can we subtype alcoholism? a latent class analysis of data from relatives of alcoholics in a...

0145-6008/96/2008-1462$03.00/0 ALCOHOUSM: CLINICAL AND EXPERIMENTAL RESEARCH

Vol. 20, No. 8 November 1996

Can We Subtype Alcoholism? A Latent Class Analysis of Data from Relatives of Alcoholics in a Multicenter

Family Study of Alcoholism Kathleen K. Bucholz, Andrew C. Heath, Theodore Reich, Victor M. Hesselbrock, John R. Krarner, John I. Nurnberger, Jr.,

and Marc A. Schuckit

We attempted to identify distinctive subtypes of alcoholics using latent class analysis with data from 2551 relatives of alcoholic probands, all participants in the Collaborative Study of the Genetics of Alcoholism. Latent class analysis is a multivariate technique using cross-classified data to identify unobserved (“latent”) classes that explain the relationships among observed variables. Data on 37 lifetime symptoms of alcohol dependence from 1360 female and 1191 male relatives were analyzed, with a 4 class solution selected as the best fitting among the 2 through 6 class solutions that were examined. We observed the following classes: class 1, nonproblem drinkers (39.6% male, 50% female); class 2, mild alcoholics (persistent desire to stop, tolerance, and blackouts) (31.8% male, 28.7% female); class 3, moderate alcoholics (social, health, and emotional problems) (18.9% male, 14.6% female); and class 4, severely alfected alcoholics (withdrawal, inability to stop drinking, craving, health, and emotional problems) (9.7% male, 6.7% female). There was little evidence for the construct of alcohol abuse; endorsement probabilities for abuse symptoms (e.g., arrests and DWIs) were very low for all classes, whereas hazardous use was common among men in class 1. In addition to those in class 3 and class 4, a majority of men in class 2 qualified for DSM-Ill-R alcohol dependence, suggesting a bimodal distribution of drinkers and alcoholics, with little nondependent problem drinking among men in this high-risk sample. We conclude

From the Department of Psychiatry (KKB. , A.C.H., T.R.), Washington University School of Medicine, St. Louis, Missouri; Department of Psychiatry (KM. H.), University of Connecticut School of Medicine, Farmington, Con- necticut; Department of Psychiatric Research (J.R.K), University of Iowa, Iowa City, Iowa; Department of Psychiatry (J.I.N.), Institute of Psychiatric Research, Indiana University Medical Center, Indianapolis, Indiana; and Department of Psychiatry (MAS . ) , Veterans Affairs Medical Center, Univer- sity of California, San Diego, California.

Received for publication April 15, 1996; accepted August 2, 1996 The Collaborative Study on the Genetics of Alcoholism (H. Begleiter, State

University of New York, Health Sciences Center at Brooklyn, Principal In- vestigator; T. Reich, Washington University, Co-Principal Investigator) includes six different centers where data collection takes place. The six sites and Principal Investigator and Co-Investigators are: Indiana University (J. Nurn- berger, Jr. and P. M. Conneally); University of Iowa (R. Crowe and S. Kuperman); University of California at San Diego and Scripps Institute (M. Schuckit and F. Bloom); University of Connecticut (K Hesselbrock); State University of New York, Health Sciences Center at Brooklyn (H. Begleiter and B. Poijesz); and Washington University in St. Louis (T. Reich and C. R Cloninger).

This national collaborative study is supported by the National Institute on Alcohol Abuse and Alcoholism ( N U ) through US. Public Health Service Grants NlAAA UIOAA08401, UIOAAO8402, and UIOAA08403.

Reprint requests: Kathleen K Bucholz, Ph.D., Department of Psychiatry, Washington University School of Medicine, 4625 Lindell Boulevard Suite 200, St. Louis, MO 63108-3729.

Copyright 0 1996 by The Research Society on Alcoholism.

1462

that, in this sample, alcoholism is not differentiated by symptom profiles but rather lies on a continuum of severity, with the possible exception of withdrawal, which characterized only class 4 individuals.

Key Words: Subtyping, Alcoholism, Latent Class Analysis, Family Studies.

UBTYPING ALCOHOLISM has been a long-standing S research interest in the alcohol field. An early re- searcher of alcoholic subtypes was Jellinek,’ who distinguished 6-alcoholism by an inability to abstain and y-alcoholism by loss of control over drinking. Cloninger,2 in his neurobiobehavioral theory of the etiology of alcoholism, drew on Jellinek’s work in his formulation of type 1 and type 2 alcoholics, wherein “loss of control” was a hallmark of type 1 and “inability to abstain” was a hallmark of type 2 alcoholism. Type 2 alcoholism was hypothesized to be more heritable, as well as to have the distinctive personality profile of high novelty-seeking, low harm avoidance, and low reward dependence. Babor et al.,3 in their specification of types A and B subtypes, extended the work of Cloninger to include other dimensions of psychopathology. In the Babor et al. formulation, type B resembles Cloninger’s type 2 in its characteristics of more childhood behavior problems, earlier onset of alcohol problems, and denser family history of alcoholism, compared with type A. However, efforts to replicate the subtypes proposed by Jellinek and others have been somewhat inconcl~sive.~-~ The question remains as to whether subtypes of alcoholism may be distinguished by specific symptom profiles as posited by Jellinek, Cloninger, and Babor et al. The alternative view- point to distinctive subtypes is that alcoholism is a disorder that exists on a spectrum of severity.’

An opportunity to investigate this question became available using alcohol symptom data from alcoholics and their relatives who participated in a National Institute on Alco- hol Abuse and Alcoholism-sponsored multicenter family study of the genetics of a l c o h o l i ~ m ~ ~ ~ Identification of subtypes of alcoholism would contribute to the principal genetic goals of the project, because discovery of specific genetic effects is more likely in homogeneous disease groups. The technique of latent class analysis (LCA)” was selected to determine whether subtypes of alcoholism

Alcohol Clin Exp Res, Vol20, No 8, 1996 pp 1462-1471

SUBTYPING ALCOHOLISM 1463

might be differentiated by alcohol-related symptom profiles in these data.

