The familial transmission of primary major depressive disorder

J. psychiat. Res,, Vol. 21, No. 4, pp. 613-624, 1987.

0022-3956/87 $3.00+ .00 © 1987 Pergamon Journals Ltd.

Printed in Great Britain.

THE FAMILIAL TRANSMISSION OF PRIMARY MAJOR DEPRESSIVE DISORDER THEODORE REICH*t, PAUL VAN EERDEWEGH*, JOHN RICE*, JOE MULLANEY*, JEAN ENDICOTT,, a n d GERALD L. KLERMAN§ *Department of Psychiatry, Washington University, Washington, 1-TheJewish Hospital of St. Louis, 216 South Kingshighway, St. Louis, Missouri 63110; ~Department of Research and Training, New York State Psychiatric Institute; §Cornell University Medical College, New York, U.S.A.

(Received 22 October 1986; revised 30 December 1986) Summary--This is a study of the familial transmission of Primary Major Depressive Disorder in the families of 235 probands with this disorder ascertained as part of the NIMH-CRB Collaborative Depression Program. Eight hundred and twenty-sixinterviewedfirst degree relatives and 109 spouses are included. Research Diagnostic Criteria have been used and interviewswere done using the SADS-L schedule. Prior analyses of these data have established the presence of strong secular trends in the age-ofonset and prevalence of Major Depressive Disorder in these families. Accordingly, new methods for the analysis of family data which incorporate secular variation were developed. Non-parametric Survival Analysis, using the Cox Proportional Hazards Model, guided the formulation of a quantitative family transmission model. Then a family analysis was conducted with the Multifactorial Model of Disease Transmission and the Tau Path Analytic Model. Using the non-parametric approach, only the sibs birth cohort, sex and affectational status of the mother were significantly related to the time of onset of illness in siblings. Proband sex, age-ofonset, and the presence of illness in the father were not significant. The quantitative analysis confirmed that more recently born cohorts of individuals had an increased expected lifetime prevalence and a decreased age-of-onset of Primary Major Depressive Disorder. Assortative mating was present and environmental factors common to siblings did not make a significant contribution to the phenotypic variance. Sex specific transmissibilities were found and the transmissibility in females (t2 = 0.62) was significantly greater than that of males (t2 = 0.28). There was a trend for the transmissibility of Primary Major Depressive Disorder to be greater in more recently born cohorts. INTRODUCTION IN TInS report the transmission o f P r i m a r y U n i p o l a r M a j o r Depressive Disorder is described in families ascertained as part o f the N I M H - C R B C o l l a b o r a t i v e Depression P r o g r a m . Pr i o r analyses o f the d i s t r i b u t i o n o f M a j o r Depressive D i s o r d e r in these families has revealed strong secular trends (KLERMAN et al., 1985). Estimates o f the lifetime prevalence o f M a j o r Depressive D i s o r d e r were m u c h higher in m o r e recently b o r n c o h o r t s a n d in a d d i t i o n the a g e - o f - o n s e t a p p e a r e d to be decreasing. Similar findings have been r e p o r t e d in o t h er family studies (GERSHON et al., 1986; WEISSMANet al., 1987) and in a large-scale c o m m u n i t y survey ( R o a m s et al., 1984). T o test h y p o t h e s e s a b o u t the n a t u r e o f the secular trend, a 6 yr p r o sp ect i v e f o l l o w - u p study o f first degree relatives is in progress. P r e l i m i n a r y results indicate t h a t the t r en d is *.Address correspondence to: Theodore Reich, Jewish Hospital, Department of Psychiatry, 216 South Kingshighway, St. Louis,: Missouri 63110, U.S.A. 613

614

TrmODOREREICHet

al.

