Michael T. Lynskey, Anne L. Glowinski, Alexandre A. Todorov, Kathleen K. Bucholz, Pamela A. F. Madden, Elliot C. Nelson, Dixie J. Statham, Nicholas G. Martin, Andrew C. Heath. Major Depressive Disorder, Suicidal Ideation, and Suicide Attempt inTwins Discordant for Cannabis Dependence and Early-Onset Cannabis Use. Arch Gen Psychiatry. 2004;61(10):1026–1032. doi:10.1001/archpsyc.61.10.1026
Previous research has reported both a moderate degree of comorbidity
between cannabis dependence and major depressive disorder (MDD) and that early-onset
cannabis use is associated with increased risks for MDD.
To examine whether associations between both lifetime cannabis dependence
and early cannabis use and measures of MDD, suicidal ideation, and suicide
attempt persist after controlling for genetic and/or shared environmental
Cross-sectional survey of twin pairs discordant for lifetime cannabis
dependence and those discordant for early cannabis use.
General population sample of twins (median age, 30 years).
Two hundred seventy-seven same-sex twin pairs discordant for cannabis
dependence and 311 pairs discordant for early-onset cannabis use (before age
Main Outcome Measures
Self-report measures of DSM-IV–defined
lifetime MDD, suicidal ideation, and suicide attempt.
Individuals who were cannabis dependent had odds of suicidal ideation
and suicide attempt that were 2.5 to 2.9 times higher than those of their
non–cannabis-dependent co-twin. Additionally, cannabis dependence was
associated with elevated risks of MDD in dizygotic but not in monozygotic
twins. Those who initiated cannabis use before age 17 years had elevated rates
of subsequent suicide attempt (odds ratio, 3.5 [95% confidence interval, 1.4-8.6])
but not of MDD or suicidal ideation. Early MDD and suicidal ideation were
significantly associated with subsequent risks of cannabis dependence in discordant
dizygotic pairs but not in discordant monozygotic pairs.
Comorbidity between cannabis dependence and MDD likely arises through
shared genetic and environmental vulnerabilities predisposing to both outcomes.
In contrast, associations between cannabis dependence and suicidal behaviors
cannot be entirely explained by common predisposing genetic and/or shared
environmental predispositions. Previously reported associations between early-onset
cannabis use and subsequent MDD likely reflect shared genetic and environmental
vulnerabilities, although it remains possible that early-onset cannabis use
may predispose to suicide attempt.
Given the high and apparently increasing prevalence of cannabis useamong youth,1 there have been continuing concernsabout whether the use of this drug may heighten risks of psychiatric illnessand, in particular, depressive or suicidal behaviors.2- 4 Clinicalstudies have reported that rates of depression are elevated among those seekingtreatment for cannabis dependence5 and, conversely,that rates of cannabis use and abuse/dependence are elevated among those beingtreated for depression6,7 andamong those making a serious suicide attempt.8 Althoughreferral biases limit the generalizability of these results, the conclusionthat cannabis abuse/dependence and depressive disorders co-occur at greaterthan expected rates has been confirmed by general population surveys.9- 11 In addition, a seriesof recent longitudinal studies has indicated that rates of subsequent depressionare elevated in young people who use cannabis12- 14 andin adults who meet criteria for cannabis abuse or dependence.15,16
Further research has also considered the possibility that comorbiditybetween cannabis use/dependence and mental health problems may reflect processeswhereby depression and other mental health problems heighten risks of substanceuse and abuse/dependence. Specifically, the self-medication hypothesis17 posits that substance use may develop in responseto attempts to ameliorate aversive experiences associated with mental healthproblems. Although intuitively appealing, this hypothesis has remained controversialand has received only limited support in the empirical literature; recentlongitudinal studies have found no significant associations between earlydepression and later risks of regular cannabis use or dependence.14,16
Thus, while there appears to be little support for the hypothesis thatearly depressive or suicidal behaviors increase risks for the developmentof cannabis dependence, recent findings have implicated cannabis use as havinga causal role in the development of depression.2 However,pivotal to this conclusion is the assumption that cannabis users and nonusersare similar, if not identical, on a range of measures predisposing to bothcannabis use and internalizing disorders. The previous studies have attemptedto address this issue by including a range of measured covariates and controllingfor these using regression or similar techniques. While Fergusson et al12 used methods of fixed-effects regression analysis,which potentially control for nonobserved sources of confounding (both geneticand environmental), no previous study has specifically assessed the influenceof genetic factors on the association between early-onset cannabis use andsubsequent risks of depression. Nonetheless, the existing evidence indicatesthat both cannabis use and dependence18- 20 andmeasures of depressive20,21 andsuicidal behaviors22,23 are moderatelyheritable. Recent findings that the genetic factors associated with cannabisuse and those associated with depression are moderately correlated20 further support the hypothesis that some componentof the apparent comorbidity between cannabis dependence and depressive andsuicidal behaviors may be due to shared or correlated genetic vulnerabilities.
