Cheng ATA, Gau S, Chen THH, Chang J, Chang Y. A 4-Year Longitudinal Study on Risk Factors for Alcoholism. Arch Gen Psychiatry. 2004;61(2):184–191. doi:10.1001/archpsyc.61.2.184
Longitudinal studies are needed to resolve inconsistencies in previous
findings regarding antecedents of alcoholism.
To investigate genetic and environmental risk factors for alcoholism.
A 4-year longitudinal cohort study.
A population-based cohort was randomly selected from 4 aboriginal groups
in Taiwan. Cohort subjects free from any alcohol use disorder at phase 1 (n
= 499) were reassessed approximately 4 years later (phase 2). The percentage
of participants who completed the study was 98.4%.
Main Outcome Measures
A standardized semistructured clinical interview for alcoholism and
other psychiatric comorbidity was used in both phases of the study. The main
outcome measure was the incidence of alcohol use disorder. Specific risk factors
examined included sociodemographic factors, family history of alcoholism,
extent of acculturation, psychiatric comorbidity, and alcohol-metabolizing
Using Cox proportional hazards regression analysis, the risk for alcoholism
was significantly higher among subjects who were male (odds ratio [OR], 2.78;
95% confidence interval [CI], 1.79-4.32), aged 15 to 24 years (OR, 5.05; 95%
CI, 2.06-6.18), unmarried (OR, 1.60; 95% CI, 1.03-2.49), and employed (OR,
2.25; 95% CI, 1.34-3.77) and had a higher educational level (OR, 1.76; 95%
CI, 1.12-2.75), a family history of alcoholism (OR, 1.73; 95% CI, 1.06-2.83),
and a higher extent of cultural assimilation (OR, 2.07; 95% CI, 1.28-3.35).
Two specific risk pathways emerged on multivariate analysis: the highest risk
was among subjects aged 25 to 34 years with anxiety disorders (OR, 16.86;
95% CI, 3.98-71.41), and the other was among men with the less active ADH2*1 gene (OR, 5.87; 95% CI, 2.73-12.60).
Based on incidence cases of alcoholism among aboriginal Taiwanese, this
study confirms the significant roles of anxiety disorders and of the ADH2*1 allele as antecedents of alcoholism among specific
age and sex groups.
Alcoholism is a complex disorder that involves multiple factors relatedto genetic, psychological, and sociocultural aspects.1,2 Previousinvestigations on risk factors for alcoholism have largely been based on cross-sectionalstudy designs, and only a few factors have been consistently identified. Theseinclude a significantly higher risk of alcoholism among adolescents and adultmen3- 6 andsubjects with a family history of alcoholism.7- 11 Inaddition, child abuse,12 marital problems,3,12- 15 unemployment,15- 19 lowereducational level,20,21 specificethnicities,6,22 and acculturation23- 27 havebeen reported to be psychosocial risk factors for alcoholism.
Studies on psychiatric comorbidity of alcoholism have been inconsistentin their findings. Most tend to provide evidence supporting anxiety disordersas antecedents28- 32 anddepression as a consequence32- 34 ofalcoholism. The effect of anxiety disorders on the risk for alcoholism hasbeen explained by the self-medication hypothesis,28- 32 althoughthis has not been confirmed in some studies.35,36
Previous investigations indicate that men are more likely than womento drink earlier and in greater quantity3- 6,37- 39 andto have an earlier onset of alcoholism and more alcohol-related problems.3- 6,37,38,40 Incontrast, women who abuse alcohol tend to have a shorter duration betweenthe onset of a drinking problem and seeking help38 anda higher rate of psychiatric comorbidity (eg, anxiety and depressive disorders)than their male counterparts.35,40- 43 Geneticmarkers for alcoholism may also play unequal roles between men and women.44- 46 However, there generallyis considered to be no sex difference in the natural history of alcoholism.47
Despite tremendous efforts to identify genetic factors underlying alcoholism,the only replicated findings are genes encoding for the alcohol-metabolizingenzymes, which are protective against the development of alcoholism.48,49 The active ADH2*2 allele is protective against alcoholism and is predominant among Asianpopulations, including the aboriginal Taiwanese,50- 53 andis rare among white ethnicities.48
To establish the temporal relationships between potential risk factorsand alcoholism, longitudinal investigations that include incidence cases ofalcoholism should be considered. However, few investigators have simultaneouslystudied genetic and environmental risk factors for alcoholism in a singlecohort until now.