not artifactual and is not due to differential mortality or simple forgetting by older respondents (RICE et al., 1987). Whether the secular trend is due to a "Cohort Effect," in which each successive cohort has a higher lifetime risk for illness, or whether it is due to a "Period Effect" i.e. increased risk across all cohorts during a specified period, is under study (LAvoRI et al., 1987). Although the increased prevalence of Major Depressive Disorder in younger cohorts has occurred too quickly to be accompanied by changes in the genetic composition of the general population, patterns of familial transmission may well have shifted. WEISSMANet aL (1984), for example, have suggested that there was specificity of transmission of Major Depression which occurred under 20 yr of age. In that study, onset at age 40 yr and over had familial loading only slightly higher than families of normal controls. To the extent that early onset depression is becoming increasingly frequent, these data would suggest that familial transmission of the disorder has greatly increased during the past few decades. The goal of the present investigation is to describe the familial transmission of Primary Major Depressive Disorder allowing for secular trends in (a) expected lifetime population prevalence, (b) age-of-onset, (c) transmissibility, and (d) environmental factors common to siblings. In order to accomplish this task, the " T a u " Path Analytic Model developed by RICE et al. (1980) was extended by indexing parameters according to the cohort of birth of probands and relatives. To guide the development of the Path Model, Survival Analysis (KALBFLEISCHand PRENTICE, 1980) was used to determine which covariates were required, and which parameters might depend on birth cohort. Subjects Probands for this study were ascertained from the National Institute of Mental Health (NIMH) Collaborative Program on the Psychobiology of Depression. This is a naturalistic longitudinal investigation of 955 probands and of 2225 first degree relatives of 612 of these probands who participated in the family study. Criteria for admission to the study were that probands had to be at least 17 yr, Caucasian, English speaking, have knowledge of their biologic parents and permit participation of their first degree relatives. Probands were interviewed using the Schedule For Affective Disorder (SADS) and the Research Diagnostic Criteria (RDC) was used to make diagnoses. All available spouses and first degree relatives for the family study were interviewed using the lifetime version of the SADS (SADS-L). In addition, the proband and the "best family informant" were interviewed using the Family History Research Diagnostic Criteria (FHRDC). Consensus diagnoses were made using all sources of information. A detailed review of the methods used in the study is available (ANDREASEN et al., 1987). For the analyses included in this report, probands and relatives were considered affected only if they were diagnosed as having Primary Major Depressive Disorder. Five relatives with Bipolar I and Schizoaffective Bipolar or Manic Disorder were counted as unaffected. Relatives with Secondary Major Depressive Disorder were also considered to be unaffected. Using these criteria, there are 235 probands and 826 first degree relatives who are included in the family analysis. One hundred and nine interviewed spouses are also included. The frequency of Primary Major Depressive Disorder in the first degree relatives and spouses of probands is shown in Table 1.

FAMILIAL TRANSMISSIONOF PRIMARY MAJOR DEPRESSIVE DISORDER

615

TABLE 1. FREQUENCY OF PRIMARY MAJOR DEPRESSIVE DISORDER IN INTERVIEWED SPOUSES AND FIRST DEGREE RELATIVES OF 235 PROBANDS WITH PRIMARY MAJOR DEPRESSIVE DISORDER Percent affected