One approach to addressing the issue of preexisting differences betweenthose who are cannabis dependent and those who are not involves the use ofa co-twin control design in which rates of depressive and suicidal behaviorsare compared within twin pairs where 1 twin is cannabis dependent and theco-twin is not. In the case of dizygotic (DZ) twins raised together, thesemethods provide optimal control for shared environmental influences, whilefor monozygotic (MZ) twins, they provide control for both genetic and sharedenvironmental factors.
Against this background, we report on a series of co-twin control analysescomparing rates of lifetime DSM-IV major depressivedisorder (MDD) and suicidal thoughts and behaviors in same-sex twin pairsdiscordant for (1) lifetime cannabis dependence and (2) the early initiationof cannabis use (prior to age 17 years). Additionally, we conducted a seriesof co-twin control analyses comparing rates of cannabis dependence in twinpairs discordant for MDD and suicidal ideation before age 17 years.
Interviewees were members of the young adult cohort of the AustralianTwin Register, a volunteer twin panel, born 1964 through 1971.18,24 Nearlyall were first registered with the panel between 1980 and 1982 by their parentsin response to approaches either through Australian school systems or viamass media appeals. The data presented in this report are derived from responsesto a single telephone interview during the period 1996 through 2000 when themedian age of the sample was 30 years (range, 24-36 years). Informed consentwas obtained from participants prior to administering the interviews, as approvedby the institutional review boards of Washington University School of Medicine,St Louis, Mo, and the Queensland Institute of Medical Research, Brisbane,Queensland, Australia.
Of 4010 pairs that could be traced, interviews were completed with bothmembers of 2765 pairs (69% pairwise response rate) and 1 member of another735 (78% individual response rate). The analyses in this article are basedon 4 subsets of the sample:
Same-sex twin pairs discordant for cannabis dependence:Individuals reporting using cannabis at least monthly were questioned, usingitems from the Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA),25 about whether they had experienced symptoms of cannabisdependence (using more frequently/for longer periods than intended; needinglarger amounts to achieve an effect [tolerance]; continued use despite usecausing emotional problems; recurrent desire to cut down on use). Those reporting2 or more of these criteria were classified as meeting lifetime criteria forcannabis dependence. While this measure did not provide formal DSM-IV26 criteria, previous analysesexploring the validity of these modified criteria indicated that they hadboth excellent sensitivity (96.7%) and specificity (94.6%) when compared with DSM-IV criteria.18 Elevenpercent of the sample (418 men [15.1%] and 269 women [7.8%]) met these criteriafor lifetime cannabis dependence, and 277 (40.2%) of these were from same-sexpairs discordant for cannabis dependence. This subset of 277 twin pairs (70female MZ pairs, 67 male MZ pairs, 55 female DZ pairs, and 85 male DZ pairs)was used in the current analyses.