The alcoholism study in the Taiwan Aboriginal Study Project22,54,55 was conducted among4 major aboriginal groups: the Ami, Atayal, Bunun, and Paiwan. In the phase1 cross-sectional survey conducted in 1986 to 1988, a random sample of individualsof both sexes 15 years and older was drawn from each of the 4 ethnic groups,with probability proportional to size (511 men and 482 women). High prevalencerates of DSM-III-R56 alcoholuse disorders were found among the 4 groups, ranging from 44.5% to 54.5% (men,70.3%; and women, 27.6%).22 The prevalencerates of alcohol abuse and alcohol dependence, respectively, were 33.9% and36.4% among men and 17.8% and 9.8% among women. The mean ages of the totalgroup and the alcoholic subjects, respectively, were 37.4 and 39.0 years formen and 39.5 and 39.9 years for women. The mean age at onset of alcoholismwas 24.1 years in men and 28.4 years in women. A phase 2 follow-up study conducted4 years later (1990 to 1992) found high rates of age-standardized annual incidenceof new alcoholic cases among the 4 groups, ranging from 2.8% to 4.9%.54
A previous study52 found low frequencies(0.02%-0.05%) of the inactive allele of the alcohol-metabolizing enzyme aldehydedehydrogenase gene (ALDH2*2) and high frequencies(0.94%-1.00%) of the active allele of the alcohol dehydrogenase gene (ADH3*1) among normal control subjects from the 4 ethnicgroups. These results indicate that neither ALDH2 nor ADH3 can be used to examine the risk for alcoholism atthe individual level.
However, 0.14% to 0.30% of normal cohort subjects across the 4 groupshave the less active ADH2*1 allele acting as a geneticvulnerability factor for alcoholism among these groups. Moreover, the kineticdifferences among ADH2 isoenzymes are much more strikingthan those among ADH3 isoenzymes,57 andthe ADH2 alleles may play a more important role than ADH3 in the risk for alcoholism.
In this study, we further investigate the individual and combined effectsof genetic, sociocultural, and psychiatric risk factors, using first incidencecases of alcoholism among the normal cohort established in the phase 1 survey.We hypothesized that the risk of becoming an alcohol abuser would be higheramong aboriginal subjects who had genetic vulnerability to alcoholism, a higherdegree of psychological distress (notably, anxiety disorders), and a lowerextent of acculturation. We also anticipated that these risk factors wouldinteract with each other to generate the highest risk for such morbidity amongthose with a synergism between gene and environment.
Details of the alcoholism study in the Taiwan Aboriginal Study Projecthave been described elsewhere.22,54,55 Onlyan outline will be given herein.
The 4 groups are of Malayo-Polynesian ethnicity. They can be distinguishedfrom each other not only by geographic distribution but also by differencesin physiognomy, language, and sociocultural institutions. The extent of acculturationamong them that took place rapidly in the past 30 years has differed considerablybetween groups and individuals. However, there have been few interethnic marriagesbetween the groups, as well as between them and the Han Taiwanese.55
The overall response rate was 98.3% in the phase 1 survey. The fieldworkconsisted of ethnographic participating observation of the survey communitiesand detailed interviews of sample subjects. The interview comprised a psychiatricassessment for alcoholism and other psychiatric comorbidity conducted by oneof us (A.T.A.C.) and collection of demographic and social environmental databy a research assistant. Additional information was obtained from key informants.
The overall response rate was 99.6% in the phase 2 survey. The normalcohort consisted of 499 subjects who did not have any lifetime DSM-III-R alcohol use disorders at phase 1.54 Twenty-six(5.2%) of them were deceased, and information related to their cause of death,drinking habits, physical and mental states, and living conditions duringthe follow-up was obtained from their key informants. The remaining livingsubjects were located and interviewed. Among them, 8 did not have completedata on sociocultural variables at the phase 1 assessment and were excludedfrom the present study. Hence, complete data for analysis were available in98.4% (n = 491) of the normal cohort. The mean ± SD duration of follow-upwas 4.3 ± 0.5 years.
