VOL. 16, NO. 2, 1990 The Role of Gender inIdentifying Subtypes ofSchizophrenia: A LatentClass Analytic Approach

263

by Jill M. Goldstein, Susan L.Santangelo, John C. Simpson,and Ming T. Tsuang

Abstract

Past literature suggests that schizo-phrenic men and women may be atdifferent risks for developing dif-ferent subtypes of schizophrenia.This hypothesis was tested usingdata from the well-known retrospec-tive cohort family studies, the Iowa500 and the Iowa non-500. Thesample consisted of 171 male and161 female DSM-IH schizophrenicpatients and 713 of their first-degreerelatives. First, bivariate tests forgender differences were conductedregarding family morbidity, age ofonset, premorbid history, season ofbirth, and expression of deficit andaffective symptoms. Restricted max-imum likelihood latent class anal-ysis was then used to test whetherthere was a subgroup of schizo-phrenic men who were more likelyto have a low familial risk forschizophrenia or schizophreniaspectrum disorders, deficit symp-toms, poor premorbid history, andbirth in the winter months, sug-gesting possible early environmen-tal insults, compared to schizo-phrenic women. Results showedthat although men were more likelyto meet these criteria, women alsomet them, thus suggesting genderdifferences in the prevalence of thesubtype. Schizophrenic womenwere more likely to express a formof the illness characterized bydysphoria, persecutory delusions,and a higher family morbidity riskfor schizophrenia than schizophren-ic men. Results for spectrumdisorders among relatives wereequivocal with regard to gender.

Gender differences in schizophreniahave been reported as far back asKraepelin (1919/1971), who describ-ed dementia praecox as a disorder of

young men. Since then, descriptiveevidence has accumulated demon-strating gender differences in age ofonset, premorbid history, early en-vironmental insults, family morbidi-ty risk, psychopathology, structuraland functional brain abnormalities,and course of schizophrenia. Schizo-phrenic men have consistently beenfound to have an earlier age of onsetthan schizophrenic women (Lewine1988). Men tend to have their peakperiod of onset from ages 18 to 25years and women from 25 to 35years. In addition, it has been foundthat approximately 3 to 10 percent offemale cases have onset of illnessafter 45 years in contrast to few, ifany, male cases. It has also beenconsistently demonstrated acrosstime period that schizophrenic menhave a significantly poorer premor-bid history than schizophrenicwomen (Gittelman-Klein and Klein1969; Goldstein 1988). Women havea better premorbid picture since theyare older and more often married atthe time of illness onset, but theytend to have better early childhoodhistories as well, as defined byschool achievement and sociability.

Further, there is evidence thatschizophrenic men and women mayexpress the illness differently aswell. Schizophrenic women tend toexhibit affective symptomatologysuch as dysphoria, irritability, andanger, as well as more paranoia andexplosivity. Schizophrenic men tendto express more flat affect and othersymptoms that resemble a negativeor deficit symptom picture (Lewine1981; Goldstein and Link 1988).

Reprint requests should be sent to Dr.J.M. Goldstein, Brockton/West RoxburyVA Medical Center, Psychiatry Service(116A), 940 Belmont St., Brockton,MA 02401.

264 SCHIZOPHRENIA BULLETIN

It is important to relate gender dif-ferences in the expression of schizo-phrenia to descriptive findings onfamily history and biological factorsdifferentiating schizophrenic menand women. A few studies havefound that schizophrenic men areless likely to have a family history ofschizophrenia than women (Bellodiet al. 1986; Goldstein et al., inpress). This may be related to otherfindings that have indicated thatschizophrenic men have more birthcomplications and neurologicaldeficits than schizophrenic women(Mednick et al. 1978; Goldberg1985). Although these organic in-sults are not specific to schizophre-nia, perhaps they contribute to theseverity of the illness for a subgroupof schizophrenic men. There is somepreliminary evidence demonstratinggender differences in the corpuscallosum and in ventricle-brainratios, suggesting more diffuse orleft hemisphere dysfunction amongschizophrenic men (Flor-Henry1978; Nasrallah et al. 1986). Some ofthe studies reported in this issue ofthe Schizophrenia Bulletin have sup-ported these findings (Kopala andClark 1990; Lewine et al. 1990); oneother has questioned them (Nas-rallah et al. 1990). This is consistentwith early work published by Gurand colleagues on gender dif-ferences in cerebral blood flow indifferent hemispheres of the brain.These findings are reviewed in thisissue (Gur and Gur 1990).

Taken together, the findings ongender discussed above suggest thatschizophrenic men and women maybe at different risks for expressingparticular forms of the illness. Pastresearch has largely been limited toexamining bivariate associationsbetween gender and the factorsmentioned previously. We were in-terested in understanding the effect

of gender by examining the joint ef-fects of age of onset, premorbidhistory, symptom expression, familyhistory, and early environmental in-sults. We hypothesized that schizo-phrenic men would be more likelyto express a form of illness charac-terized by an early age of onset,deficit symptomatology, a poorpremorbid history, low family mor-bidity risk for schizophrenia, andearly environmental insults. In con-trast, schizophrenic women wouldbe more likely to express a form ofillness characterized by the expres-sion of affective symptomatology,paranoia, and a higher family riskfor schizophrenia.

Methods

The data for this study were fromthe double-blind retrospectivecohort studies conducted in Iowa inthe 1970's, the Iowa 500 and theIowa non-500. Sample selection andprocedures for the original studieshave been fully described previously(Morrison et al. 1972; Tsuang andWinokur 1975). Probands wereselected from a blind review of indexadmissions, followup records, andinterviews of 510 consecutive admis-sions who had a chart diagnosis ofschizophrenia from the IowaPsychopathic Hospital during1934-44. Probands and relatives werethen rediagnosed by two expertpsychiatrist-diagnosticians usingDSM-1II criteria (American Psychi-atric Association 1980), although thecriterion limiting age of onset to45 years was eliminated, as inDSM-III-R (American PsychiatricAssociation 1987) (Kendler et al.1985). Diagnoses of first-degreerelatives of probands were based ona blind, personal, structured psychi-atric interview (Tsuang et al. 1980),

hospital records obtained from asystematic search of the five Iowapublic psychiatric hospitals, or acombination of the personal inter-view and hospital records. Diag-nostic procedures and reliability arereported in detail elsewhere(Kendler et al. 1985).