These analyses have four objectives: (1) to determine whether subtypes of alcoholism with distinctive symptom profiles could be discerned by using a full range of symptoms that reflect a wide range of diagnostic classification systems for alcohol dependence; (2) to interpret the classes in the context of alcohol nosology; (3) to examine the relationship of the classes to a diagnosis of alcohol dependence and to developmental milestones of alcohol dependence; and (4) to investigate the prevalence of other psychiatric disorders in each class to determine whether comorbidity distinguished some classes.

METHODS

Sample and Assessment

Data from the ongoing Collaborative Study on the Genetics of Alco- holism (COGA) were used for the study. COGA is a multistage study being conducted at university centers located in Farmington, CT; Brook- lyn, NY; Indianapolis, IN; St. Louis, MO; Iowa City, IA; and San Diego, CA. In stage I, alcoholic probands in treatment units who meet study ascertainment criteria, and their first-degree relatives are assessed with a comprehensive psychiatric interview. To qualify for stage I, probands must have met criteria for both DSM-111-R alcohol dependence” and Feighner definite alcoholism“; have two first-degree relatives living within 100 miles of the COGA centers; be free of nonalcohol-related, life-threatening illnesses: and neither have injected illicit substances within 6 months of admission nor have used them more than 30 times ever. Probands with two additional affected biological relatives are entered into stage 11, wherein all relatives of the proband (including second- and third-degree relatives) are sought not only for interview, but also for a biological protocol that includes drawing of blood for genotyping and for electrophysiological and neuropsychological testing. After completion of the stage I1 protocol, the genetically most informative families are selected for panels (stage 111) and genotyped (stage IV). The COGA sample also includes controls sampled from a variety of sources, such as dental clinic registrations and drivers’ license registries. Along with their first-degree relatives, control subjects are administered both the interview and biological protocol. In the analyses reported herein, only data from adult relatives of alcoholic probands (18 or older) in stage I to IV families were included. Details of the ascertainment criteria for stage I and stage I1 are available from the authors on request.

The assessment interview, the Semistructured Assessment for the Ge- netics of Alcoholism (SSAGA), was developed expressly for COGA and has been described in detail el~ewhere.’~,’~ The SSAGA is a comprehensive, reliable interview that elicits lifetime and current information for a range of psychiatric disorders, including alcohol dependence and abuse, dependence and abuse for five drug classes, major depression, dysthymia, mania, antisocial personality disorder (ASPD), and anxiety disorders. Several diagnostic systems are covered by the SSAGA, for alcohol dependence, these systems include DSM-III-R,l’ Feighner,” and ICD-10.15 Because DSM-IV criteria16 had not been published when COGA field work began, criteria being considered at the time for DSM-IV were used to formulate SSAGA questions. A comparison with the final DSM-IV criteria suggested that the final DSM-IV criteria were well approximated in the SSAGA.”

Several research interests guided the selection of symptoms used in the analyses. Some criterion items were divided; “persistent desire to quit or cut down on drinking” was separated from “Unsuccessful efforts to stop,” and “Adverse physical consequences” was separated from “Adverse emotional consequences.” These components of criteria that are part of several classification systems had been observed in our previous analyses” to have markedly different symptom endorsement probabilities, prompt-

ing us to treat them as separate items in this analysis. The contribution of “recurrent blackouts” to class formation was also of interest. Omitted from the DSM-111-R criteria, blackouts were explicitly included in DSM-IV as an adverse physical consequence of drinking.I6 Table 1 displays a list of symptoms included in the analysis, along with the diagnostic system each addresses and the prevalence of each by gender. All symptoms were dichotomous (yes/no) and reflected lifetime occurrence.

Data for the present analyses were from master file 26 and included all relatives aged 18 or older in stage I to IV families. Analyses were limited to those who completed the alcohol section, thereby excluding individuals with minimal exposure to alcohol, defined in the COGA project as never in their lifetime having had more than 3 drinks in a 24-hr period, or never in their lifetime having been intoxicated and never drinking at least once a month for at least 6 months. Using these criteria, 1576 individuals (38% of all relatives of alcoholic probands) were excluded from the present analyses. Of the 2551 relatives remaining for the LCAs, 1360 were female and 1191 were male.

LCA

LCA’” may be understood intuitively as a categorical form of factor analysis applied to n-way contingency tables. It is based on the assumption that the frequencies with which different symptom profiles occur in a dataset can be explained by the existence of a small number of mutually exclusive classes or subtypes, rn, with each class having a distinctive “profile” of item endorsement probabilities that is constant for all members of that particular class. A critical implication of this assumption is that, within a class, the probabilities of endorsing different symptoms are statistically independent. For a given latent class model, parameter estimates include class membership probabilities, which may be thought of as prevalence, and symptom endorsement probabilities (SEPs), which reflect the likelihood that a symptom is endorsed by an individual, given membership in that class. Classes may be characterized by SEPs, which in some ways are similar to factor loadings produced from factor analysis. The goodness-of-fit of models estimating varying numbers of latent classes may be compared by a likelihood-ratio x 2 test.

Latent class models were fitted to 36 symptoms of alcoholism plus a treatment variable by the method of maximum likelihood using a general- purpose LCA program developed by Eaves et al.19 In this study, absolute fit of the model was of less interest than relative fit, because a base model was used that did not attempt to model the causes of family resemblance. In effect, parameters were estimated under the hypothesis of no family resemblance for alcoholism class membership. Although this assumption is implausible, it was a necessary starting point for future research on modeling causes of family likeness. Furthermore, these analyses provide a baseline against which latent class models incorporating information about mode of inheritance may be compared. It also will permit eventual comparison of our results to those obtained by other investigators using samples of unrelated individuals. With this in mind, the emphasis on relative fit by using likelihood ratio comparisons was appropriate to choose among models that estimated different numbers of classes.

Separate solutions were identified for males and females. For analysis of the alcoholism milestone and comorbid psychiatric disorder data, individuals were assigned to the “most likely” class (i.e., the class for which the conditional probability of membership in this class was greatest). Although some have argued that statistics computed using this approach will be biased, because they ignore the fact that conditional probabilities for membership in other classes are greater than 0,” these problems are minimal when working with very large sample sizes, such as those in this study. An attraction of using this simplified approach, which has been standard in most latent class applications, is that it generalizes easily to the analysis of longitudinal data and to the analysis of familial similarity for class membership.