Informed consent was obtained for a follow-up assessment on alcoholismand for collection of blood samples for biological studies regarding complicationsof alcoholism, as well as for DNA preparations for molecular genetic studiesof alcoholism and related psychiatric comorbidity and behavioral traits. Confidentialityabout personal information regarding the interview records and blood sampleswas assured. For study subjects who spoke only their mother tongues, informationwas read aloud and explained to them by local aboriginal assistants. The alcoholismstudy in the Taiwan Aboriginal Study Project was ethically approved by theNational Science Council, Taipei, Taiwan.
Two psychiatric nurses conducted the fieldwork using the same instrument,with good interrater reliability. DNA preparations were initially obtainedfor 95% of this cohort, but 28% were used up in a molecular study on thalassemia.58 Nevertheless, a comparison between the total groupand those with DNA samples available did not demonstrate any significant differencein sociodemographic distributions, extent of acculturation, or psychiatricmorbidity.
At both phases, we used a standardized semistructured clinical interview,the Chinese version of the Clinical Interview Schedule,59 withan additional section on alcoholism for psychiatric assessment. The ChineseClinical Interview Schedule was developed to study psychiatric morbidity innonpsychiatric and community settings and has been widely used.60- 62 Itincludes a section with a list of 11 subjectively reported symptoms and asection with 12 clinically manifested abnormalities. The schedule uses a 5-pointscale for individual symptoms and abnormalities and for calculation of anoverall clinical severity. The interrater reliability of the Chinese ClinicalInterview Schedule has been studied and found to be acceptable.63,64
Because there are no written characters for any of the aboriginal languages,a 2-stage translation was tape recorded with the help of bilingual natives.Translations of psychiatric terminology were established through interviewswith these interpreters, and semantic equivalents were verified by local aboriginalinterpreters for respondents.
The design and standardization of the section on alcoholism were basedon preliminary observations of the drinking attitudes and behavior among the4 aboriginal groups.22 It covers drinking historyand symptoms and their duration and corresponds to the diagnostic criteriafor alcohol use disorders in the International Classificationof Diseases, 10th Revision (ICD-10),65 and the DSM-III-R.56 The assessment of interpersonal problems due to alcoholintake was based on reports from respondents' family members and from localaboriginal health personnel.
In a reliability study of the alcoholism section, the generalized κvalue for the lifetime DSM-III-R diagnoses of alcoholabuse, alcohol dependence, and absence of any disorder was 0.80.22
At phase 1, we used the Taiwan Aboriginal Acculturation Scale to measurethe extent of acculturation. The development of this scale and its reliabilityand validity have been described in detail elsewhere.23,66 Inbrief, the design of the original 54 items was based on the concept of assimilationby Gordon,67 with consideration of cross-culturalvalidity. These items were administered to 144 subjects stratified by ageand sex who were randomly sampled from the 4 aboriginal groups. Item analysisand factor analysis were applied to select an 18-item scale, including 3 subscales(factors): cultural assimilation, social assimilation, and social attitude(6 items on each subscale, with scores ranging between 0 and 18). The validityand reliability of the Taiwan Aboriginal Acculturation Scale were acceptable.23
The limited number of new alcoholic cases during the 4-year follow-updid not allow us to examine the interaction between ethnicity and severalgenetic markers for alcoholism reported in the literature.1 Wetherefore only examined the effect of the ADH2 genein this normal cohort.
According to the methods of Xu et al,68 primersHE45 (5′-AATCTTTTCTGAATCTGAACAG-3′) and HE46 (5′-GAAGGGGGGTCACCAGGTTGC-3′)were used to amplify exon 3 of ADH2, and the productswere digested with MaeIII. The products were electrophoresedon 4% NuSieve agarose gel (FMC, Rockland, Me). DNA containing β1 showeda fragment of 95 base pair (bp), whereas DNA containing β2 was cleavedby MaeIII and revealed a fragment of 65 bp. The polymerasechain reaction was carried out with 100 ng of genomic DNA, 1-fold buffer,50 µmol of each primer, 1.25 mM of deoxyribonucleoside triphosphatesolutions, and 2 units of Taq polymerase in a 50-mLreaction mixture. The polymerase chain reaction conditions comprised 35 cyclesat 95°C for 1 minute, at 50°C for 1 minute, and at 62°C for 1minute, with a final extension at 62°C for 5 minutes.
