 2 hours [2% hours |3 hours hour Controls vs. off- Standard....__ 45.0 4.1) *7°20.1 | ***22.9 | **15.2| "5.4 **10.7 spring of conjugal | Cortisone._____ 7.97 7.8 $125.9 | ***32.0 | **22.3| ***21.3 | *'16.1 diabetics. Controls vs. off- Standard._...__ 1345.4 7.11 **16.8| **14.3 sand | *tL5 7.3 spring of D X N Cortisone__.._ 5.9 4.9 10.4 77.1 **13.5 | ***15.1 | ***14.9 marriages. Offspring of conjugal —0.4| —0.3 12.3 8.1 58 4.0 3.5 diabetics vs off- 1.9 2.9 15.5 14.9 8.8 6.2 1.2 spring of D X N. 120 mean values than the controls. However, because of the larger standard deviations for the mean values of the D X D offspring, the findings in this latter group overlap almost completely the per- formance of the control individuals. F tests reveal that the variance of the D X D offspring is significantly greater than that of the controls for 10 of the 14 possible comparisons. Under these circumstances, t tests comparing mean values on the glucose tolerance curves of controls and D X D offspring must be applied and interpreted with caution. These t tests reveal apparently significant differences in the means for six of the seven points for both the standard and corti- sone tests. This point-by-point test of course fails to evaluate the significance of the consistency of the difference in the two curves. An important question is whether the lessened mean glucose toler- ance in the offspring of conjugal diabetics is due to the contribution of a relatively few highly aberrant individuals or rather results from smaller contributions from many individuals. The question cannot be answered with certainty from these limited data. However, if the present data are taken at face value, and in conjunction with the age-of-onset data for clinical diabetes, one can suspect that a sub- stantial proportion of potential diabetics exhibit minor deviations in glucose tolerance from “normal” for many years prior to the onset of clinical disease. The significance of these minor abnormalities of glucose tolerance has been discussed by ourselves (Fajans and Conn, 1959) as well as other investigators (Jackson and Woolf, 1957) previously. An important point is that in proportion to the baseline values, the composite curves for the controls and the offspring of conjugal diabetics do not move further apart after the administration of cortisone. On the average, then, cortisone does not appear to improve the discrimination between the mean curves of the predisposed and nonpredisposed at this age level. Were the correlations between corresponding individual values high, little would be added to the information on any given individual by a cortisone tolerance test. However, the fact that the correlations are relatively low indicates that the persons found in a particular quartile of the distribution of values for a point on the curve on the standard test are not necessarily the same persons in this quartile for the corresponding distribution after cortisone. Otherwise stated, since at this age level the incidence of both clinical as well as “subclinical” diabetes is relatively low even in the predisposed, a comparison of mean responses of the predisposed and nonpredisposed may hide a few important individuals in the former group in whom the cortisone-glucose tolerance test may make an important distinction. 121 735-712 0—65——9 TEST OF HYPOTHESIS We come at last to our test of hypothesis. We have derived, on the one hand, a mean GTT, with and without cortisone, for a group of young normal individuals, and similar curves for a group whose expectation of developing the disease would seem to approach 100 percent. We have also established standards for a qualitative judg- ment of the characteristics of a GTT. We now propose three ap- proaches to the question of the genetic basis of diabetes mellitus, two quantitative and one qualitative. On the quantitative side, the mean type of glucose tolerance curve encountered in individuals age 10-29 who are the offspring of diabetic X normal (D X N) marriages will be examined, to determine whether the characteristics of this curve are consistent with the concept that all diabetes mellitus is due to homozygosity for the same recessive gene, i.e., of genotype dd. This examination will consist of 1) a study of the mean curve and its position relative to the two curves already established, to determine the proportionate representation in the offspring of D X N marriages of potentially diabetic and normal children; and 2) an examination of the distribution of individual values at specified reference points in the GTT, in an effort to distinguish between unimodality and bimodality. Then, on the qualitative side, 3) data will be presented in which the types of GTT encountered in older (40-59) offspring of D X N matings have been scored according to the previously estab- lished definitions. It will be noted that we do not propose to use data on the 30-39 age group. This is felt to be a “transitional” group, which neither lends itself to the quantitative approach appropriate to the younger off- spring nor the qualitative approach appropriate to the older. In the past, investigators have tended to consider all the children of a given mating type together. It is our contention that by concentrating on the age extremes, and applying an appropriate (different) technique to each extreme, the picture may be clarified. In tables 7 and 8 are given the mean glucose tolerance curves, before and after cortisone, of 115 individuals aged 10-29 who are the offspring of diabetic X normal marriages, ascertained through the diabetic parent. The data are subdivided according to the decade and sex of the subjects. Excluded from the calculation are the results in eight individuals found grossly diabetic at the time of the glucose tolerance test. The findings in these eight persons are given in table 9. Although excluded from the derivation of the curves, they will enter into the later calculations. The rationale for exclusion is that at this level of abnormality second order effects begin to appear which render these individuals not in pari materia with the others. In addition, there were 3 siblings in this age range known to be diabetic 122 TABLE 7.—Mean glucose tolerance curves, by decade and sex, of individuals with one diabetic and one normal parent, ascertained through the diabetic parent. The significance of all differences shown has been examined by a t-test, with one, two, and three asterisks indicating significance at the .05, .01, and .001 levels respectively Hours after test meal Age Sex Value 0 1% 1 1% 2 244 3 10-19 X4ox | 90.341.5 (141.444. 2 |120.245.3 |104. 845.4 [102.94+4.0 | 95. 143.2 | 90.643. 8 o 8.4 24.0 30.5 30. 4 22.7 17.9 20.7 N 33 33 33 31 32 31 30 20-29 X+ox | 87.0+1.9 (143.8+5.4 (132. 545.6 (111. 745.2 | 98.7+4.9 [103.8+4.7 | 91.245. 4 o 8.3 24.3 25.2 23.5 21.9 21.2 23.9 N 20 20 20 20 20 20 20 10-19 F X4o% | 87.6+1.4 (130.3+5.9 (117.546.9 |104.34-6.4 | 97.14+4.0 | 98.3+4.8 | 94. 65.0 a 7.0 29.4 35.2 31.9 20.2 23.4 23.7 N 26 25 26 25 26 24 23 20-29 F X4oz | 83.3+1.2 [127.244.0 [113.9+5.5 (101.6-+4.0 | 92.243.3 | 94.4+3.4 | 86.942. 5 o 7.4 24.0 32.8 23.5 19.8 20.3 14.7 N 36 36 35 36 36 36 36 20-29 | M-F Diff. 3.7 16. 6* 18.6* 10.1 6.5 9.4 4.3 10-29 | M, Ft X+4ox | 86.940.8 (134.942. 4 (119.84+3.0 (104.9+2.6 | 97.442.0 | 97.1+2.0 | 90.3+1.9 a 8.2 26.0 31.8 27. 4 21.3 20.6 20.2 N 115 114 114 112 114 111 109 {Mean age of the subjects was 19.8 years. TABLE 8.—Mean cortisone-glucose tolerance curves, by decade and sex, of individuals with one diabetic and one normal parent, ascertained through the diabetic parent. The significance of all differences shown has been examined by a t-test, with one, two, and three asterisks indicating significance at the .05, .01, and .001 levels respectively Hours after test meal Age Sex Value 0 1 1 1% 2 2% 3 10-19 X4ox | 95.0+1.4 [150.9-+4.6 (124. 7+5.3 |111. 53.7 (105.8-£3.6 [100.5-2.9 | 98. 044.5 a 8.1 26.3 30.0 21.1 20. 4 16.3 24.5 N 32 31 32 31 32 31 30 20-29 X4ox | 94.2:+2.4 |155.346.1 (158.74+8.5 [136.4-+9.5 |115.4-£7.5 |114. 448.2 |105.2+5. 4 o 10.2 25.8 35.9 40. 4 31.0 34.6 23.0 N 18 18 18 18 17 18 18 10-19 F X+ox [100.942.8 (156. 747.8 (156. 010. 8/140. 69.9 (130. 749.6 [125.1+9.0 (115. 710.5 o 14.3 39.2 55.1 50. 6 48.7 46.1 53.7 N 26 25 26 26 26 26 26 20-29 F X+ox | 97.9:£1.8 (165.0-£6.2 |165. 49.8 [147. 149.0 |127. 546.8 [120.3+5.7 |113.0+4.5 a 10.1 34.4 54.4 49.8 37.6 31.9 24.9 N 31 31 31 31 31 31 30 20-29 | M-F Diff. -3.7 —9.7 —6.7 —10.7 —12.2 —5.9 -17.8 10-29 | M, F X+o3 | 97.141.1 |157.243.1 (149.844. 6 [133.3+4.3 |119.8+3.6 (114.7+3.3 (108.043. 4 a 10.9 32.2 48.0 43.8 36.7 34.1 34.4 N 107 106 107 106 106 106 104 TABLE 9.— The findings in 7 individuals with grossly abnormal, standard GTT’s, and for this reason excluded from the calculations of tables 7 and 8 Hours after test meal Individual Age Sex 0 1% 1 114 2 214 3 MMBI70....cvnviimunmmiiond 1 | M 272 368 437 456 452 473 444 GW-5577_ 20 | M 147 264 337 357 355 287 229 TW-5577 27 | M 210 328 405 455 451 435 392 MW-5577_ 25 | M 113 236 274 267 259 216 175 DW-5577_ 1 | M 89 175 209 221 227 210 180 TH-8156_. 2M 124 190 253 193 166 198 188 TH-8156 17 | F 153 238 288 323 323 316 312 PT-8232__ 28 | F 114 181 243 253 238 227 184 and not tested, and 35 who were not known to be diabetic who could not be tested. Again, before deriving a composite curve we must examine the data for possible age and sex differences. Firstly, for the 20-29 age group, males (as was true for the controls) exhibit higher values than females, greater in absolute value than in the control material, but significant at the 5 percent level for only two of the seven com- parisons. Again (as observed in the controls) we note the reversal of the sign of the difference after cortisone; again none of the differences is significant. By this criterion, then, the cortisone test reduces the difference between sexes. Secondly, inspection of the data again raises the possibility of age effects. All seven mean values in the standard curve of the 20-29 female group are lower than in the 10-19 group, although none of these differences is significant. More striking is the difference between the cortisone-GTT’s of 10-19 and 20-29 year old males, significant for the 1- and 1%-hour points. Study of the individual curves suggests that this age effect is due primarily to a disproportionately low response of the 10-19 M group to cortisone, as contrasted to the 20-29 M group, or to the controls. The accentu- ated response of females to cortisone may be related to a potentiation of the biological activity of this corticosteroid by estrogens (Nelson, Tanney, Mestman, Gieschen, and Wilson, 1963). In the strict sense, the data do not permit combining the various age and sex groups. Since, however, as we shall see, we view this analysis as very approximate, we have in tables 7 and 8 for subsequent purposes combined the data for both sexes and both age intervals into single curves, as was done for the controls (tables 1 and 2) and the offspring of conjugal diabetics (table 5). Table 6 reveals that most points on the resulting two curves differ significantly from the corresponding points on the two control curves. On the other hand, the corresponding differences from the curves based on the offspring of conjugal diabetics are not significant, and, in fact, the D X N offspring exceed the D X D offspring for the first two points. We are now ready for the first approximate test of hypothesis. If the marriage of diabetics and normals produces two types of indi- viduals, we might examine the data for evidence for this at those time intervals in the GT'T which provide the best test for the ability of an individual to return his blood sugar to normal levels. These are probably the 1-, 1%-, and 2-hour samples. In figure 1 are shown the findings, for these time points, in the following groups: 1 and 2) offspring of D X D, 10-29, sexes combined, before and after cortisone; 3-6) offspring of D X N, 10-29, sexes separately (because of the more marked sex difference), before and after cortisone; 7 and 8) offspring of N X N, 10-29, sexes combined, before and after cortisone. By presenting the sexes separately one avoids the danger of obscuring a bimodality because the curves being combined are somewhat out 124 of phase. The greater variability in the offspring of conjugal dia- betics than in the controls, already expressed in the standard deviations of tables 1,2, and 5, is evident. The individual blood sugar determi- nations suggest that at this age level, some of the offspring of conjugal diabetics are still essentially normal in their glucose metabolism, with all variations between this state and florid diabetes. Under these circumstances, one would not expect to demonstrate two clearly modal values among the offspring of diabetic X normal marriages, and indeed only one mode is apparent. This approach, then, has failed to provide a satisfactory test of a one-recessive-gene hypothesis. The second approach concerns itself with a comparison of the mean values observed in the three types of matings. If diabetes is inherited as a recessive trait, then one can readily calculate the proportion of affected offspring to be expected from a sample of D X N marriages, provided there are data on 1) the frequency in the population of the assumed homozygous recessive genotype and 2) the relative frequency among the phenotypically normal parents of DD, Dd, and dd, i.e., of homozygous normal, carriers, and phenotypically normal individuals who may be prediabetics or subclinical or latent diabetics. Unfortu- nately, as we have seen, neither point 1) nor 2) can be established with any degree of accuracy. A “middle-of-the-road” calculation would pro- ceed as follows: Let us assume a 0.10 frequency at birth of the diabetic predisposition, i.e., dd, which implies a frequency of the d gene of 0.317. Let us further assume that among the partners of the D X N marriages, 0.50 of those predisposed have already developed diabetes. The frequency of DD, Dd, and dd among the phenotypically normal ] 0.683% 2(0.683)(0.317) parents then becomes 1.00—0.05 4910, ~To00—0.05 4559, and 0.10—0.05 , ; 1.00=0.05 0526: From the Dd X dd marriages half of the children will be dd; assuming equal fertility, this will amount to 0.23 of all children from D X N marriages. From the dd X dd marriages all the children will be involved; again assuming equal fertility this will amount to 0.05. Total expectation of dd offspring from D X N mar- riages is thus 0.28. If we prefer to assume a frequency of the hypo- thetical dd genotype as low as 0.07, and that none of the phenotypically normal partners in the D X N cross are potential diabetics, then the proportion of affected to be expected is approximately 0.21, whereas if we assume a frequency of the predisposition of 0.20 and that the phenotypically normal parents include the diabetic genotype in the proportion of potential diabetics present in the population at birth, then the figure is approximately 0.45. Quite obviously, until a better estimate is available of the frequency of the diabetic predisposi- tion (genotype) in the population, tests of hypothesis can only be approximate. 125 Figures 2 and 3 depict the position of the mean glucose tolerance curve of the D X N offspring age 10-29 (average age 19.8 years) relative to the controls and the offspring of conjugal diabetics. From the tables thus far presented, we can, by utilizing corresponding time points (1, 1%, and 2 hours) on the appropriate curves, readily calculate the number and proportion in which ultimately diabetes would be expected to develop among the offspring of D X N marriages, assuming the potential diabetics behave as do the-offspring of D X D marriages, and the normals as the offspring of N X N marriages. For example, for the standard GTT at 1°, 1g IMT) 1108, where y is the number of potential diabetics; y=65.8 and P,=0.58. The results of these six calculations, which have an ill-defined but relatively enormous error, are given in table 10. The (unweighted) average proportion of potential diabetics among 115 persons tested by one or both types of GTT at six different time points is estimated to be 0.57. In addition, these 115 persons had, within that 10-29 year interval, 3 known diabetic siblings, 8 with newly recognized latent diabetes (see table 9), and 35 untested sibs. Assuming that half of the untested 35 would be found to be abnormal, then the proportion of individuals with the diabetic predisposition in this group would be estimated at Oates EE 0.58. This is ap- proximately twice as high a proportion as our gene frequency analysis led us to expect on the assumption of simple recessive inheritance and a frequency of 0.10 for the diabetic predisposition. Only by assigning frequencies to the homozygous recessive genotype of the order of 0.20-0.25 can we reconcile the findings with simple recessive inherit- ance. However, it must be borne in mind how rough this approach really is, as well as the indeterminate but undoubtedly very large error. For instance, in addition to the sampling errors, differences in the age and sex composition of the three samples used in the estimate might introduce errors in the calculation. In this connection, it should be noted that the mean ages of the control subjects, offspring of D X D, and offspring of D X N marriages in the 10-29 age interval, are re- spectively, 24.7, 22.7, and 19.8 years. Furthermore, as is apparent from the appropriate tables, females predominate in the D X D sample but males in the controls. While no correction for these differences seems practical, here is a further indication of the need for caution in the literal interpretation of these findings. Finally, it must be recalled that neither the N X N or D X D curves are “pure,” in that the former may contain some potential diabetics and the latter some non- potential diabetics. This dilution of the two curves would tend to bring them closer together; whether it introduces an appreciable error depends on the degree to which it occurs in each of the curves. 126 TABLE 10.— The proportion of diabetics to be expected from the offspring of D X N marriages, as computed before and after cortisone and at three different points on the glucose tolerance curve 1° 114° 2° Number | Proportion| Number |Proportion| Number | Proportion NO cortisone. ......onnneesuminms 65.8 0.58 72.4 0.65 70.5 0.62 Cortisone. 44.3 0.47 56.6 0. 64.2 0. 61 There is a second approach to estimating the proportion to be affected in a D X N mating. Among the offspring of D X N mar- riages in the families studied there are 79 in the 40-59 age range. Of these, 8 have previously diagnosed diabetes, 10 latent diabetes detected during this study, 14 abnormal cortisone-GTT’s, 21 normal GTT’s (7 of these did not have the cortisone test), and 26 are not known to have diabetes but were untested. Treating the first three types as all of the diabetic genotype, then the minimum proportion would be 30/79, or 0.38. If roughly a half of those untested have the diabetic predisposition (on the basis of the observations both on this group and the 10-29 age group), then the proportion rises to 43/79, or 0.54. Again, then, we would appear to have an excess of diabetically predisposed over expectation on the basis of a single recessive gene, given in assumptions. However, it must be borne in mind that the diagnostic standards for both the standard and corti- sone-GTT applied here were derived from a younger group; the data of the Faculty Health Appraisal described earlier suggest these standards may be somewhat inadequate at this age level, although it should be noted that taken at face value the “age effect’’ on glucose metabolism is much more marked in the seventh and subsequent decades than in the 49-59 interval. This fact might lead to an overestimate of the proportion with the diabetic genotype. On the other hand, some of those with normal GTT’s may yet develop abnormality—this would result in an underestimate. DISCUSSION The approach to the genetic basis of diabetes presented in this paper is quite laborious; the data collected to date are adequate only for very tentative conclusions. While there is no doubt of the familial nature of the disease, we have not, in the sense of adherence to genetic ratios, established that this is on a genetic basis. Rather, we believe it to be genetic because we have no other explanation for the familial constellations of the disease. In this respect, the case for a genetic basis for diabetes is no weaker than for essential hypertension or hyperuricemia. Under these circumstances, the clearest evidence for 127 a genetic etiology comes from the data of which there is need for much more, on twins, involving not only pairs, one or both of whom are diabetic, but elderly pairs, neither known to be diabetic. How- ever, if we assume a genetic basis and proceed to test the hypothesis of recessive inheritance, there appears to be an excess of diabetically predisposed among the offspring of D X N marriages over expectation on the hypothesis that the disease is due to homozygosity for recessive genes occurring at a single genetic locus, unless one is willing to assign to the diabetic genotype much higher frequencies than have been acceptable to date, of the order of 0.20-0.25. Since if recessive genes at several loci were involved expectation would be even lower, the findings are thus not consistent with this hypothesis either. Accordingly, the strongest piece of evidence for the recessive inherit- ance of diabetes is the report that all the offspring of conjugal diabetics appear destined to develop the disease. However, as we have pointed out, Post’s report as amended in this paper, does not exclude the hypothesis that only 75 percent of these offspring will develop diabetes. Other objections to the recessive gene hypothesis have been brought forward by Allan (quoted in Steinberg, 1961) and Simpson (1962). What alternative hypothesis can be suggested? The proportions given are not inconsistent with a hypothesis of simple dominant inheritance, although a precise test is again impossible because of uncertainty regarding the frequency of the diabetic genotype in the population. However, Steinberg (1959, 1961) and Simpson (1962) have advanced what appear to be rather serious objections to the hypothesis of simple dominant inheritance or to a hypothesis that regards severe, juvenile diabetes as due to homozygosity for a gene which, when heterozygous, produces in some individuals a milder, “adult-type’” diabetes. We would appear, then, to be forced either to somewhat more complex, multigenic hypotheses, involving such possibilities as a “principal gene” with modifiers, or genes at several different loci with approximately additive effect (cf. Lamy, Frézal, and Rey, 1961; Simpson, 1962) or to regarding the disease as hetero- geneous, the different entities exhibiting different inheritance patterns, a viewpoint adumbrated in Harris (1949, 1950). The evolution of genetic thought concerning diabetes would thus appear to parallel that concerning essential hypertension and hyperuricemia, for some time regarded as single gene traits but now increasingly being con- sidered as possibly due to the interaction of genes at several different loci (Hamilton, Pickering, Fraser Roberts, and Sowry, 1954; Hauge and Harvald, 1955). A multi-genic hypothesis can take a variety of forms, too many to be explored in detail at this time. Progress in refining this hypothesis is inextricably bound to advances in our understanding of the patho- physiology of the disease. It is also bound to a different approach, 128 so that instead of studying special families, the investigator examines a suitable population sample of families, to the data on which he can apply the techniques of regression, correlation, and variance analysis. The problem of age correction will remain controversial until better longitudinal data are available. However, one point requires com- ment: Under most such models, while a significant parent-offspring correlation is of course to be expected, one does not expect all of the children of affected parents to be involved. If, however, the changing environment is resulting in a progressively lower cut-off point in very recent generations on a quasi-continuous curve of diabetic predisposi- tion, then one might expect a very high proportion of the children of clinical diabetics to develop the disease themselves. But if this is the explanation of the high proportion of affected children when both parents are affected, it implies that the frequency of diabetes is increasing rather markedly. Unfortunately, our prevalence figures have until recently been so poor that this hypothesis cannot be tested, but should it be proved correct some 20 years hence, the genetic approach will have provided a clear lead. The high frequency of the genotype predisposing to so serious a disease as diabetes mellitus presents a genetic riddle. Why has natural selection not reduced the frequency of this genotype to a much lower level? Recently one of us (Neel, 1962) has suggested that the geno- type may be involved in the ability to mobilize insulin readily follow- ing the proper stimulus. It is further suggested that a genotype which might have been optimal during man’s hunting and gathering days (and then only rarely resulted in diabetes mellitus) may result in an overproduction of insulin under present conditions. This overpro- duction is countered by the development of insulin antagonists, which eventually come to dominate the picture, with resulting diabetes. This hypothesis of the role of the diabetic genotype is to a considerable extent independent of hypotheses concerning the precise genetic mechanism responsible for the disease. That is to say, the hypothesis can as well be applied to a one-gene as to a multi-gene concept of the diabetic predisposition. The genetic approach can often result in a significant clarification of a heterogeneous disease entity. A notable example of this concerns the sickle cell diseases (Neel, 1952). Equally often, however, signifi- cant progress on the genetic front must await increased data concern- ing basic pathophysiology. This would seem to be the case with respect to diabetes. The evidence suggests that the abnormalities of the GT'T of the diabetically predisposed are paralleled by, and presum- ably the result of, abnormal relationships between circulating insulin and insulin antagonists (references in Neel, 1962). Disease of the basement membrane of the blood vessels appears also to be present from an early stage (Marble, in press). Obviously longitudinal studies of the diabetically predisposed, beginning at a very early age, 129 have much to offer, their difficulty notwithstanding. Such studies should lead us to the primary lesion, upon which sounder genetic investigations can be built. SUMMARY As a disease entity, diabetes mellitus presents many impediments to a precise genetic study. An investigation is described which attempts to meet some of these impediments by 1) the use of the glucose tolerance test and the cortisone-glucose tolerance test, ‘and 2) the application of a quantitative approach to the data on the younger off- spring of certain types of marriage but a qualitative approach to the older offspring of these same types of marriage. The data obtained once again confirm the familial nature of diabetes but in the strict and formal sense are not proof of a genetic predisposition although, by exclusion, this seems highly probable. It is shown that in the dia- betically predisposed (offspring of conjugal diabetics) there are highly significant changes in the mean glucose tolerance curve in the 10-29 age interval. This fact can be utilized in the analysis of family data. Assuming a genetic basis, then the results of our analyses do not favor the hypothesis that diabetes is due to homozygosity for a single recessive gene, unless one is willing to assign a frequency of 0.20-0.25 to the recessive genotype. There appear to be valid reasons for dis- carding other simple genetic hypotheses. The data are consistent with a multigenic (multifactorial) hypothesis, although the apparently very high proportion of affected offspring from conjugal diabetics remains a troublesome point. This is explicable if the living conditions of the current generation have lowered the threshold of phenotypic expression of the diabetic genotype below that of the parental generation. REFERENCES fk . Allan, W., 1933. Heredity in diabetes. Ann. Intern. Med. 6: 1272-1274. 2. Brandt, R. L., and Tupper, C. J., 1960. Medical appraisal of elderly county hospital patients. Geriatrics 15: 233-253. 3. Chesrow, E. J., and Bleger, J. M., 1955. Diabetes case finding among 1,000 patients over 60. Geriatrics 10: 479-486. 4. Cohen, A. M., 1963. Effect of environmental changes on prevalence of diabetes and of atherosclerosis in various ethnic groups in Israel. In The Genetics of Migrant and Isolate Populations, E. Goldschmidt, ed., pp. 127- 130. Baltimore: Williams & Wilkins Co. 5. Conn, J. W., and Fajans, S. S., 1961. The prediabetic state. Amer. J. Med. 31: 839-850. 6. Fajans, S. S., and Conn, J. W., 1954. An approach to the prediction of diabetes mellitus by modification of the glucose tolerance test with cortisone. Diabetes 3: 296-304. 130 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. . Fajans, S. S., and Conn, J. W., 1959. The early recognition of diabetes mellitus. Ann. New York Acad. Sci. 82: 208-218. . Fajans, S. S., and Conn, J. W,, 1961. Comments on the cortisone-glucose tolerance test. Diabetes 10: 63-67. . Goldschmidt, E.; ed., 1963. The Genetics of Migrant and Isolate Populations, pp. xxi and 369. Baltimore: Williams & Wilkins Co. Grunnet, J., 1957. Heredity in diabetes mellitus. Opera ex Domo Biologiae Hereditariae Humanae Universitatis Hafniensis, no. 39. Copenhagen: Munksgaard. Hamilton, M., Pickering, G. W., Fraser Roberts, J. A., and Sowry, G. S. C., 1954. The etiology of essential hypertension. 4. The role of inheritance. Clin. Sci. 13: 273-304. Harris, H., 1949. The incidence of parental consanguinity in diabetes mellitus. Ann. Eugen. 14: 293-300. Harris, H., 1950. The familial distribution of diabetes mellitus: a study of the relatives of 1241 diabetic propositi. Ann. Eugen. 15: 95-119. Hauge, M., and Harvald, B., 1955. Heredity in gout and hyperuricemia. Acta Med. Scand. 152: 247-257. Hayner, N. S., Waterhouse, A., and Gordon, T., 1963. The 1-hour oral glucose tolerance test. Health Statistics (USPHS Publ. 1000), series 2, no. 3. Washington: GPO. Jackson, W. P. U., and Woolf, N., 1957. Further studies in prediabetes. Lancet 1: 614-617. Joslin, E. P., et al., 1959. Treatment of Diabetes Mellitus, edition 10:798. Philadelphia: Lea & Febiger. Lamy, M., Frézal, J., and Rey, J., 1961. Hérédité du diabétesueré. Journées Ann. Diabet. Hotel Dieu 2: 5-15. Levit, S. G., and Pessikova, L. N., 1934. The genetics of diabetes mellitus. Proc. Maxim Gorky Medico-biological Research Institute 3: 132-147. (Russian.) Marble, A. Vascular changes in prediabetes. In Conference on Small Blood Vessel Involvement in Diabetes Mellitus, M. D. Siperstein, ed. (in press). Neel, J. V., 1952. Perspectives in the genetics of sickle cell disease. Blood 7: 467-471. Neel, J. V., 1962. Diabetes mellitus: a ‘“thrifty’’ genotype rendered detri- mental by “progress”? Amer. J. Hum. Genet. 14: 353-362. Neel, J. V. The genetic basis for diabetes mellitus. In Conference on Small Blood Vessel Involvement in Diabetes Mellitus, M. D. Siperstein, ed. (in press). Nelson, D., Tanney, H., Mestman, G., Gieschen, V., and Wilson, L., 1963. Potentiation of the biologic effect of administered cortisol by estrogen treatment. J. Clin. Endocrinol. Metab. 23: 261-265. Petri, L. M., McLaughlin, C. D., and Hodgins, T. E., 1954. Mass screening for lowered glucose tolerance. Ann. Int. Med. 40: 963-967. Pincus, G., and White, P., 1933. On the inheritance of diabetes mellitus. I. An analysis of 675 family histories. Am. J. Med. Sci. 186: 1-14. Pincus, G., and White, P., 1934a. On the inheritance of diabetes mellitus. II. Further analysis of family histories. Am. J. Med. Sci. 188: 159-168. Pincus, G., and White, P., 1934b. On the inheritance of diabetes mellitus. III. The blood sugar values of the relatives of diabetics. Am. J. Med. Sci. 188: 782-790. Post, R. H., 1962. An approach to the question, does all diabetes depend upon a single genetic locus? Diabetes 11: 56-65. Randle, P. J., Garland, P. B., Hales, C. N., and Newsholme, E. A., 1963. The glucose fatty-acid cycle. Its role in insulin sensitivity and the meta- bolic disturbances of diabetes mellitus. Lancet 1: 785-789. 131 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. Rush, T., and Tupper, C. J., 1960. Two-hour postprandial glucose deter- minations in a periodic health appraisal program. Geriatrics 15: 630-636. Simpson, N. E., 1962. The genetics of diabetes: a study of 233 families of juvenile diabetics. Ann. Hum. Genet. 26: 1-21. Siperstein, M. D., ed. Conference on Small Blood Vessel Involvement in Diabetes Mellitus. (In press.) Spiegelman, M., and Marks, H. H., 1946. Sex and age variations in the prev- alence and onset of diabetes mellitus. Am. J. Pub. Health 36: 26-33. Steinberg, A. G., 1959. The genetics of diabetes: a review. Ann. York New Acad. Sci. 82: 197-207. Steinberg, A. G., 1961. Heredity in diabetes mellitus. Diabetes 10: 269-274. Then Bergh, H., 1938. Die Erbbiologie des Diabetes Mellitus. Vorliufiges Ergebnis der Zwillingsuntersuchungen. Arch. f. Rassen.-u. Gesell- schaftsbiologie 32: 289-340. U.S. National Health Survey, 1960. Health Statistics: Diabetes reported in interviews, United States, July 1957-June 1959. Pub. Health Ser. Publ. no. 584-B21, pp. 1-22. Vallance-Owen, J., and Lilley, M. D., 1961. Insulin antagonism in the plasma of obese diabetics and prediabetics. Lancet I: 806-807. West, K. M., 1960. Response to cortisone in prediabetics. Diabetes 9: 379-385. 132 Coronary Artery Disease* By Vicror A. McKusick, M.D., Professor of Medi- cine, Johns Hopkins University School of Medicine, Baltimore, Md. With regard to the relative importance of genetic and environmental factors in pathogenesis, it may be useful to think of all diseases of mankind as falling on a spectrum (fig. 1), toward one end of which genetic factors dominate and toward the other end environmental factors. Essentially all disease is to some extent genetic and to some extent environmental. Close to the genetic end lie conditions such as phenylketonuria and galactosemia, but they lie not at the extreme end because nongenetic influences can affect considerably the clinical, or phenotypic, features. Near the middle perhaps is diabetes mellitus and possibly not far beyond should be put hypertension and peptic ulcer. Toward the environmental end are infectious diseases, e.z., tuberculosis, but not at the extreme end if we can interpret the findings in twins to indicate a significant, genetically based difference in susceptibility to this disease. Where we should place atherosclerosis and specifically coronary artery disease, is the question we will be discussing. There are distinctive characteristics of the diseases located toward one or the other end of the spectrum (fig. 1): 1. Those toward the ““G’’ end tend to be rare; those toward the “E” end are common disorders, defining ‘common’ as poten- tially present in at least 1 percent of persons at some stage in life. 2. Those toward the “G’’ end tend to be monogenic (i.e., mono- meric) and relatively simple in their formal genetics. In those toward the “E’ end the genetic contribution to patho- genesis tends to be complex and almost certainly polygenic. (I shall use the term multifactorial to indicate the operation of multiple factors both genetic and nongenetic and polygenic to indicate the operation of multiple genetic factors.) *From the Division of Medical Genetics, Department of Medicine, Johns Hopkins University School of Medicine. 133 J PHENYLKE TONURIA JOABETES MELLITUS JPEPTIC ULCER G C {a ] HYPERTENSION TUBERCULOSIS Figure 1.—Schematic representation of spectrum on which all disease falls with regard to the relative importance of genetic and environmental factors in pathogenesis. In studying the genetics of disorders near one or the other end of the spectrum the questions one poses tend to be different. At the “G” end the questions concern 1) the details of the formal genetics (gene frequencies, mode of transmission, selection factors, mutation rate, etc.) and 2) the precise nature of the gene-determined basic defect which is assumed to be unitary. At the “E” end, the question tends to be, first and foremost: is there a significant genetic contribu- tion to pathogenesis? Related to this question is a second: by what mechanisms do genetic factors contribute to the pathogenesis? (In principle the questions are not greatly different; the main difference lies in the level of analysis.) The general approaches toward getting an answer to the question of the strength and mechanism of genetic factors in common diseases are of at least six types (Edwards, 1960; Edwards, 1963; Penrose, 1953): Study of familial aggregation Twin study Interracial comparisons Genetic study of component factors in pathogenesis Blood group and other associations . Genetic study of animal analogs (or homologs). Familial aggregation is essentially a sine qua non for any genetic hypothesis. Studies are of two general types depending on methods of selection of families: 1) proband studies in which index families are compared with control families, and 2) population-based studies such as the Tecumseh one (Francis, this symposium) in which familial clustering is tested for. The choice of controls is one source of difficulty in proband studies. Another is bias of ascertainment of the type that Dr. Crow (this symposium) discussed with regard to rare recessive traits. The selection biases in the latter situation have been recognized for a long time—since Weinberg. Haenzel (1959) and others before him have pointed out that in proband family studies of common disorders the same biases may operate. In elementary terms if a family with multiple cases of the disease has a greater probability of entering your series than does a family with only one case, then you cannot fail but find familial aggregation. Familial aggregation does not, of course, prove the genetic basis of the trait in question because socio-environmental factors, as well as genes, vary from family to family. Determination of the propor- tion of offspring affected according to whether both, one or neither 134 parent is affected is a further test which is usefully applied to family data, although it too is not conclusive proof of the genetic hypothesis. The proportion of affected offspring is expected to be greatest with both parents affected, less with one parent affected, least with neither parent affected. It may be worthwhile to recall the story of pellagra. In 1916, before pellagra was identified as a nutritional disease, it was thought to be due to a toxic agent, possibly infectious, suscepti- bility to which was strongly determined by genotype. In the studies of Davenport (1916) and Muncey (1916), the arguments for a promi- nent genetic factor in pellagra were 1) familial aggregation, 2) ethnic differences, specifically a lower frequency in Negroes, and 3) intra- familial similarities in the clinical pattern of the disease. Although it apparently escaped their attention, evidence against the genetic factor was provided by the proportion of children affected according to whether both parents, one parent, or neither parent was affected (see table 1). A trait which is genetically determined to a sig- nificant extent should show a decreasing proportion of affected off- spring according to whether both, one, or neither of the parents display the trait. Such is not the case in the data on pellagra. TABLE 1.—The proportion of pellagrous children according to parental mating type (Adapted from Muncey) Total Pellagrous offspring Number of parents pellagrous number of children Number Percent 413 133 32.2 752 120 16.0 120 22 18.3 Total. eas 1,285 375 20.2 It should be noted, however, that a positive “mating-type” test, although expected of a genetic trait, does not of itself prove the genetic factor. One can easily imagine a concentration of socio-environ- mental factors in families which would result in more “affected” children when more of the parents are ‘“‘affected,” and vice versa. Lilienfeld (1959) has shown that the “mating-type” test is positive for the trait “medical school attendance’; also a greater proportion of offspring are likely to attend university if the parents did than if the parents did not. Cigarette smoking in teenagers likewise shows a positive “mating-type” test (Horn et al., 1959). Twin studies are more specifically indicative of genetic factors as opposed to nongenetic ones. Dr. Harvald (this symposium) has outlined the difficulties of this approach. Of all of them, however, the twin method is perhaps the best we have for quantitating the role of genetic factors in common diseases. , 135 Illustrating his point with an example, Edwards (1963) makes the useful point that even with a four-fold increase in the risk of a common disorder the increase in concordance in monozygotic twins may be discernible only with rather large numbers. Interracial or interethnic comparisons are a hazardous basis for conclusions about genetic factors in a common disease because one can never be certain that the groups compared are identical with reference to all pertinent exogenous factors. Whenever a variant is shown to be significant to the pathogenesis of a common disease, the genetic factors involved in its variation can be studied by family aggregation, by twin study, and, for what they are worth, by interracial comparisons. For example, insofar as blood lipid variations are important to the pathogenesis of coronary artery disease, studies of the genetic basis of such variations are pertinent to the topic under discussion. Serum pepsinogen level was an example cited earlier by Dr. Fraser Roberts in relation to peptic ulcer. Dissecting out the components of pathogenesis also provides an answer to the second question in the genetics of common diseases: How do genetic factors in pathogenesis operate? Blood-group-and-disease associations have been demonstrated in several instances, most convincingly in the case of peptic ulcer and the blood type O, nonsecretor traits. The phenomenon association usually has nothing to do with genetic linkage which is the location of genetic loci fairly close together on the same chromosome. Genetic linkage produces no permanent association of traits in a population. Crossing over minces up the chromosomes so that in time no associa- tion is observed. To be sure, when the linkage is close, it may be many generations before the distribution of the two traits is random. The physical basis of most association is, therefore, not chromosomal linkage but is rather some physiologic peculiarity, of persons of group O, for example, rendering them at greater risk to the develop- ment of peptic ulcer. Since the blood groups are simply inherited we have here one of the clearest examples of constitutional, i.e., genetic, diathesis in disease. Blood groups are, of course, not the only poly- morphic traits with which association of common illness can be sought. Various serum groups such as haptoglobins, Ge, Ag and other types are further examples. Obviously if association is not demonstrated, genetic implications in the common disease are not disproved. I have put association just below genetic study of pathogenetic com- ponents because they are related matters although different in approach. The O-peptic ulcer association tells us that some feature of the type-O person is a pathogenetic component in peptic ulcer. In approach No. 4 we take a variant which, by epidemiologic or experimental means, has been shown to be a component factor in ‘pathogenesis, and study its genetic determination. In approach No. 5 we work the other way around—take a clear-cut mendelizing trait 136 and test whether it has a pathogenetic component by looking for association. If one has, in an animal other than man, a disorder seemingly iden- tical to the common disease of man, then one may be provided with insight into possible gene-conditioned pathogenetic mechanisms in man. Let us now consider some of the results of the application of these approaches specifically to coronary artery disease. Atherosclerosis and coronary artery disease cannot with validity be equated, yet much of the information on atherosclerosis in general is pertinent to coronary artery disease specifically and will be included. Familial aggregation. Clinical impression (White, 1960) suggests that one should find a considerable aggregation of cases of coronary artery disease in families—more than for the garden varieties of cancer, e.g., those of stomach, breast, and uterine cervix, which have been studied extensively. Unlike the cancer field, only a few family studies of coronary artery disease, each with shortcomings, have been reported. These include the studies of Gertler and White (1954), of Thomas and Cohen (1955), of Russek and Zohman (1958), and of Shanoff and colleagues (1961). Two family studies of modest size have been conducted by my colleagues Skyring et al. (1963) and Rose (1964). The choice of controls has been criticized in some of these studies, and problems such as better recall by proband families and biases of ascertainment are with us here as in many such studies (Schweitzer et al., 1962). These criticisms aside, the results of all these studies suggest a modest familial aggregation. These are all proband studies. The population approach under way in Tecumseh was recently reported on in a preliminary manner by Epstein (1963); the familial aggregation did not look overwhelming by any means. Twins. There are isolated reports of twins concordant (Benedict, 1958; Parade and Lehman, 1938) and discordant (Lees et al., 1963) for various aspects of coronary artery disease. But these add little. The study reported yesterday by Dr. Harvald is, of course, of great interest and is, as far as I know, unique. You will recall that among 102 monozygotic twin pairs and 155 like-sex dizygotic twin pairs the concordance rate for death from coronary artery disease was essentially the same. Several questions would be of interest: What of the sepa- rate comparison of male monozygotic twins with male dizygotic twins and female monozygotic twins with female dizygotic twins? Tt might be of interest to look at other subgroups such as those with one twin dying of coronary artery disease under 45 years of age. And a general question comes to mind about twin studies: Does not the heritability estimate derived from twin studies vary with the environmental cir- cumstances and is not the estimate interpretable only in a specific environmental context? 