The sample consisted of 332 pro-bands with a DSM-UI diagnosis ofschizophrenia (51.5% men and48.5% women). A detailed descrip-tion of the sociodemographic andonset history of the probands, aswell as a description of the relatives,is provided in a previous publication(Goldstein et al. 1989). Briefly, theIowa family studies consisted of amiddle to lower middle class, non-Hispanic white population of con-secutive admissions to a tertiaryuniversity-affiliated facility over a10-yeai1 period. Patients wereprimarily in the early stages ofdisorder on admission (i.e., 82%were first admissions) and werefollowed for an average of almost30 years.

Variables of interest in this studywere obtained from a structured in-terview developed by Tsuang et al.(1980), medical records, and rela-tives. The instrument has demon-strated reliability and validity innumerous published articles (e.g.,Tsuang et al. 1976; Kendler et al.1987). Age of onset was defined asthe onset of a psychotic symptom orbizarre behavior, which reflects theDSM-lll criterion for onset of theprodromal or active phase of illness.It was assessed using all availableinformation from probands, rela-tives, and medical charts. Reliabilityof age of onset determination washigh (0.94) (Kendler et al. 1987). Poorversus good premorbid history wasassessed by two expert psychia-trist-diagnosticians after a review

VOL. 16, NO. 2, 1990 265

of all medical chart and structuredinterview information.

The symptom variables werecoded from clinician ratings at thetime of admission to the hospital.Because of the hypotheses in thisstudy, we were interested in deficitand affective symptoms and para-noia. Deficit symptomatology wasoperationalized as flat or blunted af-fect. Dysphoria indicated affectiveexpression, and persecutory delu-sions indicated severe paranoia.Symptoms were coded dichotom-ously, as present or absent.Diagnoses in probands and relativeswere described previously.

Given that early environmental in-sults, such as perinatal complica-tions, were not measured in thisstudy, we used winter season ofbirth as a crude indicator, one that iscommonly used in the literature(Hare 1988). Winter birth wasassessed as present if a proband wasborn during December throughMarch. Although January throughMarch is often used as the winterobservation period, winter in Iowabegins in the middle to end ofNovember.

Analytic Strategy

Simple bivariate relationships be-tween gender and premorbidhistory, symptom expression, andwinter birth were examined using x2

statistics. Gender differences in ageof onset and family morbidity riskswere examined using survivalanalysis and have been presented inprevious reports (Goldstein et al.1989; Goldstein et al., in press).

Latent class analysis (LCA) wasused to examine the corelationshipsbetween variables (Lazarsfeld andHenry 1968; Goodman 1974a, 1974b,1979; Clogg 1977; Young 1983). As

with other latent variable models,like factor analysis, LCA is a pro-cedure that attempts to explaincovariation among a set of observedvariables, by modeling the covaria-tion of observed variables withunobserved (and hence latent)variables, that are fewer in numberthan the observed ones. Since it isassumed that the observed variablesare related to each other onlythrough the latent variables, thelatter have considerable explanatoryinterest. Unlike factor analysis,LCA is designed for use withdichotomous (or polychotomous)variables and assumes that thelatent variables are also categorical.LCA has therefore been referred toas the discrete analog to factoranalysis (McCutcheon 1987).

Briefly, LCA works as follows: Thedata required for input consist of thefrequencies of all possible cross-classifications of the observedvariables. LCA then uses maximumlikelihood estimation to fit one or aseries of hypothesized models to ex-plain covariance patterns among theobserved variables. Each cell in thedata table provides the number ofpatients with the joint occurrence ofthe corresponding variables. The fitof a model is measured by the likeli-hood or probability of the observedpattern of cross-classified data,conditional on the model and theparameter estimates. The parameterestimates that maximize this likeli-hood function are thus the max-imum likelihood estimates (MLEs).If the variables are unrelated, then,allowing for random variability, theobserved pattern in the cross-classification table would be com-pletely determined by the product ofthe corresponding marginal frequen-cies (i.e., would fit a model of in-dependence). If the variables arerelated, one can test whether the

observed relationships are consistentwith (or "can be explained by") thehypothesized latent variables, orclasses.

As with other maximum likeli-hood procedures, a x2 goodness-of-fit is used to test whether thehypothesized model "fits" the observedcovariances of the variables. (Fordetails of this procedure, seeLazarsfeld and Henry 1968; Good-man 1974a, 1974b; Clogg 1977; Young1983; Young et al. 1986; McCutcheon1987). An advantage of LCA is thatone can simultaneously fit a latentclass model to several independentsamples (e.g., schizophrenic menand women) and thus test the fit ofone's hypothesized model acrossgroups of interest. For example, inthis study we will test the overall fitof the subtype model in men versuswomen, and examine gender dif-ferences in specific parameters ofthe model (Clogg and Goodman1984).

There are two types of parametersin the latent class model: latent classprobabilities and latent conditionalprobabilities. The latent class proba-bilities provide information aboutthe frequency of occurrence of eachlatent class. For example, when theanalysis is stratified by gender, thelatent class probabilities indicate thenumber of latent classes and the sizeof each latent class for schizophrenicmen versus women. The latent con-ditional probabilities provide infor-mation about the degree of associa-tion between each of the observedvariables and the latent classes, andare analogous to factor loadings infactor analysis (McCutcheon 1987).As conditional probabilities, theseparameters give the sensitivity ofthe observed variables for indicatinga particular latent class. In this in-vestigation, we were interested inthe differential expression of two

266 SCHIZOPHRENIA BULLETIN

hypothesized subtypes (i.e., latentclasses) for men versus women;therefore, the parameters of thelatent class model were estimatedseparately for men and for womenwithin a single model. This enabledus to compare the frequency of oc-currence of each subtype for menversus women and to evaluate thesensitivity of particular variables foridentifying a specific subtype formen versus women.

The statistical significance of dif-ferences in parameter estimates wastested by using x2 to compare the fitof unrestricted models to modelswhere parameters were restricted tospecific values or were given equali-ty or inequality constraints. In par-ticular, values of specific parametersin each subtype were restricted tobe either equal across gender, higherin one sex than the other, or dif-ferent in magnitude across the sub-types within one sex. The good-ness-of-fit x2 of the unrestrictedmodel is subtracted from the x2 ofthe restricted model to examinewhether there is an improvement infit. The difference between the twoX2 measures is also distributed as x2

with degrees of freedom equal tothe difference in degrees of freedombetween the two models. If the in-crease in x2 is small relative to thedifference in degrees of freedom,the fit of the restricted model isfound to be acceptable and therestrictions valid. Using standarderrors to estimate confidence inter-vals for significance testing is inap-propriate, since latent class modelsdo not follow assumptions for nor-mal distributions. The LCA soft-ware, called Maximum LikelihoodLatent Structure Analysis, wasdeveloped by Clogg (1977) andadapted to personal computers byEliason (see McCutcheon 1987).

