A schematic diagram for the bivariate biometric model examining the relationship between pathological gambling (PG) and major depression (MD). Factors influencing PG and MD include genetic factors (A), shared environment (C), and unique environment plus error (E). Correlations between these factors across disorders are represented as rA, rC, and rE, respectively. The contributions of each of these factors to PG (aPG, cPG, ePG) and MD (aMD, cMD, eMD) are also indicated. CI indicates confidence interval; lowercase a, c, and e refer to path loading for factors A, C, and E, respectively.
Potenza MN, Xian H, Shah K, Scherrer JF, Eisen SA. Shared Genetic Contributions to Pathological Gambling and Major Depression in Men. Arch Gen Psychiatry. 2005;62(9):1015-1021. doi:10.1001/archpsyc.62.9.1015
Although pathological gambling (PG) and major depression (MD) frequently co-occur, little is known regarding the relative contributions of environmental and genetic factors to the codevelopment of the disorders.
To estimate environmental and genetic contributions to PG and MD as defined in DSM-III-R and the lifetime co-occurrence of PG and MD.
Survey data from the Vietnam Era Twin Registry were examined in biometric models.
Of 10 253 eligible participants, 7869 were successfully interviewed.
Main Outcome Measures
Estimated genetic, shared environmental, and unique environmental contributions to PG and MD and their lifetime co-occurrence in bivariate models.
Elevated odds ratios for MD were associated with those of PG (4.06; 95% confidence interval, 2.68-6.13), and the association remained significant following adjustment for sociodemographic and other psychiatric variables (odds ratio = 1.98; 95% confidence interval, 1.14-3.45). The best-fitting bivariate model indicated that 66% of the variance in PG and 41% of the variance in MD were owing to genetic factors, and 34% of the variance in PG and 59% of the variance in MD were owing to unique environmental factors. There was a substantial correlation between the genetic components of PG and MD (rA = 0.58), with 34% of the genetic variance for each disorder also contributing to that of the other. The best-fitting model estimated that 100% of the total overlap between PG and MD was genetic.
The correlation between PG and MD in middle-aged men appears to be largely influenced by overlapping genetic factors. Future research is needed to determine the extent to which these findings extend to other groups (eg, women), identify specific genes, and generate improved prevention and treatment strategies.
Pathological gambling (PG) and major depression (MD) frequently co-occur.1 Their co-occurrence has clinical implications. Suicidality has been associated with PG, and this association seems influenced by co-occurring disorders, including MD.2 Nonetheless, the relationship between PG and MD remains incompletely understood.
Twin studies allow for estimations of genetic contributions, unique environmental contributions (eg, different friends and individual experiences), and shared environmental contributions (eg, common familial experiences) to psychiatric disorders.3 The Vietnam Era Twin (VET) Registry data include DSM-III-R diagnoses for over 8000 twins and have been used to investigate genetic and environmental contributions to PG and frequently co-occurring disorders.4,5 These studies identified a heritable component to PG3 and genetic overlaps between PG and alcohol dependence as well as between PG and antisocial behaviors.4,5 These observations suggest that PG might be considered within the grouping of externalizing disorders.6,7 This categorization also appears to be supported by PG sharing similar phenomenological, diagnostic, and clinical features,8 including a disinhibited personality style or lack of constraint common to externalizing disorders, with substance use and antisocial disorders.9 However, prior studies supporting the clustering of common psychiatric disorders into internalizing and externalizing groups have generally not included measures of PG,6,7 and direct investigation of the most appropriate categorization of PG as an internalizing or externalizing disorder is needed.
A recent examination investigating environmental and genetic contributions to common mental health disorders categorized 7 psychiatric syndromes into 3 internalizing and 4 externalizing disorders. Within this model, 1 genetic factor loading strongly onto externalizing disorders was weakly associated with MD, and a second genetic factor strongly associated with MD and other internalizing disorders was weakly associated with externalizing disorders.7 Similarly, a unique environmental factor contributing most strongly to MD also contributed less strongly to externalizing disorders, particularly to alcohol dependence.7 No shared environmental factor loaded onto MD in the same model.7 Given the close relationship between alcohol dependence and MD10,11 as well as between alcohol dependence and PG,4 the question was raised whether a similar pattern of genetic and environmental contributions exists between MD and PG as between MD and externalizing disorders.
We used data from the VET Registry to investigate the following hypotheses. First, lifetime PG and MD frequently co-occur. Second, PG and MD correlate more substantially within monozygotic twins as compared with dizygotic twins. Third, genetic and unique environmental factors, but not shared environmental factors, contribute to the risk for PG and MD. Fourth, the factors contributing to the risk for PG correlate with those contributing to the risk for MD.