137 735-712 0—65——10 Many 0; and disregarding the large portion (1—M) of the popula- tion without the trait, for whom Y,=0. The problem arises when one cannot separate beforehand the rare class, whose members appear as a random subclass in the sample of each stratum. It received explicit treatment (Yates, 1960, p. 300; Kish, 1961; Kish, 1965, section 4.5), on which we base the following conclusions. The mean for a subclass from a stratified sample can be computed as _ 2 Yu/fn yv= > m/f my/f, =r Wan, (5) Mh where Y n=2J ¥ntis the total of the Y, variable obtained on the m,, class members, found in the sample of n, cases from the h-th stratum. Fur- ther, yn=yn/my, the sample mean in the stratum; m,=m,/ny, the pro- portion of the class; and my/fn 3 my/fy’ the stratum weights of the class found in the sample. If the popula- tion numbers My, of the class were known constants, then we would have the known weights W, instead of the variates wy; and instead of the special forms (5) and (6) we could use (1) and (2). The variance of y, can be found to a good approximation to be Wh= var (Fu) = 32 (1—f,) on [t2— TT, (FoF)? 6) where m;=mj,(n,—1)/n, and th=23 (Yn1—yw)?/my,. Examination of (6) discloses that for small subclasses, when the my (yn—yw)® are small, the gains of proportionate stratification vanish. But the gains of optimum allocation remain. This alloca- tion can be shown to be reached approximately when f,=k VM, 62/0, he the t, are approximately equal then the optimum rates are ==] ML, where k’ is a constant of proportionality. hy the population aggregate Y of the Y, variable in a subclass can be estimated with ordinary formulas, by also including all the Y,=0 values of nonmembers of the class (Hansen, 1953, p. 149-155). But it can also be estimated from the class members alone as =I Yu/fn, (7) 168 with the variance ~ 2 var (1) = (1-0) 70 ti+F2—m50)- ® ‘h c. Cluster sampling denotes selection methods in which the sampling units are clusters of elements. Contrariwise, in element sampling the elements are also the unique sampling units. Examples of cluster sampling are selecting entire families or entire city blocks to sample the persons within them; or selecting entire classes or entire schools to observe their students; or selecting entire villages to survey their inhabitants. Just what is a convenient cluster depends on the survey situation and resources, and these are matters of practical expediency. First, the elements are defined in accord with the research aims. Then the sampler must decide whether he can retain the simplicity of element sampling, or must he abandon it for some convenient clusters that permit a more economical and practical design. Cluster sampling can be useful when it decreases considerably the element cost, by lowering the costs of travel and listing. But it often increases the element variance, and it also complicates the analysis. It should be used when the lowered element cost more than compen- sates for the increase in element variance and in the cost of the analysis. The larger the average cluster the greater are these effects. When the effects are too great we may search for an economic compromise between the two conflicting effects of clustering. This search leads to subsampling or two-stage sampling; similar considerations can lead to more stages and to multi-stage sampling. Two further complications are present typically in clustered designs. First, clusters are generally selected after stratification, because this is both feasible and desirable. Second, in natural situations one must generally deal with clusters of unequal size. Even when the selection design creates clusters of roughly equal sizes, their exact equality is destroyed by nonresponse, by imperfect size measures, and by the fluctuations in the sizes of subclasses on which the statistics are often based. Despite the possible variations in complex designs, basic formulas exist, that can be used rather generally for the sample mean and its variance. They also hold for means of subclasses. The sample mean is a ratio mean, in which both numerator and denominator are random variates. If from each stratum two primary selections (a and b) are chosen at random in the first stage of selection, the ratio mean may be denoted as A (9) 169 735-712 0—65 12 Here yu, and yup, are the sample sums of the Y, variable for the two primary clusters selected independently at random from the hth stratum. Any needed weighting is incorporated in them so that the expected value for both yy, and yu, is (except for a constant) Yy, the population aggregate of the Y, variable in the Ath stratum. For epsem selections the yp, and yn, may be the simple sample sums. Similar statements hold for the x,, and the x, which usually repre- sent merely the element counts in the primary selections. The variance of the ratio mean is approximately val ()- of [= +1) D2 2 3) SD,D, | (0) where Dj, = (Fna—Ymn)®, Di, = (Xna—Xnp)?, and D, D,, = (Fna— Yun) (Xna— Xun) - When the number of primary selections in the stratum is a,, we can have more generally T= uD (9) h ap h ap with the varisnce (10) now defined by Di = Syta—3t )fan—1), and ah similarly for the other two terms. These formulas can be found justified and derived in several places (Kish, 1959; Kish, 1965, section 6.5). Clustering often has a drastic effect on the variance. A sample of a clusters appears confined, compared to an element sample which is spread over n points of the population. The distribution of variables within natural clusters is typically not random, but more or less homogeneous. The elements of clusters tend to be similar; for example, people in blocks, students in classes tend to resemble each other. This effect of clustering on the element variance can be represented approximately as [1+7ho(n/a—1)]. Here rho is the coefficient of intraclass correlation; it is generally positive in natural clusters; it is small for most variables, and its maximum value is +1. Hence the clustering effect is not great if the mean cluster size, n/a, is small. When n represents the sample total of a rare class, then n/a can be kept small, even if we use clusters rather large in terms of their complete count (Hansen et al., 1953, pp. 107-109). d. Systematic selection, an alternative to random choice, denotes the selection of sampling units with an interval of k applied to listings after a random start from 1 to £. In subsampling it can be a practical substitute for random choices. e. Two-phase sampling refers to subselecting the final sample from a larger sample that was preselected to provide information for 170 improving the final selection. The first phase can be a screening operation that cheaply separates small strata containing high pro- portions of the rare target population. SELECTION TECHNIQUES FOR RARE TRAITS A small proportion M=M/N of a population of N elements is characterized by a trait denoted with variable X,0, whereas for the vast majority (N—M) the variable X,=0. The search for elements that are rare but important can have any of three aims, but the problems and methods of selection are similar for all three. (a) We may need to know only M or M, the number or proportion of the members of the class with the trait; for example, the prevalence of a rare genetic trait. Estimating M and M represent the same problem if the population base, N, is known well enough relative to M. (b) We may need Y= Y/M= X/M, the mean of the X, variable within the class, disregarding the majority (N—M) for which X,=0. Here Y=X and Y denote the sum and the mean of the X, in the class. For example, X may denote the individual’s medical costs caused by the trait and Y the mean cost in the class. (c) Or we may need X=X/N= MY + (1—M)0, the mean in the entire population; for example, the mean cost of a trait in the population. Of course, M is only the special case of X when Y,—=1, the mere presence of the variable for all M class members. Estimating X from Y is easy when M is known well enough relative to Y. The degree of rarity and the cost of measuring the trait are the two factors that mainly determine the relative worth of different selec- tion methods, as noted in a previous discussion (Cochran, 1954). Sample surveys are used widely and successfully to obtain cheap measurements of traits that are not too rare. A screening procedure of the population may be practical to find a class if M comprises 10 or 20 percent of it, but much less so if M is 1 in 100 or 10,000. Even difficult measurements, if not too rare, have already been applied with probability sampling, as for fertility behavior (Freedman et al., 1959), and medical examinations (U.S. National Health Survey, 1964). On the contrary, very rare traits are easy to sample only if finding them is cheap. An extreme example would be a centralized and complete list of persons with a well recognized rare disease. But rare traits that are expensive to find present difficult sampling problems. Yet they may be undertaken with some of the following techniques. These may be used in combinations, and relations between them may be noted. In choosing between them consider the trait’s rarity. Consider also the available survey resources, including knowledge about large concentrations of the trait. This could mean 171 lists, or known geographical concentrations, or high correlations with other clear indicators. a. Multipurpose samples. Several rare traits can be discovered in one sample, and its costs can be divided among the cooperating investigations. This approach can also yield data about relation- ships between the traits. Some economic and market surveys are based on such cooperation. The National Health Survey may be considered a combination of many specialized purposes. b. Cumulation of a rare population can be obtained as byproduct of continuing surveys with changing samples that cover large popula- tions in time. The results must be interpreted as an average over the period covered by the surveys, but this may offer little difficulty. First, past surveys may already contain the data in an acceptable uni- form manner; though too rare in single samples, the accumulation may provide an adequate base for analysis. Second, the measure- ment can be added to future surveys if the enumerators can perform it. Even a lengthy addition adds little to the average interview length, if applied only to a small portion; but emphasis is needed to keep enumerators alert to rare events. Third, the surveys may per- form only a screening operation to identify the rare individuals; the difficult measurements on them is done separately. If the identifica- tion is imperfect it can still provide the first of a two-phase sample. c. Large clusters can decrease drastically the costs of locating and screening population elements. For example, a sample of complete blocks of a city, or of entire villages of a country, can be searched (screened) for persons with the trait. For samples of the entire population large clusters are avoided, because of their adverse effects on the variance; but they will yield only small clusters of the rare trait if this is widely spread. If, on the contrary, it is concentrated in small areas, these can be often recognized and stratified. d. Controlled selection. Sometimes the search has been confined to one or two hospitals (or other places), purposively chosen. A broader base can be secured with four or eight replication, chosen to represent a wide spectrum of backgrounds (Kish, 1965, section 14.4). But perhaps the sample can be expanded to 15, 30, or 60 hospitals (places). Then one prefers to control carefully the selection to make it represent the population along several important control variables. The num- ber of control cells may far outrun the number of permitted selections. The selection problem has been treated under “controls beyond stratification,” “two-way stratification,” “lattice sampling,” and “controlled selection” (Hansen, ef al., 1953; Yates, 1960; Hess, 1961; Kish, 1965, section 12.8). e. Disproportionate stratified sampling. It is eflicient to increase the sampling fractions from small strata if these contain large por- tions of the rare trait, demarcated from large portions of the total population containing only small portions of it. In other words, if 172 most of the rare trait can be located in small pockets of the total population it should be sampled heavily. If, for example, 90 percent of the rare trait can be located within 10 percent of the total popula- tion, then large gains can be made. But geographical strata can seldom yield high enough separations of class densities. We look for special lists of the class with the needed trait, or with another trait highly correlated with it. However, only small gains can be had from the presence of only one of the two desirable conditions: either from high concentrations of only a small portion, or from mild concentrations of a large part of the population. This can be seen in the formulas for “optimum allocation” of the sample sizes into strata, yet it is often misunder- stood. For example, U.S. cities have exclusive neighborhoods that are full of rich people; but oversampling them will not bring large gains, because most rich people live scattered elsewhere. Simiiarly, visitors to a set of clinics may contain a high concentration of the class with the trait, yet most of the class may never reach any clinic. Sometimes the visitors to the clinics may well represent the entire class; but often the two kinds of class members may differ drastically. For example, rich people in exclusive neighborhoods differ in many ways from those living outside them. An exotic example is found in interstellar space: Although almost empty compared to stellar density, it contains nevertheless a large portion of total matter, and its com- position may differ from stellar matter. A high degree of separation is needed, because optimum allocation for a rare class typically calls for selection rates about proportional to VM./J, where M,, is the proportion of the rare class, and J, is the element cost in the stratum. Large differences in M, are needed to obtain even moderate differences in /M,. f. Two-phase sampling (double sampling) may be needed to dis- tinguish, in the first phase of preliminary screening, two or more strata in which the proportions of the rare elements differ greatly. Then the methods of disproportionate sampling can be applied in the second phase. The first phase is needed when desirable stratifying variables are not readily available for the entire population. The allocation of effort, between the first phase (screening) sample and the second phase, depends on the utility of the screening and on the relative costs of the two phases. g. Screening operations underlie several methods. They are useful when we find a cheap screening test that has very few false negatives (erroneous exclusions), and not many false positives (erroneous in- clusions). Applied to the population, or to a large sample in two- phase sampling, it picks out a smaller sample of screened inclusions, the potential members of the rare class. This part of the sample is then measured with the more precise and accepted standard process to discard the false positives. 173 What about false negatives? Even good screening can be imper- fect. In the absence of overwhelming assurance of safety, a sample check of the screened exclusions is advisable on large surveys. If this finds none of the rares, it provides the error’s upper limit. But if false exclusions rise beyond a negligible proportion, then the screened exclusions become a stratum to be sampled at a lower rate. The nature of screening operations vary: a brief interview, or rapid observation by field worker, or scanning data on filed cards. Generally the screening should err more on the side of false positives than of false negatives. The unit cost for the former (some extra work) is less than for the latter (either bias or reduced efficiency). The costs and the proportions of both kinds of error should enter the design of the screening process. h. Special lists provide, when available, the best means for locating a rare target class. If they are very rare, widely scattered, and difficult to identify, a list may be the only means for finding them. The ideal list contains the entire target class (or a good sample) clearly isolated or identified, and with the information (e.g., address) needed to make the survey. Such lists can sometimes be produced with ingenuity, perhaps with tolerable redefinitions of the class, and perhaps by combining several lists. The economics of the joint use of several selection frames has been investigated recently by Hartley (1962). Snowball sampling is the colorful name (Goodman, 1961) for techniques of building up a list of a special population, or a sample, by using an initial set of its members as informants. An example given by Deming (1963) is asking an initial selection of deaf people to supply the names of other deaf persons they knew. It may be difficult to assign reliable measures of the probability of inclusion. However, members of some populations may know about each other, as the deaf do in towns and cities that are not too large; perhaps also owners of pigeons or of racing cars. Extra effort is needed if the list contains impurities, foreign elements and replicate listings. But the worst problems are caused by incompleteness, the noncoverage of elements missing from the list (Kish, 1965, sections 2.7, 11.1-11.5, and 13.3). When the list is good yet imperfect we must decide whether to go after the small missing portion or to omit it. If the missing portion is small and extremely difficult to find we may be forced to accept the “cutoff” bias. But if needed and feasible, it may be included with a supple- mental procedure in a special stratum. The decision depends on several considerations. First, the element cost is high when the missing units are rare, widely scattered and difficult to find. Hence, their inclusion should be designed with a considerably lower sampling fraction. Second, the bias of their exclusion should be judged against the other sources 174 of errors, hence also against the size of the sample. The importance of a fixed bias increases with the precision and size of the sample. The size of the bias should be balanced against the standard error, and the cost of reducing the bias against the reduction in the standard error it can buy. Third, the effect of exclusion is different for estimating frequency totals and for mean values of a trait. The exclusion of missing values has little effect on the mean, but great effect on the total, if ths missing values resemble closely the values found on the special list. Con- trariwise, if the missing values are of negligible magnitude they will affect the mean more than the aggregate. Hence one should con- sider both the number and the nature of the missing units. In that judgment he may be guided by knowledge based on theory, and on past investigations. If necessary he may conduct a pilot study to assess at least roughly the limits of the problem. i. Supplements are useful for sampling the portion missing from the special list, when that portion is small but not negligible. If the missing portion is negligible we need only the special list; if it is large we don’t need the list at all. Consider the rare elements as coming from two strata: the special list to which the sampling fraction f, is applied, and the supplement for the rest of the population sampled with the fraction f,. Because the element cost is typically much higher in the supplement, the sampling rate f, should be considerably lower, according to the optimum allocation of resources. The separation of the two strata (A and B) is complete if the supplement B excludes any element that appears on the entire special list A, whether or not it appears in the sample from that list. In that case the elements in the two samples should receive weights propor- ional to 1/f, and 1/f,. But the exclusion of A elements from the B sample may be incon- venient. Then they may also be permitted to receive the selection probability f, in the supplement, additional to their selection proba- bility f, on the special list. It is necessary and sufficient that all the A elements in the entire sample, from both the A and B samples, be identified. Then all these must receive weights proportional to 1/f,’, because f,"=(f,+f,) was their probability of selection if the two samples were independent. If duplicate appearances in the sample are excluded, than f,"=f,+f,—fufs. SUMMARY In summary, a special list that includes the rare class and nothing else is best, especially for very rare traits. If the list is good but imperfect, a supplemental sample may find the missing units. Lacking a good list we try to separate a small special stratum that contains 175 most of the rare class, from a large stratum that contains much less. The special list or stratum is sampled with disproportionately high rates. If we can’t find such special strata we try to create them with a cheap screening operation. The efficiency of the screening may be increased with double sampling, and by using large clusters. The costs of screening may be shared in multipurpose surveys, or the rare class may be accumulated in a periodic survey operation. 1. 10. 11. 12. 13. 14. 15. 16. 176 REFERENCES Cochran, W. G., 1963: Sampling Techniques, New York. John Wiley and Sons, 2d ed. Cochran, W. G., 1954: Problems Confronting Human Genetics: Discussion. American Journal of Human Genetics 6: 77-80. Deming, W. E., 1960: Sample Design in Business Research, New York. John Wiley and Sons. Deming, W. E., 1963: Chapter 1 in Rainer, J. B., Altshuler, K. Z., and Kallman, F. J. (Eds.): Family and Mental Health Problems in a Deaf Population, New York. New York State Psychiatric Institute, Columbia University. Freedman, R., Whelpton, P. K., and Campbell, A. A., 1959: Family Planning, Sterility and Population Growth, New York. McGraw-H1ll. Goodman, L. A., 1961: Snowball Sampling. Annals of Mathematical Statistics 32: 148-170. Hartley, H. O., 1962: Multiple Frame Surveys. Proceedings of the Social Statistics Section, American Statistical Association, 203-206. Hansen, M. H., Hurwitz, W. N., and Madow, W. G., 1953: Sample Survey Methods and Theory, 2 vols., New York. John Wiley and Sons. Hess, I., Riedel, D. C., and Fitzpatrick, T. B., 1961: Probability Sampling of Hospitals and Patients, Ann Arbor, Michigan. Bureau of Hospital Administration. Kish, L., 1953. Selection of the Sample Book. Chapter 5 in Festinger and Katz (eds.): Research Methods in the Behavioral Sciences, New York. Dryden Press, 175-240. Kish, L. and Hess, I., 1959: On Variances of Ratios and Their Differences in Multi-Stage Samples. Journal of the American Statistical Association 54: 416-446. Kish, L., 1959: Some Statistical Problems in Research Design. American Sociological Review 24: 328-338. Also in the Bobbs-Merrill Reprint Series in the Social Sciences. Kish, L., 1961: Efficient Allocation of a Multi-Purpose Sample. Econometrica 29: 363-385. Kish, L., 1965: Survey Sampling, New York. John Wiley and Sons. U.S. National Health Survey, 1964: Cycle I of the Health Examination Survey. National Center for Health Statistics, Washington, Series 11, No. 1. Yates, F., 1960: Sampling Methods for Censuses and Surveys, New York. Hafner, 3rd ed. The Reliability of Survey Data By Cuarues F. Cannel, Ph. D., Survey Research Center, University of Michigan, Ann Arbor, Mich. The question of the reliability of surveys can be approached by means of three propositions: 1. Some data collected in some surveys have a high level of accuracy. 2. The same data collected in other surveys are highly inaccurate. 3. Some data collected in any surveys are always inaccurate. To be symmetrical there should be a fourth proposition: Some data collected in any surveys are accurate. But this proposition is not included since even the simplest data show inaccuracies. As the propositions indicate, the major components of survey accuracy are the type of information sought and the methods used to collect it. (I am not including in this paper any considerations of sampling since these were covered by Dr. Kish.) The purpose of this paper is to examine some of the factors relating to characteristics of the information and of data collection methods, and to consider some of the variables which contribute significantly to the accuracy or inaccuracy of surveys. We turn first to a consideration of some of the characteristics of the data themselves and the degree to which the wanted information is accessible to the person who is asked to report it. Clearly, the first requisite to accurate reporting is that the respondent have these data in his possession. One point is so well known that we need spend little time on it. A person is likely to have more information about himself than he has about others. For example, in a study of the validity of the reporting of hospital episodes conducted for the National Health Survey (Health Statistics, Series D, No. 4), we found that when an individual was asked to report about his own hospitalization the under-reporting rate was 7 percent. When reporting for close relatives the under- reporting was somewhat worse, and rose to 22 percent when the 177 person being reported for was a more distant relative. Several factors are involved in this under-reporting rate, one of the most obvious being that in many cases the individual asked to report never knew the information and hence could not report it correctly. But of more importance is the inaccessiblity of information when the researcher and the respondent do not share the same concepts or even the same language. This problem is illustrated by a question- naire administered to farmers, which contained the following question: “In planning your farming operations, do you use inductive or de- ductive thinking?” This is an extreme example of the lack of shared concepts, in which the objective may be perfectly clear to the re- searcher but incomprehensible to the respondent. Problems of this type are so apparent that we avoid them almost without thinking. But less extreme examples are not so easily avoided. Frequently we would like to ask respondents about such diseases as diabetes, hypertension, or arthritis. Take the question, “Have you had rheumatic fever?” Asked of a sample of respondents, such a question will receive either “yes” or ‘no’ responses, with only a few reporting that they do not know. A ‘yes’ response may have any of several meanings. It may mean, “I have been under treatment for a con- dition which was diagnosed by a doctor as rheumatic fever,” or, “I have something like my brother had and he said he had rheumatic fever so I think I have it,” or, “I guess I must have it, but I just call it plain rheumatism.” Conversely, a “no” response may mean, “I never had anything I knew to be rheumatic fever,” or, “I know what rheumatic fever is and the doctor told me I didn’t have it,” or, “I don’t know what it is so I guess I don’t have it.” The lack of shared concepts between the researcher and the respond- ent is one of the major limitations of the survey technique as a method of providing data sought by the medical researcher. Parenthetically we might note that some surveys have been severely criticized for inaccuracy because the information obtained from patients in inter- views as to conditions from which they suffered did not check with diagnostic data obtained by direct medical examination. To indict surveys on this basis is fallacious. The fault lies not with the surveys but with the researchers who use the survey method improperly. The potentialities of survey research as a method of collecting data are limited to information which the individual possesses and to con- cepts which he understands. Unfortunately, information which the respondent thinks he has is often incomplete or inaccurate. A study conducted by the Health Insurance Plan of Greater New York under contract with the National Health Center (Health Statistics, Series D, No. 5) compared diagnostic material reported by respondents with material in doctors’ records. Major discrepancies in many categories between the records and the reports of the respondents were revealed. Our study of hospitalization data (Health Statistics, Series D, No. 4), 178 which compared hospital records with reports of patients, showed that respondents reported the diagnoses of malignant neoplasms at only 75 percent of the rate of the hospital records, whereas benign or unspecified neoplasms were reported at 150 percent. We suspect that much of this error represents not only the failure of the doctor to transmit the correct diagnosis to the patient, but also represents a perceptual distortion on the part of the respondent as to his own condition. The doctor is the major source of the respondent’s information about his physical condition. Obviously respondents cannot report informa- tion which doctors have not given them, or information which has been given them inaccurately, or information which has been given accurately but has been misunderstood. One of the primary difficul- ties in getting respondents to report what physicians have told them is caused by the esoteric language which physicians use. For example, one person reported to us that his doctor told him he had ‘hemorrhoids in the rectory.” Although the doctor is the main source of information which the respondent has about his physical condition, unfortunately for re- searchers he is not the only source. Friends, relatives, and one’s own information frequently lead to self-diagnoses. These diagnoses, reported to the interviewer as readily as the more authoritative doctor’s diagnosis, cannot be expected to be as accurate. Recently we discovered a new version of this problem in a health survey while interviewing an elderly Polish couple. The wife reported that she had not visited a doctor in 20 years because she was much afraid of doctors. Yet, when asked about her physical condition, she gave a couple of specific diagnoses unlike those commonly made by laymen. The puzzled interviewer asked her how she had arrived at these diagnoses, and she explained that her husband, who had a chronic condition, went to the doctor every other week. When she had a particular set of symptoms she related them to her husband, who reported them to the doctor as his own symptoms. The doctor then made a diagnosis and gave him medicine which he took home for his wife. Thus far we have concentrated on what one cannot expect from surveys. Returning to a positive approach, we may ask what a respondent can report and report accurately. First of all, the respond- ent can tell us how he feels. He can tell us any degree of disability he may have, whether he is able to work, whether he has been confined to bed because of illness, whether his work has been restricted, and so forth. He can report specific symptoms and can describe these symptoms in detail. He can report the frequency or duration of symptoms, their intensity, and what he does to relieve them. He can report what he considers to be the cause of the symptoms, that is, what he considers to be the diagnosis. He can go further and relate 179 the facts surrounding the diagnosis, whether it was made by himself, by his friends, or by a medical authority. He can give us the history of his conditions, illnesses, and injuries. The respondent can report these types of information plus many other types he has about himself and under certain conditions will report accurately. Tt is relevant to point out that many of these data cannot be obtained from any records whatever but must come from the respondent himself. To recapitulate: Surveys are potentially accurate or inaccurate depending upon the uses to which they are put. A major obstacle to accurate information is that the respondent may not have the data which the researcher wishes to obtain. Surveys cannot be expected to replicate data from medical records any more than medical records can be expected to provide data on undiagnosed conditions for people who have never sought medical aid. For example, if one wanted to know how many people have been hospitalized for particular diseases, surveys would be of little use. Tt would be better to obtain the information from hospital records. But on the other hand, if one wanted to know how many people had respiratory infections during a particular week, records would not help. A survey approach asking about symptoms and degree of disability would be the only source of information since only a small proportion of those persons would have been to a doctor. If one wanted to know how many people have hypertension, the only accurate procedure would probably be to examine a cross-section sample of the population, a technique which the Health Examination Survey of the National Health Survey is using currently. Assuming that the survey approach is used and that the respondent has access to the information wanted, the important question then arises: Will the information be reported accurately? There are two main obstacles to accurate reporting: memory, the ability to recall accurately, and motivation, the willingness of the respondent to report accurately. The problem of memory is a familiar one. In the study of hospitalization referred to earlier, it was found that the rate of reporting was high for episodes which occurred close to the date of the interview, but dropped consistently and with increasing rapidity the longer the elapsed time between the interview and the date of the hospitalization. For episodes occurring a year prior to the interview, the rate of under-reporting of episodes was several times the rate for the first few weeks immediately preceding the interview. If the event in which the researcher is interested is both recent and from the respondent’s peint of view significant, a simple question may be enough to bring the episode fully into his memory. But if much time has elapsed since the event, the problem is more difficult. The gradual decay of information is manifested in many ways. Events of trivial significance for the respondent may be forgotten almost as quickly as they occur. Even experiences which were once prominent 180 are likely to be forgotten if they have little relevance to the individual’s current life. The adult is unlikely to remember the age at which he developed chicken pox, even though the event may have had dramatic importance for him as a child. For such routine matters as one’s breakfast menu or the content of the previous evening’s television programs, recollection may be gone within a matter of hours. Memory is a complex function of elapsed time since the event, current cues or relevance for present affairs, and the original signifi- cance of the event to the individual. The usual effect of these process- es is a reduction in the amount and accuracy of information available to the researcher. In the extreme case, the reduction is complete, the respondent is unable to recall the event. More often, however, the reduction is partial, the respondent can recall that he had a particular experience but is unable to describe it in accurate detail. Thus Goddard et al. (1961) have found that although mothers’ evaluations of difficulty of labor and delivery agreed with the physi- cians’ records, information on the length of gestation period, feeding problems of the infant, and illnesses during infancy showed marked distortions. Unfortunately for the researcher, the process of memory decay is not uniform and orderly. The events of the past do not fade gradually from view while retaining their original dimensions. On the contrary, the process of forgetting and remembering involves considerable dis- tortion. Certain aspects drop from view, others are elevated to prom- inence, and the entire past is recalled in a way that makes it plausible and consistent in terms of the individual’s present experience. Related to this selective process of memory is the distortion of previous experience in the general directions of social acceptability or preserva- tion of one’s self-image. Wenar (1963), for example, reports that when a mother distorts developmental facts about her child she tends to do so in a way that makes her child appear more precocious; and that when she distorts child-training practices, she tends to bring them in line with Dr. Spock. Frequently, such distortions are not conscious falsification of infor- mation but result from unconscious repression of facts which are intolerable to the self. Studies of political behavior provide an interesting example of this process. In national samples voters have been asked repeatedly to name the candidate for whom they voted in the previous presidential elections. The longer the election recedes into the past, the greater the proportion of people who report that they voted for the winning candidate. When the individual is asked to report some past event, a number of complex processes occur simultaneously, and in combination they affect and even determine the availability of the information. These processes have to do with the initial importance of the event and its meaning for the respondent; they have to do with the tendency toward 181 congruence and plausibility of previous experience; and they relate to various ego defense mechanisms. We now come to respondent motivation, that is, the willingness of the respondent to report those things which are accessible to him and are not subject to the distortion of memory. Respondent motivation is coming to be recognized as perhaps the most important factor in determining the amount and the accuracy of the data available to the researcher. Thus, hospital episodes involving arthritis, deliveries, hernias, appendicitis, and gall bladder disease have shown less than 5 percent under-reporting, while episodes involving mental or per- sonality disorders have shown under-reporting of 32 percent (Health Statistics, Series D, No. 5). In general, episodes classified as non- threatening, that is, as not embarrassing or detrimental to the in- dividual’s self-image, showed an under-reporting rate of 10 percent in this study of hospitalization data. Episodes rated as threatening were under-reported 21 percent. Further, for the episodes classified as threatening which were reported, the report of the diagnosis was altered to make it more palatable or more acceptable. Other fields are replete with examples of this type. In economic studies, individuals with small savings accounts tend to overstate their size, while people with particularly large accounts tend to understate them. Men in high income groups are less likely to admit having borrowed from a cash lender than those in the lower income groups. All of these data indicate that obtaining accurate responses re- quires more than merely approaching a respondent with a set of questions. The researcher, recognizing the great value of his research, somehow expects the respondent to be equally eager to make the study a success. Yet, for many studies, there is little reason apparent to the respondent for divulging information to the interviewer. The interviewer’s request for information does not mean that the respond- ent shares the researcher’s goals even to the extent of being willing to think carefully about each question. The interview is a complex interaction of forces. The respondent has an image of himself as a particular kind of person; he has a set of social norms which tell him what behavior is appropriate and what is inappropriate. Thus, the mother may be reluctant to admit to particular kinds of illnesses on the part of her child because, to her, it implies that she has in some way been a poor mother. The husband may not be willing to report that he has been sick or has had to stay home in bed, because his image of himself is that of a healthy, self- reliant individual. Any admission of illness may be, to him, a sign of weakness or dependency. A person may be quite willing to report his appendectomy but, because of social norms, most reluctant to talk about his venereal disease. 