1464 BUCHOLZ ET AL

Table 1. Diagnostic Symptoms Used for LCA

Diagnostic symptom System

Prevalence

Females Males

Wanted to cut dowdstop 3 or more times Unsuccessful efforts at stopping or cutting down Morning drinking Craving Binge drinking and neglecting responsibility Drank more than intended Got drunk when promised would not Great deal of time drinkinghcovering from drinking Regularity of drinkinghanowing of repertoire Family problems because of drinking-3+ Lost friends because of dr inking4 + WorWschool problems because of drinking-3+ Fighting when drinking Tolerance-50% increase Gave up activities to drink Drinking interfered with responsibilities Drinking caused marital problems Thought self excessive drinker Felt guilty about drinking DWI’S/car accident when drinking-3+ Arrests while drinking-3+ Serious accidents when drinking++ Hazardous us&+ Blackouts-3+ 3R Withdrawal syndrome (shakes plus one other) Relief drinking-3+ Seizures-any Relief drinking for seizures DT’s-any Drank to relieve DT’s Continue to drink despite alcohol-caused health problen Continue to drink with serious illness Drank when on medications dangerous to mix with

Continued to drink despite psychological problems Treatment for drinking Drank nonbeverage alcohol Made rules to control drinking

alcohol

3R, 4, ICD 3R, 4, ICD F ICD 3R, 4, ICD, F 3R, 4, ICD 3R, 4, ICD 3R, 4 ICD’ 3R, F t 3R, F t 3R. F t F 3R 3R, 4, ICD 3R 3R F F 3R, F F 3R, 4, ICD 3R F, 4 3R 3R, 4, ICD F 3R, 4, ICD 3R, 4. ICD 3R. 4, ICD

ns 3R, 4, ICD 3R, 4, ICD 3R, 4

3R, 4, ICD - F F

32.1 11.6 13.8 9.9 6.3

41.8 29.2 13.5 7.8

17.1 2.9 7.1 7.1

29.3 11.3 13.4 11.2 27.5 33.2

0.7 0.9 4.2

35.1 22.1 5.7 7.6 0.6 0.2

20.2 1.6 6.1 4.9

13.3

11.3 10.1

1.7 13.5

47.2 19.1 38.4 14.9 17.1 58.8 38.8 22.9 19.0 34.3 8.4

17.3 23.3 53.1 21.6 23.2 25.5 44.8 40.3

6.7 11 .o 67.6 36.8 10.7 14.1

1.6 0.7 3.9 3.1

13.1 6.1

23.6

19.1 20.6 4.2

23.8

8.1

~~ ~~ ~~ ~~ ~~

Note: 3R, DSM-Ill-R; 4. DSM-IV; F, Feighner; ICD, ICD-10; DWI, driving while intoxicated; DT’s. delirium tremens. No longer required for ICD.

t Only once for Feighner.

RESULTS

Results of Fitting Latent Class Models Significant improvements in fit were obtained for 2-class

through 6-class models. Table 2 summarizes the results of model-fitting. Although there were significant improvements in fit with each higher class solution, the 5-class and 6-class solutions yielded some classes with very low prevalence, suggesting they were too small to be meaningful. The 4-class solution was selected as having an acceptable fit for males and females, and was chosen as the final model.

4-Class Solution Figure l a (females) and Figure l b (males) display the

class membership probabilities along with a graph of SEPs by class. Appendices A and B contain the exact SEP’s by class for females and males. As noted earlier, SEPs reflect the likelihood that an individual will be positive for a particular symptom, given membership in that class. Class prevalence estimates ranged from 6.7 to 50% for females

and from 9.7 to 39.6% for males. For both, the observed pattern was that each successive class had endorsement probabilities higher than the previous class, with class 4 representing the most affected class. Class 4 members included 6.7% of female relatives and 9.7% of male relatives. Symptom composition of the classes was generally similar for men and women. For these analyses, confidence intervals were constructed around each SEP using a bootstrap- ping technique, and nonoverlapping confidence intervals were taken to indicate distinctive SEPs.

Class I (filled square symbol) was composed of individuals with very low SEPs across most items. SEPs were rarely greater than 0.2; those items with the greatest endorsement probabilities included: “Persistent desire to stop drinking,” “Drinking more than intended,” “Feeling guilty about drinking,” and “Hazardous use.” The highest SEP of 0.4 for “hazardous use” was observed only among males. These results suggested a group of unaffected individuals, with few alcohol problems.

In Class 2 (filled circle symbol), items with moderate

SUBTYPING ALCOHOLISM

Table 2 Model Fitting

1465

Females Males

Likelihood ratio Likelihood ratio

Model -2 log-likelihood (df = 38)’ P -2 log-likelihood (df=%r P J test

2 class 24244.70 31064.54 3 class 22445.78 1798.92 0.000 28927.10 2137.44 0.000 4 class 21928.96 516.82 <0.001 28364.47 562.63 0.000 5 class 21 750.03 178.93 <0.001 28048.53 315.97 10.001 6 class 21681.73 68.30 0.001 8 27914.36 134.17 <0.001

Class prevalence

Females Males

Model 1 2 3 4 5 6 1 2 3 4 5 6 - - - - 2 class 77.5 22.5 - - - - 66.6 33.4

3 class 67.2 24.4 8.3 - - - 48.2 35.0 16.8 4 class 50.0 28.7 14.6 6.7 - - 39.6 31.8 18.9 9.7

2.4 - 34.4 29.2 21.2 11.7 3.5 - 5 class 44.9 29.2 16.8 6.6 6 class 47.9 29.2 15.4 4.1 2.2 1.3 34.9 29.4 19.7 2.9 9.9 3.1

- - - - -

* Degrees of freedom, calculated for the relative fit of the model with n + 1 classes over then class model, number 38: 37 for the parameters for the SEPs for the additional class, plus the parameter for the additional class membership being estimated.

a Wanted to stop Unable to stop

Morning drinking Craving

Binged 6 neglected resp Drank more than intended

Drunk when didn‘t want Spent time drinklng

Nanowing of reperloire Prob. w/famiiy &friends

Lost friends Problems at worklJohool

3+ hphls Tolerance

Gave up adiv. to drink Interfered w/responsibi~.