Results

Bivariate Findings: Gender andSymptom Expression, PremorbidHistory, Season of Birth, Age ofOnset, and Family Morbidity Risk.The x2 tests of gender and premor-bid history, winter birth, flat affect,persecutory delusions, and dys-phoria showed no significant dif-ferences. However, more schizo-phrenic men than women experi-enced a poor premorbid history(46% of the men vs. 39% of thewomen) and flat affect (60% of themen vs. 52% of the women). On theother hand, more schizophrenicwomen than men experienceddysphoria (54% of the women vs.45% of the men) and persecutorydelusions (59.5% of the women vs.51% of the men). An equal propor-tion of men and women were bornin the winter.

In addition, schizophrenic men ex-perienced an earlier age at onsetthan schizophrenic women. Survivalanalysis of gender differences in ageof onset showed significant resultsand were presented in a previousreport (Goldstein et al. 1989). Find-ings showed that by age 25, 53 per-cent of the men had become ill com-pared to only 32 percent of thewomen. Eight percent of womenbecame ill at or after 40 years of agecompared to only 1 percent of themen who had onset of illness at 40years. Schizophrenic men also ex-perienced a mean age of onset of24.3 (SD = 6.2) years versus 27.9 (SD= 8.3) years for schizophrenicwomen (t = 4.4, df = 287.6, p= 0.0001).

Gender differences in the familialtransmission of schizophrenia andspectrum disorders are presented indetail elsewhere (Goldstein et al., inpress). Briefly, results showed thatschizophrenic women had a signifi-

cantly higher familial risk forschizophrenia, schizophreniform,and schizoaffective disorders thanrelatives of schizophrenic men. Thefindings were due neither to selec-tion bias nor fertility effects. How-ever, when the definition of illnessin relatives was expanded to includethe full spectrum (i.e., schizophre-nia, schizophreniform disorder,schizoaffective disorder, paranoiddisorder, atypical psychosis, andschizotypal personality disorder),the difference in the risk to relativesof male versus female probands wasattenuated. This was due to theresult that relatives of male pro-bands had a significantly higher riskof schizotypal personality disorderthan did relatives of female pro-bands (Goldstein et al., in press).

Covariance Patterns. Althoughbivariate findings between genderand symptom expression, season ofbirth, and premorbid history werenot significant, we hypothesizedthat when examined in the contextof each other as well as with age ofonset and familial risk, significantpatterns would emerge. As mention-ed, the hypothesis was that therewould be a subgroup of schizo-phrenic men whose form of the ill-ness necessitated early environ-mental insults, measured as winterbirth, and that these men would ex-press the earliest age of onset, poorpremorbid history, flat affect, and alack of ill family members withschizophrenia. In contrast, schizo-phrenic women were expected to ex-press a form of illness characterizedby dysphoria, persecutory delu-sions, and a higher risk ofschizophrenia among their relatives.

First, a test of independenceamong the observed variables wasconducted to test whether theobserved variables were related to

VOL. 16, NO. 2, 1990 267

each other. Results showed that theX2 test of a model of independencewas marginally insignificant (x2 =274.91, df = 247, p = 0.11); thus,even though insignificant, it mostlikely did not fit the data well. Infact, the test of whether our hy-pothesized model fit the covariancesbetween the observed variablesdemonstrated that it was a verygood fit, as indicated by a highly in-significant x2 (x2 = 181.04, df = 22'6,p = 0.99). Table 1 presents theresults of the latent class analysiswhich showed that schizophrenicmen and women differentially ex-pressed the subtypes.

Seventy percent of the men ex-perienced the class 1 subtype.Among the men in this subtype, 60percent were born in the wintermonths, 56 percent had a poorpremorbid history, 78 percent ex-pressed flat affect, and 53 percenthad an early age of onset. In addi-tion, only 5 percent of their first-degree relatives had suffered fromschizophrenia. Only 30 percent ofthe men experienced the class 2 sub-type. Seventy-three percent of thesemen expressed persecutory delu-sions, 84 percent dysphoria, 55 per-cent early onset, and none of theirrelatives had schizophrenia.

In contrast, 63 percent of theschizophrenic women experiencedthe class 2 subtype characterized bydysphoria and persecutory delu-sions. Sixty-six percent of this groupof women expressed persecutorydelusions, 59 percent dysphoria,and 17 percent of their relatives hadschizophrenia. Thirty-seven percentof the women experienced the class1 subtype which, for women, wascharacterized primarily by poorpremorbid history alone (97% ex-perienced this).

To test the significance of theparameters for differentiating sub-

types within gender and for testingdifferences between the sexes, aseries of parameter restrictions wereimposed on the model and the fitswere compared to the unrestrictedmodel (see detailed discussionunder Methods; Clogg and Good-man 1984).

Table 2 presents a series of restric-tion tests applied to specific hy-potheses about gender differences inthe expression of the two subtypes.A model combining all importantrestrictions was then compared tothe unrestricted model (Clogg andGoodman 1984). For men, the bestdiscriminators between classes 1 and2 were winter birth, poor premorbidhistory, and flat affect for class 1,and persecutory delusions anddysphoria for class 2. When equalityrestriction tests were applied (seeTests 1 and 2, table 2), theseparameters were significantly dif-ferent between classes within men.Test 1 refers to constraining seasonof birth, premorbid history, and flataffect to be equal between the twoclasses within men, suggesting thatthese variables do not differentiatethe two classes for men. The x2 in-crease is calculated by subtractingthe x2 of the unrestricted model,which represents a hypothesis thatthese variables are not equallypresent in the two classes amongmen, from the x2 of the Test 1 modelrestrictions, which suggest thatthese variables are equally presentin the two classes among men. ForTest 1, the x2 increased 36.55 (df = 3,p < 0.0001), indicating that theparameters were not equal. For test2, equality restrictions were placedon dysphoria and persecutory delu-sions within men across classes.Again, the restricted model was apoor fit as indicated by an increasein x2 of 24.95 (df = 3, p = 1.6 x10"5), suggesting that dysphoria and

persecutory delusions were notequally expressed in the two classesamong men. Family history ofschizophrenia and early onset didnot differentiate the two classeswithin men (i.e., the parameterestimates were not significantly dif-ferent). This suggests that men hadearly onset and had few relativeswith schizophrenia, regardless of thesubtype they expressed.