N

Fathers Mothers Brothers Sisters Sons Daughters Husbands Wives

9.2 30.5 12.1 32.3 19.2 25.8 9.7 38.3

87 128 182 254 78 97 62 47

Total male relatives

13.0

347

Total female relatives

30.5

479

Methods o f analysis Survival Analysis was used to obtain Kaplan-Meier estimates of the time to onset of Primary Major Depressive Disorder and the asymptotic proportion of those to be affected. This non-parametric method was used to test for the presence of secular trends in the male and female relatives. The Cox Proportional Hazards Model (KALBFLEISCH and PRENTICE, 1980) (uni and multivariate) was used to study the effect of covariates in proband's parents and siblings on the time to onset in siblings. In applying the proportional hazards model, only the interviewed siblings were used. In modelling the effect of parental illness on the siblings, a Consensus diagnosis was used for uninterviewed parents. The offspring of probands were not used in these analyses. Covariates which were studied included: the ages-of-onset in probands and parents, the birth cohort of probands and siblings, the sex of probands and siblings and the presence of Primary Major Depressive Disorder in fathers and mothers. Age-of-onset in probands was assessed as a continuous variable and then as a dichotomous variable using three different cutoff points at ages 20, 25, and 30 yr. The Multifactorial Model of Disease Transmission (REICH et al., 1979) was then used to compute cohort and sex-specific correlations between parents and offspring, siblings and spouses. This model assumes that familial and non-familial factors which influence the development of the disorder can be summed into a single variable termed the liability, and that individuals whose liability exceeds a threshold develop the disorder sometime during the period of risk. The familial distribution of liability is assumed to be a multivariate normal and the position of the thresholds truncate a portion of the liability distribution equal in area to the sex and cohort specific lifetime population prevalence. The analysis proceeds one family at a time by estimating the probability of a family, conditional on the proband being affected, and taking into account the age and cohort of birth of each member. In this way correction for ascertainment is included and relevant population prevalences can be computed. The probability of each family is also modified by correcting for variable age-of-onset. A new method for integrating the multinormal distribution was developed to more accurately compute the probability of a family (BAIGORal et aL, 1986). Hypothesis testing is done using the likelihood ratio criterion. In these analyses

616

TrmODOR~REICHet a/.

the entire family data is used including parents, siblings, offspring, and spouses. Single ascertainment is assumed. The familial correlations were quantified using the Tau Path Analytic Model developed by RICE et al. (1980). In this model, the liability of each individual is partitioned into transmissible and non-transmissible components for each sex. The proportion of the variance in liability accounted for by transmissible factors is termed the transmissibility (t2). Transmission between parent and offspring is described by standardized path coefficients, (r), which measure the sex specific contribution of a parent to an offspring. Correlations between the liabilities of mates (Rm) and between the non-transmissible environments of siblings (c) are also included. The path parameters are used to compute correlations in liability between relatives and these in turn are used, after correcting for variable age-ofonset and ascertainment, to compute the probability of observing a particular family. This model combines path analytic structures with the tetrachoric correlations between liabilities within families. The model has been extended to include sex specific secular trends in (a) lifetime population prevalence, (b) correlations between non-transmissible factors in siblings and (c) transmissibility of the liability. This was accomplished by using a logistic function (JOHNSON and KOTZ, 1970) to relate cohort of birth to each variable. Secular trends were thus introduced into the path analytic model by assuming that individual environmental factors are a function of cohort of birth (PRovINCE and RAO, 1985). The parameters of the logistic distribution were estimated along with other cohort specific parameters when the Multifactorial Model was fitted to the data. For reasons of computational simplicity, it was assumed that there was no secular variation for individuals born since 1959 or born before 1936. This was done to minimize the effect of outliers and small numbers of observations in younger and older groups. Although our computer program allowed separate logistic functions to be estimated for each sex and each parameter, only a single logistic function (a, /3) is reported here. FN

R-L

+ L,

1 +e [a-g]

/3

where R L

= F n at right asymptote = Fn at left asymptote

g = cohort-of-birth (1966--year of birth) a, /3 = parameters of the logistic function F N = cohort indexed parameter. Variable age-of-onset was modelled by means of a quadratic regression of the mean and variance of the natural logarithm of age-of-onset on the natural logarithm of cohort of birth. Censoring of age-of-onset was taken into account. A log normal and normal distribution of age-of-onset were compared and the log normal distribution was chosen because it fitted the data better. This approach was used because preliminary analysis indicated that cohort specific ages-of-onset in relatives fit the data better than customary methods of age correction. As before it was assumed that the age-of-onset distribution was unchanged for individuals

FA.-XaI.I~ TRANSMISSIONOF PRIMARY MAJOR DEPRESSIVE DISORDER

617

born since 1959 and before 1936. The parameters of the age-of-onset distribution were estimated using in-st degree relatives as a group and were then held constant during the analyses of the family data. Correlations between the ages-of-onset of relatives and between ages-of-onset and liability were not included since these effects could not be demonstrated using the Cox Proportional Hazards Model. Using the computerized search program, GEMINI (GEMINI, 1979), the method of maximum likelihood was used to fit a generalized transmission model, including secular trends, to the family data. Then the likelihood ratio chi-square test was used to test hypotheses about the general model and produce a more parsimonious explanation of the familial transmission of the disorder. RESULTS