Marital problems Thoupht self excessive dr

Felt guilty DWI

AW3sted Accidental injuries 3+ hazardous use

3+ blackouts Withdrawal syndrome

Relief drinking seizures

Drank to stop seizures DTs

Drank to stop OTs Heaith prob from drinking Drank despite health prob

Mixed aIc 6 medication Drank despite emotni prob

Treatment for drinking Drank nonbeverage sic

Made rules to control

0 0.2 0.4 0.6 0.8 1 +Class 1 - 50.0% +Class 2 - 28.7% &Class 3 - 14.6% *Class 4 - 6.7%

b Wanted to stop Unable to stop

Morning drinking Craving

Binged 6 nspleded rev. Drank more than intended

Drunk when t i in? want Spent time drinking

Nanowing of repeltoire Prob. w/ family6friends

LO* friends Problems at wor*lschoal

3* fQMS Tolsrance

Gave up adivities to drink Interfered w/responsibil

Malilai problams ThougM self excessive dr

Felt guilty DWI

Arrested

3+ hazardous use 3+ blaCkovts

Relief drinking seuures

Drank to stop seizures DTs

Drenk to stop DTS Health pmb from drinkma Drank despite health prob

Mlxed sic 6 medication Drank despite ematnl pmb

Treatment for drinking Drank nonbeverage atc

Made rules to control

0 0.2 0.4 0.6 0.8 I

+Class 1 - 39.6% +Class 2 -31.8% +Class 3 - 18.9% *Class 4 - 9.7%

Fig. 1. SEPs by class membership: (a) females and (b) males. Abbreviations used are: resp, responsibility: prob, problem; activ, activities: responsibil, responsibilities: dr, drinker: DWI. driving while intoxicated: DT’s, delirium tremens: ab, alcohol: emotnl, emotional.

SEPs reflected loss of control over drinking and drinking excessively. There were, however, few adverse social, psychological, or physical consequences that achieved high SEPs in this class. SEPs were principally below 0.2 for most items, but a few items reached 0.4 for both men and

women. These were: “Persistent desire to stop drinking,” “Drinking more than intended,” “Becoming drunk when didn’t want to,” “Tolerance,” “Thought self an excessive drinker,” “Feeling guilty about drinking,” and “Hazardous use.” In addition, “Recurrent blackouts,” with a SEP of

BUCHOU ET AL. 1466

100

80

60

40

20

0 DSM-3R-Females DSM-3R- Males DSM-4 - Females DSM-4 - Males

BClass 1 Fig. 2. Percentage meeting criteria for

DSM-III-R and DSM-IV alcohol dependence by class and gender. n for classes 1 to 4:

474, 380,219, and 118. females = 695, 382, 191, and 92; males =

0.37, characterized men in this class, but not women (SEP of 0.21). SEPs for “Tolerance” and “Hazardous use” for men were significantly higher than those for women in this class.

Class 3 (filled triangle symbol) included all items identified herein for class 2, but with much higher SEPs (most in excess of 0.6). In addition, social and emotional problems caused by drinking received high SEPs in class 3, with SEPs ranging from 0.3 to 0.8. The following items for class 3 had SEPs equal to or in excess of 0.4, compared with class 2: “Great amount of time spent drinking,’’ “Family problems,” “Gave up activities to drink,” “Drinking interfered with responsibilities,” “Marital problems,” “Drinking when taking medication,” “Drank despite emotional problems caused by drinking,” and “Making rules to control drinking.?’ In addition, the SEPs for treatment for drinking problems was significantly higher for men and women in class 3 than it was for individuals in class 1 or class 2.

Although a comparison of results for men and women in class 3 suggested numerous similarities, some differences were apparent. The SEPs for “Morning drinking” was significantly lower (0.30) for women, compared with their male counterparts (0.67). Other items where the difference in SEPs for males and females were different included: “Narrowing of the drinking repertoiren (0.42 vs. 0.17), “Family problems” (0.75 vs. 0.53), “Recurrent fighting while drinking” (0.40 vs. 0.20), “Tolerance” (0.86 vs. 0.66), “Marital problems” (0.58 vs. 0.36), and “Hazardous Use” (0.88 vs. 0.68).

In Class 4 (open sunburst symbol), the most severely affected group, the SEPs for most items exceeded 0.6, and in many instances 0.8, for both men and women. Items in class 4 included all those in class 3, plus some items that did not distinguish class 3. Items like “Craving” and “Narrow- ing of the repertoire” (which has been eliminated from ICD-10 classification) distinguished this class from the

other three classes, as did “Withdrawal syndrome,” “Relief drinking,” “Drinking despite health problems,” and “Drinking despite emotional problems.” Despite the substantial affectedness of this class, SEPs for “Recurrent DWI’s” and “Arrests” were still comparatively low (well under 0.1) for women, and even under 0.4 for men. Class 4 alone had a substantial SEP for “Withdrawal syndrome.” This item was absent from the other three classes for both men and women. In this class, SEPs for men were still markedly higher, compared with women for “Morning drinking” (0.92 vs. 0.75), “Recurrent fighting while drinking” (0.66 vs. 0.46), “Work problems” (0.82 vs. 0.53), “DWI’s” (0.33 vs. 0.08), “Arrests” (0.30 vs. 0.08), and “Haz- ardous use” (0.95 vs. 0.82).

Relationship to DSM-III-R and DSM-IV Alcohol Dependence

Figure 2 displays by gender the proportion of those within each of the four classes who met lifetime criteria separately for DSM-III-R and for DSM-IV alcohol dependence. All men in class 3 and class 4 met DSM-III-R criteria, as did a substantial proportion in class 2 (a similar observation held for DSM-IV alcohol dependence). For both systems, a vanishingly small percentage of men in class 1 met DSM-III-R dependence criteria. Thus, among men, class 3 and class 4 were composed primarily of individuals who qualified for a lifetime diagnosis of DSM-III-R and DSM-IV alcohol dependence; whereas with class 2 individuals, 84% met DSM-III-R and nearly half met DSM-IV criteria. There did not seem to be a substantial group of problem drinkers who did not meet criteria for alcohol dependence among men.