For women, the best discrimi-nators between classes 1 and 2 werefamily history of schizophrenia,persecutory delusions, and dys-phoria for class 2, and only poorpremorbid history for class 1. Test 3placed equality restrictions on familyhistory, persecutory delusions, anddysphoria across classes withinwomen. A x2 increase of 10.9 (df = 3,p = 0.004) indicated that parameterswere not equal, suggesting thatthese variables were good discrimi-nators of the two classes amongwomen. Test 4 placed an equalityrestriction on premorbid history,which also was not equal withinwomen (x2 increase = 6.27, df = 1,p = 0.01). In addition, most womenhad onset of illness later than 25years, regardless of subtype, andthey equally experienced winterbirth and flat affect.

To test whether parameter esti-mates for early onset, winter birth,poor premorbid history, and flat af-fect for men were significantly dif-ferent from these parameter esti-mates for women in both classes,values for women in class 1 were setequal to the specific values of thesevariables for men in class 1, andclass 2 values for women and menwere also set equal. Test 5 in table 2shows that the restrictions were anunacceptable fit (x2 increased 53.58,df = 14, p = 1.5 x 10"6), suggestingthat these parameters significantlydiffered for men and women.

268 SCHIZOPHRENIA BULLETIN

Table 1. Gender differences in expression of observed and latent variables defining subtypes ofschizophrenia

Observedvariables

Men Women

Latent class probabilities

70%Class 1

30%Class 2

37»/oClass 1

63%Class 2

Familyhistoryofschizophrenia

No

Yes

0.95

0.05

1.00

0.00

1.00

0.00

0.83

0.17

Earlyonset

(<25 years)

No

Yes

0.47

0.53

0.45

0.55

0.74

0.26

0.66

0.34

Seasonofbirth

Non-winterWinter

0.40

0.60

0.90

0.10

0.55

0.45

0.52

0.48

Premorbidhistory

Good

Poor

0.44

0.56

0.77

0.23

0.03

0.97

0.94

0.06

Persecutorydelusions

No

Yes

0.57

0.43

0.27

0.73

0.55

0.45

0.34

0.66

Flataffect

No

Yes

0.22

0.78

0.77

0.23

0.44

0.56

0.49

0.51

Dysphoria No

Yes

0.68

0.32

0.16

10.84 I

0.57

0.43

0.41

0.59

Likelihood ratio x2 =181.04, df = 226, p = 0.99.

D : Boxes denote significant differences between conditional probabilities within a gender group as tested by models represented in table 2.

= : Equality signs between conditional probabilities denote nonsignificant differences within a gender group as tested by models in table 2.

VOL 16, NO. 2, 1990 269

Table 2. Confirmatory restriction tests for the hypothesized model,with schizophrenia only in relatives

Model

Unrestrictedhypothesizedmodel

Test 1

Test 2

Test 3

Test 4

Test 5

Test 6

Test 7

Test 8

Likelihoodratio x2

181.04

217.59

205.99

191.94

187.31

234.62

191.35

212.22

192.60

df

226

229

229

228

227

240

234

236

238

Increase inX2 and df

36.55, df = 3

24.95, df = 3

10.90, df = 2

6.27, df = 1

53.58, df = 14

10.31, df = 8

31.18, df = 10

11.56, df = 12

p value

0.99

5.7 x 10"8

1.6 x 10-s

0.004

0.01

1.5 x 10-6

0.24

0.0005

0.48

Test 1. Equality restrictions on season of birth, premorbid history, and flat affect across the 2classes within men.

Test 2. Equality restrictions on dysphoria and persecutory delusions across the 2 classes within men.

Test 3. Equality restrictions on family history, dysphoria, and persecutory delusions across classeswithin women.

Test 4. Equality restrictions on premorbid history across classes within women.

Test 5. Specific value restrictions on onset, season of birth, premorbid history, and flat affect forfemales whose values were set to equal males' for class 1 and class 2.

Test 6. Specific value restrictions on dysphoria and persecutory delusions for males whose valueswere set to equal females' for class 1 and class 2.

Test 7. Specific value restrictions in test 6, plus family history for males whose values were set equalto females'.

Test 8. Combination restrictions from tests 1-7: Specific value restrictions on persecutory delusionsand dysphoria; across-sex equality restrictions on family history for class 1 women and class 2 men;within-sex equality for females on season of birth and flat affect; within-sex equality restrictions ononset (within both sexes); and values for premorbid history allowed to vary, as were class 1 maleand class 2 female values for family history.

In Test 6 (table 2), parameterestimates for men regarding perse-cutory delusions and dysphoriawere set exactly equal to the valuesfor women in both classes. Theserestrictions were an acceptable fit (x2

increased 10.3, df = 8, p = 0.24),suggesting that the expression ofthese symptoms co-occurred similar-ly for men and women. However,the majority of women (63%) ex-perienced this subtype as comparedto the minority (30%) of men.

Restricting the estimates for familyhistory of schizophrenia in men toequal women's was an unacceptablefit (Test 7: x2 increased 31.18,df = 10, p = 0.0005). This suggeststhat relatives of schizophrenicwomen who expressed dysphoriaand persecutory delusions were sig-nificantly more likely to haveschizophrenia than relatives ofschizophrenic men who expressedthese symptoms. Finally, a combin-ed test of all significant restrictions,

which represent all of the hypothe-sized differences between men andwomen within and across theclasses, compared to an unrestrictedmodel that would suggest these dif-ferences do not hold, produced amodel that fit well (test 8: x2 in-creased 11.56, df = 12, p = 0.48),thus providing support for ourhypotheses about genderdifferences.