182 In some cases, however, the problem of motivation is simpler than this. The individual can report recent conditions which he has suffered but may still be unwilling to expend the effort to ensure the accuracy of dates and other relevant information. The reason for such inaccuracy is that the respondent may not share the researcher’s goals or appreciate the relevance of his participation in the interview. This problem can be solved by generating forces strong enough to overcome the negative factors. Fortunately, much of the information in which the epidemiologist is interested does not involve forces which are so negative that the task of establishing strong positive motives is impossible. You may feel that I have presented a dismal picture of the poten- tialities of surveys as a useful device in epidemiological studies. You may think further that this is an odd point of view for a person whose main activity during the past several years has been in survey research. I took this approach because survey research techniques are deceptively simple. Too many people have criticized surveys because they have used them improperly. All too often they have decided to do a study, sat down and thought up a few questions, sent out a couple of laboratory assistants to take interviews, and then have been surprised that the data they collect are not accurate. Then they tend to condemn unjustly survey research in general. My purpose has been to point out the importance of understanding the variables of reporting, and to stress the importance of using adequate techniques. Survey research is a method of measurement. As such it has the strengths and weaknesses of most measuring instruments. It also has appropriate and inappropriate applications. One can expect to get some things but not others from it. Since survey research is a relatively new technique, there is much about the methodology which we do not know as yet. In conclusion, I would like to emphasize the need for studies of methods and techniques designed to make survey research more valuable to epidemiology. The National Health Survey of the Public Health Service has an active program of methodological in- vestigations, and is to be commended for this attention. But more is needed. : There is need also for a facility for collating and evaluating the experience of researchers who have used survey methods so that others may have the benefit of their successes and failures. Minimally this means that each research report based on survey data should contain the questions and instruments used, as well as information on the types of interviewers employed and the techniques they used. Even- tually, and hopefully not too far in the future, one can envisage a facility for data retrieval to which researchers can turn for information on the methods and questions found most successful in obtaining data needed for a particular objective. 183 It is through research on methods and sharing of experience that survey research methods will become more accurate and will become more useful in epidemiological studies. REFERENCES 1. Goddard, Katharine E., Broder, G., and Wenar, C., 1961. Reliability of pediatric histories, a preliminary study. Pediatrics 28: 1011-1018. 2. Health Statistics, Series D, No. 4, 1961. Reporting of hospitalization in the heaith interview survey. U.S. Department of Health, Education, and Welfare. 3. Health Statistics, Series D, No. 5, 1961. Health interview responses compared with medical records. U.S. Department of Health, Education, and Welfare. 4. Wenar, Charles, 1963. The reliability of developmental histories—summary and evaluation of evidence. Unpublished mimeographed report, University of Pennsylvania School of Medicine. 184 A Trial of Methods for Collecting Household Morbidity Data * By KennerH R. WiLcox, Jr., M.D., Dr. P.H,, Division of Laboratories, Michigan Department of Health, Lansing, Mich. So far this morning, Dr. Kish and Dr. Cannell have presented some of the methods and problems of sampling and surveys. I would like to present a specific example of some of these methods and prob- lems. This small survey was primarily designed to compare three methods for collecting reports of acute illnesses. The purpose of presenting such a study in a meeting that is primarily concerned with chronic disease is to illustrate some of the aspects of household surveys that should be considered each time this method of collecting in- formation is used. Although some of the aspects presented are pri- marily related to acute illness surveys, some are common to most types of surveys. The study reported here was a pilot study to aid in the develop- ment of a morbidity surveillance system in Tecumseh, Michigan, as part of the Tecumseh Community Health Study. The staff of the Community Study furnished invaluable assistance in the planning and execution of the study. The surveillance itself is intended to supplement the extensive, detailed information on chronic diseases which has been collected for the past few years and which is still being collected. Since the survey method developed was to be used in a continuous surveillance of subsamples of the community, some basic requirements were imposed on it. The method chosen must be fairly simple and quick to use so that a reasonably large sample of the community might be reached each week. On the other hand, it must aim at a broad spectrum of illnesses so that unusual occur- * This study was done as partial fulfillment of the requirements for the degree of Doctor of Public Health from the University of Michigan, School of Public Health. The study was supported by grants G-26 and H-6378 from the National Institutes of Health, Public Bealth Service, U.S. Department of Health, Education, and Welfare. 185 735-712 0—65——13 rences of unexpected types of illness are not missed. The method must also obtain some basic information about the illnesses reported. Two household interviews and one daily record system were used in the study. The interviews were based on two different frames of reference for illnesses. The questions used in Interview A were based on modifications of activity that might be produced by illness as shown in figure 1. The names of the household members were placed in the spaces across the top of the form and Y or N circled in the columns below the names for the appropriate answers to all questions. A check was placed in the space below the names of the household mem- ber or members who served as respondents. Two general questions were asked about illnesses and accidents occurring during the prior four weeks for each member of the family individually. Then a series of questions were asked about “anyone in the family” con- cerning illnesses that caused them to go to bed, stay in the house, miss work or school, just feel “under the weather,” or go to a hospital. In addition, questions were asked about medicines taken or physicians Form A-2: Interview A NAME OF HOUSEHOLD MEMBER Mark R for respondent in box below name 1. Would you tell me about any times that Y N Y N Y N (you or person's name) were sick since about ? 2. Did__ (you, he, che) have any accidents or | Y N | Y N| Y N did hurt self in any way since 2 For example, did have any cuts, bruises, sprains or other injuries? 3. Were there any (other) times when anyone in| ¥ N | ¥Y N | Y N the family had to go to bed because they were sick or did not feel well since 2 L. Were there any (other) times when anyone Y N|]Y N[Y N had to stay in the house because they were sick or did not feel well since ? 5. Did anyone have to miss work or school Y N|Y N|[YN any (other) times (since ?) 6. Were there any (other) times when anyone Y N|Y N|[YN did not feel well or was under the weather but still could get around (since ?) T. Did anyone take any (other) medicines or Y N|Y NJ Y druge (since ), such as aspirin, ant- acids, home remedies or drugs prescribed by a doctor? (If YES, see other side) 8. Did anyone talk to a doctor or see a Y N|]Y N[Y N doctor for any (other) reason (since ?) (If YES, see over) 9. Did anyone (else) have to go to the Y N[{Y N| YN hospital (since ?) = Figure 1 186 visited in an attempt to elicit reports of other illnesses not yet men- tioned. The data about all illnesses reported were recorded on a separate form. The second approach to interviewing was based on a list of symp- toms and diseases modified from the work of Cartwright (Cartwright, 1959). Figure 2 shows a portion of the two sides of the form con- Form A-3: Interview B These cards have some common symptoms and sicknesses written on them. Would you take the cards and look at them one at a time. If you or any other person in the family has been bothered by the condition on the card since about , would you tell me. Mark R for res- pondent in box below name 1. Aches or pains in muscles or other parts of the body 2. Backaches 3. Unusual bleeding from any part of the body L, Breathlessness or shortness of breath Se Convulsions, seizures, or epilepsy attacks 6. Diarrhea ~ 2 =< Z, Te Dizziness or faintness Y N ¥ 8. Earache or ear trouble Y N|Y N 9. Eye trouble or trouble seeing Y N|]Y HU 10. Fever 11. Female trouble 12. Hay fever or asthma |< 2 = I 2 ~A + = 13. Headache 14. Heart trouble or heart palpitations FIGURE 2 187 Form A-3 (Continued) NAME 15. Indigestion, pain in stomach Y NJ Y or stomach ache 16. Infectious diseases or Y NJ|Y childhood diseases 17. Itching, skin rash or skin sores |Y N| Y¥ 18. Kidney trouble or trouble Y N|Y passing water 19. Pain in chest Y N| Y 20. Pain or swelling in any joints Y N| Y 21. Rectal trouble Y N| Y 22. Rheumatism or arthritis Y NJ Y 23. Runny nose, cold, sore Y N| Y throat or cough 2k, Sick to stomach or vomiting Y NJ[Y 25. Sleeplessness or unusual Y N|Y tiredness 26. Swelling of ankles Y N|¥Y 27. Trouble with teeth, gums Y nlvy or mouth 28. Unusual nervousness or Y NJ| Y irritability 29. Weakness of any part Y N| Y of the body 30. Accidents or injuries such Y NJ Y as cuts, burns, sprains 31. Has anyone used any medicines Y N Y or taken any shots ASK: Has anyone had any other ill- nesses or problems with their health that you haven't mentioned yet? COMMENTS : 188 taining the list of items. The responses were recorded on this form in the same manner as on the Interview A form. Each item was typed on a 3-inch by 5-inch file card and the deck of cards was pre- sented to the respondent at the time of the interview. As the re- spondent went through the cards one at a time, she indicated which member of the family had had the symptom or disease during the prior 4 weeks. If the respondent went through the cards too fast, the interviewer could ask those items, for which responses were not re- corded, from the printed list. In some cases, more than one symptom might contribute to a single illness. The illnesses elicited were entered on a separate form and the required data recorded. The daily record is shown in figure 3. The content of the form generally follows Interview A and is divided into four parts. The first three parts were answered with checks to indicate: 1) the occurrence of an illness or accident, 2) any changes in activities asso- ciated with such events, and 3) whether or not any medicines had been used or a doctor seen. The fourth part contained space for descrip- tion of the illness or accident that occurred. Each sheet was printed on legal size paper and contained sufficient space for 1 week’s entries for one household member. A sufficient supply of sheets to last 4 weeks for all of the family members was stapled in a folder, and a sheet of instructions stapled to the inside cover. The daily records were placed in the homes by interviewers who explained their use to the household respondent. One week later the interviewer returned to answer any further questions, and at the end of the period the diaries were picked up. The daily record used in this study is similar to the one used in the California Health Survey study in San Jose, California (Allen et al., 1954). The common form used to collect information about illnesses ob- tained from both interviews and the daily records is shown in figure 4. The two parts of this form were printed on two facing pages. Each illness was recorded on a separate line across both pages. The nature of the illness, the time of onset and cessation, consultation with a physician, and various types of disability were determined. The illnesses were coded and data processing cards were punched directly from these forms. A different arrangement appeared at the bottom of the form for recording accidents. The primary purpose of the study was to compare the ability of these three procedures to obtain reports of illnesses. The numbers of reported illnesses, or rates derived from these numbers, were used for the comparison. It was assumed that, for these acute illnesses, the method obtaining the most reports was the best. Although it ‘might have been valuable to investigate the contributions of the various component parts of the procedures and the effect of varying their order of presentation, such a study was beyond the scope of this survey. 189 061 Form A-S5: Daily record FOR (Name of person): TECUMSEH COMMUNITY HEALTH STUDY-DAILY HEALTH RECORD Week of: ==» EACH DAY, If any columns except 3 and 9 have been checked, describe what the trouble was: (a) For each sickness write in what was wrong. If a doctor was seen, tell what he said the trouble was. (b) For each accident or injury, be sure to tell what parts of the body were hurt, how each part was hurt, and how and where the accident happened. (c) If any medicines were taken, tell what the medicine was and what it was taken for. If any illness stays the same for more than one day, write in “Same”. (12) (Use the back of this sheet if more room is needed.) Give tne name of any doctor seen during this week: - EACH DAY, check (X) any wm3» EACH DAY, check (X) the columns below == EACH DAY, check of these three columns which apply: (X) in these which apply: Didn't carry on usual Carried on usual columns if: activities--had to: activities-- Yas sick, |Had an Felt as Stay Stay in Miss But not As well Any A doctor had an accident | well as in the touse | work able to as usual medicines | wag seen ailment or injury | usual haspital | at nee; | but natin | or do as or drugs or con- or did not [or felt bed school | well as were taken | tracted feel as well | effects of usual as usual previous accident or injury (1) @ (3) A 5 (6) 0 8 0 (10) (1) Mon, Tue. Wed. Thu. Fri, Sat. Sun. If in hospital as a patient this week, — give name and address of hospital: —_— Figure 3 TCHS-3/62 In an attempt to facilitate the comparison between the three pro- cedures, replicates of three families each were selected from the Tecum- seh population by matching on several family characteristics, such as, family size, number of children in the family, the age and education of the household head, the geographic location of the household within the city, etc. Each family in a matched group or replicate was randomly assigned to one of the three procedures. A total of 50 replicates were chosen so that 50 families would be contacted with each procedure. Families who had moved from the area or could not be reached were replaced from a list of alternates that had been prepared before the field work began. Four weeks was chosen to be the period over which information would be collected. The arrangement of the procedures and family groups is shown in table 1. At the end of the first 4-week period, both family groups were interviewed and the daily records placed. At the end of the second period, the first two family groups were interviewed again with the alternate procedure, and the daily records were retrieved. All of the interviews could not be done in 1 week so that there was some overlap in the actual weeks included in each period for different families. In the discussion of the results of the study, only the incidence of illnesses occurring during the 4-week periods, that is, those illnesses having their onsets within the period will be used. TABLE 1.— Arrangement of family groups and data-collecting procedures during two 4-week periods Family group I 11 III — Interview A Interview B | oooaaaee Interview B Interview A Daily record (C) Period » Before discussing the comparisons and analyses of the data, I would like to present the overall rates and types of illnesses experienced during the study. The rates for reported illnesses for the three groups of families is shown in table 2. The total incidence was approximately one-half of an illness per person per 4-week period. Note that the incidence of reported illness on the daily records was approximately twice that for either of the interviews. The illness rates obtained in the National Health Survey were compared with ours in order to determine whether or not the proce- dures used were as adequate as a standard method. The rate for those illnesses reported in our study that caused disability or a visit to a physician was expanded to obtain a rate for 13 weeks as shown in table 3. This rate was similar to the rates reported by the National 191 G61 1 3 uls le Iz 10011 [12 1314 l15]16 11718 (15]20 [21| 222324 page Name What was wrong?, or When did this|Does it stillipid talk to (illness, bother ? or see a Sex Would you mind telling me (attack, con- | IF NO doctor about about it? or dition) first|When did it |j¢? Age bother? [stop bother- | IF y5S. How did it bother ? ing ? What doctor How often was it? did it both- er you since ? (1) (2) (3) (4) (5) ce 25 [26 |27 128(29130 (31 (32133/34!35}136 37138{39 4014] 42 (4344 [45 6 47) 48. ug {SO oN N NN N N Y N Y N M F N \ I= NEE £61 Did it cause any change in daily activities? IF NO, STOP IF YE3: Did have to stay in bed In what way?|Did have Did have IF §0, to miss to stay in] STOP Did have to hat did the doctor say work the house? I[ YLS:|stav in the the trouble was?... (school)? hospital? did he give it a How jmedical name? How long? |How long? | long? |How long? Which hospital? (6) (7) (8) (9) | Qo) (11) 51152 (5354 {55|56(57|58|59 60 {61(62 63 {64 [65 [66 [67168[69 {70 171 [72 {73 | Y N Y N Y N Y N Y N Y I — Figure 4.—The Collecting Form Health Survey for the same time of year for the 4 years prior to this study. TABLE 2.—Incidence rates for all families in the study for monrecurrent illnesses having their onsets within the study period Number of | Number of Family group illnesses person- Rate* periods Li ccvnnunnnnmemimensnn vise es seen men ans 227 **453 50.1 ET 213 449 47.7 UL. sunvunmeimmi mais ie Oe SS ins 208 223 93.7 Total. . cociccicssnmesninrasonssa 648 1125 57.8 *Illnesses per 100 Derion-periods of observation. **219 persons during the first period ana 234 during the second. **#217 persons during the first period and 232 during the second. TaBLE 3.—Comparison of the present study to the U.S. National Health Surtey by rates of illness causing disability or a visit to a doctor during a 3-month perio March 15 through May Present study 1962 62.1% January April through through March June U.S. National Health Sur- | 1958 74.3 48.6 vey. 1959 68.9 54.1 1960 74.5 41.3 1961 63.0 45.2 *Rate adjusted from 4-week to 13-week period. Rates are illness per 100 person-periods. The types of illness reported were approximately the same as would be expected at this time of year, as shown in table 4. About one-third of the illnesses were respiratory, and about one-fifth were accidents. Gastrointestinal complaints were about one-tenth of the total illnesses. The specific infectious diseases contributed only a small proportion of the total illnesses. TABLE 4.— Rates for acute illnesses by category of illness Category of illness Number of Percent of Rate* illnesses total number Respiratory_._.__ 230 35.5 20.5 Accidents. _____ 127 19.6 11.3 Gastrointestinal. 74 11.4 6.6 Infectious. _______ - 37 5.7 3.3 OUNOE cocina mossrmmins sis sess rams 180 27.8 15.9 otal. conuaivinsnsossniissmsnnms 648 100.0 57.8 *Illnesses per 100 person-periods. The effect of the reported illnesses on activities is shown in table 5. Only one-third of the illnesses caused any change in the person’s activities, and only one-fifth of the illnesses were attended by a 194 physician. Of those illnesses causing some change in activity, about 40 percent caused the person to miss work or school or stay in bed, and about two-thirds of them caused the person to stay in the house when he or she would ordinarily be going out. It is clear that most of the illnesses reported were very minor. TABLE 5.— The distribution of acute illnesses by the amount of disability caused Yes No No data or not Total significant* Number Percent Number Illness caused person to: Change activities. -. - - ce comsemaness 220 34.0 416 12 648 Miss work or school - Foe 87 39.5 85 48 220 Stay in the house... 144 65.5 71 5 220 Stayin bed... --v- —_— 84 38.2 132 4 220 Seen. physielan. . .ccumvsnmumummmemmmns 127 19.6 516 5 648 *Not significant applies to those situations where the loss could not apply, e.g., a housewife could not miss work or school. The primary statistical comparison of the three procedures was done using all of the illnesses reported during the second period from the 48 replicates of families in which all three procedures were done. As shown in table 6, the daily record produced by far the most reports of illnesses. There was a significant difference between the three procedures when tested by both the analysis of variance using repli- cate families and Friedman’s nonparametric analysis of variance (Siegel, 1956). Both methods of analysis were used because the data showed significant departure from normality and inequality of vari- ance, even after logarithmic transformation. A significantly higher illness rate was obtained with the daily record than with either interview by the multiple range test of Duncan. Thus, in terms of total illnesses reported, the daily record was superior. TABLE 6.—Comparison of the rates for acule illnesses reported during the second 4-week period for the three procedures used in the study Number of Number of Number of Family group Form used illnesses persons in illnesses per group 100 persons Yims iis. Beem 116 225 1.6 Tvuunucmemsssmmmss Avponisinos 83 227 36.6 1 + 3 CT 208 223 93.3 TORE... omen rnr|emmnersssnnsns 407 . 675 60.3 In order to compare the two interviews more adequately, the 42 replicate pairs of families who were interviewed during both times with both procedures were analyzed, as shown in table 7. The rates for the two family groups were about the same. The rate for the first period was higher than that for the second period, which would be expected in the spring when the incidence of illness is generally declin- ing. The rate for Interview B is higher than for Interview A. The 195 analysis of variance using a replicated Latin square design showed the difference between the two interview procedures to be of only border- line significance (0.1>P>>0.05), (Snedecor, 1956), and a Wilcoxson matched-pairs signed rank nonparametric test on the replicate families was not significant (Siegel, 1956). This finding would indicate that if the use of this list of symptoms and diseases is the better method to elicit more illnesses, this study was not of sufficient size to demonstrate it. TABLE 7.—Comparison of the rates for acule illnesses reported during both 4-week periods for the two interview procedures Time period Family group Number of 1 2 Total rate persons Form Rate* Form Rate Loven mammaire 205 | A... 48.8 | B.______ 47.8 48.3 Tosusnuvessssnoninns 200 | B....... 60.2 | A ______ 41.3 50.7 Total........ 406 |_________ 54.4 | _________ 44.6 49.5 *All rates expressed as the number of illnesses per hundred persons. Total for form A: 45.1. Total for form B: 53.9. The remainder of the analysis was performed to determine the quali- tative aspects of the differences between the procedures used. Some analysis was also done to examine certain problems associated with surveys for illness. The changes in activity associated with the illnesses reported by the three procedures is shown in table 8. In this table, the number of illnesses reported during the second period on the daily records can be compared directly to the number of illnesses reported on the interviews. As noted above, the total number is considerably higher for the daily record. The number causing some change in activity is also higher, but this is probably due to a difference in the meaning of “change in activity’ in the daily record as compared to the interviews. For the various categories of changes in activities, the three procedures obtained about the same number of illnesses, except that the daily TABLE 8.—Changes in activity associated with the acute illnesses reported with the three procedures for the second period Family group II X III Formused.__.___________ A B C Number of persons 227 225 223 Total number of acute illnesses... ____________ 83 116 208 Number of illnesses causing: Change in activities. _____ pe 24 29 79 Missed work or school . _ 14 16 17 Confined to house... 19 23 25 Confinedtobed.._____________________ _ 13 16 8 Visit todoetor_______________________________ 28 23 28 196 record obtained fewer reports of persons confined to bed. Clearly, the daily record’s main value is for the collection of reports of minor illnesses. The difference in reporting of illnesses causing changes in activity with the two interviews was determined by comparing the number of illnesses reported during both periods for both interviews, as shown in table 9. Although a greater total number of illnesses were reported with Interview B, more illnesses causing various types of changes in activity were reported with Interview A. This can be explained by the fact that the orientation of Interview A concerned changes in activity associated with illnesses. TaBLE 9.—Changes in activity associated with the acute illnesses reported with the two interview procedures for both periods Family group. Both Both Form used... A B Number of persons... 406 406 Total number of acute illnesses_._._....._..__ 183 219 Number of illnesses causing: Changes in activities... _________________ 74 52 Missed work or school. 34 29 Confined to house..._. 64 43 Confined to bed... __ 37 31 Visth 10/80Ct0r . « . «cc cvucwinmmninmummmmn wan 54 38 Breakdown of the number of illnesses by category of illness is shown in table 10. As might be expected, the daily record produced more reports of illness in all categories except the specific infectious diseases. For both time periods, more illnesses were reported with Interview B for the four categories representing what would generally be con- sidered “illness.” This is probably due to the fact that Interview B asked specifically about various illnesses and symptoms without regard to severity. The differences in the reporting of specific infectious diseases probably represents real differences in incidence because these illnesses, most of which were measles, present a striking departure from normal health. By contrast to the general illnesses, more accidents were reported with Interview A during the second period. Accidents were asked about specifically for each individual at the TABLE 10.—The number of illnesses reported by each of the three procedures for different categories of acute illness POLIO. cuss mmm mmm oe ssp 1 2 Total 1 rE—— A B A B Family group I 11 II 1 III Number 6f Persons: .c-sscsmsessssms=sssn 219 217 232 234 223 1125 Total number of illnesses. _______._______ 109 125 88 118 208 648 Respiratory.......«ceu-e 48 53 26 46 57 230 Gastrointestinal 10 18 4 15 27 74 Musculoskeletal, mental, neurological, and special sense organs 8 20 12 24 48 112 Other... 6 11 14 14 23 68 18 5 8 6 0 37 19 18 24 13 53 127 197 beginning of Interview A; they were asked about “any member of the family’ at the end of Interview B. Again, the position and amount of emphasis placed upon the question seemed to affect the amount of reporting for these disorders. This study is too small to allow generalization, but the results seemed to emphasize the point that the manner in which the questions are asked and the emphasis of the questions may markedly influence the reporting of events. If the study utilizing interviewing is primarily focused on the impace of illness on people’s activities, questions about changes in activities should be used. If, on the other hand, the emphasis of the study is on the complete spectrum of illness, the ques- tions used should probe into the symptoms and illnesses by name. A good example of the effect of changing the emphasis is Dr. Cannell’s study on hospitalization in conjunction with the National Health Survey (Cannell, 1963). Marked improvement in the under-reporting of hospitalizations was obtained when the intensity of the questioning about these events was increased. In this study a household respondent was used for convenience. Since respondents usually do not report as completely for others as they do for themselves, the rates of illness reported on the interviews for self-respondents and respondents by proxy were compared as shown in table 11. With Interview A the rate of illness reported is about the same for respondents by proxy as for self-respondents. For Inter- view B the illness rate is higher for self-respondents than respondents by proxy. The reason for this difference cannot be ascertained from this small study, but two factors may be responsible: 1) asking the two general questions specifically about each member of the household might result in better reporting for others with Interview A; or 2) reporting of the respondent’s own minor illnesses is probably better than his reporting of other family members’ minor illnesses. It was observed in the Washtenaw County Accident Study that respondents reported more minor injuries for themselves than for others, but for major injuries the reverse was true (Velz and Hemphill, 1953). It was also shown that the rates were higher for those who responded jointly than for those who responded for themselves alone. The whole problem of obtaining information about other members of a household from a single respondent needs more study (Feldman, 1960). TABLE 11.— Rates of acute illness reported for self-respondents and respondents by proxy for both procedures and both periods Form Respondent Period 1 Period 2 | Both periods A Sell....cvvunoupan 45.1 43.8 44.6 Proxy..comsuivanes 51.2 40.8 47.2 B Sal cre sma 74.1 56.1 66. 4 Proxy... ___ 52.5 53.1 52.7 198 One of the problems associated with interviewing for illnesses is the failure of the respondent to report events because he does not remember them. One would expect illnesses occurring yesterday to be reported with more accuracy than those occurring 3 to 4 weeks ago. It is important, however, to obtain some estimate of the magnitude of this loss in reporting with the passage of time. One method which is useful over short periods of time is the comparison of the number of illnesses reported for the week prior to interview with the number reported in weeks more ramote from the interview. Such a comparison for these data using all of the interviews is shown in figure 5. The number of illnesses reported declines quite rapidly and the number reported for the fourth week prior to interview is about one-third of those reported for the week prior to interview. A loss of reporting of about the same magnitude has been reported in two other studies as shown in table 12. These studies done at different times and places and using different interview procedures show a remarkably similar loss in reporting for the four weeks prior to interview (Downes and Mertz, 1953; Nisselson and Woolsey, 1959). TABLE 12.— The decrease in reporting of illnesses with longer periods of recall in this study and two cthers Ratio of illnesses reported in prior weeks to illnesses reported for the week just prior to interview Weeks prior to interview Present Westchester | California study* County, Health N.Y. Survey S00. cps smear Sa A 0.80 0.75 0.73 TRE cc errrmmenc a fname Ens .45 .57 51 Fourth ______ .39 .37 .42 *Includes both interview procedures and both periods. This loss in reporting is not the same for all types or severity of illness. For example, hospitalizations are reported almost completely for a 4-week period (Cannell, 1961). The difference between the total illnesses reported and those illnesses causing some change in activity or a visit to a physician are shown in figure 6. During the first period, a rise in the number of illnesses during the second week prior to interview was caused by a brief epidemic of respiratory and gastrointestinal illness and the occurrence of several cases of measles in the population under survey. The number of illnesses for which a physician was visited was not influenced by this increase. During the second period the number of illnesses causing a change in activities or a visit to a physician did not vary from week to week, although the total number of illnesses shows the expected decline with increasing interval to interview. 199 NUMBER OF ILLNESSES BY WEEK OF ONSET IN RELATION TO THE WEEK OF INTERVIEW. TOTAL FOR ALL INTERVIEWS. = 1501 w wn Zz Oo Ww Oo Xx 0 = 100+ Su m n Ww wn n Ww 3 3 850 Ca Oo oc Ww m = = - 0 1 1 1 4 3 2 | WEEKS PRIOR TO INTERVIEW WEEK Figure 5 The relationship between the rates of illnesses for different weeks reported in the daily record was quite different than that reported in the interviews during the same time period, as shown in figure 7. The high rate observed during the first week of recording was probably due to initial enthusiasm and was followed by a drop in rate to the same level as that obtained by interviews for the week prior to the visit. The daily record gives more complete and consistent reporting of minor illnesses, but other problems connected with the maintenance of such records would have to be solved. Some people forget to keep the records and others do not enter enough information to allow a clear determination of the nature of the illness. Daily records will probably work best in studies of the complete spectrum of a limited group of diseases. For example, such records might be a good method to study acute respiratory illnesses. 200 NUMBER OF ILLNESSES BY WEEK OF ONSET IN RELATION TO THE WEEK OF INTERVIEW. ILLNESSES CAUSING A CHANGE IN ACTIVITY OR DOCTOR VISIT BY TIME PERIOD. A. PERIOD | 0 0 2 50 CHANGE IN y conn rr Been ACTIVITY ol — ToT ""¢<-DOCTOR VISIT 4 3 2 | WEEKS PRIOR TO INTERVIEW WEEK Ficure 6 One very important determinant of the success of a procedure is the performance of the persons actually doing the work, in this case, the interviewers. A method of assessing this effect on the results of procedures used is to examine the amount of variability between interviewers. In this study, the interviewers were assigned to the 201 735-712 0—65——14 NUMBER OF ILLNESSES BY WEEK OF ONSET IN RELATION TO THE WEEK OF INTERVIEW. COMPARISON OF THE INTERVIEWS AND THE DAILY RECORD. A. INTERVIEWS A+B B. DAILY RECORD (C) 40r BH Oo [ o——e TOTAL e— — - ALL CHANGE OF ACTIVITY o---o0 DOCTOR VISITS eens © CONFINED TO HOUSE 30+ Ww Oo T o ILLNESSES PER HUNDRED PERSONS ny Oo 20 bs L LL TN 10 10 Nf ss eS oi Be. Oe Ee a i Te or 0 8 4 3 2 | l 2 2 4 WEEKS PRIOR TO WEEKS OF INTERVIEW WEEK RECORDING Figure 7 families by a random process so that some such estimation of the variation between the results obtained by the interviewers could be made. The illness rates obtained by each of five interviewers with each of the procedures is shown in table 13. During the first period, the variances were fairly high for both procedures and were higher for B than for A. During the second period, the variances for both procedures were the same and were quite low. The variance for the daily record was quite high but this was mainly due to the contribution of one very low rate. In order to obtain a better estimate of the interviewer effect alone, the coefficient of intraclass correlation was computed for each of the procedures, as shown in table 14. Although the number of inter- viewers and interviews was small and the significance levels for rho were quite high, the rho values for the interviews done during the first period were clearly above the significance level and those for the second period were very low. The value of rho for the daily records was also low, which might be expected with this type of self-adminis- tered record. This would seem to indicate that, as the interviewers became more experienced with the interviewing procedure used, the 202 TABLE 13.—Rales of illness reported by the families interviewed by different interviewers Procedure-Time period Interviewer Overall rate A-1 B-1 A-2 B-2 C-2 0.348 0.139 0.342 0.375 1. 056 0.461 . 349 L710 . 263 .419 . 368 .420 476 .400 .314 .524 1.026 . 5562 . 500 .833 .643 . 5563 1.125 . 736 . 865 . 881 .333 L757 1.158 L791 .593 .379 . 526 L947 . 592 0448 . 0094 . 0227 .0221 .1074 . 028 variation between their results decreased. In any study involving interviewing, an estimate of interviewer variation should be done to help assess the reliability of the data collected. Such analysis will not only help judge interviewer performance, but in some cases will point to parts of the interview that need to be improved to obtain more reliable information. TABLE 14.—The interviewer variance for the three procedures during both time periods as measured by p which is the coefficient of intraclass correlation In the ‘analysis of variance table: : 2 the mean square for ‘‘among interviewers’ =sy+ks, the mean square for ‘‘within interviewers’ =sp Sa rT Period Procedure P 1 A 0.218 B . 287 A . 091 2 B —.011 C —.023 With 4 and 30 degrees of freedom and k=8: p=.05 when p=.175 p=.1 when p=.125 One other area in relation to interviewing is difficult to assess: the integrity of the interviewers. The variation in the total number of persons under study, that you may have noticed in the various tables presented, was due to one interviewer failing to complete her assigned interviews. Such an incident is unfortunate and probably rare, but it points out the need to evaluate interviewer performance. The analyses presented here are some of the approaches that can be used to evaluate an interviewing procedure. Surveys for different purposes require different approaches. For example, if independent records of the events of interest are available, these will be valuable in evaluating the interview results. Regardless of the purpose for which an interview procedure is used, the performance of the pro- cedure should be evaluated and not taken at face value. 203 Interviewing people is an instrument or tool that we use to obtain information. Like any tool we should know how to use it and what to use it for. A close analogy can be drawn between interviewing and another type of tool that we commonly use, a laboratory test. Before instituting a laboratory test, such as serum cholesterol or blood glucose determinations, we first must determine the best specific method to use. Usually there are methods described in the literature, and some- times these have even been evaluated. On occasion, we may have to develop a new method. In either case, we must determine how the test performs in our own laboratories, that is, how accurate and re- producible it is in our hands, and what variation we can expect. Once the test is established, we want to be certain that our technicians are doing a good job. Although we don’t often admit it, it is a good idea to check out our technicians once in a while to be sure that the test results are as good as we think they are. When this type of in- formation is known, we can use the test and make meaningful state- ments about the results obtained. The interviewing procedures we use should be evaluated with the same care as our laboratory tests. Furthermore, when a study that has used a laboratory test is published, the methods used are described so that the reader can better evaluate the work, and, if desired, try to repeat it. Similar information should be published about the inter- viewing techniques used in any study. In summary, a small pilot study was presented in which three pro- cedures for obtaining morbidity data from households have been com- pared. Two of these procedures were interviews; the third was a daily record. Each procedure was randomly assigned to one group of 50 families. All comparisons presented were based upon the acute illnesses having their onsets within the 4-week periods of observation. The highest rate of illness was reported on the daily records. More illnesses were reported in the interview that used a list of symptoms and diseases than in the interview that used questions about changes in activities, but the difference was of questionable significance. The rates for illnesses causing some change in activity were about the same for all three procedures. The problems presented by the use of a household respondent, the decline in reporting of illnesses with the passage of time, and the variation in results obtained by the inter- viewers were discussed. REFERENCES 1. Allen, G. I., Breslow, L., Weissman, A., and Nisselson, H. 1954. Interview- ing versus diary-keeping in eliciting information in a morbidity survey. American Journal of Public Health. 44: 919-927. 2. Cannell, C. 1961. Reporting of hospitalization in the health interview survey. Health statistics from the U.S. National Health Survey. United States Public Health Service Publication No. 584-D4. 204 10. . Cannell, C. 1963. Comparison of hospitalization reporting in three survey procedures. Health Statistics from the U.S. National Health Survey. Public Health Service Publication No. 548-D8, U.S. Department of Health, Education, and Welfare. . Cartwright, A. 1959. Some problems in the collection and analysis of morbidity data obtained from sample surveys. The Milbank Memorial Fund Quarterly. 37: 33-48. . Downes, J., and Mertz, J. C. 1953. Effect of frequency of family visiting upon the reporting of minor illnesses. The Milbank Memorial Fund Quarterly. 31: 372-390. . Feldman, J. J. 1960. The household intérview survey as a technique for the collection of morbidity data. Journal of Chronic Diseases. 11: 535-557. . Nisselson, H., and Woolsey, T. 1959. Some problems of the household interview design for the U.S. National Health Survey. Journal of the American Statistical Association. 