Same-sex twin pairs discordant for early cannabisuse: 863 (13.8%) members of the sample reported initiating cannabis use beforeage 17 years.27 Three hundred eleven (36.1%)of these were from same-sex twin pairs in which their co-twin had not usedcannabis by age 17 years. This subset of the sample (74 female and 62 maleMZ pairs; 84 female and 91 male DZ pairs) was used in a second set of analyses.
Same-sex twin pairs discordant for early MDD: TheSSAGA,25 modified for telephone administration,was used to collect information on full DSM-IV criteriafor MDD, including age of onset. Two hundred seventy-four members of the sample(4.4%) reported onset of MDD before age 17 years. One hundred fifty-six (56.9%)of these were from same-sex twin pairs in which their co-twin did not meetcriteria for MDD before age 17 years. This subset of the sample included 63female and 38 male MZ pairs and 24 female and 31 male DZ pairs.
Same-sex twin pairs discordant for early suicidalideation: Four hundred seventy-seven members of the sample (7.6%) reportedsuicidal ideation before age 17 years. Two hundred fifty-seven (53.9%) ofthese were from same-sex twin pairs in which their co-twin did not reportexperiencing suicidal ideation before age 17 years. This subset of the sampleincluded 80 female and 54 male MZ pairs and 68 female and 55 male DZ pairs.
A structured diagnostic interview designed for genetic studies on alcoholism,the SSAGA,25 was adapted for telephone usewith an Australian sample and updated for DSM-IV diagnosticcriteria.26 The interview also included assessmentsof sociodemographic factors, childhood family environment, and experiencingchildhood sexual abuse.24 These measures aredescribed next.
The modified SSAGA collected information on full DSM-IV criteria for MDD as well as separate information on whethersubjects had ever contemplated or attempted suicide. Information on age atonset of these conditions was also collected.
A number of control variables were included in the analyses.
Psychiatric disorders: DSM-IV26 conduct disorder, alcohol dependence,and nicotine dependence were assessed using the modified SSAGA, and diagnoseswere assigned by computer algorithm.
Early tobacco use: Subjects who reported smokingat least 1 day a week for a period of 3 weeks or more before age 17 yearswere classified as early tobacco users.
Early regular alcohol use: Subjects who reportedthat they started drinking alcohol at least once a month for a period of 6months or more before age 17 years were classified as early regular alcoholdrinkers.
Childhood sexual abuse: Respondents were also askeda series of questions concerning their exposure to unwanted sexual contact,sexual molestation, or rape. A composite measure was constructed, which classifiedrespondents who reported any such experiences before the age of 18 years ashaving a history of childhood sexual abuse.24
All statistical analyses were conducted using Stata.28 Conditionallogistic regression models were fitted to test for excess risk of MDD, suicidalideation, and suicide attempt in (1) individuals meeting criteria for cannabisdependence and (2) individuals who commenced cannabis use before age 17 years.Further models tested for excess of cannabis dependence in (1) individualsmeeting DSM-IV criteria for MDD before age 17 yearsand (2) individuals reporting suicidal ideation before age 17 years. The significanceof the interactions between exposure (cannabis dependence, early measuresof early cannabis use, early MDD, and early suicidal ideation) and both twinpair zygosity and sex was tested and, when nonsignificant, data were pooledacross zygosity and across sex. Analyses were repeated including the controlvariables described earlier. For analyses of early-onset cannabis use, informationon onset and duration were included so that only those reporting experiencingan episode of MDD or contemplating/attempting suicide after age 17 years wereclassified as positive for those outcomes. Additionally, in these analyses,corresponding measures of MDD and contemplated/attempted suicide before age17 years were included as potential covariates in the conditional logisticregression analyses. Stepwise regression with backward selection was conductedwith the measure of either cannabis dependence or early cannabis use forcedinto the model. Hence, for models in which there were no significant covariates,the unadjusted and adjusted odds ratios are identical.
