Co-occurrence of 12-Month Alcohol and Drug Use Disorders and PersonalityDisorders in the United States:  Results From the National Epidemiologic Survey on Alcohol and RelatedConditions FREE

Numerous studies have addressed the prevalence of personality disorders(PDs), especially antisocial PD, among alcohol and drug abusers.¹ Theyshow a high but variable rate of a broad range of PDs in alcohol and drugabusers, and several among them have demonstrated the adverse effect of thesedisorders on duration of stay in treatment and outcome.^2- 9 Studiesof alcohol and drug use disorders among patients seeking treatment for personalitypsychopathology are rare. A recent study,¹⁰ however,has found high prevalences of alcohol and drug use disorders in patients seekingtreatment for PDs. With few exceptions, psychiatric comorbidity in these clinicalstudies did not differentiate between alcohol and drug use disorders, andthese studies were conducted in predominantly male samples. That this literaturehas paid little attention to sex differences is surprising considering thatthe importance of distinguishing men and women is firmly established in thefield of substance use disorder research.

From an epidemiological perspective, however, a more serious problemwith research on comorbidity in clinical studies is that the samples of subjectsdo not represent the underlying populations. Because of this problem, it isnecessary to turn to general population samples for more accurate and preciseinformation on the comorbidity of PDs and alcohol and drug use disorders.However, large epidemiologic surveys conducted in the United States duringthe past 2 decades have focused exclusively on the prevalence and comorbidityof antisocial PD and alcohol and drug use disorders.^11- 12 Withthe exception of antisocial PD, we have very limited knowledge of the comorbiditybetween the range of PDs and alcohol and drug use disorders and whether theseassociations differ between men and women. The fact that accurate data onthe sex-specific prevalences of a broad range of PDs have not been availablein general population surveys of the United States reflects a major gap inour understanding of the processes underlying the comorbidity of PDs and alcoholand drug use disorders. The present study was designed, in part, to addressthis gap and provide the information.

Accordingly, this article presents nationally representative data onthe prevalence and co-occurrence of alcohol and drug use disorders and 7 ofthe 10 PDs defined in the Diagnostic and Statistical Manualof Mental Disorders, Fourth Edition (DSM-IV)¹³ assessed in the 2001-2002 National Institute on AlcoholAbuse and Alcoholism (NIAAA) National Epidemiologic Survey on Alcohol andRelated Conditions (NESARC).¹⁴ The NESARC isthe largest comorbidity survey ever conducted (N = 43 093). The samplesize allows for accurate estimation of current or past-year co-occurrenceof both alcohol and drug use disorders and avoidant, dependent, obsessive-compulsive,histrionic, paranoid, schizoid, and antisocial PDs among men and women.

The Wave 1 NESARC is a nationally representative face-to-face surveyof 43 093 respondents, aged 18 years and older, conducted by the NIAAAin 2001 through 2002. (A second wave will be conducted in 2004-2005.)¹⁴ The target population of the Wave 1 NESARC is thecivilian, noninstitutionalized population residing in the United States andthe District of Columbia, including Alaska and Hawaii. The housing-unit samplingframe of the NESARC was the US Census Bureau Census 2000 Supplementary Survey, ¹⁴ a national survey of more than 78 000 householdsper month conducted in 2000 through 2001. The NESARC also included a group-quarterssampling frame derived from the Census 2000 Group Quarters Inventory.¹⁴ The group-quarters sampling frame captures importantsubgroups of the population with heavy substance use patterns (eg, collegehousing) not often included in general population surveys. The sampling frameresponse rate was 99%, the household response rate was 89%, and the personresponse rate was 93%, yielding an overall survey response rate of 81%, substantiallyhigher than other surveys of this kind.

Information on race and ethnicity collected in the Census 2000 SupplementarySurvey in 2000 through 2001 was used to oversample African American and Hispanichouseholds. The oversampling procedure increased the percentage of non-Hispanic,African American households in the sample from 12.3% to 19.1% (n = 8245) andthe percentage of Hispanic households from 12.5% to 19.3% (n = 8308). Onesample person from each household or group-quarters unit was randomly selectedfor interview, and young adults, aged 18 to 24 years, were oversampled ata rate of 2.25 times that of other members in the household.