All of the participants were members of the VET Registry, a national sample of male twins, both of whom served during the Vietnam era (1965-1975) and were born between 1939 and 1957, inclusively. A detailed description of the VET Registry development and participants has been described.12 Of 10 253 eligible participants, 7869 (76.7%) of known zygosity (1125 single twins and 3372 twin pairs) were successfully interviewed in 1992 for the present study. Of the twin pairs, 1874 were monozygotic and 1498 were dizygotic as determined by responses to questionnaires regarding similarity of physical appearance and supplemental blood typing.13
The mean (SD) age of the sample was 42.0 (2.8) years. The racial/ethnic composition was 93.4% white (n = 7349), 6.2% African American (n = 489), and 0.4% other (n = 30). A high school education was reported by 31.5% (n = 2398) of the participants, and 64.8% (n = 4929) reported more and 3.6% (n = 274) reported less than a high school education. Of those with more than a high school education, 19.1% (n = 941) reported a 4-year college education and 10.4% (n = 515) reported a graduate or professional degree. Of the entire sample, 95.8% (n = 7296) reported full-time employment. A 1992 annual household income of less than $20 000 was reported by 16.8% of the participants (n = 1249), an income from $20 000 to $40 000 was reported by 49.1% (n = 3657), and an income greater than $40 000 was reported by 34.1% (n = 2539).
Lifetime diagnoses of PG and MD were determined using a computer-assisted telephone version of the Diagnostic Interview Schedule for DSM-III-R.14 Preliminary test/retest reliability for lifetime PG using the Diagnostic Interview Schedule is good (κ = 0.53; Wilson M. Compton, MD, MPE, personal communication) and comparable to other disorders.15 Project monitors oversaw lay interviewers who assessed participants following the acquisition of verbal consent, a method approved by the institutional review boards of the participating institutions. Symptoms of PG were only assessed in participants acknowledging having gambled 25 or more times in a year.
To examine the hypothesis that lifetime PG and MD frequently co-occur, odds ratios (ORs) for MD in subjects with PG were determined. Because twin pair data are not statistically independent observations, Stata software16 (Stata Corp, College Station, Tex) was used to compute Huber-White robust variance estimates to obtain 95% confidence intervals (CIs) for parameter estimates. Unadjusted ORs and ORs adjusted for sociodemographics (age, income, and education) and non-PG, non-MD psychiatric disorders (DSM-III-R externalizing disorders [alcohol abuse/dependence, nicotine dependence, drug abuse/dependence, and antisocial personality disorder] and internalizing disorders [posttraumatic stress disorder, panic disorder, and generalized anxiety disorder]) were examined by logistic regression analysis. A similar approach was used to determine the ORs for PG in subjects with MD.
To examine the hypothesis that PG and MD correlate more substantially within monozygotic twins than in dizygotic twins, tetrachoric correlations were examined using PRELIS 217 as described previously.4
To examine the hypotheses that genetic and unique environmental factors, but not shared environmental factors, contribute to the lifetime co-occurrence of PG and MD and that these factors correlate across disorders, bivariate models fitting the association between PG and MD were examined as described elsewhere.3,4 The variation in liability for each of the 2 diagnoses (PG and MD) was decomposed into that caused by additive genetic influences (A), shared environmental influences (C), and unique environmental influences plus measurement error (E) (Figure). The correlation between the 2 diagnoses was similarly partitioned into components resulting from additive genetic influences, shared environmental influences, and unique environmental influences plus measurement error (Figure). Using Mx software18 (Virginia Commonwealth University, Richmond), models were fit by the method of maximum likelihood. A series of nested submodels were each tested for their goodness of fit against a saturated model that placed no constraints on the elements of the estimated monozygotic and dizygotic twin correlation matrices. The most parsimonious model was selected as best fitting.19 The 95% CIs around parameter estimates for this model were examined to evaluate whether the genetic, shared environmental, or unique environmental contributions to PG and MD differed significantly from 0 and 1.