In a second analysis, the defini-tion of illness in relatives was ex-panded to include the full spectrum(i.e., schizophrenia, schizophreni-form disorder, schizoaffectivedisorder, atypical psychosis, para-noid disorder, and schizotypal per-sonality disorder). It was hypothe-sized that the expression of spec-trum disorders in relatives ofschizophrenic men would co-occurwith poor premorbid history, winterbirth, and flat affect in the pro-bands, even though schizophreniaalone was not expressed by relativesof men with these characteristics.This suggests a vulnerability model,in which the pathogenic gene forschizophrenia among men would beexpressed only if an environmentalinsult occurred. Thus, the expecta-tion was that the parameter estimatefor family history of spectrum dis-orders in relatives of men in class 1would be significantly higher thanits estimate in class 2 for men, andequal to the estimate of this param-eter for women in class 1 and class 2.

A test of independence indicatedthat the observed variables in thismodel were significantly related toone another (x2 = 305.80, df = 247,p = 0.006). Table 3 presents thelatent class analysis with the ex-panded definition of illness inrelatives. Findings showed that thehypothesized model was a good fit,as indicated by a nonsignificant x2:X2 = 222.84 {df = 224) p = 0.51. The

270 SCHIZOPHRENIA BULLETIN

majority of males (69%) still express-ed class 1, while the majority offemales (74%) continued to expressclass 2. Thirty-one percent of themales expressed the subtypecharacterized by dysphoria andpersecutory delusions, and 26 per-cent of the females now expressedthe same subtype as the majority ofmales (i.e., the subtype charac-terized by winter birth, poor pre-morbid history, and flat affect). Testsof the family history variable did notsupport the hypothesis.

Table 4 presents the major restric-tion tests examining the significanceof the parameter estimates. Withinmen, family history of spectrumdisorders did not significantly dif-ferentiate the two classes (Test 1:X2 increased 1.89, df = 1, p = 0.17),although only 9 percent of men whoexperienced winter birth, flat affect,and poor premorbid history hadrelatives with spectrum disorderscompared to 19 percent of the menwith persecutory delusions anddysphoria.

In contrast, 21 percent of thewomen who experienced winterbirth, poor premorbid history, andflat affect had relatives with spec-trum disorders. This was differentfrom the 9 percent of men in class 1,although borderline in significance(Test 2: x2 increased 3.45, df = 1,p = 0.06). Both schizophrenic menand women experienced this sub-type, and when across-sex equalityconstraints were placed on season ofbirth, premorbid history, and flat af-fect, the x2 did not significantly in-crease, suggesting that the valueswere equal (Test 3: x2 increased 5.62,df=6,p = 0.47). Although bothmen and women experienced thesubtype, they differed in itsprevalence: 69 percent of the menversus 26 percent of the women.

In addition, 74 percent of thewomen expressed the subtypecharacterized by dysphoria andpersecutory delusions, and 15 per-cent of their relatives expressedspectrum disorders. This was notsignificantly different from theestimate of 21 percent for familyhistory of spectrum disorder in class1 for women (Test 4: x2 increased0.33, df = 1, p = 0.57). Thus,relatives of women had spectrumdisorders, regardless of the subtypethey expressed. Parameter estimatesfor this subtype were equal for menand women as indicated by Test 5(X2 increased 5.72, df = 6, p = 0.46).

Finally, age of onset did notsignificantly differentiate the sub-types for men or women. That is,most men had onset of illness at 25years or younger, and most womenhad onset after 25 years, regardlessof what subtype they expressed.

When we subjected the model toall of the hypothesized restrictions,the model remained a good fit, sug-gesting that the observed differencesbetween men and women (describ-ed above) were significant (Test 6:X2 increased 8.82, df = 10, p = 0.61).The fact that family history is not aclear discriminator between the sub-types is illustrated by the fit of themodel in Test 7 (including within-sex equality restrictions on familyhistory) which is equally as good asthe fit of the model in Test 6. Test 8merely confirms that the propor-tions of men and women expressingthe class 1 and class 2 subtypes aresignificantly different.

To test the robustness of theestimates in the two latent classmodels (in tables 1 and 3), 10 per-cent random samples (half men/halfwomen) were eliminated in threeseparate analyses. Parameter esti-mates in both models were almost

identical to the full sample in allthree trials, suggesting that theestimates were robust and valid. Inaddition, when different startingvalues were used, estimates wereessentially unchanged, again sug-gesting the robustness of theparameter estimates.

Discussion

Findings in this study underscorethe importance of gender in under-standing the heterogeneity ofschizophrenia. Our previous workshowed that schizophrenic menhave significantly earlier onset ofillness and have a lower risk ofschizophrenia, schizophreniform,and schizoaffective disorders amongtheir first-degree relatives than dowomen. In this study, simplebivariate findings showed that menexpressed more flat affect and apoorer premorbid history, andwomen, more dysphoria, althoughthe findings were not significant. Itis interesting that although bivariaterelationships between gender andsome of the factors examined show-ed no significant effects, in the con-text of each other, patterns emergedthat significantly differentiatedcharacteristic subtypes for whichmen and women were at differentrisks. This may in part explain whythere are some inconsistencies in theliterature about gender differencesin schizophrenia. Our resultsunderscore the importance of notonly examining main effects for fac-tors, such as gender, that may beimportant for the understanding ofschizophrenia, but of also examiningtheir conditional effects, which maybe more revealing. That is, in thisstudy, not all schizophrenic menand women differed, and thereforethe bivariate relationships between

VOL. 16, NO. 2, 1990 271

Table 3. Gender differences in expression of observed and latent variables defining subtypes, withschizophrenia spectrum in relatives

Observedvariables

Men Women

Latent class probabilities

69%Class 1

3 1 %Class 2

26%Class 1

74%Class 2

Familyhistoryof Noschizophreniaspectrum Yes

0.91

0.09

0.81

0.19

0.79

0.21

0.85

0.15

Earlyonset

(<25 years)

No

Yes

0.48

0.52

0.44

0.56

0.75

0.25

0.67

0.33

Seasonofbirth

Non-WinterWinter

0.41

0.59

0.85

0.15

0.21

0.79

0.64

0.36

Premorbidhistory

Good

Poor

0.44

0.56

0.76

0.24

0.50

0.50

0.64

0.36

Persecutorydelusions

No

Yes

0.57

0.43

0.27

0.73

0.52

0.48

0.38

0.62

Flataffect

No

Yes

0.21

0.79

0.77

0.23

0.00

1.00

0.64

0.36

Dysphoria No

Yes

0.71

0.29

0.13

10.87 I

0.70

0.30

0.39

0.61

Likelihood ratio x2 = 22284, df = 224, p = 0.51.

D : Boxes denote significant differences between conditional probabilities within a gender group as tested by models represented in table 4.