In keeping with the results of other family studies, the frequency of Primary Major Depressive Disorder in the female first degree relatives is higher than in males. Rates in siblings (mean age = 39 yr) and children (mean age = 28 yr) are nearly as high or higher than those in the parents (mean age = 60 yr) even though they are on average considerably younger. Adjusting these frequencies for variable age-of-onset leads to the prediction that the illness would be most common in children when compared with siblings and parents. In order to investigate the relationship between age and prevalence of illness, Kaplan-Meier survival distributions stratified by age groups are displayed in Figs 1 and 2. In these figures interviewed male and female sibs, parents and offspring are subdivided into three groups based on their current age ( ~ 25, 26 ~ , and 45 + yr). The survival function is the proportion who remain well at each age corrected for the varying number of the relatives within each age group. If there were no secular trends, the three survival curves would fall on the same line, and this possibility can be statistically rejected at the 0.0001 level for both males and females. For the females, approximately 20°7o are affected in group 1 by age 20. SURVIVAL FUNCTION i

0.8

0.8

0.4

0.2

IlJlllJlllilllll*llrlJlllllrlltllllllltlllJtililllll|[[ ~'4

ID

0

tfl

I

0

AGE FIG. 1. Survival distribution of males stratified by age group.

6ROUP 2

~6-44

6ROUP 3

45*

TrmOOORE RE~CH et al.

618

SURVIVAL FUNCTION GROUP 1~25

0.8

'

GROUP 2

26-44

GROUP 3

45÷

""

0.6

".

0.4

0.2

0

I l i i i i l i l l l l l t l l l l l

II ill

II Illl

IIlllllill

llll

Ilil

I l l l l l l l l l l l l l l l l

I[ll]

AGE FIG. 2. S u r v i v a l d i s t r i b u t i o n o f f e m a l e s stratified b y age g r o u p .

In group 2, only 7% are affected by this age and in group 3 only 2O/o reported onsets by age 20. For the males, the corresponding figures are 17%0, 5% and 1%0. By age 35 the prevalence in the femaless is approximately 40% in those between the ages of 26 and 44 yr and about 18% over age 45. The corresponding figures for males are 20% and 2%. The magnitude of these secular trends clearly requires that this effect be taken into account when models of familial transmission are constructed. The results of the Cox Proportional Hazard Model as applied to the sibs are displayed in Table 2. Although many covariates were included, only the sibs' birth cohort, sex and the affectational status of the mother were significantly related to the onset of illness in the multivariate analysis. P r o b a n d ' s sex, age-of-onset, and the presence of illness in the p r o b a n d ' s father, did not significantly influence the time of onset of illness in siblings (SAS INsTrru~ INC., 1983). The significance of most of these variables was essentially unchanged when entered in the model one at a time, the exception being proband's age-of-onset. In the univariate analysis this was highly significant with an exponentiated beta of 0.98 and X~ of 11.48. When entered with cohort, the effect disappeared entirely. To assess the importance of the proband age-of-onset, it was included both as a continuous variable and also dichotomized at age 20, 25, and 30 yr. No significant influence of this variable was found. Because of this finding no attempt was made to include correlations between the ages-of-onset of relatives or correlations between liability and age-of-onset. As an initial step in the analysis, the mean and variance of the natural log of age-of-onset were estimated by means of a quadratic regression of these variables on the natural log of the birth cohort of first degree relatives taken together. This analysis showed that separate age-of-onset distributions were not necessary for males and females and that the relatives' birth cohort contributed significantly to the onset distribution. The quadratic terms were not significant but were included during subsequent analyses. The regression equations obtained

FAMILIAL TRANSMISSION OF PRIMARY MAJOR DEPRESSIVE DISORDER

619

TABLE 2. RESULTS OF COX MODEL APPLIED TO PRIMARY M D D FOR INTERVIEWED SIBS OF PRIMARy M D D PROBANDS (CONSENSUS DIAGNOSIS USED FOR UNINTERVIEWED PARENTS)

Brothers [exponentiated] beta* Xl2

Sisters [exponentiated] beta*

x12

Combined [exponentiated] beta* Xl2

Relative's birth cohort

1.08

8.38

1.04

10.59

1.05

17.81t

Proband's age-of-onset

1.02

0.0

0.99

0.94

1.00

0.02

3.13

22.38t

Sex o f relatives

.