Lifetime criteria for PG were met by 112 (1.4%) of the subjects, and lifetime criteria for MD were met by 755 (9.6%) of the subjects. High rates of lifetime co-occurrence were observed between PG and MD. In individuals with PG as compared with those without PG, the unadjusted OR for MD was 4.06 (95% CI, 2.68-6.13). In a logistic regression model adjusting for sociodemographic measures, the OR for MD was 4.13 (95% CI, 2.63-6.50). After adjusting for sociodemographic and externalizing variables (alcohol abuse/dependence, nicotine dependence, drug abuse/dependence, and antisocial personality disorder), the OR remained elevated at 2.41 (95% CI, 1.46-4.00). This elevation persisted at an OR of 1.98 (95% CI, 1.14-3.45) after adjusting for multiple externalizing and internalizing disorders (Table 1). The unadjusted OR for PG in subjects with MD was elevated (OR = 4.12; 95% CI, 2.62-6.49) and remained elevated following adjustment for sociodemographic measures alone (OR = 4.12; 95% CI, 2.62-6.49) or with the psychiatric variables listed in Table 1 (OR = 2.03; 95% CI, 1.19-3.47).
Tetrachoric correlations (Table 2) demonstrated greater within-diagnosis concordance for both PG and MD in monozygotic twins as compared with that in dizygotic twins. Within-twin, cross-diagnosis correlations were comparable for monozygotic and dizygotic twins, and cross-twin, cross-diagnosis correlations were more substantial in monozygotic twins than in dizygotic twins (Table 2), suggesting overlapping genetic contributions to PG and MD.
The best-fitting bivariate model (Table 3) for the relationship between PG and MD had the correlations for shared and unique environmental factors (rC and rE, respectively) set to 0 and estimated the correlation for additive genetic factors (rA) at 0.58 (95% CI, 0.40-0.77). Parameter estimates for the additive genetic, shared environmental, and unique environmental contributions to the individual disorders suggest that there are significant genetic and unique environmental contributions to each disorder (Figure). Unique environmental factors accounted for 34% (95% CI, 20%-53%) of the variance for PG and 59% (95% CI, 49%-69%) of the variance for MD. Genetic factors accounted for 66% (95% CI, 47%-80%) of the variance for PG and 41% (95% CI, 30%-51%) of that for MD. Of the 66% genetic variance contributing to PG, approximately one third (22%; 95% CI, 8%-47%) was shared with MD and two thirds (44%; 95% CI, 19%-67%) was not. Of the 41% genetic variance contributing to MD, approximately one third (14%; 95% CI, 5%-30%) was shared with PG and two thirds (27%; 95% CI, 12%-43%) was not. In other words, 34% of the genetic variance for each disorder also contributed to that for the other. All of the overlap between PG and MD was accounted for by genetic factors.
Our hypothesis regarding the observation of high rates of lifetime co-occurrence between PG and MD was supported. The ORs for MD in association with PG when adjusting for sociodemographic measures alone (OR = 4.13) or in conjunction with externalizing and internalizing psychiatric disorders (OR = 1.98) remained significantly elevated. These ORs are comparable to that of 3.3 observed for MD in association with problem gambling in data from the St Louis site of the Epidemiologic Catchment Area (ECA) study.20 These similarities are observed despite differences between the 2 studies, including varying thresholds of gambling severity (problem gambling vs PG), eras of data acquisition (ECA data were collected in 1981 and VET data in 1992), and subject characteristics. Using a problem gambling threshold similar to that used in the ECA study reduced the unadjusted OR for MD from 4.06 to 2.49, indicating a weaker association between MD and subsyndromal gambling. The VET sample largely comprises white, highly educated, middle-aged, male twins with military service whereas ECA participants were oversampled for African Americans and included approximately equal numbers of men and women from the St Louis community. Together, these data suggest that the high rates of lifetime co-occurrence observed in the present study generalize beyond the cohort of VET participants. Unfortunately, other ECA sites and large national surveys of psychiatric disorders6 have generally not included gambling measures, contributing to a relative deficiency in our understanding of the relationship between PG and other psychiatric disorders.
The finding that, when controlling for other psychiatric disorders, the OR for MD in PG is reduced from 4.13 to 1.98 suggests that some of the risk for MD in individuals with PG is attributable to psychiatric disorders that frequently co-occur with PG. Among the disorders showing increased ORs in subjects with PG were alcohol abuse/dependence, drug abuse/dependence, antisocial personality disorder, panic disorder, and generalized anxiety disorder, findings largely consistent with those from population-based and clinical samples.1,20 Surprisingly, an elevated OR for nicotine dependence in association with PG was not observed, seemingly contrasting with findings from the St Louis ECA study20 and clinical samples.21 Specific features of the VET sample, such as those listed earlier or the sample’s high prevalence of nicotine dependence,22 might contribute to this finding. The association in men between MD and externalizing disorders such as alcohol abuse/dependence appears to be mediated by influences of antisocial personality disorder,10 and internalizing disorders like MD and anxiety disorders cluster together in models of psychiatric disorders.7 Thus, it is not surprising that modeling measures of externalizing and internalizing disorders reduces the OR between MD and PG. The finding that the OR remains significantly elevated following adjustment for sociodemographic and other psychiatric measures suggests that a considerable portion of the risk for MD in association with PG is not accounted for by sociodemographics and commonly co-occurring psychiatric disorders.