54: 69-87. . Siegel, S. 1956. Nonparametric statistics for the behavioral sciences. McGraw-Hill Book Co., Inc., New York. . Snedecor, G. W. 1956. Statistical methods. Iowa State University Press, Ames, Towa. Velz, C. J., and Hemphill, F. M. 1953. Home injuries: investigation and application of home injury survey data in development of prevention procedures. University of Michigan School of Public Health, Ann Arbor, Mich. 205 PART IV Current Epidemiological Problems . - Intrafamilial Transmission of Rheumatoid Arthritis By Smioney Coss, M.D., Survey Research Center, University of Michigan, Ann Arbor, Mich. “Perhaps the emphasis should be directed to environ- mental factors and to ‘hereditary habits’ rather than to hereditary tendencies’”’ (Smith, 1928). The above quotation comes from a paper on peptic ulcer, but it seems possible that it is equally applicable to rheumatoid arthritis. The literature on the familial aggregation of this disease has been adequately covered elsewhere (Edstrom, 1941; Stecher, 1957; MecKusick, 1959; Blumberg, 1960; De Blecourt et al., 1961; Lawrence, 1962; and Burch, 1963). Almost all studies bring out evidence for familial aggregation and it is usually assumed that the cause of this is genetic rather than environmental. At the present time the evidence is inadequate to support any simple genetic hypothesis. However, the genetic notion about this disease has persisted for about 100 years; therefore it is likely that there will be as much difficulty in eradicating it as in confirming it. What is needed is a strong alternative hypothe- sis for comparison with this genetic hypothesis. The best alternative appears to be the psychosocial hypothesis. First, it is obvious that many things, in addition to genes, are passed on from father to son. Included among these things are property, religious preference and political affiliation. Of particular present relevance is the fact that personality characteristics and neurotic response patterns may be thus transmitted (Thorne, 1957; Ehrenwald, 1963). The reason for this relevance is the accumulating evidence that certain personality characteristics are fairly regularly associated with rheumatoid arthritis (King, 1955; Cobb, 1959; Moos, 1964). If, as has been suggested (Cobb, 1963a), the probability of this disease becoming manifest is increased by the interaction of these specific personality characteristics with relevant environmental situations, 209 then the rather small degree of familial clustering might possibly be explained without any genetic influence. Briefly, incompletely and somewhat imprecisely stated, the theory of the psychosocial contribution to rheumatoid arthritis runs as follows. A sensitive, lively, responsive youngster, who is raised by a family in which the parent of the same sex is intolerant of aggression and excessively and unfairly punitive or critical, may learn to hide hurt feelings, bury aggressive responses and may fail to develop a stable self-evaluation. When later in life such an individual is put into a strong conflict situation that arouses anger and threatens self-esteem, the likelihood of active arthritis is increased. Other lively children of the same family may become angry, aggressive and rebellious. When later in life such persons are forced, by the power structure of the social environment in which they find themselves, to curb their anger and its expression, they are likewise more vul- nerable to the disease. Thus we see that two kinds of personalities can be generated by the same parents and that both of these personali- ties may, under appropriate, but different circumstances, fill up with the negative self-evaluations and resentment which seem to con- tribute to rheumatoid arthritis. Since accumulation of resentment can lead to explosive and unexpectedly punitive behavior, the children of the person in question can be exposed to the same psyghological environment. Thus there might be a propagation from generation to generation giving a semblance of genetic transmission. Though these attitudes and behaviors have been demonstrated to be related to rheumatoid arthritis, as mentioned above, an adequate connection to the disease via physiological mechanisms has not been established. We do, however, know enough about the physiological alterations associated with emotional states to make a fairly strong assumption that relevant physiologic processes exist. It is therefore an important research task to seek out the relevant mechanisms, for we will never be satisfied with a causal chain until all the links have been identified. The present purpose is to see if we can explain the family clustering, not the entire causation of the disease. Our approach to this explana- tion is to select certain areas for study where the deductions from the two hypotheses, genetic and psychosocial, might give sharply contrast- ing results. So far six such areas have been identified. They are laid out in table 1. The first two areas are ones with regard to which there isalready someinformation. First, Burch (1963), has found little, if any, evidence of familial clustering in a tribe of American Indians in which the frequency of the disease is high. Second, the implication of re- cent studies of an employed population under continuous observation for 28 months is that probably everyone is susceptible to this disease and in fact will probably have at least a brief attack if he lives long enough. Furthermore, it is noted under No. 6 that husbands and their wives are both affected more commonly than one would expect by 210 chance. Each of these pointsis consistent with the psychosocial hypoth- esis, though they by no means prove it. Of the remaining areas we will focus only on No. 6 which is the principle subject of a current investigation. Incidentally information on Nos. 4, 5, and 6 will be obtained also. Table 1.—A comparison of two hypotheses about the intrafamilial transmission of rheumatoid arthritis Genetic Hypothesis Psychosocial Hypothesis Facts — . Family clustering should be found in all cultures. ~ . Susceptibility should be limited by gene frequency. 3. Given a propositus with RA the probability that a given relative will have RA should be dependent on the degree of consan, ty. 4. Given two siblings, one with RA and the other without, the choice of which is affected should be inde- pendent of the characteristics of the spouse and the marital inter- action pattern. 5. Given a propositus with RA, the probability of a sib having RA should be more dependent on the RA status of the parents than on any other characteristic of the parents. 6. Given a propositus with RA any excess frequency of RA in the spouse should be attributable to a mate selection effect and should show up in a sibling’s spouse. In those cultures in which the sources of resentment and depres- sion lie primarily outside the fam- ily there should be little familial clustering. Susceptibility should be unlimited. Given a prospositus with RA the probability that a given relative will have RA should be dependent on his psychological and social characteristics. Given two siblings, one with RA and the other without, the choice of which is affected should be re- lated to the characteristics of the spouse and of the marital interac- tion pattern. Given a propositus with RA the probability of a sib having RA should be more dependent on the punitive characteristics than on the RA characteristics of the par- ents. Given a propositus with RA the prousbiliy that the spouse will e affected should be greater than the general expectation and be de- pendent on the nature of the mari- tal interaction pattern. Little family clustering in jmerieen Indians (Burch, Susceptibility appears un- limited (Li i & Cobb, Swaim (1962) has strong clin- ical impressions that the marital interaction pattern is important. Spouses are affected more commonly than one would expect by chance (Holsti & Rantasalo, 1936; Schmid & Slatis, 1962; Cobb, 1963b). Before going on to this matter it is appropriate to review our present picture of rheumatoid disease. There seem to be eight relevant points. 1) This is a systemic disease that may affect essentially every organ in the body. It is commonly called rheumatoid arthritis because most of the complaints are referred to the joints. purposes, persons who have the disease without joint symptoms are probably rare. Fortunately for our 2) It has a continuous and very gradual severity gradient. In the first place, it seems likely that everyone is susceptible though perhaps not equally so (Lincoln and Cobb, 1963). In the second place, it is very common and only a small percentage of those who have episodes of the disease are significantly disabled. 3) It is basically episodic in nature. Only the severe cases are con- tinuously affected for long periods of time. This makes for real diffi- culty in diagnosis because a single examination will reveal only a small proportion of those who are intermittently affected. 4) It increases in frequency with age. 5) There is no difference in frequency in severe or in total disease between the sexes. common in middle-aged women. However, moderate joint complaints are more 211 6) The disease is more prevalent in spring and fall than at other seasons of the year. 7) The etiology is unknown and is probably multifactorial. 8) This is a disease that people like to talk about for there appears to be no social stigma attached to it. In fact many of its victims ap- pear to be benefited by talking to a sympathetic and patient person. On the other hand, authoritarian individuals find victims uncooperative. This tells us something about the selection and training of interview- ers for the study which we are proposing. THE DESIGN The design of the study involves people having three degrees of consanguinity and four kinds of relationships. Specifically the charac- teristics of the spouse, a sibling, a cousin and an individual unrelated to the propositus will be studied. Since the characteristics include the marital situation, it will be necessary to interview the spouse of each. Since the interview method will be used, the sample will be limited to people who can speak and read the English language. Fur- thermore, since the sample will probably not be large enough to ana- lyse Negroes separately, only Caucasians will be included. The final sample will probably be composed of equal numbers of propositi selected a) from teaching hospital arthritis clinics, b) as severe cases in screening from national samples, and ¢) as mild cases from the national screening operation. The propositi from the national samples will be selected on the basis of the index of rheumatoid arthritis and three additional questions pertaining to diagnosis, severity and duration. The relevant questions will be found in appendix A which is the screening instrument used in the first two national sample surveys. These surveys are ones that are conducted by the Survey Research Center for a variety of purposes at least four times a year and it is possible to add a few questions to each of these at moderate cost, where doing national screening for the purposes of this study alone would be prohibitively expensive. Each of these surveys is of a ran- dom sample of the continental United States and is drawn on the basis of a stratified area sampling technique (Kish and Hess, 1959). It is appropriate at this point to ask about the validity of the index of rheumatoid arthritis as a screening instrument. As mentioned above, this index is composed of the first three questions in appendix A. Tt has been independently validated in two earlier studies (Rubin et al., 1956; Cobb et al., 1959) and it has been demonstrated that its sensitivity for probable plus definite rheumatoid arthritis lies be- tween 66 percent and 77 percent, and its specificity between 95 per- cent and 98 percent. These screening characteristics are at least as good as those of the best screening tests for diabetes (Diabetes Pro- 212 gram Guide, 1960). In order to compensate for the slight lack of specificity, that is to eliminate the false positives, the second three questions in appendix A were added. These make it possible to ex- clude persons with other forms of arthritis, such as gout, and persons whose apparent joint swelling is due to peripheral edema. In addi- tion they allow us to eliminate cases who have had but a single joint swollen or who have had only brief episodes of joint swelling. At subsequent interviews greater detail will be obtained in order to get at further reasons for exclusion and to measure degree of activity and severity. Actually there will be three interviews at 4-month intervals. This will eliminate the seasonal variation and make it possible to estimate proportion of time in episode of active disease according to the method of Beall and Cobb (1961), and to collect the necessary data on the psychological and social factors. Each family cluster will be made up of the 1) propositus with rheumatoid arthritis, 2) his spouse, 3) a sibling, 4) a maternal cousin, 5) a paternal cousin and 6) an unrelated individual. Individuals Nos. 3-6 will be selected for being within 5 years of age of the spouse of the propositus so that there will be no more than 10 years difference in ages between the cluster members and the difference in age between the propositus and the cluster members will be approximately con- stant. We have found that it is not possible to select any significant number of clusters in which all the cluster members are of the opposite sex from the propositus which would eliminate the need for adjust- ment for differences between the sexes, or separate analyses by sex. In all probability the more appropriate of these two approaches will be an adjustment for sex because, as noted above, it is believed that the fundamental frequency of rheumatoid disease is about the same in the two sexes but the arthritic manifestations are more prominent in middle-aged females. In an ideal cluster all of the individuals will be married and living with their spouses and clusters will not be accepted unless the pro- positus, the sibling, the unrelated individual, and at least one of the cousins is living with the spouse. Present intentions are not to accept any propositi under 35 and it seems unlikely that there will be many propositi over 65 because most persons of this age will have lost one or more relevant persons from the cluster by death. Obvi- ously, the older clusters will be the most desirable for in them the penetrance will be most nearly complete. Very preliminary infor- mation from the first screening operation suggests that we will be able to locate slightly over 10 ideal clusters per 1,000 adults inter- viewed. The procedures for sampling patients under treatment have not yet been worked out but they should be relatively simple and will be focused on members of the American Rheumatism Association who have teaching hospital appointments. Optimum sample size 213 has not yet been determined, but it is estimated that it will be of the order of 400 clusters. MEASUREMENT In order to determine which of the proposed hypotheses is more correct it will be necessary to measure each member of each cluster on three dimensions: the degree of consanguinity, the degree of rheu- matoid arthritis, and the psychological and social characteristics. We can quickly dispose of the problem of measuring consanguinity by realizing that siblings on the average share % of their genes while cousins share only ¥% of their genes, and presumably the unrelated individuals share only a negligible proportion of their genes with the propositus. This gives us three degrees of consanguinity and if by chance we should find any instances of double cousins a fourth degree will be available. For present purposes it is appropriate to assume that the errors of accepting the apparent paternity are unim- portant in comparison to the measurement errors that will appear in the other variables. The instrument used for the identification of the members of the cluster is presented as appendix B. The measurement of rheumatoid arthritis will be divided in three parts. The first is the determination of the diagnosis. Here table 2 is relevant for it shows the relationship of various measures to the most accurate diagnosis possible on persons under continuous obser- vation for 28 months. This table speaks for itself and demonstrates clearly that the rheumatoid arthritis index used on three separate occasions is a better measure of rheumatoid arthritis than any sero- logical test or single clinical examination. The reasons for this are that rheumatoid arthritis is a remittant disease which is by no means all discovered at a single examination, and the best serological tests are as yet insufficiently sensitive to be really useful in a study of this sort. The specificity of the index will be increased by using a set of questions designed to exclude other rheumatic diseases and joint swelling due to peripheral edema. This should give specificity ap- TABLE 2.— The sensitivity and specificity of various measures of rheumatoid arthritis for the best clinical diagnosis that can be made after monthly screening for 28 months For definite RA For probable and definite RA Measure Sensitivity Specificity Sensitivity Specificity Serology IRLIP?. cu vuiecrmmoummrmrsnsmess 28 91 14 92 Sirol PLATE, cor ccnsnssmnisnansaamasis 50 90 30 93 Single Exam Prob. & Def........ coun... 13 100 16 100 S ngle Exam Poss. & Prob. & Def.____________ 69 89 51 92 RA index positive at least one out of three times... 75 91 53 95 *Brooks and Cobb, 1963. 214 proaching that of the best clinical evaluations which are also subject to some false positives, as pointed out by Dixon (1960). The second part of the measurement of the arthritis is the determi- nation of the degree of activity. This will be determined from four things: the number of joints swollen, the duration of morning stiffness, the amount of aspirin and related drugs consumed, and the proportion of time in episode with active disease. The third part is the measure- ment of the degree of disability resulting from the arthritis. This will be estimated from a series of questions about specific musculo-skeletal functions, as well as questions about time off duty, time in the hospital, frequency of visits to physicians or clinics and need for special treatments or appliances. Finally these three measures will be put together with appropriate weights to give an RA score which will hopefully have quite a wide range and considerable validity. Its validity will be checked on the propositi sampled from persons under treatment for this disease. It is not within the scope of this paper to discuss the psychosocial variables in detail, however, it is appropriate to point out that the general characteristics outlined in the psychosocial hypothesis will be studied in five categories: demographic characteristics, preception of parents, personal characteristics, marital interaction and general environmental frustration. Insofar as possible, well-established psychological measures will be used, but for certain items it will be necessary to design appropriate measurement techniques of our own. The reliability of measures of this sort is for the most part very high. The difficulties lie in the area of validity. Take for example the index of impulsiveness versus control, which has been shown to be signifi- cantly related to rheumatoid arthritis (Cobb et al., 1963). To begin with, the concept of impulsiveness is ill-defined and there is no standard measure of this. On the other hand, the questions used all have good face validity in that they are direct and apparently relevant, as the true-false question “I often act on the spur of the moment without stopping to think’; and they are all modestly intercorrelated. This means that they can be used to make a scale which is reasonably satisfactory though it may not have equal intervals (Likert, 1932). Since there is now a fairly substantial literature on the psychological and social factors in rheumatoid arthritis, it seems likely that sound and appropriate measures can be worked out that will cover the relevant variables sufficiently well to give significant results. It is useful to remember that in every field of measurement we inevitably fall somewhat short of ‘““God’s opinion.” The important thing is to determine if a given measurement is sufficiently valid and reproducible to contribute to scientific progress. The interviews will be conducted by the regular members of the Survey Research Center’s nationwide interview staff, using carefully worked out and pretested forms. Since interviewer bias could give a 215 semblance of familial clustering if each cluster were to be assigned to a single interviewer, an attempt will be made to distribute the inter- views in a cluster among several interviewers. Coding will be done in accordance with the usual standards of the center (Muehl, 1961). Transfer to IBM punch cards will make possible the complex analyses to follow. ANALYSIS In table 3 the three main lines of analysis are laid out. The first analytical job will be to put together the RA score. Then we will assemble the various psychological and social measurements into an index indicative of the psychosocial hypotheses stated above. This will be done a prior: on the basis of knowledge from previous studies rather than on the basis of their value in identifying persons with rheumatoid arthritis in this study. This will avoid the possibility of developing a spuriously strong relationship to these items. It will be necessary to work out a procedure for adjusting for the differences in reported arthritis between the sexes. This should not be difficult if a sound decision can be arrived at as to whether to use an additive or a multiplicative model. In addition it will be necessary to correct for ascertainment (Crow, 1963) and check for bias in the selection of cousins. Procedures for these purposes are being worked out. TABLE 3.— The analysis A. Mean RA score by consanguinity and psychosocial score: Degree-of-Psycho- social Characteristics Relationship Overall High Medium Low Sibs______ Cousins. _ Isolates. _ Overall . Analysis of variance in the same cells. . Mujtipls regression analysis using the following continuous variables: a. RA score. b. Degree of consanguinity. c. Psychosocial score. aw The first analysis will be to examine the mean arthritis score for the several cells in table 3. Then an analysis of variance of the rheumatoid arthritis score will be undertaken for the same table. It is recognized, of course, that this is fraught with the usual dangers of doing an analysis of variance in a situation in which the assignment to the various cells has not been made at random. However, the results are likely to prove instructive. Finally a multiple regression analysis will be undertaken using the three continuous variables: rheumatoid arthritis score, degree of consanguinity, and the psychosocial score. Separate analyses will be done by source of case, whether from the patient sample, from the national sample of severe cases or the national 216 sample of mild cases. Then all types of cluster will be analyzed together. DISCUSSION Continuing to focus only on hypothesis No. 3 in table 1, it is per- fectly clear that there is a substantial chance of an indeterminate answer when in fact the true answer should be primarily genetic or primarily psychosocial. The likelihood of this is reduced by in- creasing the validity of our measurements and by increasing the sample size (Rubin et al., 1956). Even with the best measures and optimum sample size it would be unrealistic to expect an answer on this deduction which says more than ‘“‘the facts seem to be more in line with the genetic hypothesis or more in line with the psychosocial hypothesis.” At this point examination of all the deductions will play a useful role for if the results are consistent across a whole series of deductions the conclusion will be much firmer. The results of this study will not absolutely exclude either hypothesis, though they might suggest that the contribution of one or the other is sufficiently small to be disregarded. Of course, it is possible that both hypotheses are correct and each accounts for a substantial portion of the variance, in which case the proportion of each will be estimated. This design gets us out of the problem of studying all the members of a large kinship which in fact may be thoroughly atypical. By contrast it puts us in a situation in which we get bits of data from a wide distribution by social class, subculture and geographic area. This means that any conclusions that are drawn from this kind of a study can much more likely be generalized to the population than are those derived from intensive studies of single kinships. Surely this general kind of design is applicable to a wide variety of genetic and other intrafamilial problems. SUMMARY Rheumatoid arthritis appears to cluster in families. This may be due to genetic or interpersonal factors or in part to both. A design for examining this problem has been presented. This design is applicable to a wide variety of similar problems. ACKNOWLEDGMENTS The assistance of Ernest Harburg, who managed the initial field work of this study, and of W. J. Schull, who is consultant to the study, is gratefully acknowledged. The research here reported has been supported in part by the fol- lowing grants from the National Institutes of Health: AM-06928, MH-K6-16, 709, MH-09177. 217 735-712 0—65——15 APPENDIX A AGE SEX RACE COOPERATION RATING. COL. A COL. B R1. Have you ever at any time had [OO YES (OR Oo NO arthritis or rheumatism? DON’T KNOW) R2. Do you wake up with stiffness [OO YES, USUALLY OO NO (OR or aching in your joints or OR SOME- DON'T muscles? TIMES KNOW) R3. Have you ever had any 0O YES O NO (OR swelling in any joints? DON’T KNOW) R4. INTERVIEWER—CHECK ONE) Rb. R6. R7. RS. Ro. R10. 218 0 ALL BOXES IN COLUMN A CHECKED (GO ON WITH Q. R5) 0 ONE OR MORE BOXES IN COLUMN B CHECKED (SKIP TO PAGE 47, Q. TI—DO NOT LEAVE ARTHRITIS LETTER OR FORM I) Do you think this swelling was due to arthritis? Rb5a. Why is that? What joints are or were swollen? Please tell me which ones. R6a. Any others? What was the longest period of time when your joints were continuously swollen? (INTERVIEWER—CHECK ONE) MY RESPONDENT IS MARRIED AND LIVING WITH SPOUSE (from Q. P2, Page 39) 0 YES—(GO ON WITH Q. R9) O NO—(SKIP TO PAGE 47, Q. T1I—DO NOT LEAVE ARTHRI- TIS LETTER OR FORM I) Next, we'd like to ask you, do you have any married brothers (NOT step- brothers or half-brothers)? O YES 0O NO—(GO ON WITH PAGE 45, Q. R10) R9a. What are the ages of those brothers who are now living with their wives? Do you have any married sisters (NOT step-sisters or half-sisters)? O YES O NO—(GO TO Q. R11) R10a. What are the ages of those sisters who are now living with their husbands? R11. (INTERVIEWER—CHECK ONE) 0 R HAS NO BROTHERS OR SISTERS WHO ARE MARRIED AND LIVING WITH SPOUSE—(SKIP TO PAGE 47, Q. T1— DO NOT LEAVE ARTHRITIS LETTER OR FORM I) 0 R HAS A BROTHER OR A SISTER (OR BOTH) WHO IS MARRIED AND LIVING WITH SPOUSE—(GO ON WITH Q. R12) R12. Do you have any first cousins? O YES—(CONTINUE WITH Q. R13) 0O NO—(GO TO Q. R14, BELOW) R13. Are these first cousins about your age, that is, within ten years of your age? 0 YES O NO—(GO TO Q. R14, BELOW) R13a. How many first cousins about your age are there on your mother’s side? OONE OTWO OTHREE @OFOUR OFIVE OR MORE R13b. How many first cousins about your age are there on your father’s side? OJONE OTWO OTHREE QOFOUR OFIVE OR MORE (GO ON WITH Q. R14) THE UNIVERSITY OF MICHIGAN INSTITUTE FOR SOCIAL RESEARCH ANN ARBOR, MICHIGAN RENSIS LIKERT, DIRECTOR Survey Research Center Angus Campbell, Director Research Center for Group Dynamics Alvin Zander, Director May 1963. A LETTER TO PERSONS WITH ARTHRITIS: One aim of the interview you have just completed is to find people with arthritis. Even if your arthritis is very mild, you can help a great deal in this nationwide survey. This survey is to find some of the causes of arthritis. Why is arthritis more common in some families than in others? To aid the millions who bear the pain of this illness, we must try to find answers to this problem. You can help in this search by telling us how to get in touch with other members of your family. We need to get the names and addresses of your brothers and sisters and some of your cousins. We will then select a few of them and talk with them about their health. Not many of us carry these addresses around in our heads, so I have asked your interviewer to leave the attached form with you. In this way you will have time to fill in the names and addresses for the study. Let me assure you that all the information you give us will be held confidential. Your help will be appreciated by many people—those who suffer from arthritis as well as their friends and families. Sincerely yours, Sioney Coss, M.D. Program Director. 219 APPENDIX B FORM I Page 1 1. Please print your name here (Give middle initial) sisters.) . First, we would like to ask you about your BROTHERS and SISTERS. (NOT your step-brothers or step-sisters. NOT your half-brothers or half- Simply write in the full name, age and present address of each of your brothers and sisters who are now living. 3. START HERE: 1) (2 (3) (4) Please write the full | How What is his or her name of each old present address? brother or sister, is STREET AND Is he or she married? one in each box. he, NUMBER (Check one) (Give married or CITY AND STATE name) she PHONE NUMBER OO Married, lives with spouse First name O Single Age O Widowed O Divorced or Last name PHONE NUMBER: separated 0 Married, lives with spouse First name Sei O Single Age O Widowed O Divorced or Last name PHONE NUMBER: separated 0 Married, lives with spouse First name 0 Single Age 0O Widowed O Divorced or Last name PHONE NUMBER: separated OO Married, lives with spouse First name — O Single Age 0 Widowed O Divorced or Last name PHONE NUMBER: separated (LIST ANY OTHER BROTHERS OR SISTERS ON BACK OF THIS PAGE) PLEASE GO ON WITH NEXT PAGE 6-63/RA/303 220 PART A Page 2 1. Now think about your father’s O (IF NONE, CHECK BOX AND brothers. GO TO PART B BELOW.) 2. Think about the children each of your father’s brothers may have had. These are your first cousins. O (IF NONE, CHECK BOX AND GO TO PART B BELOW.) 3. Choose one of these cousins who is closest in age to your wife or husband. a. My wife/husbandis years of age. b. The cousin I have in mind who is closest to this age is years of age, and is (check one) 0 MALE 0O FEMALE ¢. His/her full (married) name is: d. You can write to him/her at this address: NUMBER AND STREET: CITY, ZONE, AND STATE: (GO ON WITH PART B) e. The phone number is: PART B 1. Now think about your father’s O (IF NONE, CHECK BOX AND sisters. GO ON WITH NEXT PAGE.) 2. Think about the children each of your father’s sisters may have had. These are your first cousins. O (IF NONE, CHECK BOX AND GO ON WITH NEXT PAGE.) 3. Choose one of these cousins who is closest in age to your wife or husband. a. My wife/husbandis years of age. b. The cousin I have in mind who is closest to this age is years of age, and is (check one) O MALE 0O FEMALE c. His/her full (married) name is: d. You can write to him/her at this address: NUMBER AND STREET: CITY, ZONE, AND STATE: e. The phone number is: 5-63/RA/303 PLEASE GO ON TO NEXT PAGE 221 Page 3 PART C 1. Now think about your O (IF NONE, CHECK BOX AND GO TO mother’s brothers. PART D BELOW.) 2. Think about the children each of your mother’s brothers may have had. These are your first cousins. O (IF NONE, CHECK BOX AND GO TO PART D BELOW.) 3. Choose one of these cousins who is closest in age to your wife or husband. a. My wife/husbandis______ years of age. b. The cousin I have in mind who is closest to this age is. years of age, and is (check one): O MALE {OO FEMALE c. His/her full (married) name is: d. You can write to him/her at this address: NUMBER AND STREET: CITY, ZONE, AND STATE: e. The phone number is: (GO ON WITH PART D) PART D 1. Now think about your O (IF NONE, CHECK BOX AND GO ON mother’s sisters. TO NEXT PAGE.) 2. Think about the children each of your mother’s sisters may have had. These are your first cousins. O (IF NONE, CHECK BOX AND GO ON WITH NEXT PAGE.) 3. Choose one of these cousins who is closest in age to your wife or husband. a. My wife/husbandis_____ years of age. b. The cousin I have in mind who is closest to thisageis___ years of age, and is (check one): 0 MALE OO FEMALE c. His/her full (married) name is: d. You can write to him/her at this address: NUMBER AND STREET: CITY, ZONE, AND STATE: e. The phone number is: 5-63/RA/303 PLEASE GO ON TO NEXT PAGE 222 In each family there is usually a relative who is most likely to know the names and addresses of most of the members of the family. PART E 1. The person in my family who would most likely know the names and addresses of most of my FATHER’S family is: FULL NAME: STREET ADDRESS: CITY or TOWN: STATE: PHONE NUMBER: 2. The person in my family who would most likely know the names and addresses of most of the members in my MOTHER'S family is: FULL NAME: STREET ADDRESS: CITY or TOWN: STATE: PHONE NUMBER: Thank you for your kind cooperation. We will write to you soon, telling you more about our study. In the meantime, any comments you may have will be sincerely appreciated. Sioney Coss, M.D, Survey Research Center, The University of Michigan. PLEASE KEEP THIS FORM. OUR INTERVIEWER WILL RETURN TO PICK IT UP IN THE NEAR FUTURE. 5-63/RA/P.303 223 10. AL 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 224 REFERENCES . Beall, G. and Cobb, S. 1961. The frequency distribution of episodes of rheumatoid arthritis as shown by periodic examination. J. Chron. Dis. 14: 291-310. . De Blecourt, J. J., Polman, A. and De Blecourt-Meindersma, T. 1961. Hereditary factors in rheumatoid arthritis and ankylosing spondylitis. Ann. Rheum. Dis. 20: 215-223. . Blumberg, S. 1960. Genetics and rheumatoid arthritis. Arth. & Rheum. 3: 178-185. . Brooks, G. W. and Cobb, S. 1963. Studies in latex agglutination: An approach to the determination of optimum conditions for discrimination between rheumatoids and normals. Arth. & Rheum. 6: 198-207. . Burch, T. 1963. (This symposium.) Cobb, S. 1959. Contained hostility in rheumatoid arthritis. Arth. & Rheum. 2: 419-425. Cobb, S. 1963a. The epidemiology of rheumatoid arthritis. Bull. N.J. Acad. Med. 9: 52-60. Cobb, S. 1963b. Unpublished data from the Pittsburgh Arthritis Study. . Cobb, S., Chen, E. and Christenfeld, R. 1963. Psychological and social characteristics of persons with various psychosomatic diseases. (Read before Ann. Meeting Am. Psychosom. Soc., April 1963.) Cobb, S., Miller, M. and Wieland, M. 1959. On the relationship between divorce and rheumatoid arthritis. Arth. & Rheum. 2: 414-418. Crow, J. F. (This symposium.) Dixon, A. St. J. 1960. “Rheumatoid arthritis” with negative serological reaction. Ann. Rheum. Dis. 19: 209. Edstrom, G. 1941. Klinische Studien iiber den chronischen Gelenkrheuma- tismus. Acta Medica Scand. 108: 398-413. Ehrenwald, J. 1963. Neurosis in the family and patterns of psychosocial defense. A study of psychiatric epidemiology. Hoeber, New York. Holsti, O. and Rantasalo, V. 1936. On the occurrence of arthritis in Finland. Acta Med. Scand. 88: 180-195. King, S. H. 1955. Psychosocial factors associated with rheumatoid arthritis: An evaluation of the literature. J. Chron. Dis. 2: 287-302. Kish, L. and Hess, I. 1959. The SRC national sample of dwellings. Survey Research Center, Mimeograph. Lawrence, J. S. 1962. Genetic studies on rheumatoid arthritis. A.J.P.H. 52: 1689-1696. Likert, R. 1932. A technique for the measurement of attitudes. Arsh. Psychol. No. 140. Lincoln, T. A. and Cobb, S. 1963. The prevalence of mild rheumatoid arthritis in industry. J. Oce. Med. 5: 10-16. MecKusick, V. A. 1959. Genetic factors in diseases of connective tissue. Am. J. of Med. 26: 283-302. Muehl, D. 1961. A manual for coders. Survey Research Center, Ann Arbor, Michigan. Moos, R. H. 1964. Personality factors associated with rheumatoid arthritis: A review. J. Chron. Dis. 17: 41-55. Rubin, T., Rosenbaum, J. and Cobb, S. 1956. The use of interview data for the detection of associations in field studies. J. Chron. Dis. 4: 253-266. Schmid, F. R. and Slatis, H. 1962. Increased incidence of serum and clinical abnormalities in spouses .of patients with rheumatoid arthritis (Abstract). Arth. & Rheum. 5: 319-320. 26. 27. 28. 29. 30. Smith, D. A. 1928. A statistical review of gastric and duodenal ulcer. Brit. Med. J. 2: 293-297. Stecher, R. M. 1957. Heredity in joint disease. Documenta Rheumato- logica, J. R. Geigy S.A., Basel, Switzerland. Swaim, L. 1962. Arthritis medicine and the spiritual laws. Chilton, New York. Thorne, F. C. 1957. Epidemiological studies of chronic frustration-hos- tility-aggressive states. Am. J. Psychiat. 113: 717-721. U.S. Department of Health, Education, and Welfare. 1960. Diabetes Pro- gram Guide. PHS Publication No. 506. U.S. Government Print. Office, Washington. 225 Family and Genetic Studies of Rheumatoid Arthritis and Rheumatoid Factor in Blackfeet Indians By Tuomas A. Burce, M.D., M. P. H,, WiLLiam M. O’Brien, M.D., anDp JosepH J. Bun, M.D., Naticnal Institute of Arthritis and Metabolic Diseases, National Institutes of Health, Public Health Service, U.S. Depart- ment of Health, Education, and Welfare, Bethesda, Md. From 1958 to 1962 (Ziff et al., 1958; Lawrence and Ball, 1958; Bremner et al., 1959; Goldenberg et al., 1960; and Schmid and Slatis, 1962) five family studies on the occurrence of rheumatoid factor (RF) and rheumatoid arthritis (RA) in first-degree relatives (parents, siblings and children) of patients with seropositive and seronegative RA have been reported in the English literature from this country, England and Scotland. Only Bremner et al. (1959) failed to find a significant aggregation of RA and/or RF in families of seropositive rheumatoid patients. Unfortunately, important differences exist in the methods and material in these studies and in view of these fundamental differences, it is surprising that greater discrepancies in the results did not follow. The present study was designed to answer the following questions concerning rheumatoid factor and rheumatoid arthritis. 1. Is there a familial aggregation in a random sample of the selected population? This will be analyzed by contrasting the frequency of RF and RA in first-degree relatives of probands with and without RF and RA, with the frequency in a random sample of the population tested. 2. Is there any evidence of a simple genetic mechanism? This will be analyzed by comparing the observed frequency of RF and RA in offspring of phenotypically different types of parents, and in sibling 227 pairs with the frequency that could be expected from the various genetic hypotheses and by chance alone. The Blackfeet Indian population in Montana was especially suited for this type of survey for several reasons. Most of the relatives of any one individual lived in the same area; there was no reluctance to give truthful histories as to parent-child relationship and first cousin matings were very rare. MATERIAL In the fall of 1961 we conducted an area survey of Indians listed on the Blackfeet tribal roll, above age 29, residing on the Indian reserva- tion in Montana. Each respondent was questioned carefully and examined by one of three physicians. X-ray films of hands and feet were made. Serum samples were collected and tested for RF by both the Bentonite flocculation test (BFT) (Bozicevich et al., 1958) and the sheep cell agglutination test (SCAT) done by the method of Ball* (1950). From this survey appropriate probands were selected. To complete the family studies it was found advisable to return to the reservation several months later. The second session brought the completion rate of the population survey to 86 percent and the family study to 89 percent. FAMILY STUDY The probands and their first-degree relatives were classified into four groups as shown in table 1. Group A consisted of 22 probands with seropositive RA; group B had 34 individuals with a positive BFT but without RA; group C consisted of 20 Blackfeet Indians with seronegative RA and group D probands who were seronegative without RA; were matched with groups A and B probands by age and sex. An additional control group was obtained by randomly matching each of the first-degree relatives of the positive probands (groups A, B and C) with someone from the general population survey of the same age and sex without regard to their clinical or serological status. The prevalence of rheumatoid factor (tested by the BFT) and of rheumatoid arthritis in first-degree relatives of the four groups of probands and in the random sample are shown in table 1. The *In addition to the studies described above we are currently completing analysis of results of the following procedures carried out in connection with this survey: 1) X-rays of the cervical spine, hips and sacro-iliac Joints; 2) serum uric acid concentration; 3) major and minor blood group typing done by Dr. G. Albin Matson of the Minneapolis War Memorial Blood Bank; 4) ABO salivary secretor status done by Dr. E. J. Holborow of the Canadian Red Cross Memorial Hospital, Taplow, England; and 5) serum Gm groups determined by Dr. Paul Alepa in our laboratory. 228 differences between the various groups was no greater than would occur by chance (p>>.40 and >.90 respectively). When the SCAT was used instead of the BFT to test for RF the differences between the various groups were more marked but still were not significant. TABLE 1.—Prevalence of rheumatoid factor and rheumatoid arthritis in relatives of probands and in a random sample Probands 1° Relatives Group BFT RA* Number Number Positive Rheumatoid BFT arthritis* A positive. ________ Yes... __.__ 22 71 1 (1.4%) 2 (2.8%) Boisvresenovenns POSIIVEL cuca NOs aw smniin 34 77 4 % 2%) 3 (3.9%) Cc covemmsaiuige negative_________ Xos.couaniuis 20 63 4 (6.3%) 2 (3.2%) D negative... No... 55 179 5 2.8%)| 5 (2.8%) Matched controls... 211 12 (5.6%) 9 (4.3%) *Probable and definite RA. GENETIC ANALYSIS In addition to the family study just described we were able to take advantage of the completeness of the survey and analyze the data from the numerous kinships on the reservation to see if they would fit a genetic hypothesis. In order to apply the Hardy-Weinberg Law to offspring of genotypically different parents, the mating in the popula- tion must be random in regard to the characteristic under study and there should be no selection factor that would limit the size of the families. Comparison of the observed numbers of matings with neither, one, or both members affected with the numbers to be expect- ed on a random basis shows no significant difference for either RF or RA (p>>.95 for both) (table 2). Furthermore there was but little difference in the number of offspring from affected and unaffected parents. In fact spouse pairs in which one member had RA had more offspring over the age of 29 (1.57 per pair) than those with both spouses negative (1.05 per pair). TABLE 2.—Occurrence of rheumatoid factor and rheumatoid arthritis in spouse of Blackfeet Indians In both In one In neither spouses spouse spouse Rheumatoid factor: 2 32 218 1.28 33.43 217.38 1 20 254 0.44 21.12 253. 44 *Based on prevalence of 7.1 percent positive BFT in 252 couples (504 persons) tested. **Based on prevalence of 4.0 percent “probable” or “definite” RA in 275 couples (550 persons) examined. 229 The types of matings, frequency of matings and frequency of the various types of offspring in a large random mating population are shown in table 3. In this table A can be taken to represent the gene for RF or RA and a represents its allele but dominance is not neces- sarily implied. The gene frequency of A is p, and 1—p=q, the gene frequency of a. Using the frequency of RF and of RA in the Blackfeet population, the values of p and q have been calculated according to the Hardy-Weinberg equilibrium both as if the trait were a dominant and as a recessive. Frequency Dominant Recessive RF___________ 0. 05882 q?=0. 94118 p?=0. 05882 q =0. 97014 p =0. 2425 p =0. 02986 BA.cavnesunvs 0. 03966 q?=0. 96034 p?=0. 03966 q =0. 97996 p =0. 19910 p =0. 02004 TABLE 3.—Matings and offspring in a large random mating population Offspring Type of mating (reciprocals combined) Frequency of mating AA Aa aa BA KBE ocean mmm mama m—— pt pt AAX Aa pq 2p3q 2] AA Xaa___ — i 2p’ 2piq 4pq? pq? 2pq? pq? 4pq? 2pq? 2pq? qt qt 1 p? 2pq qQ? Ratio of offspring affected/total Mating phenotypes Dominant Recessive 3—2p AXA LX YR ER SR EER A ey 1.00 X A (pos. X pos.) @=p) 1 P A i TERIOR Y comms mmmmmie mmm mio mii es rn SE se te X a (pos. X neg.) ry PH p? 8 X 3 (HBF. KX DOL.)uu ui snawnwnnesenssivsesssasvinsvasmssssiosmnessmine sims 0 pe X a (neg g.) orni —p2 A LT 1000. K cs concmmisssssimissnssnniomsennnpormessammesessstms sn Hep p 2 8X? (MOG. X Too Tons g. X 7) p Tp The expected number of positive offspring for each mating class calculated both as a dominant and as a recessive trait are compared in table 4 with the observed numbers and the numbers expected if the distribution were entirely random. Chi-square tests show that only the random distribution fits the observed findings. Study of the inheritance of rheumatoid arthritis is complicated by the late and variable age of onset of the disease. In analyzing the offspring of phenotypically different types of parents, as above, it should be borne in mind that some of the younger generation might 230 TABLE 4.— Analysis of matings in Blackfeet Indians (Based on Hardy-Weinberg Law) Parents Offspring Expected number positive if— Total Observed Type of mating No. examined positive Dominant* | Recessive* Random inheritance | inheritance | distribution Rheumatoid factor: Neg. X neg. ._._____.__ 102 112 4 0. 000 2.603 5.419 12 15 1 4.547 1.786 0.725 1 0 0.453 0.610 0.048 127 288 16 7.644 12.119 13.935 9 18 0 8.356 3.880 0.870 X2 ® 6.826 1.699 Pp 0 0.076 0.64 112 128 4 0. 000 1.965 3.413 14 22 0 4.000 2.034 0. 586 130 204 10 2.960 7.730 10.723 13 35 2 9. 040 4.260 1.276 xX? ® 8.40 1.143 p 0 0.015 0. 