χ2 Test was used to assess whether the frequencies ofsociodemographic variables in men differed from those in women. The main outcomein this study was time to onset of alcoholism, ie, the duration between phase1 and the onset of alcoholism. Those who died before phase 2 without developingalcoholism or who were alive and free from alcoholism at phase 2 were treatedas censored cases.
The individual effects of sociocultural factors, family history of alcoholism,psychiatric morbidity assessed at phase 1, and genetic predisposition to ADH2 on the time to onset of alcoholism were first evaluatedusing a Cox proportional hazards regression model. A multiple regression modelincluding all significant variables identified on univariate analysis wasthen used to calculate adjusted hazard ratios and their 95% confidence intervals.All 491 cohort subjects were included in a univariate analysis of socioculturalfactors and psychiatric comorbidity. However, only the 353 cohort subjectswhose DNA was available were included in the univariate analysis of ADH2 data and in the multiple regression analysis, in whichother missing data were adjusted for by the missing-indicator method of Miettinen,Jones, and Chowl.69- 71
To assess the interactions between genetic and environmental factors,the saturated model (including 3-way and 2-way interactions and main effects)was compared with the reduced model on the basis of the backward stepwisemethod. The level of significance for deletion of variables was P<.05.
Table 1 presents the sociodemographiccharacteristics of the normal cohort by sex. The percentage of women (69.7%)was much higher than that of men (30.3%) (χ21 =75.86, P<.001). Men were significantly younger,better educated, and more likely to be employed than women. No significantdifference was observed in sex or age distribution across the 4 ethnic groups.The mean ages in the 4 groups ranged from 31 to 38 years in men and 37 to41 years in women.
Thirty-eight (11.1%) of 342 women and 41 (27.5%) of 149 men were newcases of alcoholism. Of these, 74 (93.7%) were alcohol abusers and 5 (6.3%)were alcohol dependent. The mean ages of the men and women who developed alcoholismwere 29.0 and 31.5 years, respectively. The mean age at onset was 26.5 yearsin men and 29.5 years in women.
The risk for alcoholism relative to several sociocultural factors assessedat phase 1 was then examined (Table 2).Cohort subjects aged 15 to 24 years had the highest risk for alcoholism, followedby those aged 25 to 34 years. Men had a higher risk than women, as did subjectsfrom the Bunun group.
The risk for alcoholism was also significantly higher among subjectswho were unmarried, better educated, and employed and who had a family historyof alcoholism among first-degree relatives. A higher extent of cultural assimilationwas predictive of an increased risk for alcoholism. No such trend was observedfor the other 2 subscales of acculturation.
Slightly more than 14% (14.1%) of men and 32.2% of women had 1 or more DSM-III-R psychiatric disorders assessed at phase 1. Theprevalence rates for individual diagnoses were: schizophrenia, 0.6%; mentalretardation, 0.6%; depressive disorders, 10.0% (major depression, 5.5%; anddysthymia, 4.5%); panic disorder, 5.7%; generalized anxiety disorder, 5.7%;anxiety disorder not otherwise specified, 9.8%; adjustment disorders, 2.2%;somatoform disorders, 1.6%; and primary insomnia, 2.2%.
Although there was a tendency for subjects who had any depressive disorder,anxiety disorder (including generalized anxiety disorder, panic disorder,phobic disorders, and anxiety disorder not otherwise specified), or any mentaldisorder in phase 1 to have a lower risk for alcoholism, none of the individualdiagnostic categories was a significant psychiatric antecedent of alcoholism(Table 2).
The effects of the 2 major groups of mental disorders (ie, depressiveand anxiety disorders) on the risk for alcoholism were further examined acrossdifferent age groups (Table 3).We did not find any significant association of depressive disorders with theoccurrence of new alcoholic cases. In fact, there were only 3 depressed subjectsin the alcoholic group, all older than 25 years. However, the risk for alcoholismwas significantly higher among subjects with anxiety disorders than thosewithout anxiety disorders in the group aged 25 to 34 years. No such trendwas found in the group younger than 25 or in the group older than 34.