Our hypothesis regarding higher rates of lifetime co-occurrence of PG and MD in monozygotic twins as compared with the rates in dizygotic twins appears to be partially supported. Higher rates of within-diagnosis concordance for both PG and MD within monozygotic vs dizygotic twins are consistent with prior reports from the VET data and heritable contributions to each disorder.3,4,23 The observations of largely similar within-twin, cross-disorder correlations for monozygotic and dizygotic twins and substantially larger cross-twin, cross-disorder correlations in monozygotic twins as compared with those in dizygotic twins are consistent with the possibility of significant genetic contributions to the lifetime co-occurrence of PG and MD.
The most significant finding from the present study was the identification of substantial genetic overlap between PG and MD. The findings from the best-fitting bivariate model for PG and MD were consistent with our hypothesis that PG and MD would exhibit overlapping genetic influences but not shared environmental influences. Our hypothesis that the unique environmental factors for PG and MD would correlate was not supported by the best-fitting model. These findings suggest that the environmental factors influencing PG and MD differ.
A surprising finding was the magnitude of the genetic overlap between PG and MD. The genetic correlation between PG and MD (rA = 0.58) is comparable to or larger than that observed between PG and AD (rA = 0.35)4 and those between a multiple-threshold formulation of gambling and antisocial behaviors (correlation range, 0.40-0.47).5 Thus, the correlation between genetic factors for PG and MD is as substantial as or more substantial than any such correlation previously reported for PG and any other psychiatric disorder. The genetic correlation between PG and MD is as great as or greater than those reported for other disorders grouped together within the DSM-IV-TR, eg, those between different substance use disorders (correlation range, 0.26-0.54).24 Nonetheless, the finding that 34% of the genetic variance for each disorder also contributed to the other suggests that a substantial portion of the genetic liability for PG is not shared with MD and vice versa.
Our hypothesis regarding the relationship between environmental and genetic contributions to PG and MD were based on recent studies of internalizing and externalizing disorders6,7 and considerations regarding the categorization of PG within these groups. Internalizing disorders, associated with negative emotionality and including depressive and anxiety disorders, and externalizing disorders, associated with negative emotionality and a lack of constraint and including substance use disorders and antisocial behaviors, have clustered separately in factor analyses.6,7,9 Although PG has not formally been included in these models, PG shares a lack of constraint and features of impulsivity with externalizing disorders.25 Prior analyses of the VET data have identified genetic and environmental overlaps between PG and alcohol dependence as well as PG and antisocial behaviors, which is consistent with a categorization of PG as an externalizing disorder.4,5 Genetic and unique environmental contributions have previously been described to load relatively strongly within groups of externalizing and internalizing disorders and relatively weakly across the groups.7 As such, the genetic overlap presently observed between PG and MD is more substantial than hypothesized and raises questions regarding the nature of the relationship of PG to other internalizing disorders. Investigation of the structure of mental disorders has suggested reconsideration of classifications of other psychiatric disorders, eg, that posttraumatic stress disorder might be best considered a mood disorder rather than an anxiety disorder.26 Similar studies of PG are needed to guide its most appropriate categorization.