Among female relatives, nearly all in class 3 and class 4 met lifetime criteria for DSM-III-R and DSM-IV alcohol dependence. Unlike data for men, there did seem to be a


Table 3. Milestones of Alcohol Dependence

Females

Class 1 (n = 695) Class 2 (n = 382) Class 3 (n = 191) Class 4 (n = 92) D (overall)

Age at interview 39.98.” 3 5 9 36.1” 38.3 0.0001 Age first drank regularly 21.5a,”*c 19.6a*d 18.9” 17.7C.d 0.0001 Age first intoxicated 19.8a,b,c 1 7.8a 17.1” 1 6.2c 0.0001 Age first drinking problem 22 .4a,b,c 19.5’ 19.5” 17.8’ 0.0001 Age problems clustered 26.6 24.5 25.7 23.5 0.2307

Males

Class 1 (n = 474) Class 2 (n = 380) Class 3 (n = 219) Class 4 (n = 1 1 8) D (overall) ~~~ ~~~~

Age at intelview 41 .4a 38.3a 38.8 39.7 0.0198 Age first drank regularly 19.48,b,c 1 8.2a,d,f 17.1”~”’ 15.4”d.e 0.0001 Age first intoxicated 18.1a,b#c 16.3a,d 15.4b,e 13.7c- 0.0001 Age first drinking problem 20.5a,b,c 18.08.d 17.763 15.9c,d,e 0.0001 Age problems clustered 23.6 24.2 24.3 22.3 0.138

Note: Same letter indicates significantly different (p < 0.05) (adjusted for multiple comparisons).

group of nondependent problem drinkers among the women in class 2, wherein 59% and 84% did not meet either DSM-111-R or DSM-IV criteria for alcohol dependence, respectively.

Interpreting the DSM-111-R data in terms of sensitivity and specificity yielded the following. If class 1 and class 2 were considered as unaffected, sensitivity would be 64% for women and 49% for men. Corresponding figures for specificity were 99% and 100% for females and males, respectively. However, if subjects in class 1 only were considered unaffected, sensitivity would increase to 100% for women and 96% for men, whereas specificity would drop to 75% for women and 88% for men. The analogous figures for the DSM-IV diagnosis of alcohol dependence were: sensitivity, 80% and 64.2%; specificity of 98% and 97% for scenario 1 for females and males, respectively. For scenario 2, sensitivity was 100% and 99%, and specificity was 67% and 68% for females and males, respectively.

Age at Interview and Milestones of Alcohol Dependence Table 3 displays age at interview and milestones of alco-

hol dependence by each class for men and women. Com- parison of age at interview among the four classes revealed that women in class 1 were significantly older than those in class 2 and class 3, but men in class 1 were significantly older than men in class 2 only. A significant association was observed between age and class for both men (x2 = 30.29, 3 df ,p < 0.001) and women (x2 = 33.41, 3 df ,p < 0.001). In women, the relationship was explained by a dispropor- tionately high percentage of older women in class 1; the proportion of young women aged 18 to 22 was relatively constant (-10%) across all four classes. In contrast, for men, a higher percentage of young men aged 18 to 22 was observed in the unaffected and mildly alcoholic class 1 and class 2, compared with the more severely affected class 3 and class 4 (10% and 12%, compared with 6% and 3%, respectively). Still, 90% or more of men in all four classes had passed through the high risk age for alcohol abuse. In terms of milestones, women in class 4 had significantly

earlier ages of regular drinking and of first intoxication, compared with class 1 and class 2. Class 3 and class 4 had similar ages for those milestones. In terms of age of first drinking problem, the only significant difference observed was that the first drinking problem of those in class 1 (a relatively rare occurrence) occurred significantly later than in the other classes. In men, age of first occurrence of all milestones was significantly earlier among men in class 4, compared with men in the other classes.

Psychiatric Comorbidity The prevalence of other psychiatric disorders in each

class was investigated. The expectation was that prevalence of other disorders would increase with class severity. Using the Mantel-Haenszel test for linear trend:’ this hypothesis was tested. As can be seen in Table 4, this was borne out for all disorders but depression for females, and for all disorders except panic, depression, and agoraphobia for men. Prevalence of specific drug dependence was calculated among users of that substance, thus appropriately taking into account only those who would be eligible for such a diagnosis by virtue of their exposure to a drug class.

When data were examined with respect to male to female ratios, a decrease in these ratios from class 1 to class 4 was observed for ASPD, marijuana dependence, and opiate dependence. For depression, where the prevalence in females exceeded that in men, the ratios approached 1 in class 4 (i.e., the disparity in the male and female prevalence became less among those severely alcoholic). Furthermore, post-hoc tests by gender revealed that males had significantly less depression than their female counterparts for classes 1 to 3 (all p values < O.OOOl), but males in class 4 had the same rate of depression as females in class 4 (p = 0.44). Post-hoc tests also revealed that males in class 4 had significantly higher rates of depression than their counterparts in class 2 ( p = 0.03) and class 3 ( p = 0.02), and there was a trend in that direction for the comparison between class 4 and class 1 ( p = 0.06).

BUCHOLZ ET AL 1468

Table 4. Lifetime Prevalence of Psychiatric Disorder by Class

Diagnosis Class I Class 2 Class 3 Class 4 P* Females

ASPD Child conduct disorder (3+ behaviors) Cocaine dependencet Marijuana dependencet Opioid dependencet Sedative dependencet Stimulant dependencet Major depression Obsessive-compulsive Panic Social phobia Agoraphobia

ASPD Child conduct disorder (3+ behaviors) Cocaine dependence? Marijuana dependencet Opioid dependencet Sedative dependencet Stimulant dependencet Major depression Obsessive-compulsive Panic Social phobia

Males

0.9 2.5

19.7 6.1 8.3

13.6 11.1 22.8

1 .o 4.0 2.5 1 .o

4.4 12.0 18.0 14.2 12.1 0 9.4

10.6 0.6 0.9 0.6

1.3 5.8

21.9 14.3 5.6

10.0 23.8 25.6

1.8 5.8 3.7 1.3

11.6 23.4 28.2 27.7 11.9 10.1 19.0 9.5 0.3 2.9 0.8

4.7 10.5 39.3 24.5 9.5 4.7

25.0 25.0

1 .I 5.8 3.2 3.2

20.7 33.8 45.6 53.4 26.4 16.9 28.9 9.7 0.5 0.9 3.3

15.2 19.6 64.9 50.0 43.2 43.1 52.0 22.8

5.5 12.1 8.8 4.4

35.6 44.9 60.8 54.5 46.3 33.3 42.2 18.0 4.3 3.4 5.2

<o.ooo <o.ooo <0.001 10.001 <0.001 <0.001 10.001

0.70 0.01 8 0.004 0.010 0.005

<o.ooo <o.ooo <0.001 <0.001 <0.001 <0.001 <0.001

0.08 10.008

0.08 0.0002

Agoraphobia 1.1 0.8 1.9 2.6 0.162

’ Mantel-Haenszel test for linear trend. t Among users only.