= : Equality signs between conditional probabilities denote nonsignificant differences within a gender group as tested by models in table 4.

272 SCHIZOPHRENIA BULLETIN

Table 4. Confirmatory restriction tests for the hypothesized model,with schizophrenia spectrum in relatives

Model

Unrestrictedhypothesizedmodel

Test 1

Test 2

Test 3

Test 4

Test 5

Test 6

Test 7

Test 8

Likelihoodratio x2

222.84

224.73

226.29

228.46

223.17

228.56

231.06

230.56

256.63

df

224

225

225

230

225

230

234

234

236

Increase inx2and df

1.89, df = 1

3.45, df = 1

5.62, df = 6

0.33, df = 1

5.72, df = 6

8.22, df = 10

7.72, df = 10

33.79, df = 12

p value

0.510.17

0.06

0.47

0.57

0.46

0.61

0.66

0.0007

Test 1. Equality restriction on family history within men

Test 2. Equality restriction on family history for class 1 across sex.

Test 3. Equality restrictions on season of birth, premorbid history, and flat affect for classes 1 and 2across sex

Test 4 Equality restrictions on family history within women.

Test 5 Equality restrictions on family history, dysphoria, and persecutory delusions for classes 1and 2 across sex.

Test 6. Combination restrictions from tests 1-5' Equality restrictions on family history, season ofbirth, premorbid history, and flat affect for class 1 across sex; equality restrictions on family history,dysphoria, and persecutory delusions for class 2 across sex; within-sex equality restrictions on ageof onset; and restrictions on the latent class probabilities equating class 1 males with class 2females and class 2 males with class 1 females.

Test 7. Same as test 6 except that equality restrictions on family history are within sex instead ofacross sex.

Test 8. Identical to test 6 except in the latent class probability equality restrictions. Here class 1males were set equal to class 1 females and class 2 males and females were equated.

gender and certain characteristicswere not always significantly dif-ferent. Rather, our results suggestthat schizophrenic men and womenmay have similar subtypes; how-ever, the prevalence of the subtypesamong men and women significant-ly differ.

The results in this study demon-strated that schizophrenic men wereat higher risk for expressing a sub-type characterized by poor premor-bid history, winter birth, and flat

affect. These men also had fewrelatives with schizophrenia or spec-trum disorders. In contrast, schizo-phrenic women were at higher riskfor a form of the illness character-ized by dysphoria and persecutorydelusions. Women also expressedthe first subtype, although theprevalence among women was low(26%). In addition, first-degreerelatives of women were at risk forspectrum disorders regardless of thesubtype they experienced. Relatives

of women were at risk for schizo-phrenia alone, for the subtypecharacterized by dysphoria andpersecutory delusions.

Our findings raise the question ofwhether schizophrenic women havea higher genetic loading and higherthreshold for the illness. However,this would result in a lower preva-lence of less severe forms of the ill-ness among men, which is in con-trast to previous literature on genderdifferences in the expression andcourse of schizophrenia (Salokangas1983; Goldstein 1988; Lewine 1988).On the other hand, our resultsdemonstrate that there may be asubtype of the illness for whichnonfamilial factors play an impor-tant role. Winter birth is a crude in-dicator, but it has been associatedwith perinatal complications, vul-nerability to viruses, and nutritionaldeficiencies (Hare 1988). These en-vironmental insults may in part leadto early brain damage, resulting in apoor premorbid history and flat af-fect. Our findings are also consis-tent with the idea that the patho-genic gene is not expressed in thissubgroup without an environmentalinsult, or there may be a sporadicform of the illness. Further, men areat higher risk for a subtype ofschizophrenia in which nonfamilialfactors are significant.

Our results are consistent with thehypothesis of a neurodevelopmentalform of the illness (Seidman 1983;Weinberger 1987; Lewis 1989;Murray et al. 1989), and extend thishypothesis by demonstrating thatmen may be at higher risk for its ex-pression. This is reminiscent ofKraepelin's (1919/1971) descriptionof dementia praecox as an illness af-flicting young men.

In addition, schizophrenic womenare at higher risk for a form of theillness in which affective symptoms

VOL. 16, NO. 2, 1990 273

and familial factors are prominent.Our previous work has shown thatrelatives of schizophrenic womenare at higher risk for unipolardisorder than those of schizophrenicmen, although the findings did notquite reach significance (Goldstein etal., in press). Future research mustexamine how familial factors, affec-tive expression, and gender arerelated to each other by examiningthe modes of transmission and theirrelationship to gender and phenom-enology. Our results suggest that therole of affective symptomatology inschizophrenia may provide insightsinto understanding the hetero-geneity of the illness, and thatgender may provide clues as towhere to begin to search for possiblepathological mechanisms.

Finally, findings showed that ageof onset did not differentiate the twosubtypes. That is, most men becameill at or before 25 years and mostwomen became ill after 25 years,regardless of subtype. Results of thesubtype model suggest that age ofonset is not differentially related tofamily morbidity risk, winter birth,poor premorbid history, or symptomexpression. Thus, age-of-onset dif-ferences between men and womenare most likely explained by someother factors. As suggested in otherwork (see Seeman and Lang 1990),likely etiological candidates are hor-monal differences between men andwomen.

There are some limitations to themodel presented in this study. Someof the observed indicators are crude,such as winter birth to measure en-vironmental insults. Since crudermeasures are less reliable indicators,the expectation is that they would beless likely to correlate with other fac-tors. In this study, this was not aproblem given that winter birth wasa significant discriminator betweensubtypes. Our current work assessesperinatal complications and severe

illness or injury before age 1, insteadof relying on winter birth, whichshould facilitate replication in thefuture.

In addition, the latent class modelis not fully specified, since thereare limitations on the number ofobserved variables it can handle.Current software allows up to eightdichotomous variables, although sixare thought to be more reasonablegiven the nature of the statisticalprocedure (C.C. Clogg 1988, per-sonal communication). Althoughthere are limits to the allowable in-dicators, this would not invalidatethe ability of the chosen indicatorsto differentiate meaningful classes.Further, latent variable models arehighly dependent on the variabilityof the observed indicators. In thisstudy, however, the robustness ofthe estimates was established, giventhat they remained stable after threetrials of eliminating random sub-samples. Finally, it may be arguedthat results in this study are whollyattributable to the perceptual bias ofclinical raters regarding women andmen. Although this argument maybe plausible in considering differen-tial symptom expression, it is notplausible for the assessment ofwinter birth, family morbidity risk,or premorbid history.