The finding of substantial genetic overlap between PG and MD has multiple research and clinical implications. First, more research is needed to identify specific genes involved in the pathophysiology of both MD and PG. Although candidate genes implicated in PG overlap with those implicated in MD,27 large-scale, genome-wide genetic investigations in PG have not yet been published to our knowledge and are needed. Gambling assessments should be included in genetic studies of MD and vice versa. Although many widely used diagnostic assessments, including the Structured Clinical Interview for DSM-IV, have not included PG measures, the recent availability of a Structured Clinical Interview for DSM-IV– compatible module for PG should minimize this concern.28 Second, the existence of genetic overlap suggests that PG and MD share biological mechanisms that could include abnormalities in stress-response pathways,29,30 biogenic amine systems,27,30 or impulse control networks31 identified in preliminary studies of PG and implicated in MD.32 Third, the findings suggest that treatments effective for MD might be helpful for individuals with PG.33 As pharmacological trials involving subjects with PG have generally excluded individuals with co-occurring MD, it will be important to evaluate the clinical utility of specific treatments, both pharmacological and behavioral, in individuals with co-occurring MD. Fourth, overlapping genetic features between PG and MD suggest that identification of specific allelic variants might help to target treatments. For example, variants of the μ-opioid receptor gene predict naltrexone treatment outcome for alcohol dependence.34 Similar approaches might guide treatment selection in co-occurring PG and MD. Fifth, the findings raise questions regarding the categorization of PG. It is currently classified in DSM-IV-TR as an impulse control disorder, although categorizations as an addiction without the drug or a disorder lying along an obsessive-compulsive spectrum or mood continuum have also been described.35 Inclusion of PG measures in studies of the structure of mental health disorders could guide the most appropriate categorization of PG.6,7
The current study has multiple limitations.3- 5 First, the VET Registry sample comprises middle-aged, male twins who are largely well educated and white. Thus, the findings might not generalize to other populations including adolescents, women, and other racial/ethnic groups. As the average age of the subjects was 42 years, late-life onset of PG or MD may be poorly represented. The absence of women seems particularly salient given their high rates of MD and sex-related differences in etiologies for depression36 and types/patterns of PG.37 Some observations suggest that the present findings might extend towomen. Specifically, the presence of multiple psychiatric disorders appears to increase the risk for MD largely similarly in women and men,11 and genetic and environmental contributions to externalizing and internalizing disorders are largely similar across sexes.7 However, these studies did not include gambling measures, and direct investigation is needed to assess the generalizability of the present findings. Sex-related differences in mood disorders and gambling disorders (eg, higher rates of the former in women and the latter in men) and the observation of sex-related differences in heritability of certain forms of gambling38 suggest a sex specificity and emphasize the need for additional investigation. Second, the sole availability of lifetime diagnoses precludes investigation of temporal co-occurrence of MD and PG. Third, as previously described,4,5 co-occurrence of multiple psychiatric disorders with PG confounds interpretation of the results. Fourth, the data were collected in 1992. The availability and social acceptance of legalized gambling have been increasing,8 thus limiting generalizability to the current gambling environment. Fifth, only subjects acknowledging having gambled 25 or more times within a year were queried regarding PG. Subjects with less frequent gambling might have met criteria for PG and been incorrectly classified, potentially altering the results. Sixth, DSM-III-R criteria were used to assess PG. Multiple changes have occurred in the diagnostic criteria, including the number of inclusionary criteria needed to fulfill PG (4 of 9 in DSM-III-R and 5 of 10 in DSM-IV-TR), and the specific criteria used. Perhaps most relevant to the present study, the current criterion of gambling “as a way of escaping from problems or of relieving a dysphoric mood (eg, feelings of helplessness, guilt, anxiety, depression)” was absent in the DSM-III-R criteria. This criterion is frequently endorsed among those with PG behavior,39 and further investigation is needed to determine the extent to which its inclusion might influence the current findings. Seventh, the bivariate models are based on assumptions (eg, the equal environment assumption3,13) that might overestimate genetic contributions. Consequently, environmental contributions to PG and MD might be underestimated.
The finding of genetic overlap between PG and MD highlights the need for closer examination of PG in individuals with MD and vice versa. Future investigation is needed not only to identify specific genetic factors that PG and MD have in common, but also to translate these findings into advances in prevention and treatment of the disorders. The identification of specific genes could facilitate the development of improved treatments, such as those targeting specific gene products.
Correspondence: Marc N. Potenza, MD, PhD, Connecticut Mental Health Center, 34 Park St, Room S-104, New Haven, CT 06519 (email@example.com).
Submitted for Publication: October 13, 2004; accepted March 17, 2005.
Funding/Support: This study was supported by grants MH60426, DA04604, and DA00167 from the National Institutes of Health, Bethesda, Md, a Research Enhancement Award Program grant from the US Department of Veterans Affairs, Washington, DC, and a MIRECC-VISN1 grant from the US Department of Veterans Affairs Mental Illness Research Education and Clinical Center. The US Department of Veterans Affairs has provided financial support for the development and maintenance of the VET Registry.
Acknowledgment: We gratefully acknowledge the continued cooperation and participation of the members of the VET Registry and their families. Without their contribution, this research would not have been possible. We gratefully acknowledge Joel Gelernter, MD, for helpful suggestions. Numerous organizations have provided invaluable assistance in the conduct of this study, including the Department of Defense (Washington, DC); National Personnel Records Center (St Louis, Mo); National Archives and Records Administration (College Park, Md); Internal Revenue Service (Washington); National Opinion Research Center (Chicago, Ill); National Research Council (Washington); National Academy of Sciences (Washington); and Institute for Survey Research, Temple University (Philadelphia, Pa).