DISCUSSION

This application of LCA to alcoholism symptom data from relatives of alcoholic probands was undertaken to inform the debate about subtypes of alcoholism; whether, as some have suggested, alcoholism lies on a spectrum of severity, with groups distinguished not by type but by number and severity of symptoms, or whether independent subtypes exist distinguished by particular symptoms. These data provide evidence in favor of the former view. We did not observe any patterns that suggested certain symptoms were present in one class but not in others. Rather, almost all symptoms occurred with increasing likelihood (as reflected in the SEPs) with each successive class. This is expected if alcoholism is conceptualized to be on a spectrum of severity.

The possible exception to the interpretation herein is the withdrawal state, which we view as encompassing both the withdrawal syndrome and relief drinking. The likelihoods of endorsing the withdrawal syndrome and relief drinking were small in classes other than class 4. Therefore, withdrawal might define a subtype of alcoholism. This notion has received some attention in nosology. For example, in the DSM-I11 classification system, alcohol abuse and dependence were not mutually exclusive (as in DSM-III-R and DSM-IV). Evidence of tolerance or withdrawal was required for the diagnosis of dependence. In DSM-IV, one may specify a subtype “with physiological dependence” (i.e., with either tolerance or withdrawal). Our data would tend to support a differentiation of a subtype of alcoholism based on withdrawal, but not tolerance.

Our findings should not be interpreted as modeling

stages of development of alcohol dependence. As the age analyses suggest, by the time of COGA assessment, all relatives were well into the period of risk for developing alcohol dependence with mean age in the late thirties, so that most of those who would become alcoholic have al- ready done so.

Relationship to Alcohol Dependence Criteria Analysis of the latent classes in light of a diagnosis of

DSM-III-R and DSM-IV alcohol dependence revealed a strong association. As expected, all or nearly all members of class 3 and class 4 met criteria for alcohol dependence. This was true for both men and women. What was not expected was the very high percentage (84%) of men in class 2 who also met criteria for DSM-III-R alcohol dependence. We were surprised not to observe a class of nondependent problem drinkers among the male relatives of alcoholic probands, although we did observe such a group among the female relatives. We speculate that the nature of our sample as one at high risk for alcohol dependence leads to this sort of bimodal distribution, wherein a stage of problem, nondependent drinking is brief in the male relatives. However, all of these data are cross-sectional. It will be interesting to examine these findings in light of new data from the 4-year follow-up study, which will be underway soon.

Nosological Implications In terms of nosological implications, we observed that,

for all classes except class 4, SEPs for “Persistent desire to


stop drinking” were much higher than for “Unsuccessful efforts to stop drinking.” However, each addresses the same diagnostic criterion item in DSM-111-R, DSM-IV, and ICD-10 systems. A similar observation was made for “Ad- verse health consequences from drinking,” which was endorsed to a substantial degree only among those in class 4. This item is paired in the DSM and ICD systems with “Adverse emotional consequences from drinking,” which in our data was found to have a similar SEP in class 3 and class 4. The present findings suggest that, in future nosological systems, these items should be uncoupled.

We were interested in the inclusion in DSM-IV of “Re- current blackouts” as one operationalization of the criterion “Adverse physical health consequences.” Although not incorporated into the summary criteria for DSM-IV substance dependence, ‘‘blackouts” were specified as an illus- tration of adverse health consequences in the accompany- ing text in DSM-IV elaborating upon the criteria.16 This was a change from the DSM-111-R system. However, “blackouts” were included in the DSM-I11 classification system as an indicator of criterion A, “Pathological use.”” The LCAs suggested that the SEP for “Recurrent blackouts” markedly exceeded those for its so-called parent criterion of “Adverse health consequences.” “Adverse health consequences from drinking” was rare in class 1 and class 2; in contrast, “Recurrent Blackouts” was relatively common in class 2 for both males and females. These data suggest that including “Recurrent blackouts” as part of the criterion of “Adverse physical consequences” in the DSM-IV may weaken this otherwise severe criterion.

Comorbidity In terms of psychiatric comorbidity, there was evidence

of a linear trend in prevalence [i.e., those in the most severe class (class 4) had the highest lifetime prevalence of an additional psychiatric disorder]. This was especially true for other substance use disorders. We limited our prevalence calculations to individuals who had used a particular class of substance more than casually (i.e., more than 10 times).

In terms of nonsubstance disorders, we observed a similar trend for ASPD and panic disorder for women, and for ASPD and social phobia for men. We did not observe a relationship with major depressive disorder for women, wherein about one-quarter of each class qualified for a DSM-111-R diagnosis of major depression. For these analyses, episodes caused by alcohol or other substances were not included in the diagnosis.

Results were somewhat different for men. There was an increase in major depression among men in class 4, with 17.4% of men in this class qualifying for a diagnosis, compared with -10% in classes 1 to 3. Post-hoc tests of this association were significant, suggesting that the prevalence of depression in men in class 4 was significantly higher than for the other three classes. Furthermore, the ma1e:female ratio for major depression was close to 1 (0.8) among those

in class 4, whereas the ratios were well under 0.4 for those in classes 1 to 3 (meaning a deficit of depression in men). As was true for ASPD and some substance dependence disorders, the excess of disorder in males was much less pronounced in class 4, wherein the rates in women were similar to those of men. This was not so in other classes. This effect has been reported elsewhere using as a severity measure the requirement that individuals meet criteria for several diagnostic definition^.^^ In general, severity of alcoholism seems to have an equalizing effect on comorbid conditions among men and women.

We examined lifetime prevalence of ASPD in the sample as a whole, not counting behaviors that were exclusively caused by being under the influence of alcohol or other substances. We observed a marked excess of ASPD as class severity increased. A second method of analysis examined the subset of the sample with three or more childhood conduct problems before age 15. We were interested in knowing whether the excess in observed rates of ASPD reflected the proportion of the sample with three or more childhood conduct problems. Even among those who are eligible for the diagnosis of ASPD by virtue of having at least three conduct problems, those in class 4 still had significantly higher rates of ASPD than their equally eligible counterparts. It may be that alcoholism, which we have shown to start at a significantly earlier age in class 4, makes it more likely for some people to engage in ASPD behaviors when they are under the influence of alcohol and may also lead to life-styles wherein antisocial behaviors are common. Unless carefully screened, these behaviors may mimic nonalcohol-related sociopathy. This relationship will be more fully examined in future analyses.