56 *The theoretical frequency is estimated from the assumed gene frequency which is based on observed prevalence of R.F. (5.882 percent) and R.A. (3.966 percent) in the 1961 survey. **The theoretical frequency, adjusted proportionally to equal the number observed. have been destined to develop RA but this, of course, would not be apparent at the time the data were collected. In other instances, by the time the children reached sufficient age to develop RA, the parents might be dead. Obviously a method of analysis based on a single generation, who would be approximately the same age, would be a superior method of studying the inheritance of an age dependent trait. Such a method was devised by Penrose (1935) to study linkage and was extended by Cotterman (1937) to study unit factor inheritance. This is an exten- sion of the Hardy-Weinberg law to cover the special case of sibling pairs. The frequency of each combination of sibling pairs for all possible types of matings is shown in table 5 with the trait considered in one instance dominant and in another recessive. The number of pairs obtained from a family of size n is the number of combinations of n things taken two at a time or n(n—1)/2. Hence a family of 4 siblings would consist of 4X3/2=6 sibling pairs and 5 siblings would have 5X4/2=10 sibling pairs, etc. A total of 1,066 such sibling pairs were identified in the survey population. The number of pairs with neither, one, or both members affected is com- pared in table 6 with the number that would be expected assuming the trait were dominant, recessive or randomly distributed. As in the previous analysis the random distribution of affected individuals is the only one compatible with that actually observed. 231 TABLE 5.—Matings and sibling pairs in a large random mating population Frequency of sibling pairs in offspring Frequency Type of of Considering A as dominant Considering A as recessive mating mating RA-RA RA-Norm |Norm-Norm| RA-AR RA-Norm |Norm-Norm (S++) (8+0) (Seo) (S++) (S+0) (Soo) 1/2pq*(3+q) 1/4q*(1+-q)? 1/4p*(p+1)? 1/2p*(1-p) 3+p) TABLE 6.—Analysis of sibling pairs in the Blackfeet Indians (Based on Hardy-Weinberg Law and Penrose Formula) Expected number with— Type of pair Observed Dominant* Recessive* Random inheritance inheritance distribution Rheumatoid factor: Both negative________________________ 850 854. 836 847.171 850. 633 One positive. - - 85 52.214 64. 588 83.324 Both positive. ______________________ | 28. 950 21.241 2.040 X22 47. 597 23.782 0. 563 Pp <0. 00001 <0. 00001 0.45 Rheumatoid arthritis** Neither. __ 983 973. 562 996. 644 981. 442 One... 81 59. 466 54.160 82.828 Both. _ 2 32.91 15.196 1.737 X? 36. 980 24.936 0.081 Pp <0. 00001 <0. 00001 0.78 *The theoretical frequency is estimated from the assumed gene frequency which is based on observed prevalence of R.F. (5.882 percent) and R.A. (3.966 percent) in the 1961 survey. **Probable or definite R.A. according to the A.R.A. criteria. DISCUSSION It is thus apparent that at least among the Blackfeet there is no evidence of familial aggregation of RF or RA and that the observed data are consistent with a randomly distributed condition and not consistent with simple dominant or recessive inheritance. A complex type of inheritance due to two recessive genes with a low degree of penetrance, has not been excluded but the evidence makes such types of inheritance very unlikely. Secondly, there could be a rare in- herited form but, since there is no way of distinguishing this hypo- thetical inherited RA from the usual form of the disease, it is impossible to test this hypothesis. It is difficult to reconcile the conflicting observations reported by other workers and those reported here but they may well be due to 232 differences in methods and material. In none of the published studies on familial aggregation, for instance, were the probands, relatives and controls all from exactly the same population. In only two studies (Stecher et al., 1953, and Neri Serneri and Bartoli, 1956) was an at- tempt made to test genetic hypotheses and in both of these the method of ascertaining affected individuals among the relatives was inadequate. We have recently completed a similar survey on the Pima Indians of southern Arizona. A preliminary perusal of the data indicates that they also do not have a familial aggregation of RA or RF. If the final analysis of these data is similar to that reported here for the Blackfeet, it will be necessary to search for a nongenetic etiology for rheumatoid arthritis and rheumatoid factor. CONCLUSIONS 1. There is no evidence of familial aggregation of RF or RA in the Blackfeet Indians of Montana. 2. The frequency of offspring with RF and RA from “pheno- typically” different types of parents and in sibling pairs was con- sistent with a randomly distributed condition and not with a simple dominant or recessive inheritance. ACKNOWLEDGMENTS We wish to acknowledge with gratitude the valuable help of Dr. John S. Lawrence, Director, Empire Rheumatism Council Field Unit, who participated in the examination of the respondents. We are grateful to the Division of Indian Health of the U.S. Public Health Service and the Blackfeet Tribal Council for their effective assistance and encouragement. REFERENCES 1. Ball, J., 1950. Serum factors in rheumatoid arthritis agglutinating sensitized sheep red cells. Lancet 2: 520-523. 2. Bozicevich, J., Bunim, J. J., Freund, J. and Ward, S. B., 1958. Bentonite flocculation test for rheumatoid arthritis. Proc. Soc. Exerp. Biol. and Med. 97: 180-183. 3. Bremner, J. M., Alexander, W. R. M. and Duthie, J. J. R., 1959. Familial incidence of rheumatoid arthritis. Ann. Rheum. Dis., 18: 279-284. 4. Cotterman, C. W., 1937. Indication of unit factor inheritance in data com- prising but a single generation. Ohio J. Sci. 37: 127-140. 5. Goldenberg, A., Singer, J. M. and Plotz, C. M., 1960. Hereditary factors in rheumatoid arthritis. Arthritis and Rheumatism 3: 446. 6. Lawrence, J. S. and Ball, J., 1958. Genetic studies on rheumatoid arthritis. Ann. Rheumat. Dis. 17: 160-168. 233 7356-712 0—65——16 10. 11. . Neri Serneri, G. G. and Bartoli, V., 1956. Ricerche sui fattori ereditari del reumatismo cronico primario. Acta Genet. Med. (Roma) 5: 402-425. Penrose, L. S., 1935. The detection of autosomal linkage in data which consist of pairs of brothers and sisters of unspecified parentage. Ann. Eugen. 6: 133-138. Schmid, F. R. and Slatis, H., 1962. Increased incidence of serum and clinical abnormalities in spouses of patients with rheumatoid arthritis. Arthritis and Rheumatism, 5: 319. Stecher, R. M., Hersh, A. H., Soloman, W. M. and Wolpaw, R., 1953. The genetics of rheumatoid arthritis: analysis of 224 families. Am. J. Human Genet. 5: 118-138. Ziff, M., Schmid, F. R., Lewis, A. J. and Tanner, M., 1958. Familial occur- rence of the rheumatoid factor. Arthritis and Rheumatism 1: 392-399. 234 PART IV—Continued Longevity and Coronary Artery Disease Family Patterns of Longevity and Mortality* By Bernice H. Congn, Ph. D., Department of Chronic Diseases, Johns Hopkins School of Hygiene and Public Health, Baltimore, Md. INTRODUCTION Among the numerous defects in scientific research, two major types seem to predominate. On the one hand, many investigations are not original —they repeat the same studies, same methods, same ap- proaches. And, on the other hand, many researchers accept too readily, old, supposedly established concepts based on insufficient or unreliable evidence—concepts which by sheer repetition have become “axioms.” On just such an insecure foundation rests our present knowledge of the role of heredity in determining life span. Space does not permit a comprehensive review of the literature; a detailed, critical analysis will be published elsewhere. My purpose here is: Firstly, to define the problem and the assumptions that will be considered acceptable; Secondly, to refer briefly to some of the previous and current studies of family patterns of mortality, with particular reference to methodological problems; Thirdly, to outline two studies currently in progress in the Depart- ment of Chronic Diseases at the Johns Hopkins School of Hygiene, and to describe not only their objectives, but also, and primarily, their procedures: how they attempt to avoid some of the pitfalls of previous studies, what problems are involved in these current inves- *This investigation was supported in part by Public Health Service research career program award number 5-K3-GM-5590 (formerly GM-K3-5590; GSF-486) from the Division of General Medical Sciences and Public Health Service research grant number HE-06023 (formerly H-6023). Some of the computations were done in the Computing Center of the Johns Hopkins Medical Institutions which is supported by Research Grant FR-00004 from the National Institutes of Health. In addition, since this paper was presented, support for continuation of the Special Group Study has been received from the Public Health Service (Contract No. PH-86-64-18 and later Grant No. CD-00043-01). 237 tigations, and what improvements in methods for future planning might be suggested; and Lastly, to present some preliminary results from these studies. THE PROBLEM AND ASSUMPTIONS While it is clear that each species has its own characteristic life span and there is substantial evidence in experimental organisms for genetically determined variation within species (Pearl, 1928; Chali, 1959; Hartwig, 1959; Smith, 1959; Miihlbock, 1957; Comfort, 1956; Strong, 1936), one cannot extrapolate directly from experimental organisms to man, nor carry out studies of similar nature in man, where neither test matings nor laboratory controlled environmental conditions are possible. For the human geneticist, the direct epi- demiological approach is the realistic one. The first step in such an approach is to determine whether there are differences in human mortality that follow family patterns. If there is a genetically controlled variability within the human species (whether it involves hereditary disease susceptibilities, patterns of cellular degeneration, or somatic mutation rates), there should be familial aggregation in regard to length of life. Certainly, the demon- stration of familial aggregation does not imply a genetic factor unequivocally, since this may result from nongenetic causes. Never- theless, positive findings of familial aggregation are the usual pre- requisite for evaluating the role of heredity, whereas the absence of familial aggregation would contraindicate the testing of specific genetic hypotheses, except under unusual circumstances. PREVIOUS AND CURRENT STUDIES OF FAMILY MORTALITY The concept of familial longevity and ‘‘brachybioty’’ is certainly not a new one. Many family studies of life span have been carried out, a few dating back to the 19th century (Beeton and Pearson, 1899). Some of the published studies are summarized in table 1, where they are presented in chronological order. These reports can be classified into two broad groups: firstly, those based on genealogy records, including both large individual kinships (the Hyde family, the Peirce family) and collections of genealogy records (Foster’s Peerage, Burke’s Landed Gentry); and secondly, those based on samples from special subgroups of the popu- lation (e.g. workingmen’s families, families of living aged persons, senescent twins, insurees, etc.). The results of the genealogy studies, both single kinships and collections, appear to show distinct familial patterns of mortality. 238 There are, however, certain problems inherent in the use of old genealogy records and other defects that may have led to biases in the data. Even if the validity of recorded ages and persons is ac- cepted (and there is no objective means of verification or reliability), the statistics have proved inadequate for other reasons, such as: (1) the inadvertent omission of some family members, with the completeness of ascertainment varying by sex ! and by age; ? (2) the pooling of family data covering several centuries without allowance for the decline in mortality rates with calendar time; i.e. the cohort effect (Yuan, 1931; Jalavisto, 1951); (3) the failure to adjust for combining of different socioeconomic classes which are known to vary in mortality experience (Jalavisto, 1951); and, (4) the absence of any reference population or comparison group for the time period studied, or any other time period. It is difficult to estimate the effect of these biases on the reported correlation coefficients, average ages at death, proportionate mortality, and life expectancies reported from studies of genealogy records, but certainly the results would be considerably altered, and, in most cases, in the direction of indicating spurious relationships. Studies of family mortality patterns which have involved the investi- gation of samples from special groups in the population, however, have not entirely eliminated all these problems either. While the use of population segments is more desirable than the investigation of genealogy records, incompleteness of data and unsystematic or irregu- lar sampling procedures can provide biases in population studies just as serious as those found in genealogy studies. Such was the situation in Pearl’s investigation of the FHR series (family history records extracted from those collected by the Department of Biology Labora- ory for various other projects) and in his investigation of the families of nonagenarians, who had been ascertained through newspapers and friends (Pearl, 1931; Pearl and Pearl, 1934). Kallmann ef al. (1956) and Kallmann’s (1956) studies of senescent twins and their families, aside from questions of acertainment and zygosity diagnosis, and irrespective of the quality of the data, pertain to only a very unique segment of the population. Because twinning per se may be related to fertility and fertility to longevity, it is difficult to extrapolate from twin studies to the general population. There is, in fact, no ! Data recorded in Burke's Landed Gentry and Foster's Peerage (Beeton and Pearson, 1899) as well as the southern Chinese family studies by Yuan (1931) and others were almost entirely confined to males. Yet to limit the analysis to one sex eliminates perhaps critical evidence (e.g., the sex differences suggested by Jalavisto, 1951.) ? Many of the early records (Beeton and Pearson, 1899; Bell, 1918) showed a marked deficiency of those dying in infancy and late adult life with more complete reporting of young adult deaths. As a consequence, distributions of deaths show marked distortion when compared with vital statistics of the time (Carlisle’s 18th century life tables and Glover's life tables for Massachusetts males 1890). Inclusion of all deaths ir- respective of year of birth relative to the date of conclusion of study period contributed at least in part to this artificial distribution. 239 0¥e TABLE 1.— Family studies of mortality and life span Investigators Date published Source of data Time peried Classification of data Method of analysis and results covere Beeton and Pearson___| 1899. ______________ a) Old English Genealogy | Dating from 17th | Genealogy records—upper so- | Correlation coefficients. Records: century to end cial classes, confined mainly Results: Positive correlation between 1) Foster’s Peerage of Great | of (?) 19th to males. length of life: parents—children and Britain & Ireland. century. pairs of sibs. Greater similarity 2) Burke’s Landed Gentry Same as above. between collaterals, but in general, low of the British Empire. coefficients with some negative. 1901... 3) Society of Friends Rec- Genealogy records of small ords. traders & yeomen. b) Family Data—Colleagues Histories reported by fam- at Univ. College. ily members—professional classes mainly. Bolles oosmmmmmmnnesss M918, cc ce cic The Hyde Family Descend- | Mid-1600’s to 1864.| A single genealogy. __._______. Percentages, distributions, average duration ants of William Hyde who of life for offspring of various parental died 1681 in Norwich, Conn. combinations and for parents of offspring Publ. 1864. of different classification. Results: An almost direct relationship between life span of offspring and parents noted. Wilson and Doering...| 1926_______________ Peirce family descendants in | 1637-1880__________ Single family genealogy... ____ Correlation coefficients; distributions; and male line from John Pers. average ages of death for Ofspring of Publ. 1881. different parental combinations. Analysis confined to those who lived to be 20 years and could have reached 95 by end of study. Results: Father-son combinations (con- fined to those sons who were also fathers) yielding high correlation. Most of correlation coefficients higher than in other studies. Holmes..____.________. 1028 oo eeenenses Allstrém’s Dictionary of | “Earliest times to | Genealogy records of royalty..| Correlation coefficient methods similar to Royal Lineage. 20th century.” Beeton and Pearson. Separation by centu- ries. Results: Low correlation coefficients. Pearl... ______ 1924, 1931... _____ Five American Genealogical | 1678-1913__________ Genealogy records......__.___. Correlation coefficients. Works; Whitney Family, Results: No significant deviation from Hyde Family, Spooner ‘“‘zero’’ correlation. Family, Smedley Family, Durham Town History. 398) cceneanannnnns Family History Records | 1800s to 1920’s?___| Personal interview data con- | Life tables—standard actuarial methods. (Working men’s families in Balto. & surrounding area) referred to as FHR series. cerning families of working men. Results: Greater life expectancy for parents of children dying over 50 than for those of parents dying under 50. Mean after life times of children were differentiated from one another in accordance with age of death of parent. 176 Family Histories of Non- agenarians. Probably late 1700’s to 1920. Data on families of nonagen- arians (mailed question- naire). Life tables. TIAL=total immediate ances- tral longevity (sum of ages at death of 2 parents, 4 grandparents) higher in non- agenarians than for younger living propositi of FHR series. Chinese Genealogy.._.________ 1365-1914 Single family genealogy. .__._. Life table functions. Survivorship curves (1x); expectation of life (ex) and probability of dying (qx). All persons born after 1819 and dying before 20 years of age omitted. Results: Small amount of material, small differences, but consistency in that parental life expectancy greater among long lived sons and sons’ life expectancy greater if parents were long lived. Stoessiger..........__.. Pearson’s Family Data and families of Macpherson’s private patients. 19th century primarily? Collected families: 2/3: Pearson’s family data (i.e. genealogy records). 1/3: family history interview data from Dr. Macpher- son’s private patients, friends and colleagues in lasgow. Reanalysis of Pearson’s and Macpherson’s data excluding deaths from accidents or infection—correlation coefficient method. Results: Similar to those of Beeton and Pearson. Twin pairs—old aged homes, institutions, population. 19th century to present. Senescent twin pairs and their families. Twin analysis—mean intrapair differences in survivorship of MZ vs DZ twins. Mean life spans of sibs of twins classified by parental age at death. Results: Mean intrapair life span differ- ences greater for DZ than for MZ. Sib life span increased with those of parents. Jalavisto_...___________ Genealogy Records Finnish and Swedish Middle Class & Nobility. Old genealogy records of numerous families. Mean ages of death of offspring of parents with different life spans. Results: Both maternal and paternal longevity associated with mean life span of offspring. Maternal effect greater than paternal with daughters showing practically no paternal effect. Insurance Studies: Specialized Mor- tality Investi- gation. Actuarial Society of America (based on 34 life companies) involving 300,000 insured men. 1869-1899. Insured persons and their families. Death rates at successive ages among male ©: licy holders classifi according to ongevity of parents. Mortality of men whose parents lived to 75 was lower than those whose parents died under 60. Life expectancy greater for those with long lived parents. ove TABLE 1.— Family studies of mortality and life span—Continued Investigators Date published Source of data Time periad Classification of data Method of analysis and results covere Symonds... 1B. Mutual Life Ins. Co., 116,590 | ? to 1909__________. Insured persons and their | Mortality experience using a percent scale men insured 1891-1899 traced families. based on average family history (including to 1909. grandparents). Results: Mortality of those with most favorable family history 25 recent better than those with unfavorable one. Hunter......coconvnsnens 1032. ccna New York Life, 314,000 persons | ? Mid-1800’s ? to | Insured persons and their | Mortality experience (ratio of actual to associated with standard 1931. families. expected deaths) of policy holders grouped male risks insured 1914-1917 by rating scheme for good, average, poor, traced to 1931. family histories. Results: Small variation in 3 groups but mortality ratios were slightly better among those with good family histories at all ages up to 50 years at policy issue. Marshall... .. cone 1932 (1929 study Provident Mutual 30,000 pol- | ? Mid-1800’s ? to | Insured persons and their | Mortality experience (ratio of actual to pub. 1932). licies issued Dec. 1908-Aug. 1928. families. expected deaths) of insured men and their 1912 traced to 1928. sibs classified by parental history. Results: Mortality rate lower among insurees both of whose parents were living at policy issue than when parents dead. Dublin & Marks___.___ 1041-1042 _________ Metropolitan Life Ins. Co., | 2t01939___________ Insured persons and their | Ratio of actual to expected deaths of insured 118,000 insured white men, Ordinary Dept. living 20-64 in 1899-1905 traced to 1939. families. men grouped by parental living-dead status and age at death. Results: Survivorship of insurees asso- ciated with living status of parents at time of issue and apparently inde- pendent of age at death when both parents dead. meaningful reference population in any of these studies of special subgroups. In contrast to all of the investigations which have indicated positive familial associations in regard to life span, some of the insurance company studies suggest a relationship of the mortality experience of insurees with the living or dead status of their parents at the time of policy issue, irrespective of parental age at death. As indicated by Dublin and Marks (1941), however, there are many defects in life insurance material, including “an appreciable amount of misstatement, intentional or unintentional, of ages and causes of death’ of family members, with no attempt to verify these statements on the part of insurance companies. There is also the definite bias introduced by medical selection of persons for insurance—a selection often influenced by family history, and thus leading to more rigid screening on physical criteria for those with poor family mortality experience. Moreover, where no actual body of recorded data was utilized either as a source, or for verification [as was probably the case in Pearl’s FHR and nonagenarian series as well as Pearson’s family data from col- leagues, Stoessiger’s collected families (1932-33), the insurance data and others], human longevity reports have depended largely on “unsupported memory and tradition in a field where the emotional premiums of exaggeration are high” (Comfort, 1956). Thus, though there are much data on the familial patterns of life span, their interpretation is questionable, since the suggested positive relationships as well as any lack of relationships are all explicable by one bias or another that is involved in the selection, completeness, or the source of the data. PLAN OF STUDIES IN PROGRESS In view of this situation, two studies of patterns of mortality and aging were initiated in the Department of Chronic Diseases at the Johns Hopkins School of Hygiene to determine: firstly, whether there are family patterns of mortality; secondly, if they do exist, the nature of those patterns in terms of age at death, cause of death, and cause- age interaction; and thirdly, to identify in living subjects possible indicators related to family mortality patterns. The Population Study: For the first and main study, a sample of deaths (deceased probands) occurring in Baltimore City to U.S.-born white residents has been selected and is being matched with living controls of the same sex, race and neighborhood, and U.S. born within 5 years prior to the deceased probands. Detailed personal and family data are being obtained concerning each proband and each living control. The mortality experience of relatives (mothers, fathers, sisters, brothers, 243 spouses and spouses’ families) of the deceased probands will be com- pared with that of the corresponding family members of the matched controls. Intrafamilial as well as interfamilial mortality patterns will be examined. The influence and interaction of various biological and social factors (such as size of sibship, birth order, parental age at birth of subject and number of offspring, as well as number of marriages, divorces, occupation, education, religion, etc.) will be evaluated for a better comprehension of the relationship of the extrinsic to the intrinsic factors. To gain an insight into the possible effect of assorta- tive mating on familial mortality experience, the mortality of spouses and family members of spouses of dead probands will be compared with spouses and corresponding family members of living controls. To obtain patterns characteristic of the general population, a sample of total deaths of the community was selected. It was esti- mated that a sample of deaths and a sample of living controls in excess of 520 each would be required for the detection of certain reasonable differences. [Samples of 520 should detect at least a 10 percent difference between dead probands and living controls in the percentage of dead mothers, dead fathers, sisters and/or brothers at the 5 percent level of significance with a 90 percent probability, even if the smaller percentage is 40 percent (i.e., assuming a level of maximum standard error). When total sibs are pooled (on the assumption that the number of sibs would be similar to that ob- served in a pilot study described below) a difference of 5 percent should be detected at a 5 percent level of significance with 80 percent probability. Pooling of parents still assures detection of only a 10 percent difference but with 95 percent probability at the 1.0 percent level of significance.] Thus with allowance for an attrition rate of 5 percent due to refusals, inability to locate, and other factors, a sample of 550 de- ceased probands and 550 living controls would be adequate (to detect differences at the above levels). These are only approxi- mations, however, since, in the analysis, life table factors will be used, while the estimates of standard errors of the above are based on binomial percentages. The sample of deaths stratified by age was selected from the 6,535 deaths in Baltimore City in 1960 meeting the specifications for eligibility: U.S.-born white residents of Baltimore City and/or Baltimore County (table 2). In each of the age groups indicated, 110 deaths were selected by a combined random number and systematic sam- pling procedure with the method of selection taking into account seasonal variation in deaths. Table 3 shows the distribution of the sample of deaths according to age and seasonal quarter of year. Table 4 presents the death certificate causes of death of the proband sample as compared with distribution of causes of death for Baltimore City residents in 1960. 244 TABLE 2.— Distribution of Baltimore deaths and selected sample by age group (1960) Deaths of Baltimore City and Baltimore County residents in Baltimore Eligible* Sample Age group City 1960 size deaths Nonwhite White 605 687 | |e 336 356 297 110 468 622 536 110 646 1,349 1,127 110 1,280 4, 925 3,467 110 211 1,746 1,108 110 TOt.....cocciinnsvininsninsunmmmann aimee 3, 546 9, 685 6, 535 550 * Eligible =meeting specifications of U.S.-born white residents of Baltimore City and/or County, over 20 years of age. To minimize spurious correlations within families due to hetero- geneity of socioeconomic groups, the living control group consists of the nearest neighbors of the deceased probands matched on the criteria already indicated. A systematic method for selecting such neighbor- hood controls was developed, including maps, forms, and a detailed outline of procedure. The neighborhood control selection sheet as shown in figure 1 gives the necessary data for matching, and space for the screening record. The Neighborhood Control Selection Route (figure 2) shows the walking pattern to be followed in selecting a city neighborhood control. The proband’s household is marked with an “X” in Block 1. A sample of a control selection route mapped for a “city” control is given in figure 3. Note the indication of census tract boundaries. In Baltimore City, all controls are being matched within the census tract of the proband. If a census tract boundary intervenes, it becomes a stopping point for screening in that direction. Then the searching pattern is enlarged, fanning out in other directions within the proband’s census tract. For county screening, election district boundaries are used, in- stead of census tract boundaries. An outline map of the whole county divided into election districts (figure 4), as well as a map of the election district shape and boundaries for the specific proband to be matched, are provided with each control selection sheet for county controls. A sample “county” control is mapped in figure 5—the “X” indicates the location of the proband household. Variation in socioeconomic factors not taken care of through match- ing by place of residence is to be evaluated by the analysis. Another pitfall in some studies, the influence of secular trends observed in investigations encompassing large spans of calendar time, should be made negligible by means of the matching of probands with controls of similar cohorts. To avoid incompleteness and selection bias, all family members are being included, whether living or dead. Also, several respondents are 245 9%6 TaBLE 3.—Sample selection—1960 deaths to U.S.-born white residents of Baltimore City and/or County dying in Baltimore City, selected by quarters of year (Jan.-Mar., Apr.-June, July-Sept., Oct.-Dec.) Based on total observed deaths to eligibles—1960 Ptal January-March April-June July-September October-December otal eligible Age group deaths Second Second Second Second in 1960 Dx fraction X 110 adjusted Dx fraction X 110 adjusted Dx fraction X 110 adjusted Dx fraction X 110 adjusted sample sample sample sample 3/27/61 3/27/61 3/27/61 3/27/61 297 85 | .2862 | 31.48 31 72 | .2424 | 26.66 27 64 | .2155 | 23.71 24 76 | .2559 | 28.15 28 536 157 | .2029 | 32.22 32 112 | .2090 | 22.99 23 123 | .2295 | 25.25 25 144 | .2687 | 29.56 30 1,127 311 .2760 | 30.36 30 290 | .2573 | 28.30 28 247 | .2192 | 24.11 24 279 | .2476 | 27.24 27(28%) 3, 467 1,019 2039 | 32.33 32 811 | .2339 | 25.73 26 751 .2166 | 23.83 24 886 ( .2556 | 28.12 28 1,108 344 3105 | 34.16 34 229 | .2067 | 22.74 23 229 | .2067 | 22.74 23 306 | .2762 | 30.38 30 6,585 || 1,000 [ccna omnmsine 159 || 1,514 | mmm ci 27 | nad]... ceeelemmmanes 120 || 1,691 |-oo__|-—o__. 143(144) *One extra needed for total 110, added. L¥3 TaBLE 4.—Comparison of proband causes of death with reported causes of death in Baltimore City (1960), Baltimore City Health Depart- ment Annual Report Infectious Neoplasms CVRD Accidents Suicides Diabetes Others Total Age Source No. Per- No Per- No. Per- No. Per- No. Per- No. Per- No. Per- No. Per- cen cent cent cent cent cent cent cent 8 7.27 23 | 20.90 37 | 33.63 17 | 15.45 8 7.21 1 0.91 16 | 14.54 110 20. 00 20 7.63 53 | 20.22 74 | 28.24 4 | 16.79 21 8.01 6 2.29 44 | 16.79 262 3.44 4 3.64 27 | 24.54 48 | 43.63 8 7.27 2 1.82 1 0.91 20 | 18.18 110 20.00 38 7.81 121 | 24.89 200 | 41.15 33 6.79 14 2.88 3 0.61 77 | 15.84 486 6.38 5 4.54 33 | 30.00 54 | 49.09 2 1.82 0 0.00 3 2.73 13 | 1.81 110 20.00 70 6. 50 262 | 24.34 547 | 50.83 38 3.53 13 1.21 17 1.58 120 | 11.98 | 1,076 14.14 2 1.82 20 ( 18.18 67 | 60.90 0 0.00 1 0.91 5 4.54 15 | 13.63 110 20.00 202 4.79 780 | 18.71 | 2,639 | 62.59 77 1.83 27 0.64 133 3.15 349 8.27 | 4,216 55. 42 5 4.54 11 | 10.00 82 | 74.54 4 3.63 0 0.00 2 1.82 6 5.45 110 20. 00 68 4.34 139 8.87 | 1,217 | 77.66 41 2.61 2 0.13 22 1.40 78 4.98 | 1,567 20.59 24 4.36 114 | 20.72 288 | 52.36 31 5.63 11 2.00 12 2.18 70 | 12.72 550 | 100.00 398 5.23 | 1,364 | 17.93 | 4,677 | 61.48 233 3.08 77 0.10 181 2.38 677 8.90 | 7,607 | 100.00 *Deceased Probands. **Reported causes of death for Baltimore City, including foreign born. NEIGHBORHOOD CONTROL SELECTION SHEET Proband Reg. No. Sex: Male Female Address: Birth Date: Month/ Year Death Date: _ Control Month/Day/Year Sex: Male Female Birth Date: From to Month/Year Month/Year Age: From To Census Tract: Street Telephone Ages Ficure 1.—Neighborhood control selection sheet. usually interviewed for each family to prevent omissions of family members—even those dead in infancy. Besides provisions for assuring completeness of records, validation procedures for assuring accuracy are incorporated. To verify the ages of the living, households of all living relatives are being contacted, and to verify ages at death and causes of death for deceased relatives, death certificates and other vital records are obtained. The types of data being collected are outlined in appendix A. In addition to the detailed information obtained on the biological relatives of the probands and controls for comparison of their respective mortality patterns, the same types of data are recorded on spouses and spouses’ relatives in order to examine the role of assortative mating and adult environment shared by the spouses as compared to the influence of genetic factors and childhood environ- ment shared by the sibs. 248 Reg. No. MAP FOR CONTROL SELECTION ROUTE —_— nL OLLEVIION ROUTE Each square equals one s quatre block Check blocks 1-13 to determine whether they are within same X equals h quals home: of census tract. If so: proband 1. Proceed in numbered order 2. When starting a new block (3), begin with block closest to proband's home. 3. If a dead end or wooded area interrupts the normal order, proceed to the next consecu- tive number 4. If there is no answer at a home, mark this on Neighborhood Control Selection Sheet for a return call. Proceed onward and obtain control. Mark this control "Tentative Control" until all closer dwellings are checked out, If on a return call, a con- trol is found closer to the home of the proband, he will be the control, and the 2 "Tentative Control" will be removed or used as a secondary control, according to a subsequent standardized routine for secondary con- trols. —e 5 A. 5. If no control has been found at the end of block 13, return for further instructions. = == = v=" 0 =} —& — If any of the blocks fall out- side the census tract of the deceased proband, cross these out and outline the area of the pattern within the given census tract, Proceed with only those blocks within the census tract in their numbered order. FiGure 2.—Map for neighborhood control selection route. Progress of Population Study: Actual scheduling of the various aspects of the study has been subject to revision according to the relative progress of different phases. A brief step-by-step summary of the investigation since its inception may help clarify the application of some of the procedures under discussion. Before any work was begun, a pilot study was carried out on re- spondents for a sample of 50 consecutively recorded Baltimore City 249 735-712 O0—65——17 CITY CONTROL PII264 N EASTERN AVENUE on ° c Oo «qm o £ LANCASTER X 800 S. BROADWAY =~ > wm < WwW - i E o 3 [=4 lal om wn) » of od zd = ol 3 on) wv [58 x oO PROBAND LIVED AT 800 S.BROADWAY Figure 3.—Sample map for a ‘‘city’”’ control. “X’’ marks location of proband’s residence. 250 BALTIMORE COUNTY ELECTION DISTRICTS PewnsyrLvania BALTIMORE City Anne Arundel Co X indicates location for P- 00167 Ficure 4.—Map of Baltimore county election districts as they surround Baltimore. deaths occurring in March 1959 to white persons over 20 years of age. A trial interview schedule was developed for this purpose. The problems involved in tracing, interviewing, questions, etc., were noted, and suitable adjustments in the interview schedules and 251 COUNTY CONTROL P 00167 DISTRICT Bird River PROBAND LIVED AT LONG GREEN & MANOR ROADS Figure 5.—Map for a sample county control. Proband’s residence at Long Green and Manor Roads is indicated by an “X’’ mark. procedures were made. On the basis of the findings it was decided that only U.S.-born probands would be included in the population study, since the quality and quantity of information on relatives of the foreign born were inadequate and verification by death certificates not feasible, whereas the data obtained from Telotives of native-born whites were quite satisfactory. In this pilot study, information was collected on 593 persons derived from families of 38 dead persons (12 probands having been rejected for various reasons, such as place of residence, refusal, etc.), 252 thus providing an estimated 15.6 persons per proband to be expected in the family study—the basis for determination of the required sample size as already indicated. After the pilot study, the main investigation was initiated in the spring of 1961. It can be considered as consisting of three phases. The operations included in the first phase are listed in the appendix B-1. While procedures for the total study (including control matching and interviewing) were being established in this first phase, the major areas of concentration were in revising the interview schedules into their final form and in the selection of the deceased proband sample, a necessary prerequisite to control matching. Priority in time was advisable for the collection of proband data for several reasons. Since the deceased proband’s address was the starting point for searching for respondents, it was considered important to make the contact as soon as possible after the proband’s death before the nearest relatives moved away or died. Also, because living controls were matched to deceased probands, it was desirable to have deceased proband data concerning such factors as exact home address, transient or permanent resident status, and actual inclusion in the study, before searching for the control match. The next phase of the study continued with many aspects begun in the preceding phase but here the emphasis shifted (1) from initiating to final revision of procedures; and (2) to the primary objective: the major data collection on probands and controls. The various aspects of this phase are listed in appendix B-2. The last phase of the study (appendix B-3) will consist of analysis of the collected data and reporting of the results. Special problems: Several methodological problems were encountered which may be of interest to those considering this type of study. Though matching of probands with ‘nearest-neighbor’ living controls is theoretically desirable to minimize differences in socioeconomic status, it is a very expensive and time-consuming procedure, especially in the older age groups. Our statistician, Dr. Earl Diamond, has recently calculated that for a 50-percent probability of obtaining a match for an 80-year- old male, one would need to screen 90 households in Baltimore. Consequently, for future studies where suitable population controls are needed, Dr. Diamond has suggested a ‘balancing’ procedure, the details of which will be published. In the revised design, a random number type of household sampling is used where the sampling probabilities are based on the age, sex and neighborhood distribution of the proband population to be matched. This method, therefore, results in the selection of a control group which has the same age, 253 sex and neighborhood composition as the proband group without individual proband-control matching. Another difficulty encountered has been locating respondents for elderly deceased probands and determining the exact dates and places of death of their relatives, especially when the index case was the last surviving member of the family or one of the few surviving mem- bers, the others being ill or senile. Moreover, even when suitable family respondents are available, often some of the relatives died so long ago that the recall of survivors for exact details is poor and/or the official records are inadequate. The handling of death certificate verification locally and out of state for the proband and control families has developed into a sizeable aspect of the study. For obtaining death certificates, a minimum amount of accurate information is required. Often contacting the cemetery of burial to obtain the exact date of death and other informa- tion has proved helpful. The degree of searching that each state will carry out around approximate dates depends upon its staff, facilities and the organization of its vital statistics divisions. These are highly variable from state to state. Moreover, even within individual states—and Baltimore City itself—the filing system for death records has changed numerous times in the past century. When death certificates are not available from state health departments, they can sometimes be obtained from local county, city or borough divisions. Some useful pamphlets include the Rand MeNally Geographical Handbook, a keyed index of the United States locating counties and all cities and towns of over 100 population, the Cemetery Directory of the American Cemetery Association and “Where to Write for Birth and Death Records, United States and Outlying Areas” PHS Publication No. 630 A-1, to mention a few. Another problem not to be overlooked in studies where respondents have entered the sample without their own knowledge, interest or motivation is that of obtaining cooperation and maintaining rapport. Special Group Study: The second investigation of family mortality is very different in design from the first. Planned to gain clues concerning manifestations of family mortality patterns in living subjects, it deals with test responses, measurements and family mortality experience in a special group. This special group comprises about 500 males from 24 to 100 years of age who have volunteered to participate in a longitudinal study of aging being carried out by the Gerontology Branch of the National Heart Institute. About every 18 months these subjects spend 2% days at Baltimore City Hospitals going through a series of physiological, biochemical, and psychological tests and anthropometric measurements. In addition, our staff is obtaining the following: 254 (1) Personal and family history data: Information similar to that being collected in the population study just described is being obtained by means of a combined interview and self-administered questionnaire. (2) Smoking histories comparable to those used by Haenszel, Shimkin and Miller (1956) are being filled out by subjects with the assistance of interviewers. (3) Ability to taste phenylthiourea (PTC) is being tested. The results of the test responses, anthropometric data and the PTC tests, smoking and other personal data will be correlated with each other and with the sociobiological factors as well as with the family mortality data in order to gain clues as to possible indicators related to family mortality. Although a volunteer group is not the most satisfactory type of sample, nevertheless, the detailed longitudinal study of this coopera- tive and well-motivated group makes available a unique combination of data which would be unobtainable in the larger, community-based investigation. A less than ideal approach, therefore, seems justifiable to gain possible insight into the ‘natural history of mortality patterns,” an area where there is very little concrete information at this time. PRELIMINARY FINDINGS Both of the studies discussed here, the population study and the special group study, are in the data collection phase. However, some preliminary findings are available for the population study, though they are limited to an analysis of the first 101 matched pairs (101 probands and 101 controls). Moreover, these results must be con- sidered with caution since they constitute less than 20 percent of the sample, with the information not yet totally verified. In religious affiliation, probands and matched controls show almost identical distributions (table 5). On the other hand, levels of educa- tion differed slightly. As shown in table 6, more probands had been graduated from high school or gone beyond high school than controls, with the difference residing in those dead under 50 years of age, (i.e. born after 1910) and no difference in those over 50 years of age. TABLE 5.