The NESARC data were weighted to reflect the probabilities of the selectionof primary sampling units (PSUs) within strata and for the selection of housingunits within the sample PSUs. The PSUs are mutually exclusive categories ofpersons or units of interest identified in the first stage of the multistageNESARC sample. The PSUs consisted of geographic units representing the entireUnited States defined in terms of sociodemographic criteria. The data alsowere weighted: (1) to account for the selection of 1 sample person from eachhousehold; (2) to account for oversampling of young adults; (3) to adjustfor nonresponse at the household level and person level; and (4) to reducethe variance arising from selecting 2 PSUs to represent an entire stratum.The weighted data were then adjusted to be representative of the US civilian,noninstitutionalized population for a variety of socioeconomic variables includingregion, age, sex, race, and ethnicity using the 2000 Decennial Census of Populationand Housing¹⁴ and statistics on births, deaths,immigration and emigration, and the size of the Armed Forces.

Approximately 1800 experienced lay interviewers from the US Census Bureauadministered the NESARC using laptop computer–assisted software thatincluded built-in skip, logic, and consistency checks. On average, the interviewershad 5 years' experience working on census and other health-related nationalsurveys. All NESARC interviewers completed a 5-day self-study at home andparticipated in a standardized 5-day in-class training session at 1 of thebureau's 12 regional offices. The NESARC training supervisors from each regionaloffice also were required to complete the home study and to attend a centralizedtraining session prior to fielding of the survey, where they completed thein-class training under the direction of NIAAA sponsors and Census Field andDemographics Survey Division headquarters staff.

Regional supervisors recontacted a random 10% of all respondents forquality-control purposes. In these quality-control interviews, a series ofquestions were reasked to verify that respondents had received the entireinterview and that the questionnaire had been administered properly. Therewas no case in which it was determined that the interview had been conductedin any manner that was inconsistent with the interviewer's extensive training.In addition, 2657 respondents were randomly selected to participate in a reinterviewstudy after completion of their NESARC interview. Each respondent was readministered1 to 3 sections of the survey assessment instrument. These interviews notonly served as an additional check on survey data quality but formed the basisof a test-retest reliability study of new modules of the survey instrument.¹⁵

Diagnoses presented in this article were made by the NIAAA Alcohol UseDisorder and Associated Disabilities Interview Schedule–DSM-IV Version (AUDADIS-IV),¹⁶ a state-of-the-artstructured diagnostic interview designed to be used by lay interviewers. TheAUDADIS-IV included an extensive list of symptom questions that separatelyoperationalized DSM-IV criteria for alcohol and drugabuse and dependence including 10 classes of drugs: sedatives, tranquilizers,opiates (other than heroin or methadone), stimulants, hallucinogens, cannabis,cocaine (including crack cocaine), inhalants/solvents, heroin, and other drugs.Consistent with the DSM-IV, current (in the last12 months) dependence diagnoses required the respondent to satisfy at least3 of the 7 DSM-IV criteria for dependence duringthe last year. The withdrawal criterion of the alcohol-dependence diagnosiswas measured as a syndrome, requiring at least 2 positive symptoms of withdrawalas defined in the DSM-IV alcohol-withdrawal category.The AUDADIS-IV diagnoses of alcohol abuse required a respondent to meet atleast 1 of the 4 criteria defined for abuse in the 12-month period precedingthe interview and not meet criteria for dependence. The drug-specific diagnosesof abuse and dependence were derived using the same algorithm and were aggregatedto produce measures of any drug use disorder, any drug abuse, and any drugdependence.