Thus, findings in this study pro-vide convincing evidence of thevalue of gender in providing clues todelineate subtypes of schizophrenia.Given the limitations on the specifi-cation of the model, it is importantto relate results in this study tofindings in previous research sug-gesting gender differences in struc-tural and functional brain abnor-malities (Gur et al. 1985; Nasrallahet al. 1986; Gur and Gur 1990),differential neuroleptic response(Seeman 1985), and neurocognitivefunctioning (Goldberg 1985; Haas

1987). For example, it is possible thatearly environmental insults maycause structural brain damage. Inview of gender differences in braindevelopment—for example, testoster-one has a retarding effect on the lefthemisphere of the brain (Geschwindand Galaburda 1985; DeLisi et al.1989)—men may be at higher risk forconsequences from perinatal com-plications than women. It may bethat the explanation of significantgender effects in schizophrenia in-volves factors inherent to being aman or a woman, such as the effectsof estrogen and differences in braindevelopment. It may also be thatexplanatory factors consist ofsecondary consequences of being aman or woman, such as differentialexposure to head injury. In eithercase, findings in this study suggestthat gender can provide importantclues to understanding the etio-logical heterogeneity of schizo-phrenia, and should be specificallyconsidered in the selection ofsamples and evaluation of data instudies of schizophrenia.

References

American Psychiatric Association.DSM-11I: Diagnostic and StatisticalManual of Mental Disorders. 3rd ed.Washington, DC: The Association,1980.

American Psychiatric Association.DSM-I1I-R: Diagnostic and StatisticalManual of Mental Disorders. 3rd ed.,revised. Washington, DC: TheAssociation, 1987.

Bellodi, L.; Bussoleni, C ; Scorza-Smeraldi, R.; Grassi, G.; Zacchetti,L.; and Smeraldi, E. Family study ofschizophrenia: Exploratory analysisfor relevant factors. SchizophreniaBulletin, 12:120-128, 1986.

Clogg, C.C. "Unrestricted andRestricted Maximum Likelihood

274 SCHIZOPHRENIA BULLETIN

Latent Structure Analysis: A Manualfor Users." Working Paper 1977-09.Pennsylvania State University,Population Issues Research Center,1977.

Clogg, C.C., and Goodman, L.A.Latent structure analysis of a set ofmultidimensional contingencytables. Journal of the AmericanStatistical Association, 79:762-771,1984.

DeLisi, L.E.; Dauphinais, I.D.; andHauser, P. Gender differences in thebrain: Are they relevant to thepathogenesis of schizophrenia? Com-prehensive Psychiatry, 30:197-208,1989.

Flor-Henry, P. Gender, hemisphericspecialization, and psychopathology.Social Science and Medicine,12:155-162, 1978.

Geschwind, N., and Galaburda,A.M. Cerebral lateralization:Biological mechanisms, associations,and pathology. I. A hypothesis anda program for research. Archives ofNeurology, 42:428-459, 1985.

Gittelman-Klein, R., and Klein, D.F.Premorbid asocial adjustment andprognosis in schizophrenia. Journalof Psychiatric Research, 7:35-53, 1969.

Goldberg, S. Positive and negativefamily history for schizophrenia:Differentiating variables. In: Alpert,M., ed. Controversies in Schizophrenia.New York: Guilford Press, 1985.pp. 317-329.

Goldstein, J.M. Gender differencesin the course of schizophrenia.American Journal of Psychiatry,145:684-689, 1988.Goldstein, J.M.; Faraone, S.; Chen,W.; Tolomiczenko, G.; and Tsuang,M. Gender differences in the famil-ial transmission of schizophrenia.British Journal of Psychiatry, in press.Goldstein, J.M., and Link, B.G.Gender differences in the clinical

expression of schizophrenia. Journalof Psychiatric Research, 22:141-155,1988.Goldstein, J.M.; Tsuang, M.T.; andFaraone, S.V. Gender and schizo-phrenia: Implications for under-standing the heterogeneity of the ill-ness. Psychiatry Research,28:243-253, 1989.

Goodman, L.A. The analysis ofsystems of qualitative variableswhen some of the variables areunobservable. American Journal ofSociology, 79:1179-1259, 1974a.

Goodman, L.A. Exploratory latentstructure analysis using both iden-tifiable and unidentifiable models.Biometrika, 61:215-231, 1974b.

Goodman, L.A. On the estimationof parameters in latent structureanalysis. Psychometrika, 44:123-128,1979.Gur, R.E., and Gur, R.C. Genderdifferences in regional cerebralblood flow. Schizophrenia Bulletin,16:247-254, 1990.

Gur, R.E.; Gur, R.C; Skolnick, B.;Caroff, S.; Obrist, W.; Resnick, S.;and Reivich, M. Brain function inpsychiatric disorders: III. Regionalcerebral blood flow in unmedicatedschizophrenics. Archives of GeneralPsychiatry, 42:329-334, 1985.

Haas, G. "Gender and Neurocogni-tion in Schizophrenia." Presented atthe 140th Annual Meeting of theAmerican Psychiatric Association,Washington, DC, 1987.Hare, E. Temporal factors andtrends, including birth seasonalityand the viral hypothesis. In: Tsuang,M., and Simpson, ]., eds. Handbookof Schizophrenia: Vol. 3. Nosology,Epidemiology, and Genetics. Amster-dam: Elsevier Press, 1988.pp. 345-377.Kendler, K.S.; Gruenberg, A.M.;and Tsuang, M.T. Psychiatric illness

in first degree relatives ofschizophrenic and surgical controlpatients: A family study usingDSM-I1I criteria. Archives of GeneralPsychiatry, 42:770-779, 1985.

Kendler, K.S.; Tsuang, M.T.; andHays, P. Age at onset inschizophrenia: A familial perspec-tive. Archives of General Psychiatry,44:881-890, 1987.

Kopala, L., and Clark, C. Implica-tions of olfactory agnosia for under-standing sex differences in schizo-phrenia. Schizophrenia Bulletin,16:255-261, 1990.

Kraepelin, E. Dementia Praecox andParaphrenia. (1919) Translated byR.M. Barclay. New York: Robert E.Krieger Publishing Co., Inc., 1971.

Lazarsfeld, RE, and Henry, N.W.Latent Structure Analysis. Boston:Houghton Mifflin, 1968.