This study is not without its limitations. The sample may not be representative of all relatives of alcoholics or of all persons with significant alcohol problems. All alcoholic probands in the COGA study met stringent diagnostic, geographic location, comorbid medical illness, and family size criteria. These sampling requirements are appropriate for a genetic study of alcoholism, but do not necessarily ensure selecting a representative sample of alcoholics in treatment or from the general population. In fact, only -6% of those screened in all sites’ treatment facilities were accepted for stage I of the Furthermore, the symptom data are self-report only, with no collateral verification of symptoms used in these analyses. We have treated the data in our analyses as independent observations, although given the familial nature of alcohol dependence, this is not the case. Still, we believe that this analysis will contribute to the debate on whether alcoholism is composed of subtypes or is on a continuum of severity. The findings of the present study support the latter view.

ACKNOWLEDGMENTS

We gratefully acknowledge the expert manuscript preparation by Janet Konrad, and programming assistance from Michael

1470 BUCHOU ET AL

Appendix A SEPs for 4-Class Solution: Females

Class 1 Class 2 Class 3 Class 4

Wanted to stop 3 or more times 0.0564 0.3762 0.8553 0.8886 Unable to stop 0.001 3 0.0322 0.3686 0.7843 Morning drinking 0.0160 0.1214 0.3055 0.7472 Craving 0.0067 0.0314 0.2133 0.8221 Binges 0.0001 0.0029 0.1259 0.6580

Drunk when did not want to be-repeatedly 0.0414 0.3793 0.701 1 0.8943 Great amount of time spent drinking 0.0122 0.0442 0.3789 0.9036 Narrowing of repertoire 0.0001 0.0218 0.1683 0.7040 Drinking caused recurrent problems with family or friends 0.0044 0.1041 0.5289 0.9157 Lost friends because of drinking 0.0001 0.0001 0.01 61 0.3935 Drinking caused recurrent problems at work or school 0.0001 0.0245 0.1960 0.5328 Fights when drinking-repeatedly 0.001 3 0.0346 0.1 993 0.4616 Tolerance 0.0427 0.3954 0.6553 0.9268 Reduced activities to drink 0.0030 0.01 91 0.3424 0.8304 Drinking interfered with responsibilities 0.0043 0.0462 0.4015 0.8936 Marital problems because of drinking 0.0019 0.0318 0.3622 0.7296 Thought was excessive drinker 0.0232 0.3249 0.7715 0.8584 Felt guilty about drinking 0.0809 0.4212 0.7669 0.8697 DWI 0.0001 0.0050 0.0001 0.0769 Arrested because of drinking 0.0014 0.0001 0.0201 0.0769 Accidental injury when drinking 0.0018 0.0060 0.1 089 0.3498 Hazardous use 0.1248 0.4662 0.6826 0.8181 Recurrent blackouts 0.01 64 0.2130 0.631 0 0.8830 Withdrawal syndrome 0.0001 0.0049 0.0696 0.6732 Relief drinking 0.0012 0.0106 0.1323 0.7991 Seizures 0.001 5 0.0001 0.0001 0.0769 Drank to stop seizures 0.0001 0.0001 0.0001 0.0329 DT's 0.0001 0.0001 0.0050 0.31 86 Drank to stop DT's 0.0001 0.0001 0.0001 0.2417 Health problems caused by drinking 0.0001 0.0090 0.1400 0.5676 Continued to drink despite health problems 0.0057 0.0508 0.0748 0.3120 Mixed alcohol and medication 0.0221 0.1 128 0.3057 0.6722 Continued to drink despite emotional problems 0.0001 0.0312 0.3070 0.8888 Treated for drinking problem 0.0022 0.0244 0.2797 0.7743

Drank more than intended-repeatedly 0.0916 0.6446 0.8249 0.9999

Drank nonbeverage alcohol 0.0029 0.0104 0.0001 0.1 868 Made rules to control drinking 0.0053 0.1468 0.3093 0.6662

DWI, driving while intoxicated; DT's, delirium tremens.

Hodge and Doris McGartland Rubio. In addition, this manuscript has benefitted from the careful review of Dr. T.-K. Li.

REFERENCES 1. Jellinek EM: The Disease Concept of Alcoholism. New Brunswick

N.J., New College and University Press and Hillhouse Press, 1960. 2. Cloninger CR: Neurogenetic adaptive mechanisms in alcoholism.

Science 236:410-416, 1987 3. Babor TF, Hofmann M, DelBoca FK, Hesselbrock V, Meyer RE,

Dolinsky ZS, Rounsaville B: Types of alcoholics. I. Evidence for an empirically derived typology based on indicators of vulnerability and severity. Arch Gen PsychiatIy 49:599-608, 1992

4. Schuckit MA, Anthenelli R, Bucholz K, Hesselbrock VM, Tipp J: The time course of development of alcohol problems in men and women.

5. Schuckit MA, Tipp J, Smith TL, Shapiro E, Hesselbrock VM, Bucholz KK, Reich T, Nurnberger JI: An evaluation of type A and B alcoholics. Addiction 901189-1203, 1995

6. Pokorny AD, Kanas T, Overall JE: Order of appearance of alcoholic symptoms. Alcohol Clin Exp Res 5:216-220, 1981

7. Park P: Developmental ordering of experiences in alcoholism. Q J

8. Delboca F, Hesselbrock MN: Types of alcoholics: A reanalysis of

9. Skinner H A Spectrum of drinkers and intervention opportunities.

10. McCutcheon AL: Latent Class Analysis. Newbury Park, CA, SAGE

J Stud Alcohol 56~218-225, 1995

Stud Alcohol 34:473-488, 1973

risk, severity and gender. Alcohol Clin Exp Res 19(Suppl.):56A, 1995

Can Med Assoc J 143:1054-1059, 1990

Publications, 1987

11. American Psychiatric Association: Diagnostic and Statistical Man- ual of Mental Disorders, ed 3-rev. Washington, D.C., APA, 1987

12. Feighner JP, Robins E, Guze SB, Woodruff RA, Winokur G, Munoz R: Diagnostic criteria for use in psychiatric research. Arch Gen Psychiatry 26:57-63, 1972

13. Bucholz KK, Cadoret R, Cloninger CR, Dinwiddie SH, Hessel- brock VM, Nurnberger JI, Reich T, Schmidt I, Schuckit MA: A new, semi-structured psychiatric interview for use in genetic linkage studies: A report of the reliability of the SSAGA. J Stud Alcohol 55:149-158, 1994

14. Bucholz KK, Hesselbrock VM, Shayka JJ, Nurnberger JI Jr, Schuc- kit MA, Schmidt I, Reich T Reliability of individual diagnostic criterion items for psychoactive substance dependence and the impact on diagnosis.