Table 4 shows the effectof the ADH2 genotype on the risk for alcoholism stratifiedby sex. Male subjects with genotypes comprising 1 or 2 alleles of the lessactive ADH2*1 gene had a significantly higher riskfor alcoholism. This effect was not found in women or in the total group.
Following multiple Cox proportional hazards regression analysis of allthe significant risk factors (Table 2, Table 3, and Table 4), 2 interactions were retained in the final model, one betweenage and anxiety disorders and the other between sex and ADH2 (Table 5). The riskfor alcoholism was highest among subjects aged 25 to 34 years with anxietydisorders, followed by subjects aged 15 to 24 without anxiety disorders. Itwas also high among male subjects with 1 or 2 alleles of the less active ADH2*1 gene, followed by men without such inheritance.This effect was not found in women.
For aboriginal subjects aged 25 to 34 years, the population attributablerisk for alcoholism due to anxiety disorders was 31.1% (preventive fraction).For men, the population attributable risk due to ADH2*1 alleles was 34.7%.
This longitudinal study investigated the causal relationships betweengenetic and environmental factors and the onset of alcoholism in a community-basedrepresentative normal cohort from 4 aboriginal groups in Taiwan. We simultaneouslyassessed the contributions of genetic, sociocultural, and psychiatric factorsusing standardized instruments.
Despite these strengths, there are some limitations that require considerationin the interpretation of the findings. First, our phase 2 follow-up was 4years in duration, and some normal cohort subjects may have been too youngto test positive for alcoholism, resulting in a substantial quantity of false-negativecases. On the other hand, a short follow-up can reduce recall bias.72 Second, the limited number of incidence cases inthis study may not have sufficient statistical power to identify risk factorsfor alcoholism and to detect possible interactions between these factors.However, given a 90% statistical power and a 5% significance level, with 16.1%of the total cohort as incidence cases, the required sample size is approximately340, fewer than in this cohort. Therefore, we believe the statistical poweris sufficient to test the association.
Third, although this study has satisfactory internal validity, its externalvalidity for other ethnic groups remains to be examined. Fourth, DNA preparationswere not available for 28% of cohort subjects in this study. However, therewas no difference in demographic and psychosocial factors between those withand without DNA preparations, and the missing-indicator method69- 71 toadjust for missing data on genotype was applied. Last, an information biasmay have arisen from data collected on the deceased subjects. However, thepercentage of deceased was low (5.3%), and every effort was made to minimizethe bias by gathering all the relevant information from key informants. Becauseresidents in these aboriginal villages are very close to each other, the effectof such potential bias is believed to be negligible.
Our finding of a male excess in incidence cases of alcoholism is consistentwith 2 other longitudinal community studies,73,74 andall 3 studies reported the highest incidence among adolescents and young adults.The protective effect of marriage from becoming an alcohol abuser was observedin this study and in several others.3,15,16
The significantly higher risk for alcoholism among our cohort subjectswith jobs or with higher educational levels was contradictory to findingsin previous studies.16- 20 Theprevalence of alcoholism in the phase 1 cross-sectional survey was, however,significantly higher among the less educated and the unemployed.22 Itis likely that demographic factors affect prevalence and incidence differently.For instance, age may have a confounding or modifying effect, as young aboriginalTaiwanese nowadays are more likely to be employed and be better educated.Those who are less educated might have a longer duration and a poorer outcomeof their alcoholism, resulting in a higher prevalence.
Extent of acculturation may have generated different levels of alcoholuse problems among various ethnic groups.22,24- 26 However,the confounding effect of age may explain the increased risk for alcoholismamong subjects with greater cultural assimilation, because young aboriginalTaiwanese who are better educated frequently work in cities. There may besome potential risk factors contributing to the highest incidence and prevalenceof alcoholism among the Bunun.22,54 Genetically,a significant association between alcohol dependence and the tryptophan hydrogenasegene was only found in the Bunun,75 suggestingthe likelihood of ethnic heterogeneity among the 4 groups. Environmentally,the Bunun were the last among all Taiwanese aboriginal groups to make contactwith modern civilization.55 Further cross-ethnicstudies are needed to better explain our finding.