The reliability of AUDADIS-IV alcohol and drug use disorder measureswas assessed in several large test-retest studies conducted in clinical andgeneral population samples.^17- 21 Thereliability of alcohol and drug abuse and dependence in these studies wasexcellent, exceeding κ = 0.74 for alcohol diagnoses and κ = 0.79for drug diagnoses. The discriminant, concurrent, convergent, construct, andpopulation validities of the AUDADIS-IV alcohol and drug use disorder diagnosesalso have been well documented,^22- 35 includingin the World Health Organization/National Institutes of Health Reliabilityand Validity Study.^36- 40 Inthese studies,³³ alcohol and drug use disorderdiagnoses were found to be significantly and highly correlated with importantvalidators, including substance use, social/occupational dysfunction and disability,and family history (convergent validity),^{24- 26,32- 34} andthese results were shown to generalize to other populations (population validity).²⁷ These studies also demonstrated that abuse and dependencediagnoses also were related to different sets of validators (discriminantvalidity). Alcohol and drug use disorder diagnoses defined by DSM-III, DSM-III-R, DSM-IV, and the International Classification of Diseases,10th Revision (ICD-10)⁴¹ criteria alsowere shown to be highly concordant (convergent validity).^{22- 23,28- 30,37} Concordancebetween AUDADIS-IV alcohol and drug use disorders and those assessed withthe Schedule for Clinical Assessment in Neuropsychiatry⁴² was high (concurrent validity),^36,39 and the construct validity of thesediagnoses has been supported by both exploratory and confirmatory factor analyses.^32,35,38

The diagnosis of PDs requires an evaluation of the individual's long-termpatterns of functioning.^13(p630) Diagnosesof PDs made using the AUDADIS-IV were made accordingly. Respondents were askeda series of personality symptom questions about how they felt or acted mostof the time throughout their lives regardless of the situation or whom theywere with. They were reminded on 20 occasions throughout the PD section notto include times when they were depressed, manic, anxious, drinking heavily,using medicines or drugs, experiencing withdrawal symptoms (defined earlierin the AUDADIS-IV), or times when they were physically ill.

To receive a DSM-IV diagnosis, respondentsneeded to endorse the requisite number of DSM-IV symptomitems for the particular PD and at least 1 positive symptom item must havecaused social or occupational dysfunction. Multiple symptom items were usedto operationalize the more complex criteria associated with certain PDs. Thefollowing number of symptom items were used to assess each PD: avoidant (n= 7); dependent (n = 8); obsessive-compulsive (n = 10); paranoid (n = 9);schizoid (n = 10); histrionic (n = 11); and antisocial (n = 30). Because oftime and space constraints, not all DSM-IV PDs wereassessed in the Wave 1 NESARC. The decision to exclude borderline, schizotypal,and narcissistic PDs was based on the larger number of symptom items requiredto operationalize the disorders relative to those PDs assessed in Wave 1 (ie,borderline, 18 items; schizotypal, 16 items; and narcissistic, 19 items).However, in the follow-up Wave 2 of the NESARC, borderline, schizotypal, andnarcissistic PDs will be included.

The reliability of AUDADIS-IV PDs was assessed in a test-retest studyconducted as part of the NESARC survey proper.¹⁵ Arandom subsample of 282 respondents was reinterviewed with the antisocialPD module, and another subsample of 315 respondents was reinterviewed withthe AUDADIS-IV modules containing the remaining PD measures. These reinterviewswere conducted approximately 10 weeks after the NESARC interviews. The reliabilityof the PD diagnoses in these community samples ranged from fair to good, from κ= 0.40 for histrionic PD to κ = 0.67 for antisocial PD. Reliabilitiesof the AUDADIS-IV PD diagnoses are as good as or better than those found forsemistructured personality interviews in short-term test-retest studies conductedin treated samples of patients.⁴³

The validity of AUDADIS-IV PDs was assessed in a series of linear regressionanalyses, using the NESARC data, that examined the associations between eachPD and 3 Short Form 12v2⁴⁴ disability scores,controlling for age, all other PDs, and 12-month comorbid DSM-IV substance use disorders and anxiety and mood disorders. TheShort Form 12v2, a reliable and valid measure of generic quality of life usedin large population surveys, yields 10 component summary and profile scoresassessing various dimensions of disability and impairment. In the presentanalyses, the focus was on 3 Short Form 12v2 scores: the mental componentsummary score; the social functioning score, reflecting limitations in socialfunctioning; and the role emotional function score, measuring role impairmentdue to emotional problems. All PDs, except histrionic, were shown to be highlysignificant (P < .01 to P <.001) predictors of the mental component summary, social functioning, androle emotional scores. Respondents with those PDs had significantly greaterdisability and social/occupational dysfunction than respondents who did nothave the PD.