Finally, a series of bivariate genetic Cholesky models were fitted tothe data on both onset of cannabis use and lifetime cannabis dependence andrisks of MDD to assess the extent of correlation between genetic influencesassociated with cannabis use and those associated with MDD. These models werefitted to the data using the program Mx,29 usingthe method of maximum likelihood estimation.
Power was estimated using the conditional logistic model, using parametersderived from our sample. Our results indicated that power would be 80% orbetter for an odds ratio (OR) higher than approximately 1.9 for the analysesof cannabis dependence and depression, ORs higher than approximately 2.4 forcannabis dependence and suicidal ideation, and ORs higher than approximately2.8 for cannabis dependence and attempted suicide. When early cannabis usewas used as the exposure variable, power was 80% or better for ORs of approximately1.3 or higher with depression, ORs of approximately 1.6 or higher with suicidalideation, and ORs of 2.8 or better with suicide attempt. Power was more than80% for ORs higher than approximately 2.2 for the analyses of early suicidalideation and for ORs higher than approximately 2.4 for the analyses of earlydepression.
Table 1 presents estimates ofthe lifetime prevalence (percentage) of MDD, suicidal ideation, and attemptedsuicide for those meeting lifetime criteria for cannabis dependence and fortheir co-twins who did not meet these criteria. The strength of the associationbetween cannabis dependence and these outcomes was assessed using the conditionalOR, both before and after control, for a range of measured covariates. ForMDD, these estimates are shown separately for MZ and DZ twin pairs, becausethere was evidence of a significant (P<.05) interactionbetween zygosity and cannabis dependence. However, for the remaining 2 measures,these interactions were nonsignificant (P>.20) anddata were therefore pooled across zygosity. The results in Table 1 show:
Relative to their nondependent co-twins, thosewho met criteria for cannabis dependence had elevated lifetime rates of MDD,suicidal ideation, and attempted suicide. The unadjusted conditional ORs indicatedthat, in individuals who were cannabis dependent, the odds of MDD, suicidalideation, and attempted suicide were 1.3 to 3.4 times higher than in theirco-twins who were not cannabis dependent. Importantly, there was evidencethat cannabis dependence was associated with increased risks of MDD in DZtwin pairs discordant for cannabis dependence but not in MZ twins.
After further controlling for known risk factorsassociated with increased risk of MDD, suicidal ideation, and attempted suicide,rates of these outcomes were still significantly elevated in individuals meetingcriteria for cannabis dependence.
The results of analyses comparing rates of MDD, suicidal ideation, andattempted suicide (all required to have onset after age 16 years) in twinpairs discordant for initiation of cannabis use before age 17 years are summarizedin Table 2. Again, there was a significantinteraction between early cannabis use and zygosity for risk of MDD (but notfor risks of either suicidal ideation or attempt), and therefore, data arepresented separately by zygosity for this outcome. The results in Table 2 show:
Relative to their co-twins who had not used cannabisby age 17 years, those who had used cannabis by this age had significantlyelevated rates of both suicidal ideation and suicide attempt. However, inthe case of MDD, this association was significant only for discordant DZ pairs.
Controlling for known risk factors substantiallyreduced these associations. After such control, there was no significant associationbetween early cannabis use and either MDD or suicidal ideation. However, comparedwith nonearly users, those who initiated cannabis use before age 17 yearshad 3.5 times the odds of reporting a subsequent suicide attempt.
Application of techniques of genetic model fitting confirmed the keyfindings of these analyses; there were significant correlations with risksof MDD with genetic liability for age of onset of cannabis use (men, r = 0.23 [95% confidence interval (CI), 0.05-0.43];women, r = 0.35 [95% CI, 0.18-0.57]) andlifetime cannabis dependence (men, r = 0.44[95% CI, 0.17-1.00]; women, r = 0.69 [95% CI, 0.30-1.00]), confirmingthat a significant component of the comorbidity between cannabis use and MDDarose from the influence of shared or correlated genetic liabilities.