.

.

.

Mother affected (Primary MDD)

3.43

7.99

1.82

6.48

2.07

12.96I

Father affected

1.50

0.28

0.42

1.45

0.63

0.82

*For a dichotomous variable this corresponds to relative risk. For a continuous variable this is the multiplicative change in relative risk associated with a unit change in the covariate. t P < 0.001.

TABLE 3. MEAN AND VARIANCE OF AGE-OF-ONSET OF PRIMARY MAJOR DEPRESSIVE DISORDER FOR INDIVIDUALSLIVING TO AGE

80 YR Mean LN (age-of-onset) = 3.01 - 0.15z + 0.101z2 Variance LN (age-of-onset) = 0.06 + 0.05z + 0.05z2 z= LN (1966--cohort of birth).

in this fashion are displayed in Table 3 and illustrated in Fig. 3 where the age-of-onset d i s t r i b u t i o n o f P r i m a r y M a j o r Depressive D i s o r d e r is d i s p l a y ed f o r t w o b i r t h c o h o r t s . The regression equations estimated the expected m e a n a n d variance distribution o f ages-ofonset for individuals living to age 80 yr. Young individuals, however, are early in their period o f risk a n d so the age-of-onset distribution represents a prediction o f onsets under the assumptions described above. The precision with which these predictions can be m a d e is low for m o r e recently b o r n cohorts who are early in the period o f risk. In Fig. 3 the age-of-onset distribution is displayed to illustrate the magnitude o f the decrease in age-of-onset in two cohorts. It is i m p o r t a n t to note that the variance o f age-of-onset appears to have decreased as well as the mean, considerably shortening estimates o f the period o f risk for the disorder. This effect is statistically significant (likelihood ratio test). While fitting the age-of-onset distribution to grouped family d a t a the parameters o f the logistic function were also estimated. The p a r a m e t e r ~ represents the mid-point o f the logistic function a n d was f o u n d to be 28 and the p a r a m e t e r B was 1.797.

620

THEODORE REICH et al. 8 7

E6 ~5

g 0

1

0

10

20

30

40

50

60

68

AGE FIG. 3. Age-of-onset distribution of Primary Major Depressive Disorder in two birth cohorts. TABLE 4. PARAMETERSOF THE " T A u " PATH ANALYTICMODEL OF DISEASETRANSMISSION FOR PRIMARY MAJOR DEPRESSIVE DISORDER COMPARING THE PRESENCE AND ABSENCE OF A SECULAR TREND IN TRANSMISSIBILITY No secular trend in transmissibility R

mate

Secular trend in transmissibility

0.27

0.33

0.28

0.22 0.25 - 0 . 6 4 0.65

0.62

0.47 0.51-0.99 1.0'

Lifetime population prevalence of males born before 1936 born 1 9 3 6 - 1942 born after 1942

0.08 0.10-0.22 0.23

0.08 0.09-0.18 0.19

Lifetime population prevalence of females born before 1936 born 1 9 3 6 - 1947 born after 1947

0.26 0.29-0.51 0.52

0.25 0.27-0.44 0.45

c

0.00y

0.00t

z

0.50t

0.50t

t 2 male born before 1936 born 1 9 3 6 - 1947 born after 1947 t 2 female born before 1936 born 1 9 3 6 - 1947 born after 1947

Loglikelihood *Parameter reached a bound. ~Parameters fixed at these values. x~= 5.54.