The significant effect of a family history of alcoholism on the developmentof alcoholism reported in previous studies7- 11 wasonly found on univariate analysis in this study. A possible explanation isthat the social drinking pattern in these aboriginal communities may havediluted the effect of the family environment, and the genetic influence mayhave been expressed through other biological markers, including alcohol-metabolizinggenes.
Although findings in this study did not confirm previous findings ofan association of anxiety disorders with alcoholism among women,40- 43,76 theylend support to the self-medication hypothesis regarding anxiety disorders,28- 30,32,36 notablyamong adults aged 25 to 34. Subjects in this age group may have underlyingmechanisms of alcoholism that are different from those of alcoholic subjectswith early (around age 20 years) and late (>34 years) onset. One possibleexplanation is that the youngest alcoholic subjects (<25 years) have anincreased genetic vulnerability and the older subjects (>34 years) have greaterenvironmental stresses.
In contrast to findings in previous studies,15,32,34,77 wedid not find an association between depression and incidence of alcoholism.Similar to some other studies,32,34 ourfindings suggest that depression mainly affects the course of alcoholism,including the transition from alcohol abuse to dependence, and depressionitself is frequently a psychological complication of alcoholism. In this study,the higher risk of depressive disorders among nonalcoholic subjects may reflecta higher risk of such morbidity among cohort subjects older than 24 years(none of the alcoholic subjects aged >24 years had a comorbid depressive disorder,in contrast to 6.3% and 11.9% of normal subjects aged 25-34 and >34 years,respectively).
Although the ALDH2*2 and ADH3*1 alleles do not have a protective effect against alcoholism amongTaiwanese aborigines, the ADH2*2 allele is protectiveagainst alcoholism among men of this ethnicity. This is consistent with findingsin previous studies52,78 and inthe Australian Alcohol Challenge Twin Study,45 suggestingthat the alcohol dehydrogenase genotype may not contribute to alcoholism amongwomen. The ongoing phase 3 (16-year) follow-up is expected to identify morefemale alcoholic subjects to examine the role of ADH2*2 among the female Taiwanese aborigines, as well as to evaluate whethersex modifies the effect of ADH2 on the risk for alcoholism.
In summary, findings in this study have lent considerable support toour hypothesis regarding the significant roles of genetic vulnerability (ie,the less active ADH2*1 gene) and psychological distress(ie, anxiety disorders) on the risk of developing alcoholism. The associationof greater extent of acculturation with development of alcoholism that wasonly found on univariate analysis is probably the result of a confoundingbias of age. Our findings do not support any synergistic effect between factors;however, interactive effects were found between genetic vulnerability andsex and between psychological stress and age.
The findings in this study suggest that early identification and treatmentof anxiety disorders may prevent alcoholism and its possible psychiatric complications,including depressive disorders, among subjects with a genetic vulnerabilityto alcohol-metabolizing enzymes and with sociocultural risk factors for alcoholism.In addition, as specific protective genetic markers against alcoholism areidentified,52,79,80 moleculargenetics and genetic epidemiologic measures may be used to identify specificenvironmental targets for primary prevention, particularly among the geneticallyvulnerable.81,82 Our results alsoindicate that findings from molecular genetic studies on alcoholism amongsevere alcoholic patients need to be verified in longitudinal studies amongrepresentative cohort subjects from the community.
Corresponding author and reprints: Andrew T. A. Cheng, MD, PhD, FRCPsych,Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan (e-mail: email@example.com).
Submitted for publication November 15, 2002; final revision receivedApril 14, 2003; accepted April 22, 2003.
This study was funded by grants NSC 83-0203-B001-102 and NSC 84-2331-B001-046from the National Science Council and by grants DOH-80-03 and DOH-81-111 fromthe Department of Health, Taipei, Taiwan.
This study was presented in part at the Biennial Symposium of the WorldPsychiatric Association Section on Epidemiology and Public Health; June 3-6,2001; Baltimore, Md.