Table 3 compares rates of lifetimecannabis dependence between members of twin pairs discordant for early MDDor for early suicidal ideation. Comparisons are shown separately for MZ andDZ twin pairs because the interaction between zygosity and early internalizingfor risk of cannabis dependence was significant for MDD (P<.05) and marginally significant for suicidal ideation (P<.10).
Relative to their co-twins who did not report MDD or suicidal ideationbefore age 17 years, rates of cannabis dependence were significantly elevatedin (1) DZ twins who met DSM-IV criteria for MDD beforeage 17 years (OR, 9.5 [95% CI, 2.21-40.78]) and (2) DZ twins who reportedsuicidal ideation before age 17 years (OR, 5.5 [95% CI, 1.90-15.96]). In contrast,there were no significant differences in rates of cannabis dependence in MZtwins discordant for either early MDD or early suicidal ideation.
Analyses including statistical control for observed covariates are notreported herein, because once the effects of internalizing problems were takeninto account, there were no statistically significant associations betweenthese measures and later risks of cannabis dependence. Therefore, the additionof these covariates did not materially influence our conclusions.
Results of our co-twin control analyses indicated a moderate degreeof comorbidity between cannabis dependence and measures of mental health:individuals who met lifetime criteria for cannabis dependence had odds ofMDD, suicidal ideation, and attempted suicide that were 1.3 to 3.4 times higherthan their non–cannabis-dependent co-twins. These results are consistentwith previous findings that rates of depressive and suicidal behaviors areelevated in those who meet criteria for cannabis dependence.9- 11 Importantly,however, there was evidence that a substantial component of the associationbetween cannabis dependence and MDD could be attributed to shared geneticvulnerabilities. Specifically, the finding of a significant interaction betweenzygosity and cannabis dependence on risks of MDD, with cannabis dependencebeing associated with significantly elevated risks of MDD (relative to theirnon–cannabis-dependent co-twin) in DZ but not in MZ twins, implies thatshared or correlated genetic vulnerabilities make substantial contributionsto the comorbidity between cannabis dependence and MDD. In contrast, the findingof no significant interaction between zygosity and either early-onset cannabisuse or lifetime cannabis use on risks of suicidal behaviors implies that geneticinfluences are relatively less important in explaining the observed associationsbetween these sets of behaviors.
Further analyses indicated that individuals who initiated cannabis usebefore age 17 years were at increased risk of subsequently attempting suicide,but there was no association between early cannabis use and subsequent measuresof either MDD or suicidal ideation. These results provide only limited supportfor the hypothesis that early cannabis use is a risk factor for the subsequentdevelopment of MDD and suicidal behaviors, as has been reported by a numberof previous studies.12- 16 Theapparent discrepancy between our results and previous results may be partiallydue to the fact that, by using co-twin control methods, we were able to providemore effective control for background predisposing factors (both genetic andshared environmental) than had previously been attained. Indeed, the findingof a significant interaction between early cannabis use and later risks ofMDD again suggests that a substantial component of this association may bedue to shared genetic vulnerabilities. While our results are consistent withthe hypothesis that early cannabis use increases risks for subsequent suicideattempt, we cannot exclude the possibility that both these outcomes resultfrom aspects of the nonshared environment that were not controlled in ouranalyses. In addition to the pharmacological effects of cannabis, there area number of other potential mechanisms that may underlie any association betweencannabis use and risks of suicidal behaviors. In particular, the social contextwithin which cannabis is used and obtained may also promote access to otherdrugs27 and a range of life events and circumstances,including risks of legal sanctions against possession and use of cannabis,30 that may predispose cannabis users to suicidal behaviors.