- 2017.28

- 2014.51

FAMmt~ TRANSMISSIONOF PSaMARYMAJORDEPRESSIVEDISORDER

621

The parameters described above were used in the analysis of family data. The parameters of the multifactorial model reported here use the age-of-onset distribution in Table 3. The findings of the analysis of the family data parameterized using the " T a u " model are displayed in Table 4. The family data was used to test the hypothesis that the transmissibility of the disorder was different in more recently born cohorts. Under this model the parameters which we found are displayed on the right side of Table 4. It can be seen that estimates of the transmissibility in females born before 1936 (#=0.47) is less than that for females born after 1959 (t2~ 1). For males the equivalent figures are 0.22 and 0.65. Although there was a trend for the transmissibility in younger individuals to be greater, this result did not reach statistical significance. The more parsimonious model includes only a single transmissibility for each sex and is displayed on the left hand side of Table 4. The transmissibility in females (#=0.62) is significantly greater than that of males (t2=0.28). Estimates of the expected lifetime population prevalence of males and females are also included in Table 4 and their distribution is illustrated in Fig. 4. Although many versions of the " T a n " model were fitted to these data, estimates of the lifetime population prevalence were quite stable. The expected lifetime rates were higher in more recently born cohorts and the prevalence of affected females is decreasing. The cohort specific correlations between the non-transmissible factors of siblings were estimated by including these parameters in the family analysis. This was done to test the hypothesis that familial non-transmissible environmental factors might have a greater influence in younger sibling pairs. Estimates of the correlations between non-transmissible environmental factors common to sibs varied between 0.05 and 0.40. They were, however, not significantly different from 0 and hence, are not included in the model displayed in Table 4.

o.o/F

0.50

O

.E_

z

>~

Females

0.30

UJ

z O

Males

0.20 0.10 / 0.0519351940 I 1945 I 1950 I and

before

J 1960 J

1955

YEAR OF BIRTH

FzG.4. Expectedfifetimepopulation prevalenceof primarymajor depressivedisorder found by fitting the Tau path analytic model to family data.

622

TI-IEODOaE REICH et al.

During the multifactorial analysis of family data, sex specific logistic parameters were fitted separately for transmissibility and population prevalence. Likewise, age-of-onset parameters were also fitted. These extensive computer runs were undertaken to test the hypothesis that transmissibility was greater in more recently born cohorts. As noted earlier, a stable statistically significant effect was not found. In no case was the transmissibility less in younger cohorts suggesting that the increased prevalence in recent cohorts is not due to non-familial sporadic cases. The age-of-onset distribution varied little. Attempts were made to vary the parameter " T a u " (0, but its value did not deviate significantly from 0.5. The presence of a maternal effect was studied by comparing the value of r from mother to offspring with r from father to offspring. They did not significantly differ from each other or from the value 0.5. Accordingly a maternal effect was ruled out. The parameters c and z are fixed at their respective values in Table 4. The correlation in liability between mates varies between 0.27 and 0.38 and was significantly different from 0. DISCUSSION

Although the mean and variance of age-of-onset was strongly dependent on cohort of birth, the p r o b a n d ' s age-of-onset did not influence the frequency of illness or age-of-onset in the siblings. Similarly, the correlation in age-of-onset between probands and parents was not different from 0. We, therefore, were not able to confirm the strong relationship between onset age and familial aggregation reported by WEISS~',~ et al. (1984). Our results were consistent whether age-of-onset was treated as a continuous or dichotomized variable. Since the criteria for diagnosis in the Weissman study was somewhat more stringent than ours, the Cox Proportional Hazards Model was run, using severity of the proband's illness and proband's age-of-onset as covariates. Severity was variously defined as the presence of incapacitating depression; depression with psychotic features; recurrent depression; and score on the Global Assessment Scale for worst level of functioning during the index episode. In no case did proband's age-of-onset become a significant covariate. The presence of a specific severe early onset form of Primary Major Depressive Disorder was ruled out. RIcE et al. (1987) have found secular trends in the age-of-onset distribution for Bipolar I and Schizobipolar Disorder, but secular trends in the lifetime population prevalence were not noted. The frequency of Bipolar I Disorder in the relatives was strongly related to the proband's age-of-onset leading to the conclusion that the proband's age-of-onset was correlated with the underlying liability distribution. Although we found strong secular trends in the age-ofonset distribution of Primary Major Depressive Disorder, no relationship between age-of-onset and liability was found when cohort of birth was taken into account. In addition, we find a strong secular trend in lifetime prevalence of Primary Major Depressive Disorder. These differences suggest that variable age of onset in bipolar and non-bipolar disorders may be due to different mechanisms and that unlike bipolar I disorder, the age-of-onset of Primary Major Depressive Disorder is under the control of non-familial factors. Furthermore, factors which influence the lifetime prevalence of these disorders are also different. Estimates of the transmissibility of Primary Major Depressive Disorder were higher in more recently born cohorts, but this finding did not quite reach statistical significance. It seems likely, therefore, that in spite of prominent secular trends in the prevalence of the disorder,