Cross-tabulations were used to calculate prevalences and comorbidityrates of PDs and alcohol and drug use disorders. A series of univariate logisticregression analyses was used to study associations between PDs and alcoholand drug use disorders. The β coefficients from these analyses were transformedinto odds ratios (ORs) for ease of interpretation. Differences in the associationsof PDs and alcohol and drug use disorders between men and women were examinedby comparing sex-specific β coefficients derived from the logistic regressionanalyses. Because of the complex survey design of the NESARC, variance estimationprocedures that assume simple random sampling cannot be used. The stratificationof the NESARC sample will result in standard errors much larger than thosethat would be obtained with a simple random sample of equal size. To takeinto account this NESARC sample design component, all standard errors and95% confidence limits (CIs) presented here were generated using SUDAAN (ResearchTriangle Institute, Research Triangle Park, NC),⁴⁵ asoftware program that uses appropriate statistical techniques to adjust forsample design characteristics.

The 12-month prevalences of any alcohol use disorder and any drug usedisorder were 8.5% and 2.0%, respectively (Table 1). Rates of abuse exceeded those for dependence for bothalcohol and drug use disorders. The most prevalent PD in the general populationwas obsessive-compulsive PD (7.9%), followed by paranoid PD (4.4%), antisocialPD (3.6%), schizoid PD (3.1%), avoidant PD (2.4%), histrionic PD (1.8%), anddependent PD (0.5%).

As indicated in the top row of Table2, 28.6% and 47.7% of respondents with a 12-month alcohol use disorderand drug use disorder, respectively, had at least 1 PD. Rates of any PD weregreater among respondents with any drug abuse (37.8%) and any drug dependence(69.5%) than among respondents with alcohol abuse (19.8%) and alcohol dependence(39.5%). The prevalence of antisocial PD (12.3%), obsessive-compulsive PD(12.1%), and paranoid PD (10.2%) were the highest among respondents with analcohol use disorder. These also were the most prevalent PDs among respondentswith any drug use disorder, but the rates were much higher. The prevalenceof specific PDs was much greater among respondents with dependence on alcohol(2.5%-18.3%) or drugs (10.1%-39.5%) compared with respondents with alcoholabuse (0.3%-9.5%) or any drug abuse (2.0%-22.3%).

As indicated in Table 3,16.4% of the respondents with at least 1 PD met criteria for a current alcoholuse disorder and 6.5% met criteria for a current drug use disorder. The prevalenceof any alcohol use disorder was greatest among respondents with histrionic(29.1%), antisocial (28.7%), dependent (21.6%) and paranoid (19.5%) PDs. Similarly,the rate of any drug use disorder was greatest among respondents with dependent(18.5%), antisocial (15.2%), and histrionic (12.8%) PDs. Prevalences of alcoholabuse (2.5%-9.5%) and drug abuse (2.0%-8.4%) were consistently lower amongrespondents with specific PDs than the corresponding rates for alcohol dependence(7.4%-21.3%) and any drug dependence (2.3%-12.9%). The only exception to thispattern was among respondents with antisocial PD, where the prevalence ofany drug abuse (8.4%) exceeded the rate for any drug dependence (6.8%).

Associations between alcohol and drug use disorders and PDs are shownin Table 4 in the form of ORs.The overall pattern of ORs is overwhelmingly positive, with 88% of the disorder-specificORs being positive and statistically significant. The association betweenany PD and any alcohol use disorder (OR, 2.6) was weaker than the associationfound for any drug use disorder (OR, 5.5), a pattern also found when specificPDs were examined. Specific PDs were more strongly related to alcohol dependence(ORs, 2.2-7.5) and drug dependence (ORs, 4.8-26.0) than to alcohol abuse (ORs,0.5-2.2) or drug abuse (ORs, 1.5-8.2). Although histrionic PD (OR, 1.7) andantisocial PD (OR, 2.2) were significantly associated with alcohol abuse,the associations between alcohol abuse and avoidant, dependent, obsessive-compulsive,paranoid, and schizoid PDs were not significant. All specific PDs, however,were strongly and consistently related to any alcohol use disorder (ORs, 1.7-4.8)and any drug use disorder (ORs, 2.4-11.8). Dependent, histrionic, and antisocialPDs were more strongly related to both alcohol and drug use disorders thanany of the other PDs.