Finally, we found that early-onset MDD and early-onset suicidal ideationwere both associated with increased risks for the development of cannabisdependence in DZ twins but not in MZ twins from discordant pairs. These resultsqualify recent reports that early internalizing behaviors are not associatedwith increased risks for cannabis dependence2,14,16 byindicating that there may indeed be an association, albeit one that is largelyexplained by common genetic, as opposed to environmental, factors predisposingto both outcomes. Again, this result is consistent with recent evidence ofa moderate overlap in the heritable factors associated with cannabis dependenceand those associated with MDD.20
These analyses also identified a range of other risk factors, includingchildhood conduct disorder, exposure to childhood sexual abuse, and earlysubstance use, that were associated with increased risks for the developmentof MDD and suicidal behaviors. These findings are consistent with an accumulatingliterature demonstrating the effect of these factors on risks for psychopathology31,32 and highlighting that end pointssuch as suicide attempt or MDD are often the outcomes of a life path characterizedby multiple disadvantages and dysfunction in which a range of adverse exposures,which by themselves increase risks only slightly, accumulate to increase riskssubstantially.33- 36
By quantifying the extent to which associations between both early-onsetcannabis use/dependence and measures of MDD, suicidal ideation, and suicideattempt can largely be explained by common genetic and/or environmental riskfactors, our analyses highlight that the development of these behaviors isintertwined. It therefore appears likely that risk-factor based interventionsmay have benefits in reducing risks of a range of outcomes.32,37 However,while our results indicate that comorbidity between these disorders largelyarises from the influence of multiple risk factors on both sets of outcomes,it remains possible that continued cannabis use may exacerbate the prognosisfor those with existing depression and other conditions. Although this issuehas received relatively little attention, there is a parallel literature indicatingthat continuing cannabis use may exacerbate the clinical course of those diagnosedwith psychosis.38,39
In interpreting these results, a number of potential limitations shouldbe considered. First, the sample was based on a volunteer sample of twins,and the nature of this sample may potentially have introduced some selectionbiases into the sample. Nonetheless, based both on relatively high recruitmentand retention rates and on previous comparisons with characteristics of theAustralian population, we are confident that any biases introduced by thevolunteer nature of the sample would be slight. Second, we have relied onthe use of cross-sectional data and retrospective reports of lifetime behaviors,including age of onset for different behaviors. Finally, our measure of early-onsetuse did not take into account the frequency of early use, and it is possiblethat increased risks for the development of MDD and other problems may occuronly in those who use cannabis frequently at young ages. Indeed, there issome evidence to support this conjecture to the extent that previous researchhas demonstrated a dose-response relationship between the frequency of earlycannabis use and risks for mental health problems.12 Whilewe have not been able to address this issue fully, the current analyses doaddress a pivotal issue in public policy: the extent to which early-onsetcannabis use is associated with increased risks and, concomitant with this,the extent to which delaying the onset of cannabis use may decrease subsequentrisks. While acknowledging these potential limitations, our results highlightthat early cannabis use—or lifetime cannabis dependence—constitutesonly one of many potential risk factors predisposing to MDD and suicidal behaviors.
Correspondence: Michael T. Lynskey, PhD,Missouri Alcoholism Research Center, Department of Psychiatry, WashingtonUniversity School of Medicine, 40 N Kingshighway, Suite 1, St Louis, MO63108 (firstname.lastname@example.org).
Submitted for Publication: October 20, 2003;final revision received March 26, 2004; accepted April 21, 2004.
Funding/Support: This study was supported bygrants AA07728, AA09022, AA10249, AA11998 (Dr Heath), AA 12640, DA 14363,DA 14632 (Dr Bucholz), DA00272, DA12854 (Dr Madden), AA-12232 (Dr Todorov),and AA00277 (Dr Nelson) from the National Institutes of Health, Bethesda,Md, and grants 951023 and 981351 from the National Health and Medical ResearchCouncil, Canberra, Australia (Dr Martin).
Acknowledgment: We thank the Australian TwinRegistry and the twins for participating in this research.