—Religion of probands and controls Probands Controls No. Percent No. Percent 47 46.5 49 48.5 36 35.6 37 36.6 16 15.8 15 14.9 1 199 Temnmsimmesssnslsnrsanseonses 1 99 lonemanummnsns|snamssermanes 10) [eecnrvmmmenans OL Jenmenmnnmmnsns TABLE 6.— Education of probands and controls <50 504 Total Age Probands Controls Probands Controls Probands Controls Per- Per- Per- Per- Per- Per- No. | cent | No. | cent | No. | cent | No. | cent | No. | cent | No. | cent High School and beyond... 40 | 56.3 31 | 43.7 7]23.3 712.3 47 | 46.5 38 | 37.6 than High School Graduates. .____________ 30 | 42.3 40 | 56.3 23 | 76.7 23 | 76.7 53 | 52.5 63 | 62.4 Unknown ._._____________ 1 [eescan wk mmm [rms] foes we | craeas lememes [manne mmm joie Total..cocnvcessnns A} cons nn looses 80 |-ueees 80 |oeriis 1); 1 FRC 101 |o_... Regarding parental survivorship, as shown in table 7 and figure 6, there were no differences between parents of probands and parents of controls in survival to 50 years of age. Fathers of probands had the same life table probability of surviving to 50 as fathers of controls (.739 and .740 respectively); and similarly the mothers of probands and mothers of controls (.837 and .818 respectively). There is, however, a suggestion of a lower probability of surviving to 70 years of age for fathers of probands (.323) than for fathers of controls (.393) with again a similar trend—but smaller difference—for mothers (.527 versus .571). Probands and controls had the same proportion of total living fathers (26/101, 25.7 percent, and 25/101, 24.8 percent, respectively) with probands having a slightly higher proportion of fathers living under 70 than controls (19.8 percent versus 15.8 percent). Of TABLE 7.—Survivorship of parents and sibs of probands and controls Fathers Mothers Probands Controls Probands Controls 101 101 101 101 .739 . 740 . 837 .818 +. 062 +. 062 +. 052 +. 055 85 86 88 88 3 2 5 4 .323 .393 . 527 L571 +. 050 +. 053 +. 057 =+. 057 25 35 35 47 20 16 30 21 Brothers Sisters Probands Controls Probands Controls Total Number. 190 194 187 188 Cum Pg. 700 827 . 806 827 S.E_. +.037 +. 031 =+.033 +. 031 Number reaching 50 years. 80 64 76 Number alive under 50... _ 80 88 96 87 Cum Prp..oon visnmnns 605 579 L722 697 8. Bcc snusnvansmivess +. 0567 =+.056 +. 050 +.052 Number reaching 70 years. .__. 3 11 9 6 Number alive under 70... ____________________ 131 141 147 150 256 PARENTAL SURVIVORSHIP FATHERS MOTHERS CumP | TO AGE 50 TO AGE 70 TO AGE 50 TO AGE 70 P = PROBAND C = CONTROL Figure 6.—Life table probabilities of surviving to age 50 years (Cum Ps) and to age 70 years (Cum Py) for fathers and mothers of probands and controls. course, very few living fathers were under 50. On the whole, pro- bands had slightly more living mothers (48/101, 47.5 percent, versus 40/101, 39.6 percent), and thus more living mothers under 70 years (30/101, 29.7 percent) than controls (21/101, 20.8 percent). As for sibs, no real differences were noted in numbers of sibs of probands and controls. Probands had 190 male sibs and 187 female sibs as compared to 194 male and 188 female sibs for controls (table 7). Of these, slightly fewer male sibs of probands were alive and under 70 years of age than male sibs of controls (68.9 percent and 72.7 percent respectively). As for female sibs, probands had more living sisters under 50 than controls (51.3 percent versus 46.3 percent) though probands and controls had the same percentage alive under 70 years (78.6 percent versus 79.8 percent). Sib survivorship experience followed a pattern similar to that of parents. As shown in table 7 and figure 7, the male sibs of probands had a significantly lower probability of surviving to 50 years than those of controls (.700+4.037 as compared to .827 +.031) and possibly less likelihood of surviving to 70 years (.505 versus .579) although the latter group involved very small numbers. Among female sibs, there were no real differences; probability of survival to 50 years being .806 and .827 for female sibs of probands 257 SIBLING SURVIVORSHIP BROTHERS SISTERS Cum P TO AGE SO TO AGE 70 TO AGE 50 TO AGE 70 i | .90¢ .800} .700| .600F P = PROBAND C= CONTROL Freure 7.—Life table probabilities of surviving to age 50 years (Cum Pj) and to age 70 years (Cum Py) for brothers and sisters of probands and controls. and controls, respectively. The numerical reversal in direction for survivorship to 70 years (.722 versus .697), though appearing to be inconsistent with the other findings, involved only 9 and 6 persons, respectively. It appears that whatever suggestion there is of survivorship differ- ences between corresponding family members of probands and con- trols, these are limited to the male relatives. The female relatives of probands seem to be similar in survivorship experience to those of the controls. These findings, however, provide only a preliminary glimpse of a very small part of the data. Tt is not possible to predict whether the same trends will appear when all the validated information is included and more refined statistical techniques are utilized. At this stage of the study it would not be safe to venture any conclusions nor specula- tions. It is hoped, however, that both the population and the special group studies may ultimately yield clues as to hypotheses (genetic or other) for further, more definitive types of investigation. Since such studies require sizable samples, and preferably a variety of population segments (geographical, ethnic, urban, rural, ete.) more and larger population samples, perhaps collaborative projects at 258 several research centers, are required both for the retrospective family data analyses and for the longitudinal followup of living subjects under study. ACKNOWLEDGMENTS I wish to express my appreciation to Dr. Nathan Shock for per- mission to use the special group of subjects participating in his longitudinal study of aging and to Mr. Arthur Norris for his coopera- tion and invaluable assistance in administrative problems. I also want to thank Dr. Abraham Lilienfeld and Dr. Margaret Bright for their most careful reading of the manuscript and their helpful sugges- tions. REFERENCES 1. Actuarial Society of America, 1903. Specialized Mortality Investigation Ezperience of 34 Life Companies upon 98 Special Classes of Risks. New York. (Cited by Dublin, L. I., Lotka, A. J., and Spiegelman, M. 1949. Length of Life, rev. ed. New York: Ronald Press.) 2. American Cemetery Association Bulletin. Cemetery Directory. 329 E. Broad St., Columbus 15, Ohio. 3. Beeton, M., and Pearson, K., 1899. Data for the problem of evolution in man. II. A first study of the inheritance of longevity and selective death rate in man. Proc. Roy. Soc. 65: 290-305. 4. Beeton, M., and Pearson, K., 1901. On the inheritance of the duration of life, and the intensity of natural selection in man. Biometrika 1: 50-89. 5. Bell, A. G., 1918. The duration of life and conditions associated with longevity. A study of the Hyde genealogy. Washington: Judd & Detweiler, Inc. 57 pp. (Study based on R. H. Walworth, “Genealogy of the Hyde Family” 1864.) 6. Chai, C. K., 1959. Life span in inbred and hybrid mice. J. Hered. 50: 203- 208. Comfort, A., 1956. The biology of senescence. New York: Rinehart & Co. Diamond, E., 1963. Personal communication. Dublin, L. I., and Marks, H. H., 1941. The inheritance of longevity—a study based upon life insurance records. 7T'r. A. Life Insur. M. Dir. America 28: 89-120. 10. Haenszel, W., Shimkin, M. B., and Miller, H. P., 1956. Tobacco smoking patterns in the United States. Public Health Monograph No. 45. PHS Publication No. 463. Washington, D.C.: Government Printing Office. (Issued concurrently with Nov. 1956 issue Publ. Health Rep. 71, No. 11, May 1956.) 11. Hartwig, W., 1959. Lifespan of cattle and horses under various climatic conditions and the reasons for premature culling. In: The Lifespan of Animals (Ciba Foundation Colloq. on Ageing, Vol. 5), pp. 57-71. 12. Holmes, S. J., 1928. Age at parenthood, order of birth, and parental longevity in relation to longevity of offspring. Univ. California Pub. in Zoology 31 (15): 359-375. 13. Hunter, A., 1932. Note on the effect of family history and longevity. Tr. Actuarial Soc. Amer. 33 (87-88): 405-408. 14. Jalavisto, E., 1951. Inheritance of longevity according to Finnish and Swedish genealogies. Ann. Med. Int. Fenn. 40: 263-274. ew 259 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. Kallmann, F. J., 1956. The genetics of aging from the neurologic and psychiatric aspects of the disorders of aging. Proc. Assoc. Res. Nerv. & Ment. Dis. 35: 95-111. Kallmann, F. J., Aschner, B. M., and Falek, A., 1956. Comparative data on longevity, adjustment to aging and causes of death in a senescent twin population. Novant’ Anni delle Leggi Mendeliane, pp. 330-339. Marshall, E. W., 1932. Parental history and longevity. Tr. Actuarial Soc. Amer. 33 (87-88): 395-404. Maurizio, A., 1959. Factors influencing the lifespan of bees. In: The Life- span of Animals (Ciba Foundation Colloq. on Ageing, Vol. 5), pp. 231- 246. Miihlbock, O., 1957. The use of inbred strains of animals in experimental gerontology. In: Methodology of the Study of Ageing (Ciba Foundation Colloq. on Ageing, Vol. 3), pp. 115-130. Pearl, R., 1924. Preliminary account of an investigation of factors influenc- ing longevity. J.A.M.A. 82: 259-264. Pearl, R., 1928. The rate of living. New York: Alfred A. Knopf. Pearl, R., .1931. Studies on human longevity. IV. The inheritance of longevity. Preliminary report. Human Biology 3: 245-269. Pearl, R., and Pearl, R. D., 1934. The ancestry of the long-lived. Baltimore: Johns Hopkins Press. Rand McNally Geographical Handbook. 1962. A keyed index of the United States locating counties and all cities and towns of over 100 population. New York: Rand McNally & Co. Smith, J. M., 1959. The rate of aging in Drosphila subobscura. In: The Lifespan of Animals (Ciba Foundation Colloq. on Ageing, Vol. 5), pp. 269-281. Stoessiger, B., 1932-1933. On the inheritance of duration of life and cause of death. Ann. Eugen. 5: 105-178. Strong, L. C., 1936. Production of the CBA strain of inbred mice: long life associated with low tumor incidence. Brit. J. Exptl. Pathol. 17: 60-63. Symonds, B., 1913-15. Some studies in family history. Proc. Life Ins. M. Dir. America 23: 4-57 Wilson, E. B., and Doering, C. R., 1926. “The Elder Peirces.” Proc. Nat. Acad. Sci. 12: 424-432. Yuan, I-Chin., 1931. Life tables for a southern Chinese family from 1365 to 1849. Human Biology 3: 157-179. Yuan, I-Chin., 1932. A critique of certain earlier work on the inheritance of duration of life in man. Quart. Rev. Biol. 7 (1): 77-83. Where to Write for Birth and Death Records, U.S. and Outlying Areas. PHS Publication. No. 630A1 and others. APPENDIX CoMmMUNITY BASED PoPULATION STUDY . Types of data being collected 1. On dead probands only Complete death certificate information including age at death, cause of death, year of death, place of burial, etc., on the dead probands. These death certificate data are then verified with respondents). 2. On dead probands and living controls (information on familial, social and miscellaneous factors including): a. Size of sibship. b. Birth order. 260 c d e. f g h. 3. On all spouse a. b. . Parental ages at birth of proband. . Marital history: number of marriages, divorces, separations, ete. Fertility history: offspring, miscarriages, stillbirths. . Occupation. . Education. Religion. relatives (parents, sibs, offspring, spouse, parents and sibs of for both study groups—dead probands and living controls): For each individual: name, sex, birthplace, year of birth (in ad- dition: birth order for sibs, spouses and spouses’ sibs). For living relatives: age and address (for follow-up checking and possibly longitudinal survivorship study). . For dead relatives: place and year of death, cemetery, respond- ent’s information concerning age at death and cause of death to be checked with death certificates. . Families of dead probands are also checked with respect to con- sanguinity and for overlapping of family members by use of Living and Dead card files. B. Progress of the study 1. First phase a. b.- C. h. The selection of the study population (tables 2, 3, 4). Search of death certificates of the 550 selected probands. Pretest of interview schedule in field trials and final revision of schedule to incorporate all changes deemed feasible on the basis of the pilot study and pretest period. . Design of plan for matching living controls to deceased probands (figures 1-5). . Compilation of an Interviewer’s Manual as a standardized guide line for procedures and decisions. . Selection and training of initial staff. . Beginning of a record-keeping system with cross-index files for relatives, as well as probands and controls; and design of various forms to be used in recording data. Initiation of interviewing of respondents for deceased probands. 2. Second phase a. (Personnel) Recruiting and (training of qualified personnel (interviewers, neighborhood screeners for matching living controls and clerical staff with replacements as required). The length of training period for individual interviewers depended on their experience, qualifications and aptitude, usually requiring at least 3 to 4 com- bined group and individual instruction sessions. Screeners re- ceived similar, though somewhat less extensive, training than interviewers. . (Procedures and record system) Establishment of standardized procedures for each phase of study: Sample selection, interviewing, control matching, obtaining, classifying, recording, verifying and coding data. . (Obtaining proband interview data and selection of matching controls) Searching for and interviewing respondents for deceased pro- bands: (1) As many respondents are interviewed as can be located to obtain complete data indicated in interview schedule. As 261 of May 31, 1963, interviews concerning 485 of the 550 probands had been at least initiated. Of these, data concerning 274 were completely obtained. Information on an additional 202 was still incomplete and 9 interviews were in return processing (z.e., being edited). (2) Screening neighborhoods of deceased probands for the purpose of matching living controls to those probands, with additional procedures for “no contact’ households. d. (Status of Control Search) As of May 31, 1963, the following controls had been located: 426 matched controls 339 true controls 56 secondary controls 31 census-tract matched controls 65 unmatched controls. e. Interviewing of living controls and/or any additional respondents required for obtaining family, spouse and spouse family data indicated in the interview schedule. The present status of control interviewing is as follows: 319 true and secondary control interviews 201 complete 74 incomplete 44 return processing 23 census-tract matched interviews 26 unmatched control interviews. f. (Editing, completion of individual records and verification) (1) Obtaining additional family data by mail and telephone from out of city family members, friends, institutions, etc., as well as local institutions, funeral homes, ceme- teries, hospitals, physicians as needed. (2) Verification of information concerning probands and living and dead family members (living/dead status, age, cause of death, etc.). Information about dead probands: (a) Death certificate information is being checked with one or more respondents who are nearest relatives of the deceased proband. (b) Hospital and other medical records are also to be checked to verify cause of death but this phase has not yet been initiated. Different sources of information, such as Medical Exam- iner’s Reports, etc. are utilized. (3) Information about relatives of dead probands and living controls: (a) For dead individuals, death certificates are being obtained to validate data on these deaths. (b) For living individuals, verification of birth dates and “living status” is being carried out by telephone, by mail and in person. The status is considered verified if the individual himself or a member of his present household or 2 respondents from his side of the family have confirmed it. g. Follow up of “no contact” households and recontact of initial 262 refusals by mail and in person in order to reduce the number of final refusals and of no contact households. h. (Coding and data processing) (1) Establishment of a code suitable for punching on IBM cards and programming for electronic computers. (This code not only utilizes individually specific tabulation codes designed for this particular study but also in- corporates such standardized forms as U.S. Census Population Occupational Code and Industrial Code and the International Classification of Causes of Death.) (2) Coding of completed records after editing of interviews and verification of information. (3) Development of computer programs for processing of data. 3. Third phase a. Analysis of the collected data b. Reporting of the results. 263 Coronary Heart Disease in Relation to Blood Pressure and Cholesterol Levels in Population Studies’ By Freperick H. Epstein, M.D.,> AND MaR- cus O. KseusBerG, Pu. D., Department of Epidemiology and Department of Biostatistics, University of Michigan School of Public Health, Ann Arbor, Mich. Studies of familial aggregation of disease may be carried out among various subdivisions of the general population. A total com- munity within the general population may be the site of operation. A random sample of a defined population may be drawn. Index cases and their kin may be identified from a number of possible sources and compared with a suitable control group. Each of these ap- proaches carries advantages and shortcomings and there is no single approach which is better than all others, the choice depends on the aims. The subsequent discussion will be mostly confined to familial aggregations of coronary heart disease in the population at large. While previous studies have suggested differences in familial clustering of coronary disease between the relatives of index cases and controls, the picture as it presents itself in the general population is largely unknown. In order to illustrate the community approach, a descrip- tion of our own investigations will be given and compared with some calculations based on population models. 1 This study was supported by the Cardiovascular Research Center, University of Michigan, under program project grant H-6378 from the National Heart Institute, National Institutes of Health, U.S. Public Health Service, and prepared in collaboration with the research staff, Cardiovascular Center, University of Michigan, and the Tecumseh Community Health Study. 2 Supported by a Research Career Award (HE-K6-6748) from the National Heart Institute, National Institutes of Health, Public Health Service, U.S. Department of Health, Education, and Welfare. 265 [735-712 O—65——18 SOME OBSERVATIONS ON CORONARY HEART DISEASE IN A TOTAL COMMUNITY The community which is the site of our studies is the town of Te- cumseh, Michigan, some 30 miles from Ann Arbor (Epstein, 1960a; Francis, 1961; Napier, 1962). Within this study area, about 90 percent of all inhabitants-have received intensive medical examinations during 1959 and 1960; a second round of examinations is now in progress. The initial examinations included 8,641 persons. In such a setting, the investigation of familial aggregations of disease or, for that matter, health is a natural component of the total operation. Great flexibility is possible in using variously defined case and control groups within this entire universe. The clustering of variables or events within kindreds, families, households, and other population subgroups can be observed against a firm background of age and sex- specific distributions in the total group. A study of this kind in a total, natural community provides the ultimate opportunity to weave man and his environment into an all-embracing picture of human ecology. The advantages of total community studies, as contrasted with studies among population samples, are apparent. Their short- comings revolve around difficult problems of adequacy in numbers, methods, strategy and logistics. Complexity is now taken for granted in many fields of research. Some problems of population biology require equally complex approaches. There are problems of human ecology which can only be answered within the encompassing frame- work of an operation of which the Tecumseh Study is an example. The description of such a framework is presented as one approach to the identification of genetic and environmental influences and their effect on the maintenance of health and susceptibility to illness. One of the first tasks in Tecumseh was the charting of kindreds and the development of a system to assemble individuals related to each other in various ways, so that the attributes of these persons could be compared. Kindred charts were prepared by hand and each member of the kindred was given a 10-digit code which identified his position within the kindred.® Households were identified independently. This system makes it possible to analyze the data on living and de- ceased individuals by biological relatedness, households, and examina- tion status. The population in Tecumseh could be assigned to a total of some 4,500 overlapping kindreds. Since a major interest centers on coronary heart disease, one of the first questions was whether clinically manifest coronary heart disease could be shown to cluster in families in the Tecumseh population. During the first round of examinations, 248 men and women were identified as having a probable diagnosis of coronary disease by cri- 3 Samples of kindred charts are obtainable from the authors. 266 teria serving the needs of this particular analysis. Since most of these persons were middle-aged or older, many of their parents were de- ceased while many of their children were too young to show clinically detectable disease, and diagnostic methods for pre-clinical disease are, at present, unsatisfactory. Pending more accurate data on causes of death among the parents and more sensitive methods of clinical and pre-clinical disease detection among the children of index cases, a preliminary look at the data was limited, therefore, to the siblings of index cases. A sizeable proportion of the siblings under consideration live outside the study area; in order to obtain additional, direct in- formation, we have interviewed but not yet examined those siblings of index cases who live within a radius of some 30 miles of Tecumseh and have analyzed the information on deaths among the siblings of index cases. The composition of these sibships in terms of examined per- sons, interviewed persons, siblings living beyond immediate reach and deceased siblings is shown in table 1. Thus, some 54 percent of all living siblings (466 out of 869) have either been examined or live within reach. With regard to the remainder, information already available from the family history must be verified through mail in- terview and contact with private physicians since it would be beyond our means to send a roving team around the country to examine the 403 siblings involved as well as appropriate controls. TABLE 1.— Distribution of brothers and sisters of 248 cases with coronary disease (CHD) in Tecumseh, Michigan, by examination status, diagnosis and location Examination status Diagnosis of CHD Male Female Total Percent N N N of total Examined cases. _._________.__ Probable..........ccccviauuias 142 106 248 20. 4 Examined Sibs......cowsuiiwanan BHSpPOLt.. conus ae waa 7 13 A |insinnenns None.....- = 20 24 44 |i Subtotal _ _ 27 37 64 5.3 Interviewed sibs (living near- | Probable.__ 8 1 YY: eomeirmen by). Suspect... 5 10 15 | __ None... _____________________ 64 66 150! cde Subtotal _ _____________________ 7 77 154 12.7 Noninterviewed sibs (mot | -______________________________ 214 189 403 33.2 living nearby). Sibs who died below age 30____| Death due to: Heart Disease__ 2 2 4 |icosasues Other Causes. _ 71 64 138 occa Sibs who died at age 30 or over.| Death due to: Heart Disease. _ 34 22 [| J] PE —— Other Causes. _ 75 75 0 | ccrenenz Subtotal ._____________________ 182 163 345 28. 4 AN sibs. oe 642 572 1,214 100.0 These data have been reviewed in detail because they raise a funda- mental problem in the study of familial aggregations among general populations in the United States. The mobility of the population in this country makes it difficult to assemble complete and accurate in- formation on kindreds. The distribution presented gives a quantita- tive assessment of this problem in a general population. Multiple cases among examined persons were observed in six sib- ships so that, in 236 of the sibships, the index case was the only ex- amined individual with coronary disease in the sibship whereas the 267 remaining 12 index cases occurred in pairs in the remaining six sib- ships. In addition, there were 10 sibships in which an interviewed brother or sister of an index case also reported “probable” coronary heart disease; in one of these 10 sibships, two other brothers had died of “heart trouble”. There were another 43 sibships in which a sibling of an index case reportedly died of heart disease. Thus, there re- mained 133 sibships of index cases in which there was neither another living sibling with “probable” coronary heart disease among persons examined or interviewed nor deaths from this disease as far as could be ascertained. It may be asked to what extent incompleteness of kindred data might limit the inferences which can be drawn. If, among all living family members, those available were a random sample of the total group, the analysis could be confined to those within reach. In fact, the available family members may well be different since, in some studies, the more mobile segments of the population tend to have more disease (Syme et. al., 1964). Whether these persons differ also in their genetic constitution is, however, another question which would be difficult but possible to test. As far as the feasibility of carrying out familial studies of chronic disease in the general popula- tion in the United States is concerned, one has to face the facts and develop statistically valid epidemiological methods to solve the prob- lem as it presents itself in a country such as this. This appears pref- erable to taking a nihilistic approach and deciding that such studies can only be done in less mobile populations. Since the disorders under discussion no doubt result from both genetic and environmental in- fluences, the etiology of these conditions must be studied as part of the environment in which they occur and observed in different geographic locations 1ather than through extrapclation of findings from one area to another. AGGREGATIONS OF CORONARY HEART DISEASE IN FAMILIES As a first approach, it might be useful to view familial aggregations of coronary heart disease in terms of a hypothetical model, for com- parison with the situation encountered in the field. A disease as common as coronary heart disease must inevitably coexist sometimes in several members of the same family by chance alone. This fea- ture, as is well known, causes difficulty when familial aggregations are studied by sampling index cases and controls, the “fallacy of the index case” phenomenon; this particular problem does not exist in studies among total populations. Let us assume that all middle- aged men in a population are brother-pairs (table 2). In a popula- tion of 2,000 men, there would, therefore, be 1,000 brother-pairs. 268 The prevalence of manifest coronary heart disease among middle- aged men in this country is approximately 5.5 percent (Epstein, 1960b) so that there will be 110 cases and 6 of these will be expected to occur in 3 pairs (0.055 times 0.055 being 0.0030). Thus, if there were no familial aggregations of coronary heart disease, an epide- miological study of 1,000 living middle-aged men and their brothers would yield 3 instances in which both brothers are affected simul- taneously. At the present state of knowledge, it is not possible to say to what extent the true coexistence of disease among living male siblings might exceed chance expectation. However, even if the frequency of concordant pairs were four times greater than expected, a sizable excess, no more than 12 out of 1,000 sets of brother-pairs would be affected simultaneously. The data may be viewed in another way. Thus, with four times random expectation, 24 out of the total of 110 cases, i.¢., a little more than 20 percent, occur in pairs, as com- pared with a random expectation of about 5.5 percent (6 out of 110). TABLE 2.—Frequency of coronary heart disease in a hypothetical population of 1,000 brother-pairs Total number of sibships Total number of cases Number of Number of examined Total number of | cases per members per | persons in sibships | sibship Random 4 times Random 4 times sibship expectation random expectation random expectation expectation Diseases 2,000 (1,000 2 3 12 6 24 sibships). 1 104 86 104 86 0 893 902 0 0 Total...) 2000. .ccnusesmsusnsnmssmmsdns 1, 000 1,000 110 110 Frequencies are based on a prevalence of 5.5 percent for coronary heart disease among middle-aged men. Detecting familial aggregations would be easier if the disease prevalence were greater. A prevalence of around 5 percent encom- passes only those cases of coronary heart disease which are clinically detectable by presently available means. The true prevalence of severe lesions, based on autopsy findings, is in the region of 20 percent. In this situation, four times random expectation (40 concordant brother-pairs) would yield 160 brother-pairs simultaneously affected among a total population of 2,000 brothers. Against this background, some very preliminary calculations based on the Tecumseh data may be discussed. We are not yet ready to present definitive data on familial aggregations of coronary disease in Tecumseh, but merely intend to discuss ways to approach the prob- lems of analysis. We wished to calculate whether myocardial infare- tion occurred more often than could be expected by chance in the sibships of persons examined in Tecumseh. This pilot analysis was limited to sibships in which all examined members were aged 40 years or older. In order to increase the yield, persons with a history of both 269 “probable” and “suspect” myocardial infarction were combined into a single category, “myocardial infarction”. By contrast, the term “coronary heart disease’, as used earlier, included persons who had either a history of probable myocardial infarction, symptoms of probable angina pectoris, electrocardiographic evidence of myo- cardial damage, or a combination of these. Each man and woman in the sibships under consideration was assigned a probability of having myocardial infarction, as defined. The probabilities assigned were simply age- and sex-specific prevalence rates for probable or sus- spect myocardial infarcation in the total Tecumseh population. A computer program was developed to calculate the number of sibships, according to the number of examined persons in each sibship, in which multiple cases would be expected to coexist by chance alone. These expected numbers were compared with the actual numbers observed (table 3). TABLE 3.—Frequency of observed and expected* number of sibships** and cases with a diagnosis of myocardial infarction (probable and suspect); Tecumseh, Michigan, 19569-1960 Total number of Total number of cases Number ofexamined| Totalnumber of [Number of sibships members per examined persons in | cases per sibship sibship sibship Observed | Expected | Observed | Expected Blam ta is 264 (132 sibships).___ 2 3 .8 6 1.6 1 18 17.1 18 17.1 0 111 114.1 0 .0 For nrmes Bremen 81 (27 sibships)______ 3 0 .0 0 0 2 0 .5 0 1.0 1 1 5.1 1 5.1 0 26 21.4 0 .0 Bo mrmiin on ie 16 (4 sibships)_______ 4 0 .0 0 .0 3 0 .0 0 .0 2 0 .2 0 .4 1 0 «9 0 .9 0 4 2.9 0 .0 Total .________ 361 (163 sibships).___|____________ 163 163.0 25 25.7 *Expected frequencies based on age-sex-specific prevalence rates for Tecumseh population. **Including only those sibships with all examined members age 40 or over. There were 361 men and women in 163 sibships including 2, 3 or 4 examined members. Of the two-member sibships, 0.8 sibships were expected to show both members to be diseased against an observed number of three sibships in which both members had a diagnosis of probable or suspect myocardial infarction. The remaining observed and expected numbers agreed quite closely. Viewed differently, it may be seen that 6 of the observed 25 cases were, in fact, multiple cases within the same sibships, as compared with an expected fre- quency of 1.6 cases; whereas the remaining cases occurred singly in the sibships of various sizes. It may be noted that the prevalence of probable and suspect myocardial infarcation in this particular group is about 7 percent (25 cases in 361 persons), as compared with a prevalence of 5.5 percent in the hypothetical model discussed before. 270 The lesson learned up to this point, using either the hypothetical model or the preliminary test in Tecumseh, suggests that there is a significant problem relating to the adequacy of numbers in searching for familial aggregations of coronary heart disease in the general population. This problem exists not only in the case of coronary heart disease but also for other chronic diseases. Actually, familial ag- gregations of coronary disease are probably more marked than these calculations would suggest. It is likely that such aggregations could be more readily demonstrated if information on all family members were more accurate and complete. Thus, how can one be sure that a person who is labelled as being free from disease might not have severe coronary atherosclerosis, diagnostic methods being so insensi- tive? In a person who has died, how sure can one be of the accuracy of a diagnosis of arteriosclerotic heart disease as a cause of death and how many such cases may have been misclassified under another label (Beadenkopf et. al., 1963)? Another uncertainty relates to the difficulties in obtaining accurate information on persons who cannot be directly interviewed and examined. A further problem arises because the frequency of the disease is highly dependent on age, so that freedom from manifest disease at a younger age would not have the same significance as the same finding in an older person of the same sex. This makes it difficult to compare individuals and groups differing in age. Similarly, freedom from disease at a given age has not the same significance in men and women. Therefore, sibship comparisons should be age- and sex-specific, is—difficult to ac- complish on account of the problem of numbers; alternatively expected and observed rates, adjusted for age and sex, may be contrasted, as illustrated earlier. There is also a question whether data on living and deceased persons can be combined in an analysis as has been done in some studies. Thus, would it be reasonable to say that a man whose brother or father has just died of a myocardial infarction at the age of 75 or 80 has more of a “family history’ of the disease than another man whose brother or father is well and alive at that age? In other words, for a disease which is almost by necessity the major cause of death in older people, is it reasonable to give the same weight to these deaths, regardless of the age at which they occur? It is not possible, therefore, to combine data on living and deceased persons without giving special consideration to the problem of age at death. At the same time, data on living family members should not be considered without taking into account persons in the kindred who have died. Failure to detect familial aggregations of coronary disease among living members might be due, for instance, to the fact that relatives with the disease have died prior to the study. It is clear, therefore, that analysis should not be confined solely to the survivors. An attempt to take into consideration deceased relatives 271 has been made in the analysis of the data from the Tecumseh study, discussed earlier. The nature of these problems makes it unlikely that there is a single approach to their solution. In the field, a number of different ways of analyzing the data must be used to gain a composite picture from several pieces of evidence. One approach has already been illustrated in terms of comparing observed and expected frequencies. Other approaches might be briefly considered. 1. The mortality experience of the parents or siblings of propositi and controls may be compared. The proportion of parents or siblings living and deceased and the frequency distribution of the ages at death may be ascertained for the two groups. Life tables on cohorts might be constructed for the parents or siblings in the two groups and the survivorship curves contrasted. Such an analysis might sug- gest that changing environmental factors affect the relatives of pro- positi differently from those of controls. 2. In a study like the one in Tecumseh, one can compare morbidity and mortality among relatives living outside the study area. A greater frequency of disease among the relatives of index cases might indicate that one or more familial factors are operative. 3. As a further test for familial aggregations, the frequency of “suspect” disease among relatives of propositi and controls can be estimated. We will report on some of these approaches in the Tecumseh popu- lation when a more complete report can be given. In the meantime, it must be stressed that the problems discussed so far relate to the situation where data on a population include only currently present conditions and retrospective information. Longitudinal, prospective observations, as will be mentioned, will add greatly to the potential of such studies. PREDISPOSING FACTORS AND FAMILIAL AGGREGATIONS Up to this point, attention has been confined to the question whether coronary heart disease aggregates in families, regardless of the mechanisms involved in causing these aggregations. In fact, on a priori grounds, there can be no doubt that coronary disease must show familial clustering since the disease is clearly associated, in individuals, with three variables which are, in part, determined by heredity: serum cholesterol, blood pressure and glucose tolerance, using these terms in preference to hypercholesterolemia, hypertension and diabetes. In view of these relationships, it is rather surprising that familial aggregations of coronary disease are, as was shown, so difficult to demonstrate in a general population. A recent review of 272 familial studies of coronary heart disease has pointed out the short- comings of existing data and the need for new studies, using both the index case-control and community type of approach (Epstein, 1964). The subsequent discussion will be confined to serum choles- terol and blood pressure levels even though there is a great need to search for possible genetic influences which may exist independently of these two recognized associations. In familial studies, it would facilitate some types of analysis if persons could be properly subdivided into those with and without hypercholesterolemia or hypertension. The data from Tecumseh support the view that such subdivisions are artificial (Johnson et al., 1965). While the total evidence tends to favor the opinion that both serum cholesterol and blood pressure levels are determined by multiple genes and multiple environmental influences, it is convenient to divide the distributions arbitrarily into an upper and lower range by suitable cutting points, in order to estimate the effect of these variables on the occurrence of coronary heart disease. Toward this end, some further calculations based on the hypothetical population model of 1,000 brother-pairs used earlier are instructive. It will be assumed, for the purpose of these calculations, that hypercholesterolemia and hyper- tension are both inherited as dominant characters. The probable incorrectness of this assumption has already been pointed out but, from a practical point of view, both of these variables can be con- sidered to behave as if their inheritance were dominant, using a suitable cutting point between ‘normal’ and “abnormal.” This was illustrated by Platt (Platt, 1959), using the data of Pickering and his associates (Hamilton et al., 1960) on blood pressures which are, in the opinion of most workers, unimodally distributed. At face value, even uni- modally distributed data can be manipulated in such a way that the result is compatible with a dominant mode of inheritance, as discussed and criticized earlier by Oldham and associates (Oldham et al., 1960). For the purpose of the calculations which follow, it will be assumed, therefore, that blood pressure and also serum cholesterol levels, which show a similar familial pattern, are distributed in the general popula- tion as if each were determined by a single, dominant gene; whether this assumption is correct or not has no influence on the broad inter- pretation of the subsequent calculations. It is further understood that hypercholesterolemia and hypertension do not appear to be significantly associated (Dawber et al., 1957). Before turning to the concrete situation, regarding the interrelation- ship between coronary disease and elevated serum cholesterol and blood pressure levels, let us assume that the population can be sub- divided into persons with a factor, to be designated as X, inherited as a dominant character, which carries an increased risk toward the development of coronary disease, and persons without this predisposing factor, to be designated as x. Knowing the frequency of X in the 273 population of 1,000 brother-pairs, it can be calculated under the assumption of random mating by the use of the appropriate formulae (table 4) how many brother-pairs are characterized by the combina- tions X-X, X-x, and x-x. Taking the frequency of X as 30 percent, these pairs will occur with a frequency of 190, 219 and 591, respectively, among the 1,000 brother-pairs (table 4). Next, we will again set the prevalence of coronary heart disease among these men at 5.5 percent and assume that the disease is three times as frequent among men with X as among those with x. On these premises, one can calculate from the formulae in table 4 the distribution of cases of coronary disease in the three types of sibships X-X, X-x and x-x according to whether the brothers are concordant or discordant for X and x. It is found that 3.5 (2.04-0.8+0.7) of the 1,000 sibships will show both brothers affected, representing 7 cases (4.0+1.6+1.4) among the total of 110 cases (table 4). It will be recalled that the random expectation of simultaneously affected pairs was 3 per thousand. Surprisingly, no more than an additional 0.5 pairs with both brothers affected were added above random expectation. TABLE 4.—Frequency of coronary heart disease in a hypothetical population of 1,000 brother-pairs, according to presence or absence of a characteristic X Number Total Type of sibship Total number of sibships [of cases of Total number of sibships number CHD per of cases sibship X-X oe Ni1=N(p/4) (4+4pa+pq ?) 2 | kK?RIN1=2.0 4.0 =190 (380 persons) 1 | 2kR(1-kR)N1=35.0 35.0 0 | 1-kR)!N=1525 ~~ [_________ Xsswinnissms Ni=N(pq?/4) (6+2q) 2 | kR2N3=0.8 1.6 =219 (438 persons) 1 | [R(1—kR)+kR(1-R)]N3=28.5 28.5 0 | 1-kR)(1-R)N:=1900 | _________ pS SU N3=N(q?¥4)(1+q)? 