15. WHO. The ICD-10 Classification of Mental and Behavioural Dis- orders: Diagnostic Criteria for Research. Geneva, World Health Organi- zation, 1993

16. American Psychiatric Association: Diagnostic and Statistical Man- ual of Mental Disorders, ed 4. Washington, D.C., APA, 1994

17. Schuckit MA, Hesselbrock V, Tipp J, Anthenelli R, Bucholz K, Radziminski S: A comparison of DSM-111-R, DSM-IV and ICD-10 substance use disorders in 1922 men and women subjects in the COGA study. Addiction 89:1629-1638, 1994

18. Heath AC, Bucholz KK, Slutske WS, Madden PAF, Dinwiddie SH, Dunne MP, Statham DB, Whitfield JB, Martin NG, Eaves LJ : The assessment of alcoholism in surveys of the general community: What are we measuring? Some insights from the Australian Twin Panel Interview Survey. Int Rev Psychiatry 6:295-307, 1994

19. Eaves U, Silberg JL, Hewitt JK, Rutter M, Meyer JM, Neale MC,

J Stud Alcohol 56:500-505, 1995

SUBTYPING ALCOHOLISM

Appendix B SEPs for 4-Class Solution: Males

1471

Wanted to stop 3 or more times Unable to stop Morning drinking Craving Binges Drank more than intended-repeatedly Drunk when did not want to be-repeatedly Great amount of time spent drinking Narrowing of repertoire Drinking caused recurrent problems with family or friends Lost friends because of drinking Drinking caused recurrent problems at work or school fights when drinking-repeatedly Tolerance Reduced activities to drink Drinking interfered with responsibilities Marital problems because of drinking Thought was excessive drinker Felt guilty about drinking DWI Arrested because of drinking Accidental injury when drinking Hazardous use Recurrent blackouts Withdrawal syndrome Relief drinking Seizures Drank to stop seizures DT's Drank to stop DT's Health problems caused by drinking Continued to drink despite health problems Mixed alcohol and medication Continued to drink despite emotional problems Treated for drinking problem Drank nonbeverage alcohol Made rules to control drinking

class 1

0.0900 0.0047 0.1122 0.0022 0.0001 0.2026 0.0937 0.0048 0.0001 0.0165 0.0001 0.0022 0.0321 0.1627 0.0051 0.0026 0.0001 0.0375 0.0818 0.0044 0.0001 0.0028 0.3938 0.0458 0.0001 0.0001 0.0001 0.0001 0.0001 0.0001 0.0017 0.0055 0.0372 0.0034 0.0059 0.0001 0.0544

Class 2

0.5650 0.0842 0.3328 0.0492 0.0465 0.7234 0.3954 0.0941 0.0528 0.2503 0.0102 0.0368 0.2166 0.6041 0.0925 0.0829 0.1420 0.5325 0.4565 0.0217 0.0294 0.0529 0.7904 0.3704 0.0229 0.0275 0.0018 0.0001 0.0027 0.0001 0.0252 0.0364 0.1944 0.0794 0.0908 0.0294 0.2366

Class 3

0.7930 0.3728 0.6739 0.2374 0.2827 0.9028 0.6654 0.4475 0.4216 0.7524 0.0986 0.3464 0.4007 0.8559 0.4423 0.5178 0.5813 0.8370 0.6932 0.1 775 0.1229 0.1886 0.8761 0.6725 0.1 338 0.2301 0.0085 0.0001 0.0123 0.0042 0.2509 0.1405 0.4646 0.3494 0.4173 0.0304 0.41 76

Class 4

0.9285 0.7939 0.9224 0.7650 0.8940 0.9484 0.8637 0.9696 0.7947 0.9672 0.5439 0.8158 0.6597 0.9874 0.8626 0.9064 0.8481 0.9005 0.8201 0.3279 0.2984 0.4886 0.9526 0.9022 0.6554 0.7709 0.1217 0.0597 0.3140 0.2684 0.6454 0.1771 0.6133 0.8527 0.8198 0.2377 0.5370

DWI, driving while intoxicated; DT's, delirium tremens.

Pickles A Analyzing twin resemblance in multisymptom data: Genetic applications of a latent class model for symptoms of conduct disorder in juvenile boys. Behav Genet 23:5-19, 1993

20. Clog CC: In Handbook of Statistical Modeling for the Social and Behavioral Sciences. Arminger G, Clogg CC, Sobel ME, (eds): New York, Plenum, 1995

21. Kleinbaum DG, Kupper LL, Morgenstern H: Epidemiologic Re- search: Principles and Quantitative Methods. Toronto, Ontario, Canada, Lifetime Learning Publications, 1982

22. American Psychiatric Association: Diagnostic and Statistical Man- ual of Mental Disorders, ed 3. Washington, D.C., M A , 1980

23. Bucholz KK, Hesselbrock VM, Reich T: An analysis of the overlap of diagnostic classification systems with respect to psychiatric comorbidity: Data from the COGA study. Paper presented at the Annual Meeting of the Research Society on Alcoholism, Maui, HI, June 1994

24. Nurnberger JI, Bucholz KK, Crowe R, Hesselbrock VM, Reich T, Schmidt I, Schuckit MA, Van Eerdewegh P, Begleiter H: Systematic ascertainment of families with multiple cases of alcoholism for the Col- laborative Study of the Genetics of Alcoholism. Alcohol Clin Exp Res 19(Suppl.):72A, 1995

can we subtype alcoholism? a latent class analysis of data from relatives of alcoholics in a...

Documents