Similar to the pattern observed in the total sample, the associationsbetween current alcohol and drug use disorders and PDs among men and womenwere overwhelmingly significant and positive, with the exception of the associationsbetween avoidant, dependent, obsessive-compulsive, paranoid, and schizoidPDs and alcohol abuse (Table 5).With respect to any drug use disorder, drug abuse, and drug dependence, associationsremained the strongest for antisocial, histrionic, and dependent PDs amongmen and women. The same pattern was observed for any alcohol use disorderand alcohol dependence among men and women.

Significant sex differences in the associations between alcohol anddrug use disorders and PDs also were observed. The relationship between obsessive-compulsive(P < .02), histrionic (P <.04), and antisocial (P < .006) PDs and alcoholdependence was significantly greater for women than men. With regard to anydrug abuse, the associations with obsessive-compulsive (P < .03), schizoid (P < .009), histrionic(P < .02), and antisocial (P < .002) PDs were greater for women than for men. In contrast,the association between drug dependence and dependent PD was significantlygreater (P < .04) among men than women.

The co-occurrence of DSM-IV current alcoholand drug use disorders and DSM-IV PDs is pervasivein the US population. Among individuals with a current alcohol or drug usedisorder, 28.6% and 47.7%, respectively, had at least 1 PD. While the proportionof individuals with a PD who also had an alcohol or drug disorder was lower,a considerable proportion of those with PDs did meet criteria for alcoholor drug abuse or dependence. Overall, 16.4% of individuals in the generalpopulation with at least 1 PD had a current alcohol use disorder, and 6.5%had a current drug use disorder. The strong associations between most PDsand alcohol and drug use disorders were generally consistent when examinedseparately among men and women. Consistent with clinical research on comorbidityof Axis II disorders and alcohol and drug use disorders,¹ thisstudy found greater associations between PDs and drug use disorders comparedwith alcohol use disorders.

Comorbidity in the general population is often lower than comorbidityin treated samples since individuals with more than 1 disorder have a greaterprobability of seeking treatment (ie, Berkson bias). However, a striking findingin this study was that the prevalence of any PD and antisocial PD (one ofthe most extensively studied PDs in treated samples) among individuals withcurrent alcohol and drug use disorders was similar to the median rates observedin samples of patients receiving treatment for alcohol and/or drug use disorders,as assessed with other standardized assessment instruments (ie, the Structured Clinical Interview for DSM-III-R Personality Disorders [SCID-II]⁴⁶ andthe Diagnostic Interview Schedule).⁴⁷ For example,the median rate of any PD among patients receiving treatment for an alcoholuse disorder assessed with the SCID-II^48- 49 was39.0% compared with 39.5% found among individuals in this study with currentalcohol dependence. The median rate of any PD among patients receiving treatmentfor drug use using the SCID-II^{4,7,10,49- 57} was59.0% compared with the 69.5% rate found among individuals with current drugdependence. The prevalence of antisocial PD among respondents with currentdrug dependence was 39.5%, a figure midway between the median rates foundin studies of drug treatment samples using the semistructured SCID-II,^{50,52- 53,55,58- 60} (21.0%)and the fully structured Diagnostic Interview Schedule^61- 64 (49.0%)assessment instruments. The rate of antisocial PD among individuals with currentalcohol dependence was 18.3%, somewhat lower than the median rate of 37.5%found among patients in alcohol treatment settings using the Diagnostic InterviewSchedule.^65- 72 Itis likely that the prevalences of PDs among individuals with alcohol and druguse disorders in this study would have been greater if all DSM-IV PDs had been assessed. If all PDs had been assessed, we mightexpect the reported rates of PDs using the fully-structured AUDADIS-IV tohave slightly exceeded the rates presented earlier for semistructured interviews,as would be predicted by the literature.