FAMILIALTRANSMISSIONOF PRIMARY MAJOR DEPRESSIVEDISORDER

623

it is strongly familial in both older and younger cohorts. The threshold for illness may have been changing over the past few decades but the underlying continuous liability distribution has been invadant. Thus, the great increase in the population prevalence of the affective disorders is most pronounced in the offspring of affected individuals, and not due to the onset of sporadic or nonfamilial cases. The Proportional Hazards Analysis, reported earlier, failed to find an effect of Major Depressive Disorder in the fathers on the risk to the siblings. By contrast, the Multifactorial Analysis found that the correlation between fathers and offspring was not zero. The difference is likely due to the inclusion of proband's offspring in the Multifactorial Analysis. In addition, uninterviewed parents, including a preponderance of fathers, were diagnosed using only FHRDC and used in the Proportional Hazards Analysis. This procedure underreports Major Depressive Disorder when compared with a direct interview. In this analysis two different approaches have been used together. Non-parametric methods were used to delineate important covariates which were then included in a parametric model. We have attempted to combine the advantages of both approaches to maximize the information which can be obtained from the family study. In general, the non-parametric approach requires fewer assumptions and in this sense is more robust. The parametric approach, however, offers quantitative predictions which can be tested in future studies. With specific references to models of familial transmission, the quantitative approach enables us to produce a general transmission model which can be extended over several generations and to more distant relationships. Unlike the non-parametric model, it does not require the assumption of conditional independence of siblings and can therefore be used more effectively to quantify the risk for illness in unaffected family members taking into account the affectational status of sibs as well as more distant relatives. Including secular variation in the Tau model of familial transmission has required many assumptions about the complex relationship between secular trends in lifetime population prevalence, transmissibility, and age-of-onset. We are conducting computer simulation studies to more fully explore the relationship between these variables and to determine the statistical power of time dependent familial transmission models to test hypotheses about underlying casual relationships. In particular, the relationship between cohort indexed population prevalences and transmissibilities, and the distribution of cases in relatives needs to be explored to provide a more direct test of the goodness-of-fit of the model to pedigree data. Acknowledgements--Completed with the participation of the collaborative program investigators: G. L. Klerman (Chairperson) (New York); R. M. A. Hirschfeld (Project Director and Co-Chairperson) and P. Griffith (Washington D.C.); M. B. Keller and P. Lavori (Boston); J. A. Fawcett and W. A. Scheftner (Chicago); N. C. Andreasen, W. Coryell, G. Winokur and P. Wasek (Iowa City); J. Endicott, P. McDonald-Scott and J. E. Loth (New York); J. Rice, T. Reich (St. Louis). Other contributors include: P. J. Clayton, J. Croughan, M. M. Katz, E. Robins, R. W. Shapiro and R. Spitzer. Supported in part by USPHS Grants MH-25430, MH-37685, MH-31302, AA-03539, the Endownment Fund of The Jewish Hospital of St. Louis and the MacArthur Foundation Research Grant. REFERENCES ANDREASE~, N., RICE, J., ENDICOTT, J., CORYELL,W., GROVE,W. and REICH,T. (1987) Familial rates of affective disorder: A report from the NIMH Collaborative Study. Arch. gen. Psychiat. 44, 461-469. BAIcOPaU,A. R., VAN EERDEWEGH,P. and REICH, T. (1986) Error bounded integration of multivariate normal densiUes

over rectangular regions. (Submitted.).