Lewine, R.R.J. Sex differences inschizophrenia: Timing or subtype?Psychological Bulletin, 90:432-444,1981.

Lewine, R.R.J. Gender and schizo-phrenia. In: Tsuang, MX, andSimpson, J.C., eds. Handbook ofschizophrenia: Vol. 3, Nosology,Epidemiology, and Genetics. Amster-dam: Elsevier Science Publishers,1988. pp. 379-397.

Lewine, R.R.J.; Gulley, L.R.; Risch,S.C; Jewart, R.; and Houpt, J.L.Sexual dimorphism, brain mor-phology, and schizophrenia.Schizophrenia Bulletin. 16:195-203,1990.

Lewis, S.W. Congenital risk factorsfor schizophrenia. PsychologicalMedicine. 19:5-13, 1989.

McCutcheon, A.L. Latent ClassAnalysis. Sage University Papers, Serieson Quantitative Applications in theSocial Sciences, No. 07-064. BeverlyHills, CA: Sage Publications, 1987.

VOL. 16, NO. 2, 1990 275

Mednick, S.A.; Schulsinger, E;Teasdale, T.W.; Schulsinger, H.;Venables, P.H.; and Rock, D.R.Schizophrenia in high risk children:Sex differences in predisposing fac-tors. In: Serban, G., ed. CognitiveDefects in the Development of MentalIllness. New York: Brunner/Mazel,1978. pp. 169-197.

Morrison, J.; Clancy, J.; Crowe, R.;and Winokur, G. The Iowa 500: I.Diagnostic validity in mania, depres-sion, and schizophrenia. Archives ofGeneral Psychiatry, 27:457-461, 1972.

Murray, R.M.; Owen, M.J.; Good-man, R.; and Lewis, S.W. A neuro-developmental perspective on someepiphenomena of schizophrenia.In: Cazzullo, C.L.; Invernizzi, G.;Sacchetti, E.; and Vita, A., eds.Plasticity and Morphology of theCentral Nervous System. Lancaster:MTP Press, 1989.

Nasrallah, H.A.; Andreasen, N.C.;Coffman, J.A.; Olson, S.C.; Dunn,V.; Ehrhardt, ].; and Chapman, S. Acontrolled magnetic resonance im-aging study of corpus callosumthickness in schizophrenia. BiologicalPsychology, 21:274-282, 1986.

Nasrallah, H.A.; Schwarzkopf, S.B.;Olson, S.C.; and Coffman, J.A.Gender differences in schizophreniaon MRI brain scans. SchizophreniaBulletin, 16:205-210, 1990.

Salokangas, R.J.R. Prognostic im-plications of the sex of schizophrenicpatients. Psychiatry, 142:145-151,1983.Seeman, M.V. Interaction of sex,age, and neuroleptic dose. Com-prehensive Psychiatry, 24:124-128,1985.Seeman, M.V., and Lang, M. Therole of estrogens in schizophreniagender differences. SchizophreniaBulletin, 16:185-194, 1990.

Seidman, L. Schizophrenia andbrain dysfunction: An integration ofrecent neurodiagnostic findings.Psychological Bulletin, 94:195-238,1983.Tsuang, M.T.; Dempsey, G.M.; andRauscher, F. A study of "atypicalschizophrenia": Comparison withschizophrenia and affective disorderby sex, age of admission, precipi-tant, outcome, and family history.Archives of General Psychiatry,33:1157-1160, 1976.

Tsuang, M.T., and Winokur, G. TheIowa-500: Field work in a 35-yearfollowup of depression, mania, andschizophrenia. Canadian PsychiatricAssociation journal, 20:359-365, 1975.Tsuang, M.T.; Woolson, R.F.; andSimpson, J.C. The Iowa StructuredPsychiatric Interview: Rationale,reliability, and validity. AdaPsychiatrica Scandinavica,62(Suppl.):283-287, 1980.Weinberger, D.R. Implications ofnormal brain development for thepathogenesis of schizophrenia.Archives of General Psychiatry,44:660-669, 1987.Young, M.A. Evaluating diagnosticcriteria: A latent class paradigm.Journal of Psychiatric Research,17:285-296, 1983.Young, M.A.; Scheftner, W.A.; Kler-man, G.L.; Andreasen, N.C.; andHirschfeld, R.M.A. The endogenoussubtype of depression: A study ofits internal construct validity. Britishjournal of Psychiatry, 148:257-267,1986.

Acknowledgments

This article was, in part, presentedat the International Congress onSchizophrenia Research, San Diego,CA, April 1989, and the AnnualMeeting of the American PsychiatricAssociation, San Francisco, CA, May1989. The research was supported by

a grant to Dr. Goldstein from theScottish Rite Foundation, Schizo-phrenia Research Program, andUSPHS grant MH-42604 from theNational Institute of Mental Healthto Drs. Tsuang and Goldstein. Theauthors thank Dr. Robin Murray forhis helpful comments on an earlierdraft of the manuscript.

The Authors

Jill M. Goldstein, Ph.D., is AssistantProfessor of Psychiatry, HarvardMedical School, Department ofPsychiatry, Massachusetts MentalHealth Center, Section on Psychi-atric Epidemiology and Genetics;and Psychiatry Service, Brockton-West Roxbury VA Medical Center.Susan L. Santangelo is ResearchAssociate, Harvard Medical School,Department of Psychiatry, Massa-chusetts Mental Health Center, Sec-tion on Psychiatric Epidemiologyand Genetics; and Psychiatry Serv-ice, Brockton-West Roxbury VAMedical Center. She is also apredoctoral NIMH fellow in Psychi-atric Epidemiology, Harvard Schoolof Public Health. John C. Simpson,Ph.D., is Instructor in Psychiatry,Harvard Medical School, Depart-ment of Psychiatry, MassachusettsMental Health Center, Section onPsychiatric Epidemiology andGenetics; and Psychiatry Service,Brockton-West Roxbury VA MedicalCenter. Ming T. Tsuang, M.D.,Ph.D., is Professor of Psychiatry andDirector of Psychiatric Epidemiology,Harvard Schools of Medicine andPublic Health; Chief of PsychiatricEpidemiology and Genetics, Har-vard Department of Psychiatry,Massachusetts Mental HealthCenter; and Chief of Psychiatry,Psychiatry Service, Brockton-WestRoxbury VA Medical Center.