The PDs most strongly associated with alcohol and drug use disorderswere antisocial, dependent, and histrionic PDs. The degree of diagnostic overlapbetween DSM-IV PDs has long been recognized,^73- 74 and it may be responsible for thestrong relationship observed between histrionic, antisocial, and dependentPDs and alcohol and drug use disorders. For example, individuals with antisocialPD share certain tendencies with individuals with histrionic PD to be impulsive,seductive, superficial excitement seeking, reckless, and manipulative, butindividuals with histrionic PD do not characteristically exhibit antisocialbehaviors.¹³ Individuals with dependent PDand histrionic PD are excessively dependent on others for praise, guidance,and nurturance, but individuals with dependent PD do not characteristicallydemonstrate the flamboyant emotional features of histrionic PD. Although multivariatestudies^75- 77 havebeen conducted on item-level criteria of DSM PDsin search of the factor structure underlying PD diagnoses, the findings ofthis study suggest that this search be expanded to include criteria of AxisI substance use disorders along with the components of PDs that are most closelyassociated with them. The results of these future studies might elucidatesubtypes of alcohol and drug use disorders, refine the classification of bothtypes of disorder, and increase our understanding of the pathological processesunderlying their comorbidity.

A number of the PDs examined in this study were more strongly associatedwith alcohol and drug use disorders among women, including antisocial PD.However, a stronger association between dependent PD and drug dependence wasobserved among men. Although reasons for these observed sex differences areunknown, these findings highlight the need to examine a broader set of factorsthat affect the prevalence and co-occurrence of PDs and alcohol and drug usedisorders, including age, socioeconomic status, and, importantly, primarysubstance of abuse. In the current study, the stronger associations observedbetween antisocial PD and alcohol and drug use disorders among women may bethe result of differential mortality or incarceration. That is, men who arehighly comorbid for antisocial PD and alcohol and drug use disorders are morelikely to die young or be incarcerated than women and thus less likely tobe represented in general population surveys. This explanation is consistentwith the findings that men are overrepresented in jail and prison populationsand that substance use disorders occur in about 90% of individuals with antisocialPD who are incarcerated.⁷⁸

In light of the extensive comorbidity between PDs and alcohol and druguse disorders found in this study, there would appear to be great value inassessing a broad range of PDs among substance abuse patients. This more comprehensiveassessment can guide treatment planning. For example, patients with comorbidalcohol and drug use disorders and PDs can be expected to require treatmentthat is more extensive and of longer duration. In this regard, modified psychoanalyticpsychotherapy focused or targeted on particular features of PDs might holdgreat promise for successful recovery among comorbid individuals.^79- 81 The trend towardintegrating 12-step programs into rehabilitation programs also appears promisingin that 12-step programs require individuals to examine their relationshipto others, overcome feelings of helplessness, gain an internal locus of control,encourage self-examination, address defects in character, and promote honestrelationships.^82- 83 More clinicalresearch is needed to examine the role of these and other approaches targetedat treating substance use disorders (eg, contingency management, motivationalenhancement therapy, cognitive behavior therapy) in improving the chancesof recovery and the lives of individuals with comorbid alcohol and drug usedisorders and PDs. This work will be formidable, because some of these componentsof treatment are on uncertain grounds in terms of efficacy and mechanismsof action (eg, 12-step programs and psychoanalytic treatments). Attentionin this clinical work on the effects of sex, substance of choice, and otherfactors that affect treatment outcome and eventual recovery might furtherrefine treatment planning.

This national study of comorbidity represents a landmark study in thearea of PDs. Previous psychiatric epidemiology studies were too small to addressthese important relationships in detail. Personality disorders are not onlypervasive and associated with substantial disability,⁴⁴ theyare very common among those with alcohol and drug use disorders. Further workin many directions is indicated by the results of this study, including adissection of the components of the 2 types of disorders that are most closelyassociated, the factors giving rise to the associations, and the treatmentand prevention implications of these disorders when comorbid.

Corresponding author and reprints: Bridget F. Grant, PhD, PhD, Laboratoryof Epidemiology and Biometry, Division of Intramural Clinical and BiologicalResearch, Room 3077, National Institute on Alcohol Abuse and Alcoholism, NationalInstitutes of Health, MS 9304, 5635 Fishers Ln, Bethesda, MD 20892-9304 (e-mail: bgrant@willco.niaaa.nih.gov).

Submitted for publication August 25, 2003; final revision received October31, 2003; accepted November 19, 2003.

The views and opinions expressed in this article are those of the authorsand should not be construed to represent the views of any of the sponsoringorganizations, agencies, or the US government.