2 | R3N3=0.7 1.4 =591 (1,182 persons) 1 | 2R(1-R)N3=39.5 39.5 0| A—-R)N3=5510 = __._..... Total...... N=Ni1+N2+N3=1,000 N=N1+N2+N3=1,000 110.0 (2,000 persons) = |..._..._.. anameies are based on a prevalence of 5.5 percent for coronary heart disease (CHD) among middle- aged men. Calculations assume dominant inheritance for X and a frequency for X of 30 percent. X-X indicates that both brothers of the pair have characteristic X; X-x indicates that one brother of the pair has characteristic X; x-x indicates that neither brother of the pair has characteristic X. Qspropottion of x individuals in the population. =prevalence rate for CHD among the x individuals, kR =prevalence rate for CHD among the X individuals, The numerical values used were N =1,000, q?=.70, R=.0344 and k=3. With some hesitancy, let us consider that X stands for the combined frequency of hypercholesterolemia and hypertension * in a population of middle-aged men. In fact, in such a population in this country, these conditions do occur with an approximate frequency of 30 percent, as assumed in the foregoing calculations, this figure being slightly lower than found in Framingham (Dawber et al., 1957) but slightly 4 Serum cholesterol 260 mg. percent or over; blood pressure 160 mm Hg. systolic or over and/or 96 mm Hg. diastolic or over. 274 higher than the Tecumseh experience. On the basis of this highly simplified hypothetical model, therefore, we have attempted to explain why familial clustering of coronary disease—and this applies to other chronic diseases as well—is difficult to demonstrate in a general popu- lation in a one-time, cross sectional study. Are these statements in contrast to the relatively few studies in which relatives of index cases showed, perhaps, a three- or four-fold excess of coronary disease as compared with relatives of control subjects (Epstein, 1964)? In view of the fact that these data include or are based on information on deceased family members, and on histories obtained from family members rather than direct examinations, there is no need to postulate that there is necessarily any contradiction. A final look at the population model reveals that 69.1 (4.0+35.04 1.6+28.5) of the 110 cases, or as many as 63 percent occurred among men who might be considered as belonging to hypercholesterolemic or hypertensive families. In the Tecumseh population, in the five out of the six sibships with multiple examined cases of coronary disease, there was at least one person with at least one of these variables in the upper range. Moreover, among 81 men with coronary disease aged 40-54 in Tecumseh, 65 percent showed serum cholesterol, blood pressure or both in the upper range, the corresponding figure among 52 women being 75 percent. It is suggested that the factors, both genetic and environmental, which contribute to elevated serum cholesterol and blood pressure levels, account in considerable part for aggregations of coronary heart disease in families. It is of interest to look at the population model also from the point of view of the investigator who would sample index cases and controls at random from this universe of 2,000 middle-aged brothers. In a 10 percent random sample of index cases and controls, 0.7 cases of coronary disease would be expected among the 11 siblings of index cases—a prevalence of 6.4 percent, as compared with a prevalence of 3.6 percent among 186 siblings of controls. This difference, though less than twofold, is nevertheless not negligible. OUTLOOK AND CONCLUSIONS In the foregoing discussions, an attempt was made to show that the search for familial aggregations of coronary heart disease and other chronic disorders of similar frequency of occurrence in the general population, on the basis of a single survey, will yield relatively few instances of multiple cases within sibships or, presumably, kin- dreds. Awareness of this problem is essential in planning such studies. Fortunately, seen in perspective, the problem has a solution. First of all, in a prospective study of coronary heart disease, new cases will accumulate rapidly. Since the prevalence of the disease among 275 middle-aged men in this country is about 5 percent and the yearly incidence about 1.5 percent (Epstein, 1960b), after an observation period of 3 or 4 years as many new cases will accumulate as were identified at the beginning. Thus, with each year of observation, the chances increase to detect the occurrence of new events where they would be expected to arise preferentially—i.e., among the rela- tives of the index cases identified at the outset. Secondly, the primary interest relates not to the detection of established disease within families but to the familial clustering of predisposing factors since, from the overriding point of view of disease prevention, it is the early detection of susceptibility to the future development of overt illness which commands major attention. The prevalence of these predisposing factors, such as elevated serum cholesterol and blood pressure levels as usually defined, is considerably greater than the frequency of the associated disease itself. The problem of rela- tively small numbers is thus reduced. There is much hope, therefore, that the extent of familial clustering of coronary heart disease and predisposing factors in the population at large will become known within the next few years. SUMMARY Aggregations of coronary heart disease in families are difficult to demonstrate in one-time, cross-sectional epidemiological studies of total populations. The index case-control type of approach may be misleading in this regard unless index cases and controls are repre- sentative of the universe from which they are derived. An actual population study and a population model have been used to illustrate and discuss the methodological problems involved in the study of familial clustering of coronary heart disease and, by implication, other chronic disorders. Even though two major factors predisposing to coronary atherosclerosis, elevated serum cholesterol and blood pressure levels, show definite familial trends, their effect on familial concentra- tions of coronary disease was shown to be surprisingly small on the basis of some theoretical calculations. Tt is suggested that prospec- tive rather than cross-sectional studies are needed to assess the true degree of familial aggregations of coronary heart disease. REFERENCES 1. Beadenkopf, W. G., Abrahams, M., Daoud, A. and Marks, R. U., 1963. An Assessment of Certain Medical Aspects of Death Certificate Data for Epidemiologic Study of Arteriosclerotic Heart Disease. J. Chronic Dis. 16: 249-262. 2. Dawber, T. R., Moore, F. E. and Mann, G. V., 1957. Coronary Heart Disease in the Framingham Study. Am. J. Pub. Health 47. Supple- ment to No. 4: p. 4-24. 276 10. 1% 12. . Epstein, F. H., 1960a. An Epidemiological Study in a Total Community: The Tecumseh Project. University of Michigan Medical Bulletin 26: 307-314. . Epstein, F. H., 1960b. Epidemiology of Coronary Heart Disease. In: Modern Trends in Cardiology, edited by A. Morgan Jones. London, 1960. Butterworth. . Epstein, F. H., 1964. Hereditary Aspects of Coronary Heart Disease. Am. Heart J. 67: 445-456. . Francis, T., Jr., 1961. Aspects of the Tecumseh Study. Public Health Reports 76: 963-965. . Hamilton, M., Pickering, G. W., Fraser Roberts, J. A., and Sowry, G. S. C., 1954. The Etiology of Essential Hypertension. IV. The Role of In- heritance. Clin. Sci. 13: 273-303. . Johnson, B. C., Epstein, F. H., Kjelsberg, M. O., 1965. Distributions and Familial Studies of Blood Pressure and Serum Cholesterol Levels in a Total Community—Tecumseh, Michigan. J. Chronic Dis. (In Press). . Napier, J. A., 1962. Field Methods and Responses Rates in the Tecumseh Community Health Study. American Journal of Public Health 52: 208- 216. Oldham, P. D., Pickering, G., Fraser Roberts, J. A., and Sowry, G. S. C., 1960. The Nature of Essential Hypertension. Lancet 1: 1085-1093. Platt, R., 1959. The Nature of Essential Hypertension. Lancet 2: 55-57. Syme, L., Hyman, M., and Enterline, P. E., 1964. Some Social and Cultural Factors Associated with the Occurrence of Coronary Heart Disease. J. Chronic Dis. 17: 277-289. 217 Inherited Antigenic Differences in Serum Beta Lipoproteins in Relation to Coronary Artery Disease and Diabetes By Baruch S. BLumBerg, M.D., BARBARA B. Misael, B.A., anp Sam Visnice, B.S, National Institutes of Health, Public Health Service, U.S. Department of Health, Education, and Welfare, Bethesda, Md. INTRODUCTION The method of genetic analysis described in this paper originates from the age-old observation that people differ in their susceptibility to illness. All those exposed to the tuberculosis organism do not develop clinical disease nor does everyone in an environment where rheumatic fever is common fall prey to it. The same may be said for a large number of infectious and noninfectious diseases. Barring those cases in which chance or divine intercession are operating, this differ- ence in susceptibility must be due to inherent differences between individuals, and some of these differences are inherited. The study of normal inherited variation may, therefore, help in differentiating those human differences which bear on disease resistance and suscepti- bility. Biochemical polymorphic traits are particularly useful for this purpose. In polymorphisms two or more discontinuous forms of an inherited trait are common in a population, and the frequency of the least common form is higher than could be expected from recurrent mutation alone (Ford, 1956). It has been theorized that a poly- morphism is maintained as a result of differences in selective value of the genotypes; and that the heterozygote is at a selective advantage as compared to either of the alternate genotypes. Selection is ex- pressed as differential fertility, that is, the relative contribution of the different genotypes to the genetic makeup of the next reproducing generation. Selection can take place at any time in the reproductive cycle or before the completion of childbearing, and any diseases which 279 have a differential selective effect on the genotypes could influence the gene balance. A list of polymorphic traits which have been studied in human populations is given in table 1 (Blumberg, 1961). TABLE 1.—Some polymorphic traits in humans Red blood cell types Serum proteins ABO Haptoglobins. MNS Transferrin. P Gamma globulins. Rhesus Beta-lipoprotein. Lutheran Group-specific substance. Kell Serum cholinesterase. Lewis Duffy Kidd Diego Sutter Sex-linked Other cell types Miscellaneous Hemoglobin BAIB, urinary excretion. G6PD Secretor of ABH blood group sub- stance. ‘White blood cell Taste of PTC. platelet GENERAL OUTLINE OF THE METHOD OF INVESTIGATION 1. The detection of discontinuous, biochemical differences between individuals. “Discontinuous” indicates that the qualitatively different forms of a trait are readily distinguishable from each other. This makes it possible to clearly designate an individual as of one or another type, and ambiguities in classification would occur only rarely. In con- tinuous traits, the differences are quantitative and classification is dependent on arbitrary divisions of the spectrum. In some cases, there can be two curves of continuous distribution with a definite division between them, and for practical purposes the trait can be considered as discontinuous. Many discontinuous biochemical differences between individuals have been found by the application of protein separation and identi- fication techniques to nonpooled blood samples. When Smithies (1955) developed the technique of starch gel electrophoresis, he was able to separate serum protein into more fractions than had been possible with previous methods. He noted that the protein patterns were not the same in all individuals; this led to the discovery of the haptoglobin polymorphism (Smithies and Connell, 1959), and later the transferrin polymorphism (Smithies and Hiller, 1959), the post albumin polymorphism (Smithies, 1959), the thyroxine-binding prealbumin differences in Macaca mulatta monkeys (Blumberg and Robbins, 1960), and others. Similarly, the development of rapid and 280 sensitive methods of immunoelectrophoresis (Grabar and Williams, 1955) led to the discovery of the Ge group-specific serum system (Hirschfeld and Beckman, 1960). The discovery of discontinuous differences is not limited to blood studies. Electrophoretic differences observed in cattle milk proteins resulted in the identification of the B-lactoglobulin polymorphism (Aschaffenburg and Drewry, 1955) and later in Fulani and Indian cattle, the alpha-lactalbumin system (Blumberg and Tombs, 1958). Studies on urine have revealed the polymorphism for the excretion of the amino acid, beta-amino iso- butyric acid (BAIB) (Gartler, 1959). One of the most rewarding means for detecting discontinuous bio- chemical differences is by the study of inherited differences in sensi- tivity to drugs, that is, by pharmacogenetics. The pseudocholines- terase polymorphism (Kalow, 1959) and the very important G6PD polymorphism were discovered in this manner (Hockwald et al., 1952; Beutler, 1959). 2. To determine if the discontinuous differences are inherited. a) Is the trait an inherent characteristic of an individual? In order to determine if an individual’s phenotype is always the same, or, if not, under what conditions it might vary, the phenotype is determined over the course of several days, weeks, or years, and if possible, under varying conditions of nutrition, climate, and any other factor which could conceivably alter it. For example, an individual’s erythrocyte sedimentation rate may vary depending on his clinical condition; it could not be considered an unvarying characteristic of the person by which he could be identified at any time. = His hemoglo- bin type (i.e., A/A, A/S, S/S) would nearly always be the same and detectable, barring some extraordinary pathological change. b) Age dependence. The late development of an inherited trait often makes genetic analysis difficult. A well known example in medical genetics is Huntington’s chorea, an inherited disease which may become apparent only late in life. Pedigree analysis of families with many young children may show a marked difference from the predicted frequencies, and from the frequencies determined in families in which the offspring have all reached the age at which the disease can develop. Age dependence is usually not a difficult problem for the analysis of biochemical polymorphic traits, although caution must be observed. The haptoglobins of many infants are not fully developed until ap- proximately 4 months of age and reliable phenotyping cannot be done before that time. A child’s Gm (gamma globulin) type is not expressed until sometime after birth; furthermore, the mother’s specificity may be transmitted transplacentally and detected in the child’s blood. There are, however, some instances when the age dependence may extend over a longer period of time. Steinberg and Giles (1959) have 281 735-712 O—65——19 described an inherited polymorphic trait with a striking age dependence. c) Sex distribution. Most of the polymorphisms are autosomal and independent of sex. However, the sex dependence has not been carefully studied, even for some of the better understood traits. At least two sex- linked biochemical polymorphisms are now known: the G6PD defi- ciencies and the Xg* blood group systems. For a newly discovered discontinuous trait, sex dependence can be determined by appropriate population and family investigations. d) Twin studies. Twin studies are very useful in determining if a discontinuous trait is inherited, although they do not give very much information on the mode of inheritance. If all identical twins are concordant, if non- identical twins are either concordant or discordant, and if the differ- ence is significant, then the likelihood that the discontinuous trait is inherited is very high. Tt is particularly useful if the trait can be studied on stored material, such as serum, for then a single collection of blood samples from a twin population can be used to study a variety of traits. Sera collected from twins for one population study (McDon- ough et al., 1963) have been retained in our laboratory for several years and used for subsequent investigation of serum polymorphisms. Twin studies are a rigorous test of a genetic hypothesis. If even one of the proven identical twin pairs are discordant for the trait, this will cast doubt on a simple genetic hypothesis or, depending on how strongly held the convictions are, on the monozygosity of the twins. e) Family studies. To learn the mode of inheritance, family and population studies are necessary. For many biochemical traits it is not possible to de- termine if a particular family is useful for genetic testing until the biochemical study is actually performed. However, for polymorphic traits, if two or more of the forms are relatively common and particu- larly if the heterozygote can be detected, then all families studied contribute to the genetic analysis. The method of analysis proposed by Smith (1956) is particularly helpful in efficiently utilizing the in- formation, and this has been used in several studies. For many of the traits, it is possible to use stored material. In our laboratory a file of family sera is maintained for use in the appropriate investigations. f) Racial differences. The frequency of the genes controlling polymorphic traits may vary enormously in different populations. The extreme of this is seen in traits which are essentially absent in some populations, but poly- morphic in others: i.e., the Diego blood group is relatively common in some Asian and American Indian populations, but extremely rare in populations of European and African origin (Levine et al., 1956); 282 the Js red blood cell group is common in Africans and populations of African origin, but rare or absent in other populations (Giblett, 1958). For many of the biochemical traits, it isn’t possible to ascertain the phenotype without actually performing the biochemical test. Hence, it is difficult to know which families to study to obtain critical matings for testing the genetic hypothesis. The usual practice is to collect as many families as possible in a given population, determine the phenotypes of each entire family, note the pattern of segregation, and from this propose and test a genetic hypothesis. In order to decide in which population to conduct the family (and twin) studies, the phenotype frequencies in several populations are determined. (For this purpose, collections of sera from different populations are extremely useful; some of the material available in our own laboratory is shown in table 2.) With this information, an intelligent selection of the appropriate population for family studies can be made. For example, if there are two forms of a trait, A; and A, and in population I, A, is 99 percent and A, 1 percent, but in population II, A, is 50 per- cent and A, 50 percent, then population II would be far more con- venient for the study of the genetics of the trait. TABLE 2.— Populations represented in serum bank America: Evans County, Georgia Eskimo: New York, New York Eskimos, Alaska Tangier Island, Maryland Indian: Africa: Athabascans, Alaska Fulani, Nigeria Athabascans, Canada Yoruba, Nigeria Haida, Canada Asia: Montagnais, Labrador Arabs, Jordan Naskapi, Labrador Chinese, Formosa Quechua, Peru Chinese, San Francisco Sioux, South Dakota Israelis, Israel Tlingit, Alaska Japanese, San Francisco Negro: Vietnamese, South Vietnam Evans County, Georgia Pacific Sapelo Island, Georgia ‘Washington, D.C. White: Bethesda, Maryland Boston, Massachusetts Aborigines, Australia Micronesians, Majuro Micronesians, Rongelap Micronesians, Utirik Polynesians, Bora Bora 3. Biochemical nature of the inherited trait. An advantage of studying biochemical polymorphic traits is that the investigator is supplied with some ‘hints’ as to the physiological or pathological role of the polymorphism, while this may not be the case when investigating morphological traits. For example, when Smithies determined that the haptoglobin types were inherited, a large body of information on the clinical, pathological, and physiologi- cal characteristics of the protein was available to provide a starting point for investigation. Jayle and his colleagues (Jayle et al., 1952) had for years been studying the variation in the amount of this pro- tein in a variety of illnesses, and it was known that the haptoglobins were responsive to inflammatory diseases. It was also known that haptoglobin combined with hemoglobin, and that this might be one of its physiological functions. This led to investigations of the role 283 of haptoglobin in hemolytic diseases such as the hemoglobinurias, malaria, thalassemia, sickle cell disease, and others (for discussion see Blumberg et al., 1963). Similarly, the role of the transferrins or siderophilins in iron binding was known before the discovery that these proteins were polymorphic. This gave an immediate clue as to the physiological function which might be first studied. There is, however, no good evidence at present that the iron-binding function is implicated in the selective forces which control this complex in- herited trait. A further application of biochemical polymorphisms is in the study of the genetic control of protein synthesis. They provide readily identifiable common traits which can be used in combined biochemical and genetic studies. The brilliant investigations of Smithies and his colleagues (1962) and Parker and Bearn (1963) on the genetic control of haptoglobin synthesis are excellent examples. Polymorphisms also provide suitable systems for the study of population genetics of human cells in which these biochemical traits can be used as markers. The studies of Gartler et al. (1962) and others, using the G6PD trait, are examples of the application of this method. 4. Relation of the polymorphic traits to disease. a) Direct association of a genotype with disease. In some instances one of the genotypes may be directly associated with a disease entity. An example of this is the sickle cell homozy- gote. In some polymorphic systems there is no apparent direct association with illness, but when individuals with one or more of the genotypes are exposed to special hazards, symptoms may occur. Under normal conditions, bearers of the G6PD deficiency appear to be quite normal. However, if they are exposed to a special hazard, such as eating fava beans, or taking any of a large number of drugs, they can develop hemolytic anemia. Sickle cell heterozygotes appear to be clinically normal, but when exposed to a special hazard such as the low pressures experienced in high-flying nonpressurized aircraft, they are more likely to suffer splenic infarct than individuals of differ- ent genotype. Similarly, individuals with the rare pseudocholines- terase variant are under ordinary circumstances quite normal; however, if they are given certain muscle relaxants, such as suxameth- onium, a prolonged apnea will result. This group of polymorphic traits are of particular interest because they lend themselves to practical use in preventive medicine. By means of simple chemical tests, individuals who have an increased inherited susceptibility to known environmental hazards can be detected and steps taken to avoid exposure. b) Association of rare alleles or phenotypes with an illness. Another point of contact of polymorphic systems and clinical disease is the association of a rare allele or phenotype with an illness. There are few well-documented cases of this kind. Ahaptoglobinemia, 284 which is rare in most non-African populations, may be associated with a variety of hemolytic illnesses, as noted above. Many rare alleles do not appear to be associated with illness, how- ever. For example, most of the rare transferrin phenotypes have been reported in apparently normal persons (see, for example, Robin- son el al., 1963). ¢) Statistical association of polymorphic traits with disease. For most polymorphic systems the disease association is not as direct as those described above. However, for many there is a statis- tical association which in some cases may be quite striking. The systems which have been most studied in this regard are the blood groups. These have been reviewed by Roberts in this symposium and will not be discussed here. AN EXAMPLE OF A POLYMORPHISM STUDY: THE Ag SYSTEM, AN INHERITED DIFFERENCE IN THE g- LIPOPROTEIN The first part of this paper was devoted to a general discussion of how biochemical polymorphic traits are detected and analyzed in respect to their genetics, biochemistry, and disease association. In the present section, these methods will be illustrated by the use of a polymorphic trait that has been studied in our laboratory. The Ag system was discovered as a result of a search for a poly- morphism rather than as a chance observation. Experience over the past 10 years has shown that there are large numbers of serum protein polymorphic systems. These include the haptoglobins, transferrins, Gm gamma globulin groups, Ge group-specific alpha, globulin system, and others. The frequencies of the genes controlling some of these systems are such that if a patient were to receive even a small number of transfusions, the probability of transfusion with a serum protein slightly different from his own would be very high. If this were the case, then the patient might develop an antibody against the foreign protein. The method of double diffusion in agar gels was used in the investiga- tion. The serum of a transfused patient was placed in the center of a 7-hole Ouchterlony pattern. The sera of normal individuals who could have been, but were not necessarily donors for the patient, were placed in the peripheral wells. After approximately 15 patients had been studied, one was found who gave a strong precipitin line with some but not all the donor sera (Blumberg, 1961). It was immedi- ately apparent that this antiserum could serve as a reagent to study what appeared to be a new polymorphic system. Biochemical and immunological studies showed that the antibody present in the patient’s serum was a 7S gamma globulin and that the antigen in the normal serum with which it reacted was the low-density beta-lipopro- tein (Blumberg, Dray, and Robinson, 1962). The ability of a serum to react was found to be an inherent characteristic of an individual. 285 Sera obtained from subjects over the course of several years and under different conditions were tested and found always to give consistent results (Allison and Blumberg, 1961; Blumberg, Bernanke; and Allison, 1962). Furthermore, serum which had been frozen and stored for up to four years still retained the ability to react with the antiserum. This permitted the use of stored serum from our serum bank for many of these studies. Twin studies indicated that it was highly probable that the reactor specificity was inherited (Blumberg, Bernanke; and Allison, 1962), and the family studies were consistent with the hypothesis that the reactor specificity was inherited as a simple autosomal dominant trait (Blumberg, Bernanke; and Allison, 1962; Blumberg, 1963). The reactor phenotype was called Ag(a+), the nonreactor phenotype Ag(a—), the gene controlling the reactor phenotype Ag*, and the alternate gene whose product had not been discovered Ag. Individuals homozygous or heterozygous for the dominant gene Ag* are phenotype Ag(a+) and those homozygous for Ag are phenotype Ag(a—). Since the discovery of the first precipitin-containing serum, nearly 20 others have been found. (Blumberg, 1963; Blumberg and Riddell, 1963; Blumberg, Alter, Riddell, and Erlandson, 1963). By im- munological and epidemiological studies, it has been shown that these can distinguish several different specificities. The genetics of two of these specificities have been studied in some detail, but the exact genetic interrelationship of the two has not been determined. An interesting problem in the study of the genetics became apparent as more of the antisera were discovered. For several of them (i.e., New York and J.B.) nearly all the sera from U.S. whites and Negroes were reactors. The probable explanation for this is that patients who lack an antigen common in the donor population are likely to develop an antibody against the common antigen as a result of repeated transfusions. As a consequence of this, it is necessary to conduct the family and twin studies in populations where the antigens are less common so that suitable families for testing a genetic hypothe- sis can be more readily found. The results of our surveys in several populations are shown in table 3. These indicate that American Indian and Oceanic populations are most suitable for genetic studies with several of the antisera. Having determined that the antigenic difference was inherited, the next step was to determine what, if any, was the association with disease. For the first antibody (anti-Ag(a-+)), there was no indication of any direct association of any of the phenotypes with overt disease. There is less information about the other antisera, but further investigations are in progress. A priori it was difficult to know which diseases to study, but it was considered logical to investigate illnesses in which the g-lipoprotein 286 TABLE 3.—Reactors with antisera in different populations C. de B. New York J. B. A. DiB 8.1L. Anti—A Anti—B Num- | Per- | Num- | Per- | Num-| Per- [| Num-| Per- | Num- | Per- ber | cent+ ber | cent ber | cent+ ber | cent+ ber | cent+ ‘Whites (U.S.A.)._. 120 59 164 100 168 98 144 7 157 97 Micronesians (Rongelap)....___ 194 98 185 45 54 98 45 24 50 24 Naskapi- Montagnais (Labrador)... 234 97 103 85 234 93 91 73 93 95 was known to be abnormal. These include coronory artery diseases, arteriosclerosis, diabetes, a variety of familial lipodystrophies, and others. The results of several preliminary investigations will be given here. 1. Coronary artery disease. Through the kindness of Dr. T. R. Dawber, Heart Disease Epidemi- ology Study, Framingham, Mass., we were able to obtain serum specimens suitable for a preliminary investigation. The design of the Framingham study has recently been discussed by Dawber et al. (1963). Briefly, approximately 13 years ago, a health survey of a large portion of the adult population of Framingham was conducted. Most of these people have been followed and those who have developed coronary artery disease, as well as other forms of heart disease, have been detected. Dr. Dawber sent to us sera from 69 individuals who had developed coronary artery disease and sera from 67 individuals of the same sex and approximately the same age who had not. Most of these specimens were drawn several years before the testing, but the test and control sera have approximately the same history of storage and handling. All the sera were phenotyped using three of our antisera. The results are shown in table 4. There is no significant difference between the control group and the patients for any of the antisera which were studied. The frequencies of reaction with New York antiserum and J.B. antiserum are lower than those previously reported, but this is probably due to the long storage of the sera (Blum- berg and Riddell, 1963). However, since both the patient and the control samples have been handled in the same manner, it is unlikely that this would have significantly altered the results. Although this study could hardly be said to rule out the possibility of an association, it does suggest that there might be more profitable avenues to follow. 2. Diabetes. For the preliminary study, American Negro patients with known diabetes were casually selected from the Diabetes Clinic at the Washington Hospital Center over a 4-month period. For controls, sera were obtained from Negro employees of the Center during the 287 TABLE 4.— Association of Ag systems with disease Human anti-human antisera C.de B. New York J.B. Anti—Ag(a+) Anti—Ag(b+) Per- Per- Per- Total | cent ? Total | cent ? Total | cent ? POS POS POS Coronary artery disease_____ 68 60.3 1 67 88.1 3 57 75.4 13 COMPO). vnc wnvevimemmmmnns 66 59.1 x 66 77.3 2 56 76.8 11 DIaboles. coc cnnninsicrnmmnns 208 71.6 0 205 96.6 2 205 97.1 2 Control. .___________________ 166 55.4 1 159 98.1 2 154 94.8 5 Rheumatic fever____________ 165 83.6 0 37 | 100 0 35 | 100 0 Control.________ Bec seins ii 152 82.2 0 34 | 100 0 34 | 100 0 routine preemployment medical examination over the same period of time. There was a higher frequency of C. de B. reactor (Ag(a+)) individuals among the diabetics than among the controls, and this difference is significant (x*=10.4, P<.001). The differences using the other antisera were not significant (table 4). We are now evaluating the meaning of this finding and have received some comfort in our agony from the words of the late Prof. Sir Ronald A. Fisher. Follow- ing a lecture on correlations at the National Institutes of Health several years ago, I excitedly presented to him some data which showed a definite correlation between certain inherited traits. When TI asked him if he could comment on the value of these results, he said “Blum- berg, you have something they can never take away from you.” Before making any conjectures on the meaning of these results, it is clear that additional studies should be done. If these show the same striking difference, then a further examination of the genetic and biochemical significance of the correlation would be warranted. 3. Rheumatic fever. Susceptibility to rheumatic fever is not generally associated with abnormalities of lipoproteins. However, there have been some studies which indicate that the serum lipid-containing fractions are abnormal in this disease. Our major reason for studying this disease was the ready availability of sera from patients with rheumatic fever and their spouse controls. These were presented to us through the kind- ness of Dr. Thomas D. Dublin and his coworkers. The method of ascertainment and selection of patients and controls is described elsewhere in this conference. There is no difference in the frequency of reactors with C. de B. in the patient and control group. To conserve antiserum, a smaller number of patients and controls were studied with the other two antisera, and they also showed no differ- ences (table 4). 288 4. Lipodystrophies. The sera of patients with a variety of lipodystrophies were pre- sented to us through the courtesy of Dr. D. Fredrickson. In one of these, the C. de B. antiserum reacted with the patient’s serum to form a precipitin line; the line did not stain with Sudan Black, but did do so with Azocarmine. This suggests that the antigen is not a lipoprotein. We have seen only one other instance in the thousands of sera tested with C. de B. in which the antigen did not contain a lipid-staining material (Blumberg, Alter and Riddell, 1963), and in this other case it was not possible to examine the subject. We are continuing our investigation to determine if this rare phenotype is in any way connected with the patient’s illness. Patients with acanthocythosis lack all or nearly all serum beta- lipoprotein. We have studied four such patients and the immediate families of two of them. There is no convincing evidence at present that a gene for absent beta-lipoprotein exists. However, in two of the cases we found that the serum of these patients formed a precipitin line with a lipid-containing protein present in the serum of some but not all normal persons. One of these abetalipoproteinuria patients had received a single transfusion ten years prior to the examination, and the second had never had a transfusion (Blumberg, 1963). This suggests that if in fact this material is an antibody, the patients de- veloped it as a result of immunization with some nonhuman material. The other possibility is that some form of “autoimmunity” is involved. SUMMARY A method of genetic analysis of disease susceptibility using poly- morphic traits is described. The beta-lipoprotein Ag system is used to illustrate the technique. REFERENCES 1. Allison, A. C., and Blumberg, B. S. 1961. An isoprecipitation reaction distinguishing human serum protein types. Lancet 1: 634-637. 2. Aschaffenburg, R., and Drewry, J. 1955. Occurrence of different beta- lactoglobulins in cow’s milk. Nature 176: 218-219. 3. Beutler, E. 1959. The hemolytic effect of primaquine and related com- pounds. A review. Blood 14: 103-139. 4. Blumberg, B. S. 1963. Iso-antibodies in humans against inherited serum low-density beta-lipoproteins, the Ag system. Ann. N.Y. Acad. of Sciences 103: 1052-1057. . 5. Blumberg, B. S. 1961. Inherited susceptibility to disease. Its relation to environment. Arch. of Environ. Health 3: 612-636. 6. Blumberg, B. S., Alter, H. J., Riddell, N. M., and Erlandson, M. 1964. Multiple antigenic specificities of serum lipoproteins detected with sera of transfused patients. Vox Sanguinis 9: 128-145. 7. Blumberg, B. S., Bernanke, A. D., and Allison, A. C. 1962. A human lipoprotein polymorphism. J. Clin. Investigation 41: 1936-1944. 289 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 25. 26. 27. 28. Blumberg, B. 8., Dray, S., and Robinson, J. C. 1962. Antigen polymorphism of a low density g-lipoprotein. Allotypy in human serum. Nature 194: 656-658. Blumberg, B. 8., Kuvin, 8. F., Robinson, J. C., Teitelbaum, J. M., and Contacos, P. G. 1963. Alterations in haptoglobin levels. J. Amer. Med. Assoc. 184: 1021-1023. Blumberg, B. S., and Riddell, N. M. 1963. Inherited antigenic differences in human serum beta-lipoproteins. A second antiserum. J. Clin. Invest. 42: 867-875. Blumberg, B. S., and Robbins, J. 1960. Thyroxine-serum protein complexes: single dimension gel and paper electrophoresis studies. Endocrin. 67: 368. Blumberg, B. 8., and Tombs, M. P. 1958. Possible polymorphism of bovine a-lactalbumin. Nature 181: 683-684. Dawber, T. R., Kannel, W. B., and Lyell, L. P. 1963. An approach to longitudinal studies in a community. Ann. N.Y. Acad. of Sciences 107: 539-556. Ford, E. B. 1956. Genetics for Medical Students. London, Methuen. Gartler, S. M. 1959. A metabolic investigation of urinary B-aminoisobutyric acid excretion in man. Arch. Biochem. and Biophys. 80: 400-409. Gartler, S. M., and Gandini, E., Ceppellini, R. 1962. Glucose-6-phosphate dehydrogenase deficient mutant in human cell culture. Nature 193: 602-603. Giblett, E. R. 1958. Js, a “new” blood group antigen found in Negroes. Nature 181: 1221-1222. Grabar, P., and Williams, C. A., Jr. 1955. Méthode immuno-electrophorétique d’analyse de mélanges des substances antigéniques. Biochem. et Biophys. Acta 17: 67-74. Hirschfeld, J., and Beckman, L. 1960. A new group-specific serum system (GC groups) in relation to blood and serum groups. Acta Genet. 10: 48-53. Hockwald, R. 8., Arnold, J., Clayman, C. B., and Alving, A. S. 1952. Status of primaquine. 4. Toxicity of primaquine in Negroes. J. Amer. Med. Assoc. 149: 1568-1570. Jayle, M. F., Boussier, G., and Badin, J. 1952. Electrophoresis de I’hapto- globine et de son complex hemoglobinique. Bull. Soc. Chim. Biol. 34: 1063-1069. Kalow, W. 1959. Cholinesterase types. In CIBA Found. Symposium. Biochemistry of Human Genetics. Boston, Little, Brown and Co., Levine, P., Robinson, E. A., Layrisse, M., Arends, T., and Dominques Sisco, R. 1956. The Diego blood factor. Nature 177: 40-41. McDonough, J. R., Hames, C. G., Greenberg, B. G., Griffin, L. H., Jr., and Edwards, A. J., Jr. 1963. Observations on serum cholesterol levels in the twin population of Evans County, Georgia. Circulation. In press. Parker, W. C., and Bearn, A. G. 1963. Control gene mutations in the human haptoglobin system. Nature 198: 107-108. Robinson, J. C., Blumberg, B. 8., Pierce, J. E., Cooper, A. J., and Hames, C. G. 1963. Studies on the inherited variants of blood proteins. II. Familial segregation of transferrin B,—,B,. J. Lab & Clin. Med. 62: 762 765. Smith, C. A. B. 1956. A test for segregation ratios in family data. Ann. Hum. Genet. 20: 257. Smithies, 0. 1959. An improved procedure for starch-gel electrophoresis. Further variations in the serum proteins of normal individuals. Biochem. J. 71: 585-587. 290 29. 30. 31. 32. 33. Smithies, O. 1955. Zone electrophoresis in starch gels: group variations in the serum proteins of normal human adults. Biochem. J. 61: 629. Smithies, O., and Connell, G. E. 1959. Biochemical aspects of the inherited variations in human serum haptoglobins and transferrins. In CIBA Found Symposium: Biochemistry of Human Genetics. Boston, Little, Brown and Co. Smithies, O., Connell, G. E., and Dixon, G. H. 1962. Chromosomal rear- rangements and the evolution of haptoglobin genes. Nature 196: 232-236. Smithies, O., and Hiller, O. 1959. The genetic control of transferrins in humans. Biochem. J. 72: 121. Steinberg, A. G., and Giles, B. D. 1959. A genetically determined human serum factor selected by its effect on a mating reaction in yeast. Amer. J. of Human Genetics 11: 380-384. 291 PART IV—Continued Regional Variations in Morbidity and Mortality Regional Variations in the Incidence of Spina Bifida By Davio Hewrirr, M.A., Department of Social Medicine, Oxford University, Ozford, England. The contributors to this symposium, and especially to the workshop, were enjoined ‘“‘to emphasize the experimental design of (their studies) with . . . particular emphasis on how (they) proposed to define familial factors and distinguish between the genetic and the non- genetic. . ..” For the purpose of such a methodological discussion the study which I have to describe is a nonstarter. It is merely a piece of descriptive epidemiology, based entirely on official population statistics, and thus employing no information—genetic or nongenetic— about the individual families in which cases of spina bifida occur. However, it did give rise to a somewhat cloudy hypothesis concerning a genetic aetiology for spina bifida, and possibly for other related malformations of the central nervous system. It will be for others to judge whether this hypothesis deserves to be taken seriously and, if so, to suggest an appropriate design for a further study to test the hypothesis. So far from starting with a problem and then setting up appropriate machinery for studying it, I started with certain machinery—es- sentially a desk calculator and a medical library—and then looked for a problem to which this equipment might usefully be applied. A great abundance of second-hand epidemiological data is available in the publications of the National Office of Vital Statistics and in par- ticular in their series of annual reports on the “Vital Statistics of the United States” (National Office of Vital Statistics, 1947-62). This suggested the feasibility of a study of regional—specifically of interstate—variations in mortality. To avoid the problems of inter- pretation which arise when there is a possibility of selective migration of sick persons, it seemed prudent to focus on mortality occurring very early in life. Two other desiderata affecting the choice of a particular topic were: 295 (i) that the statistics should indicate more than trivial differences between States, and (11) that the reported differences should not be artefacts produced by interstate differences in diagnostic practice. Tables 1 and 2 show that, according to certain criteria, spina bifida meets both these requirements. In table 1 are given values of a statistic intended to measure the amount of interstate variation in the mortality of white infants during the decade 1950-59. This statistic is analagous to the coefficient of variation in common use, but has been defined in such a way ! as to measure only the relatively permanent differences between States, excluding differences due to chance or other short-term factors. It will be seen that, relatively speaking, the infant death-rates from spina bifida are more than twice as variable from State to State as are the infant death-rates from all causes, and more than five times as variable as the infant mortality from congenital malformations as a whole. TABLE 1.