624

TrmODORE REICH et al.

GEMINI (1979) Technical report No. 14 Department of Biophysics and Medical Computing, University of Utah. GlmsnoN, E. S., HAMovrr, J. H., GUROFF, J. T. and Num,mERCER, J. N. (1986) Birth cohort changes in manic and depressive disorders in relatives of bipolar and schizoaffective patients. Archs gen. Psychiat. (in press). JOHNSON, N. L. and KOTZ, S. (1970) Distributions in Statistics: Continuous Univariate Distributions-2. J. Wiley, New York. ~ H , J. D. and PRENTICE,R. L. (1980) The StatisticalAnalysis o f Failure Time Data. Wiley and Sons, New York. KLERMAN, G. L., LAVORI, P. W., RICE, J. P., REICH, T., ENDICOTT, J., ANDREASEN,N. C., KELLER, M. B. and HII~SCI-I~mLD,R. M. A. (1985) Birth cohort trends in rates of major depressive disorder among relatives of patients with affective disorder. Archs gen. Psychiat. 42, 689-693. LAvoPa, P. W., KLERMAN,G. L., KELLER, M. B., REICH, T., RICE, J. P. and ENDICOTT, J. (1987) Age-period-cohort analysis of secular trends in onset of major depression: findings in siblings of patients with major affective disorder. J. Psychiat. Res. 21, 23-35. PROVINCE, M. A. and RAo, D. C. 0985) Path analysis of family resemblance with temporal trends: applications to height, weight, and quetelet index in northeastern Brazil. A m . J. Hum. Genet. 37, 178-192. REICh, T., RICE, J. P., C"LO~m~t3~R,C. R., W~rl~, R. and JAMES,J. (1979) The use of multiple threshold and segregation analysis in analyzing the phenotypic heterogeneity of multifactorial traits. Ann. hum. Gen. 42, 371-390. RICE, J. P., CLO~INOER, C. R. and REICH, T. (1980) General causal models for sex differences in the familial transmission of multifactorial traits: an application to human spatial visualizing ability. Soc. Biol. 27, 36-47. RICE, J. P., ENDICOTT, J. KNESEVICH, M. A. and ROCHBERa, N. (1987) The estimation of diagnostic sensitivity using stability data: an application to major depressive disorder. J. psychiat. Res. 21, 335-343. RICE, J. P., REICH, T., ANDREASEN, N. C., ENDICOTT, J., VAN EERDEWEGH, M., FISHMAN, R., HIRSCHFELD, R. M. A. and KLERMAN, G. L. (1987) The familial transmission of bipolar illness. Archs gen. Psychiat. 44, 441-447. ROBINS, L. N., FIELZER, J. E., W~SSMAN, M. M., ORVASrmL, H., GRU~mER6, E., BuRrm, J. D. and REGmR, D. A. (1984) Lifetime prevalence of specific psychiatric disorders in three sites. Archs gen. Psychiat. 41, 949-958. SAS INSTrrUTEINC. (1983) Proportional Hazards General Linear Models. SUGI Supplemental Library Users Guide. SAS Institute Inc., Cary, North Carolina. WEISSMA~, M. M., WICrd~MARATNE, P., MERIKANOAS, K. R., LEcr,m.~,~, J. F., PRUSO~, B. A., CARUSO, K. A., KtDD, K. K. and GAmaO~, G. D. (1984) Onset of major depression in early adulthood: increased familial loading and specificity. Arehs gen. Psychiat. 41, 1136-1143. WEtSSMAN, M. M., KIND, K. K. and PRUSO~, B. P. (1987) Variability in rates of affective disorders in relatives of depressed and normal probands. Archs gen. Psychiat. 39, 1397-1403.