—“Reduced” coefficient of variation of mortality rates (based on white infants of each State and of the District of Columbia, 1950-54 and 1956-59) All causes: Percent Ages 0 MORNE 1011 MOBEHE.... « «.cvomin cnn sms ssumimmmimie mses SHEET EES a as emma mma 12.5 Ages under 1 month Ages 1 to 11 months Congenital Malformation: & irculatory system. ___ Other malformations The credibility of this reported interstate variation of spina bifida mortality was tested by correlating the State death-rates with an index of the quality of death certification in each State. The con- struction of this index has been described elsewhere (Hewitt, 1963). As would be expected of such an index it tends (see table 2) to be low where post-neonatal infant mortality as a whole is high. It also has a significant positive correlation with the reported mortality from cer- tain types of malformation, suggesting that, in some States at least, the reported rates may be low merely because of inadequate diagnosis or certification. However, for spina bifida the correlation is as near zero as one could wish. TABLE 2.—Correlation between an index of diagnostic quality and the reported mortality of white infants in each State, 1960-69 All causes: ARES 110 TI MNOMENS. coors ss nssmmmmins beens —————————————— —0. 481 Congenital malformation: Spina bifida _____ irculatory syste Hydrocephalus*_______ a. 4 Other malformations*________________________________ TTT +0. 428 *Including some deaths after the age of 1 year. 1 (Co-variance of 1950-54 rate with 1955-59 rate) Mean of 1950-54 and 1955-59 rates 296 Hence spina bifida is a “good” subject for armchair epidemiology, because the geographical variation is both large and credible. The next question is, how can this geographical variation be sum- marized? Following the precedent set by Wesley’s (1960) study of congenital malformations it was decided in the first instance to correlate the State mortality rates with measures defining the geo- graphical position of the States, thus testing the possibility that the risk of spina bifida might vary in continuity over some or all of the country. In Wesley’s work a prior interest in the possible relevance of cosmic-ray energy flux caused him to use, as his one and only measure of position, the magnetic latitude; that is, a particular joint function of geographical latitude and geographical longitude. In the present case it seemed preferable to give independent consideration to latitude and to longitude. The result was striking. Spina bifida mortality in the United States has no correlation with latitude (r=-—0.046). On the other hand, longitude of the State capital is a very good predictor of spina bifida mortality (r=—0.839), and may be sald to “account for’ just over 70 percent of the variance between States, or more than twice as much as remains to be accounted for by all those factors, including sampling error, which are not correlated with longitude. The regression equation estimates that between the extremes of longitude spanned by the conterminous United States, risk of fatal spina bifida differs by a factor of nearly three. Figure 1 shows how this relationship appears on the map. This map differs from the one drawn by Alter (1962) by incorporating 4 more years’ data and, more importantly, by excluding nonwhites, whose low average rates and disproportionate concentration in the southern States complicate the picture. When only white infants are considered, this picture is rather simple. Of the 24 States with rates below the national median, 20 lie entirely west of the Mississippi, and 3 of the 4 exceptions are contiguous with the low area. Only Connecticut is “out of place”. What happens outside this range of longitude? It would be a naive extrapolation which suggested that the rate in Alaska should be lower than for any of its 48 seniors and that in Hawaii the rate should be lower again, but both these suggestions turn out to be correct! Moreover, Puerto Rico, which lies somewhat east of Maine, has a rate higher than all but one of the present 50 States. The evidence for east-west variation is greatly strengthened by data from Canada (Dominion Bureau of Statistics, 1957-61). This is shown in table 3, where the Provinces have been arranged in order from east to west. Except that the rate for Quebec is too high, this arrangement also corresponds to their ranking by spina bifida mortality. The cases of spina bifida which enter into the numerators of these death-rates are far from being all that occur. Some cases will die 297 735-712 O—65——20 Rank Order of States 1-12 (highest mortality) 13-24 25-36 37-48 (lowest mortality) FiGure.—Mortality of white infants attributed to spina bifida and meningocele in each State of the U.S.A., period 1950-59. TABLE 3.— East-west variation in Canada of mortality attributed to congenital malformation (Per 10,000 live births, 1955-59) Spina Bifida Other Malformations Newfoundland and Maritimes 9.8 51.1 QUEDRE. cc nc cn 11.7 57.4 Ontario__ 8.0 50.0 Manitoba. __ 6.9 45.8 Saskatchewan. 5.3 45.1 Alberta___________ 4.6 48.0 British Columbia 4.1 44.6 in utero and not appear in the cause-of-death statistics, while others which do appear in these statistics will be placed in a different category because of coexistent hydrocephalus. It has to be allowed that inclusion of these lost cases might modify the geographical pattern to some extent. But there is reason to think that they would reinforce rather than weaken the general east-west trend. Thus, post-natal mortality certified as due to hydrocephalus is also heavier in eastern than in western States and—a significant detail—this is especially true of females. Also the data on medical causes of stillbirth in the Provinces of Canada (Dominion Bureau of Statistics, 1957) indicate that intra-uterine death attributable to spina bifida is commoner in Quebec than in Ontario, and commoner in Ontario than in any of the four western Provinces. No parallel can be traced in the mortality of nonwhites. Analysis of the data on nonwhites is complicated by the fact that American Indians, Negroes and other nonwhites have reported mortalities from 298 spina bifida in approximately the ratio 6:3:1. These contrasting groups have very different geographical distribution, but cannot always be separated out in the statistics. A striking difference between the white and Negro infants dying of spina bifida is that among the former (in Europe as well as in America) there is a great preponderance of girls; among Negroes there is a more normal and distinctly masculine sex-ratio. When such a basic epidemiological characteristic as the sex-ratio among affected individuals differs so markedly between the races, one should perhaps not expect the geographical patterns to be the same. At first look the east-west gradient of the white death-rates may seem nonsensical. The strong correlation with latitude found in multiple sclerosis strikes one as quite reasonable, even while the reason for it remains unknown, because of the variety of conceivably relevant physical factors which also vary with latitude (Acheson, Bachrach and Wright, 1960). But no east-west polarity exists in nature. Hence, if we are to make any sense of the finding, we need to turn up some factors in the history or culture of the human popula- tion which—in this continent?’—could express themselves as differ- ences between east and west. Lacking any data of a more appropriate type, we can only pursue the search for such factors by correlating the State death-rates with whatever other average characteristics of the State populations happen to be known. In this way we may hope to find a principle of classification of the States which will separate out those with high and low mortality rates from spina bifida, which will necessarily correspond to some extent to the classification by longitude, but which will be more directly meaningful than is degrees west. Table 4 summarizes some of the classifications which have been tried so far. In each column of the table there appear 4 figures; each figure is the average of 12 State mortality rates from spina bifida (Alaska, Hawaii and the District of Columbia have been excluded), and each group of 12 States has been composed according to the ranking of the States by the characteristic named, with the highest on line A and the lowest on line D. In the first column the States have simply been ranked by spina bifida mortality itself, to provide a standard of comparison for the other columns of the table. Column 2 recapitulates the finding on longitude. One fairly safe generalization about western areas of the United States is that they have been under white settlement for a relatively short time. In case this should be thought relevant, column 3 shows the effect of classifying the States by a rough index of ‘‘community age’, namely, the order of admission of the States to the Union. ? In Europe, too, some of the variations in frequency of spina bifida as well as of anencephalus conform to an east-west pattern, but with the higher levels in the west (Penrose, 1957; Lamy and Frézal, 1959). PAY) TABLE 4.— Mortality rates of white infants from spina bifida averaged over the States lying in the first, second, third and fourth quartiles on certain rankings 1 2 3 | 4 | 5 6 7 6.05 5.50 4.94 4.47 4.34 5.44 5.44 4.58 4.89 5.00 5.39 3.96 4.63 3.91 3.74 3.78 3.78 4.17 4.10 3.82 3.89 2.46 2.65 3.11 4.42 4.42 2.94 3.60 Key to rankings: . By recorded mortality from spina bifida. . By longitude of State capital. By order of admission to the Union. By ratio of 1890 to 1960 population. . By percentage of 1959 mothers foreign-born. . By percentage of naval recruits lifetime residents of State. . By a feature of the birthweight distribution (see text). Noma w= Pursuing the same line of thought, the newer communities tend to be those whose populations have expanded most rapidly in the last two generations. Accordingly, column 4 uses a classification by the ratio between the 1890 and the 1960 population. This is fairly successful in picking out the States with the lowest rates. One element in the 20th-century population expansion has been new immigration from overseas, but, among the parents of today’s children, first-generation immigrants are not particularly numerous, and they do not make up any larger proportion in the western States generally, or in the States with low spina bifida mortality (see column 5). Probably a more important factor in the differential growth-rates of State populations has been the westward tide of internal migration, in consequence of which the young adult populations of western States contain a relatively high proportion of people who have moved some distance from their parental homes. Adapting figures given by Edwards and Palmer (1963) for the proportion of white naval recruits who were lifetime residents of the State in which they enlisted, we obtain the ranking of States shown in column 6. This index of the “rootedness,” as against mobility, turns out to be almost as good a predictor of spina bifida mortality as is the longitude. Why should such an index have any predictive value? One possible explanation might be that matings in the more mobile populations more frequently involve outcrossing between stocks with contrasting allelic frequencies at a number of loci, so that the offspring tend to have a higher pro- portion of loci in a heterozygous state. According to one of the theses put forward by Lerner (1954), and previously discussed in relation to human malformations by Neel (1958), this could reduce the risk of abnormal morphological development. One attraction of Lerner’s thesis in this context is that it suggests a perspective on regional and temporal variations of spina bifida much broader than that of America in the 1950’s. Though the rates on the Atlantic seaboard may have been high by American standards, yet they were lower, for example, than contemporary rates in any major section of the British Isles, from which much of the parental stock (in particular, that of Canada’s 300 Maritime Provinces) was drawn. We must also find a place in the picture for the downward secular trend of spina bifida mortality (MacMahon ef al., 1951; Gittlesohn and Milham, 1962; Hewitt, 1963). It may be that what we have to explain is not an excess of affected individuals in certain parts of North America, but a falling incidence in the white population of the world, with particularly low rates in those regions where there is the most intermarriage between formerly separate subpopulations such as the European national groups. What independent evidence is there of greater outbreeding in the western States and Provinces than elsewhere, or of any connection between this and frequency of spina bifida? I shall offer two findings consistent with these ideas. The first was suggested by a paper of Morton’s (1958) in which he stated that on theoretical grounds relatively intense inbreeding would be expected to reduce the mean and increase the variance of the distribution of birthweights. As it happens, this distribution is known for liveborn white infants of each State through the years 1955-59. Because scatter towards the lower end of the birthweight distribution could be due to a variety of irrelevant factors, I decided to try a measure based on the scatter at the upper end of the distribution, namely the ratio number of live births weighing 5,001 g or more number of live births weighing 4,501 g or more This somewhat esoteric index appeared to perform rather well (see column 7 of table 4), in that it grouped the States into 4 sets whose average mortality rates were in exactly the predicted rank order (P=1/24). The second and more direct piece of evidence relates to Canada, and is derived from a special tabulation of legitimate live births in 1950 according to the “national origin” of the parents in combination (Dominion Bureau of Statistics, 1955, table 34). Matings between persons of the same ‘national origin’ ranged from a low of 37.5 percent of all matings in British Columbia up to a high of 91.7 percent in Quebec. As previously shown, these two Provinces were respec- tively the lowest and the highest for spina bifida mortality. Between these extremes the five other divisions of Canada shown in table 3 preserve almost the same ranking for spina bifida as for this measure of endogamy, the only discrepancy being the reversal of Manitoba and Ontario. The composite hypothesis suggested on the basis of this study is therefore as follows: (1) that the incidence of spina bifida in white populations is in- fluenced by the average degree of homozygosity; and (ii) that both the downward time-trend of recent years and the downward geographical trend as one proceeds westward across America are due to greater outbreeding. 301 This is clearly akin to the hypothesis of recessive inheritance proposed by Polman (1950) on the basis of certain inbred Dutch communities in which he considered there was an excessive frequency of several types of central nervous system malformation. One followup to the present study which may seem worthwhile, whether or not the suggested hypothesis is taken seriously, is to determine whether anencephalus is also commoner in the east than in the west of the United States. A positive or a negative finding would be equally interesting. A second type of followup would be to enumerate the number of “national origins’ represented in the immediate ancestry of malformed and control infants, possibly along the lines of the study already proposed by Bresler (1962) for testing a hypothesis about the total fertility of marriages between members of different religious groups. REFERENCES 1. Acheson, E. D., Bachrach, C. A. and Wright, F. M. (1960). Some comments on the relationship of the distribution of multiple sclerosis to latitude, solar radiation, and other variables. Acta psychiat. scand., Suppl. 147, 35, 132-147. 2. Alter, M. (1962). Anencephalus, hydrocephalus and spina bifida. Arch. Neurol. (Chic.), 7, 411-422. 3. Bresler, J. P. (1962). The relationship between the fertility patterns of the F, generation and the number of countries of birth represented in the P, generation. Amer. J. Phys. Anthropol., 20, 509-514. 4. Dominion Bureau of Statistics (1955). Vital Statistics for the Year 1950. Queen’s Printer and Controller of Stationery, Ottawa. 5. Dominion Bureau of Statistics (1957). Causes of Stillbirth 1949-55. Ref- erence Paper No. 79. Queen’s Printer and Controller of Stationery, Ottawa. 6. Dominion Bureau of Statistics (1957-61). Vital Statistics for the Years 1955-59. Queen’s Printer and Controller of Stationery, Ottawa. 7. Edwards, P. Q. and Palmer, C. E. (1963). Nationwide histoplasmin sensitivity and histoplasmal infection. Publ. Hlth. Rep. (Wash.) 78, 241-259. 8. Gittlesohn, A. M. and Milham, S. (1962). Declining incidence of central nervous system anomalies in New York State. Brit. J. prev. soc. Med., 16, 153-158. 9. Hewitt, D. (1963). Geographical variations in the mortality attributed to spina bifida and other congenital malformations. Brit. J. prev. soc. Med., 17, 13-22. 10. Lamy, M. and Frézal, J. (1959). In Coll. int. malformations congénitales de ’encéphale. Paris: Masson. 11. Lerner, I. M. (1954). Genetic Homeostasis. Wiley, New York. 12. MacMahon, B., Record, R. G. and McKeown, T. (1951). Secular changes in the incidence of malformations of the central nervous system. Brit. J. soc. Med., 6, 254-258. 13. Morton, N. E. (1958). Empirical risks in consanguineous marriages: birth weight, gestation time and measurements of infants. Amer. J. Hum. Genet., 10, 344-349. 302 14. National Office of Vital Statistics (1947-62). States for the Washington. 15. Neel, J. V. (1958). 4-15. 17. Polman, A. (1950). 18. Wesley, J. P. (1960). 25, 29-78. malformation. Vital Statistics of the United Years 1945-49. U.S. Government Printing Office, A study of major congenital defects in Japanese infants. Am. J. Human Genet., 10, 398-445. 16. Penrose, L. S. (1957). Genetics of anencephaly. J. Ment. Defic. Res., 1, Anencephaly, spina bifida and hydrocephaly. Genetica, Int. J. Radiat. Biol., 2, 97-116. Background radiation as the cause of fatal congenital 303 Vital Record Incidence of Congenital Malformations in New York State By A. M. Grrrewsonn, Ph. D., M.P.H,, and S. Miuawm, Jr.,, M.D., M.P.H., Director of Health Statistics and Development Consultant, New York State Department of Health,* Albany, N.Y. A concomitant of the growth of the fields of human genetics, ex- perimental teratology and cytogenetics, has been an increased interest in the incidence of congenital malformations in human population groups. As stated causes of death per se, congenital defects account for almost one out of five deaths in infancy and early childhood. Mal- formations are also assigned an important role in abortion and fetal loss. The recent thalidomide experience suggests that the interval of more than one year between appearance of drug-associated cases and removal of the product from distribution might have been re- duced had incidence data been available on a central basis. The wide geographic and temporal distribution of phocomelia cases among the total population of births reduced the opportunity for any one physi- cian to observe a significant number of cases. Since 1940, the New York State birth certificate has contained an entry for recording congenital malformations. Analysis of the data has been sporadic and no routine procedure was developed for tabu- lating the data. The present investigation was undertaken from sev- eral viewpoints. The objectives were to delineate the usefulness and limitations of information on congenital defects obtained through the vital record system; for those defects reported with sufficient quality, to describe patterns in incidence relative to available geographic and demographic characteristics; and to record the “background” inci- dence against which secular changes could be measured. The population of events consists of over 2 million live births and fetal deaths registered in upstate New York or occurring to residents *Present address of senior author: Department of Biostatistics, School of Hygiene and Public Health, Johns Hopkins University, Baltimore, Md. 305 of the jurisdiction during the 11 years between 1950 and 1960. (Up- state refers to New York State exclusive of New York City.) Through a routine procedure, birth records are updated by death in- formation on infants and children up to 5 years of age. The combined birth and death data are available for producing survivorship tables and mortality rates in the perinatal, infancy, and the early childhood periods of life. Malformation cases hence can be ascertained through birth and/or death reports. Because of the inadequacy of the WHO Statistical List for coding malformations, all fetal death certificates reporting a congenital defect were searched and recoded according to a more specific classification. Live birth and death records with mal- formations of the CNS or with multiple malformations were similarly treated. Completeness of caseascertainment is a central problem in epidemio- logic studies of incidence rates. Rate in this connection is defined in terms of the relative frequency of specified events within a defined population during a given time period. Registration is required for all products of conception of over 20 weeks gestation. Thus, early fetal deaths and abortions are not available for observation and analy- sis. The completeness of birth registration is estimated to be well above 99 percent. By contrast, the recording of malformation data on birth certificates is highly variable, depending primarily on recognizability of the defect by external examination. In order to assess completeness, a number of tests were carried out. Beginning with a group of children receiving treatment for specific congenital defects under the State Aid Program, examination of corresponding birth records revealed a wide range of reporting variability (table 1). As anticipated, internal defects were generally not noted at the time of birth while external defects tended to be picked up by the system. Eighty percent of 240 spina bifida cases, as contrasted with less than 10 percent of 226 circulatory system malformations, were recorded. The incomplete- ness would tend to be overstated under such circumstances as the children with the more severe forms of the defect, incompatible with extended postnatal survival, would not enter the program. TABLE 1.—Children Treated in State Aid Program by Type of Defect and Vital Record Statement Number in | Defect noted | Noted death Percent Type of defect State Aid on birth record only recorded program record Cy bifida... 240 188 4 79.9 eft lip and/or palate 1, 096 817 3 74.8 Hypospadias_..______ 97 43 | _ _ 44.3 Club 00%. cacvvnesae 68 BO) enmnentaiinaiians 4.1 Polydactyly...._ oe 52 2 |essnsnansunsse 40.4 Hydrocephalus — 79 12 15 34.3 Syndactyly........-- = 19 | T_T 31.6 All circulatory defects... coeur 226 |... 19 8.4 306 As one of the more important malformations of the central nervous system, anencephalus has long been recognized; its appearance is so marked that it is unlikely not to be noted. A certain number of cases undoubtedly are not recorded on the birth record, and others are described under generic terms such as “monster” and are thereby lost. In the course of developing sibship data for probands with anencephalus or spina bifida the diagnostic indexes of the three maternity hospitals in one city were exhaustively searched (Milham, 1962). In tracing the birth records of the 150 cases so ascertained, it was found that more than 96 percent had the defects noted. In so far as the diagnostic indexes are complete, the birth certificates for at least one community accurately reflect their content for the two neural tube anomalies. In a similar approach to evaluating reporting of clefts of the lip and palate, the diagnostic indexes of three large teaching hospitals in the State were searched (Milham, 1963). All cases reported among the 79,536 births occurring in the three institutions between 1950 and 1960 were collected. Comparison of the two information sources revealed the following: Cases of Cleft Lip and/or Palate by Type of Ascertainment Number of Cases cases 100,000 births Birth certificateonly___________________ 104 131 Hospital indexes only__________________ 117 147 TBOUIY HOTTTOBE cosmos sm smssisososiitisisit A R 143 180 Total DIPUDS. ccc wnoninnismions snnsmmmssspsssne sin ssn sean nese 79, 536 On the basis of both sources, the incidence was 180 cases per 100,000 births. This is a minimum figure, since there may be cases not recorded on both birth certificate and diagnostic index. Ascertain- ment by birth certificate alone would have led to a 27 percent lowered incidence, while ascertainment by hospital indexes alone would have resulted in a loss of 18 percent of the cases. Since the three institu- tions were teaching hopitals with active obstetric, pediatric, and plastic surgery services, the record systems would be expected to be above average and the agreement between birth certificates and hospital information probably represents an overstatement of the statewide pattern. In addition to the problem of completeness of ascertainment of cases through vital registration is the question of specificity of the morpho- logic description of the malformation. For clefts of the lip and palate this is a problem of particular importance. In carefully constructed case series, it has often been noted that isolated cleft palate behaves differently from cleft lip with or without palatal involvement. Sex incidence for the former defect is predominantly female, while for the 307 latter it is predominantly male. Experience has shown that the statement ‘cleft palate’’ on the birth record may or may not exclude lip involvement. * A similar situation pertains to the statement ‘hare lip.” Anencephalus may be variously described under such terms as acephalus, acrania, aplasia of brain, or monster. Birth records are not designed for obtaining precise anatomic detail on which specific classifications can be traced. A further complication is introduced by the use of the present WHO Statistical Classification. The code category No. 750 includes such diverse entities as anencephalus, hemicephalus, exencephalus, syn- cephalus and monster, not otherwisestated. Adactylismislumped with polydactylism, amelia and phocomelia. To this end, a new classifica- tion system allowing for greater specificity had to be developed and each record recoded thereby. In general, the vital record system is a useful mechanism only for ascertainment of gross external malformations readily apparent at birth. A tentative listing of those defects which we feel are recorded with sufficient frequency and precision to be useful for certain limited purposes, is as follows: Anencephalus Conjoined twins Spina bifida Exomphalos Imperforate anus Polydactyly Exstrophy of bladder Gross deformities of limbs In general, the quality of information on vital records is highly variable. The major virtue of the system is to be found in the com- pleteness of birth and/or death registration from a population stand- point. Beyond the simple fact of birth or death and statements of identification, there is a lack of uniformity in the underlying observa- tions pertaining to pregnancy, labor, and the condition of the child at birth. VITAL RECORD INCIDENCE OF CONGENITAL MALFORMATIONS The possibilities for analyses of incidence data derived from vital records include spatial and temporal distributions, survivorship, and case rates by parental ages, parity, sex, and race. The present purpose is to summarize briefly a few of the findings. During the period 1950-60, in upstate New York, a total of 26,728 children with one or more congenital malformations were recorded on fetal death, live birth and/or death certificates. Counting children rather than malformations, the total incidence was 1,305 cases per 100,000 births. The defects ranged in severity from anencephaly to polydactyly. When two or more malformations were coexistent 308 in a single individual, coding assignment was to the more severe. Information on multiple malformations was preserved for future analysis. Table 2 shows the number of cases by source of vital record infor- mation. For the CNS defects, the majority of cases was reported on birth records. Over one-half of the anencephalus and hydrocephalus cases were stillborn and of those born alive, most died during the first year. For spina bifida, a smaller proportion was stillborn (13 percent), and over one out of four were not known to be dead after 5 years. By contrast, information on internal malformations involving the circulatory, digestive, G.U., or respiratory system was largely ascer- tained through the death certificate only, the corresponding birth record containing no indication of the defect. TABLE 2.—Number of Malformations by Type and Source of Data, Upstate New York, 1960-60 Total | Fetal | Deaths | Deaths| Not |Percent|Percent cases | deaths | first 1-4 |known| fetal [on D.C. year years | dead | deaths | only All malformations. ____________________ 26,728 | 2,853 | 8,014 736 | 14,225 10.8 16.2 CNS: Anencephalus 1, 504 912 585 2 5 60.5 3.9 Spina bifida__ 2,025 272 | 1,103 92 558 13.4 5.1 ydrocephal 1,580 853 526 95 106 54.0 16.8 Other CNS 711 220 280 56 55 30.9 21.0 Circulatory: Tetralogy of Fallot ______________________ 95 0 73 19 8 |scencies 77.8 Patent ductus arter. _ 190 7 170 7 6 3.6 73.6 IV septal defect____ = 266 24 216 22 4 .9 81.2 IA septal defect__ 200 14 167 11 8 7.0 68.0 OINOE «cnn mame macnn msm msn 90 | 2,553 145 324 2.9 66. 1 Digestive: Cleft lip/palate___ 24 261 25 | 1,047 1.1 N | Pyloric en: U l.wmmmp 66 1 4 beeccanns 91.5 300 104 8 183 1.7 5.7 797 65 189 2.1 17.6 97 6 10 7.4 38.5 117 13 | 1,568 1.0 % 130 15 63 [_______. 46.2 216 29 | 2,440 8 feeeeeae 50 2 22 12.9 [ccvesmms 183 11 718 4.7 2.3 75 81 LM [soccasnswsmmmens 43 5 J —— «5 u 74 18 56 196 5 Respiratory system _ 4 118 5 Webbed digits... 310 1 11 2 Other skin______ -| 2,263 222 664 27 MOoNgOHSIN.... conc vicuvmmniiinaessesenn mann 772 35 250 51 During the decade, there has been a consistent decline in the total malformation incidence rate from 1,428 to 1,144 cases per 100,000 births (fig. 1). An appreciable part of this reduction is due to a marked decline in the CNS malformations and club foot. A special effort was made to obtain data for the 1940 decade, and it was possible to construct incidence rates back to 1945 for this one group of defects (table 3). In the 15-year period between 1945 and 1960, each of the 309 150, 150 Spina Bifida Cleft Lip/Palate 100, 100] An 504 Anencephalus 504 Hydrocephalus A i v 1 V 1 50 55 60 50 55 60 A ooo TT Club Foot Total Malformations 1004 1000 50 Vongolisn 500, All Circulatory System Defects — te —— Figure l.—Incidence of Selected Malformations by Year, Upstate New York, 1950-60 (Cases per 100,000 total births) three major CNS malformations has decreased by as much as one-half. It is doubtful that reporting artefact could account for a downward secular trend of this magnitude, since it has occurred during a time of increasing awareness of, and interest in, the subject. During the same period the incidence of other defects, notably cleft palate and heart malformations, has remained fairly stable (table 4). The declining rates for CNS and club foot defects make certain hypotheses being currently entertained less tenable. The 10- or 15-year period for which the trends have been measured is of short 310 TABLE 3.—Incidence of Anencephalus, Spina Bifida, and Hydrocephalus by Year, Upstate New York, 1946-60 Cases per 100,000 births Total Year births (1,000’s) Anen- Spina Hydro- cephalus bifida cephalus 110 120 192 141 139 123 166 94 159 116 165 103 150 114 150 89 152 117 156 92 153 95 140 90 164 98 138 105 173 75 130 78 175 72 101 96 183 78 90 73 190 73 92 77 194 69 93 81 203 65 84 68 204 70 79 65 204 60 83 60 205 64 81 68 2,758 83 117 83 TABLE 4.—Number of Malformations by Type and Year, Upstate New York, 1960-60 Number Cases per 100,000 total births of Average cases 1950-60 1950- 1952- | 1954- 1956~ 1958 1960 All malformations...______ 26, 728 1,305 1,453 1,423 1,333 1,256 | 1,196 1,174 CNS: Anencephalus. ______________ 1, 504 74 97 73 76 67 65 64 Spina bifida.......ccauex 2,025 99 139 116 91 88 81 81 ydrocephalus. 1, 580 77 98 87 75 74 63 68 Other CNS__.......... 711 35 46 37 32 32 30 34 Clronlatory........ccvonssnssunm 3,863 189 199 187 190 196 174 183 Digestive: Cleft lip/palate.. ..__._______ 2, 259 110 121 109 116 100 110 105 Pyloric stenosis._.__... _.___ ua 4 6 5 5 3 2 Imperforate anus. ___________ 300 15 16 17 16 11 156 1 Other G.I. _________________ 1,074 52 58 58 50 50 50 50 Genito-urinary: Polycystic kidney. ._________ 122 6 5 6 7 5 6 8 External genitals. ___ 1,716 84 84 76 82 85 91 84 HHO Gl oc ccssmsnssnnmin 10 8 8 10 12 13 9 Bones and joints: Club foot... _________ 2,693 132 166 156 132 124 106 105 Chrondrodystrophy. ._ 85 4 4 6 4 3 5 3 Malformations of skull 957 47 56 63 42 38 39 44 Digits. 1,249 61 60 64 61 60 60 68 Other. 604 29 36 32 33 27 25 27 Other malfo Hemangioma and nevi. _____ 1,451 71 88 70 59 64 73 58 Hernia of abdominal cavity_ 747 36 39 28 36 39 38 38 Webbed i system.____._____ 164 8 8 7 8 7 10 9 bed Shgets and toes ____ 310 15 17 13 16 14 13 19 bb skin... ______________ 2, 263 110 74 169 146 113 80 61 Mongols. ...c cn so smssmssusuuse 772 38 31 39 40 44 48 39 Total Births (1,000’s).......ccunce.- 2,047 186 317 348 373 398 408 205 enough duration as to make unlikely the operation of factors as pre- dominant causes which remain fairly constant in the environment; i.e., background radiation and certain meteorological variables. It is difficult to reconcile a simple genetic hypothesis with such short term changes. The pattern suggests that whatever the genetic predispo- sition to congenital defect may be, the phenotypic manifestations of that predisposition are very significantly influenced by varying 311 environmental factors, a point already clear from the well-known relationship between maternal age and frequency of defect. Malformation rates are generally higher at the extremes of the maternal age range (figure 2, table 5a—5g). A similar pattern is noted for practically all malformations except polydactyly. In mongolism, the maternal age curve assumes its most striking form, empirical risk figures rising over tenfold with increasing age. Despite the better understanding of mongolism in terms of abnormal karyotype, the mechanism of the maternal age effect remains unclear. 150 , 15Q, Spina bifida Cleft palate/lip 100 J 1004 50 | Anencephalus sal Hydrocephalus T mg ™ T T T T Tv gm v v T <20 20- 25- 30- 35- 40+ <20 20- 25- 30- 35- 40+ 350 , 300 Malformations of 150, Club foot Circulatory System 250 4 200 4 10Q4 150 4 100 A 504 Polydactyly 50 4 Mongolism <20 20- 25- 30- 35- 40+ 0.05 Undetermined P>0.05 of dying from lung cancer either in association with, or independently of, the history of cigarette smoking. As an indication of the possible influence of the environmental factor common to members of married pairs, spouses of the probands and those of the index controls were compared with respect to lung cancer mortality. There were no significant differences in such mortality among spouses, including both wives and husbands. Since the index controls were selected within the adjacent neigh- borhood and the majority of the relatives were also residing in the same geographic areas, the possible effects on lung cancer of such additional environmental factors as atmospheric pollution, socio- economic status and ethnic background are assumed to be fairly 346 homogeneous. From the totality of these observations it would appear that some inherent factor may play an important role in the etiology of lung cancer. III. Analysis of the Data on Smoking: It has been established that the lung cancer patients are much more likely to be cigarette smokers than those who do not have such disease. According to our data, 98.4 percent of the male probands were actually cigarette smokers. This is significantly greater than 81.6 percent that would be expected if the age-specific proportions of cigarette smokers among the male index controls were applied. From this observation several questions may be asked regarding the smoking habits of relatives of the index subjects. First, are the relatives of the probands also more likely to be smokers than the counterparts of the controls? | Second, is the smoking pattern in the families of the lung cancer group | different from that of the control group? And, third, is there an | association between the smoking habits of offspring and that of parents? | We will attempt to answer these questions with particular emphasis on the application of several different methods that have been devel- oped in epidemiology and genetics. It should be noted that the data on smoking among relatives were not ascertained through smokers. Although the majority of the lung cancer probands were smokers, the difference in the proportion of smokers and nonsmokers among the index controls was much less marked. In computing the proportion of smokers, both living and deceased individuals were considered; however, those for whom smoking information and/or age was not given were excluded. A. Inter-Familial Aggregation of Smokers: To determine the presence of inter-familial aggregation of smokers, i.e., excess frequency of smokers among the proband relatives as compared with the control relatives, it is necessary to take into account the possible differences in sex, age, and generation of the two family populations which may be associated with the smoking habits. For this purpose we computed the number of smokers that would be expected among the proband relatives by applying the age-sex-generation specific proportion of smokers among the control relatives. The expected number of smokers was then compared with the observed number within the proband relatives. A summary of these calculations is given in table 6. As shown in this table, the observed value was significantly greater than expected in both sexes, i.e., both male and female relatives of the probands manifested a greater risk of being smokers than those of the controls. The data further show that this excess frequency of smokers was actually accounted for by siblings of the index subjects. 347 TaBLE 6.—Observed and Expected Number of Cigarette Smokers Among Case Relatives Number of smokers Numberof |__| Significance relatives test Observed Expected* 261 70 58.4 | P>0.05 560 363 322.5 | P=0.008 215 141 150.8 | P>0.05 1, 036 574 531.7 | P=0.05 268 7 5.4 | P>0.05 585 160 129.9 | P=0.04 199 116 91.1 {| P=0.01 Al 1OMB108. cc cnnirm res sm msm Tae EE 1,052 283 226.4 | P=0.005 Parents... 529 77 63.8 | P>0.05 Siblings. ___ 1,145 523 452.4 | P=0.02 Children... 414 257 241.9 | P>0.05 AIL PIatiVeS.. ccna sun nnmnn ns suman 2,088 857 758.1 | P=0.0009 *Computed on the basis of the age-specific smoking frequency of the control relatives. Note: Those with unknown smoking status, unknown ages, or those under 20 years of age are excluded- A similar trend was also observed among parents and children; how- ever, the difference was not statistically significant. B. Intra-Familial Aggregation of Smokers: The intra-familial aggregation of smokers was determined by com- paring the proportion of smokers among those relatives whose index subjects are also smokers with that among those relatives whose index subjects are nonsmokers. This comparison was possible in each of the proband and control family populations as well as in the combined population. The analysis of intra-familial aggregation of smokers was made for parents, siblings, and children of the index subjects and the results are summarized in table 7. Let us first consider the combined popu- lation. The data clearly indicate that there is significant aggregation of smokers in families. Specifically, those parents and siblings whose index subjects are smokers were much more likely to be smokers than those whose index subjects are nonsmokers. Although more than twice as large a proportion of the male as the female relatives are smokers, the same mode of aggregation was revealed in both sexes. The data also show a striking difference when the proband and control families were considered separately. As mentioned earlier, the overall frequency of smokers among the proband relatives was significantly greater than that among the control relatives. No such relationship was found among spouses of the index subjects. It is of interest to note that about 40 percent of proband relatives reported a history of regular smoking regardless of the smoking status of the index subjects. This generally increased and “diffuse” frequency of smokers among the proband relatives suggests that there may be 348 TABLE 7.— Percentage of Cigarette Smokers among Relatives, According to Cigarette Smoking Status and Case Control Identity of the Index Subject Case group Control group Combined group Category of relatives Index Index Index Index Index Index subjects subjects subjects subjects subjects subjects who are who are who are who are who are who are smokers |nonsmokers| smokers |nonsmokers| smokers [nonsmokers Father. _________________ 27.9 15.0 30.4 12.0 28.9 12.5 Mother. -- 2.8 0.0 3.1 0.0 2.9 0.0 Brother___ 65.0 62.9 66.9 46.7 65.7 49.0 Sister. ..... 27.6 20.7 27.4 14.1 27.5 15.0 Son....... 66.0 63.0 72.8 67.6 68.7 66.7 Daughter 58.3 58.3 47.1 42.3 53.7 45.3 Total males... 55.7 51.2 58.3 43.1 56.7 44.4 Total females... 26.8 27.4 24.7 17.5 26.0 19.1 Total parents____ 15.2 7.5 16.7 6.0 15.8 6.2 Total siblings.___ 45.8 43.8 47.6 31.5 46.4 33.2 Total children. __ 62.3 60.8 60.2 55.0 61.4 56.2 Total. ..covscwiinsiauvas 41.1 40.0 41.8 30.7 41.3 32.2 Note: Percentage of smokers is computed against those for whom smoking information is given. Significance test Case group Control group Combined group -| P<0.0001 | Total relatives. _____| P<0.0001 P<0.0001 | Males.__.______ -| P<0. 0001 P=0.005 Females... -| P=0.002 P=0.0002 | Parents___ -| P<0. 0001 -| P<0.0001 | Siblings___ «eu £<0.0001 P>0.05 Children... cc... P>0.05 something inherent in the biological makeup common to all “blood” relatives of the probands. It has been pointed out that the control relatives were much less likely to be smokers than the proband relatives as a whole. The data further show that approximately 40 percent of those control relatives whose index subjects were smokers, as compared with about 30 per- cent of those control relatives whose index subjects were nonsmokers were actually smokers; a difference which is statistically significant. This “selective” intra-familial aggregation of smokers may be due to genetic factors, or common environmental factors, or both. C. Parental Smoking and Smoking among Offspring: In analyzing the smoking behavior of offspring in relation to that of parents, two generations, i.e., parents and siblings of the index subjects, were considered. Since a number of children of the index subjects were still under the age of 20, this generation was not included. The number of male and female offspring and the proportion of smokers according to the smoking status of their parents are given in table 8. The relative frequency of matings in which both parents are smokers was about the same in the proband and control families. However, the like frequency in which neither parent is a smoker was much more common in the control than in the proband families. 349 TABLE 8.—Number and Proportion (9%) of Cigarette Smokers among Offspring According to Parental Smoking Habits (A) CASE FAMILIES Num- Male offspring Female offspring All offspring Parental mating | ber type fami- lies | Total [Smokers| Percent | Total |Smokers| Percent | Total |Smokers| Percent smokers smokers smokers Both smokers__._ 7 16 15 93.8 14 11 78.6 30 26 86.7 One smoker___.__ 171 509 380 74.7 412 119 28.9 921 499 54.2 Neither smoker. 85 274 196 71.5 171 42 24.6 445 238 53.5 Total..c.-- *263 799 591 74.0 597 172 28.8 | 1,396 763 54.7 X? Test X2=4.28 X2=18.67 X2=12.73 (d.f.=2) P>0.05 P<0.001 0.001