Worldwide, the lifetime prevalence of suicidal ideation (SI), suicide planning, and attempt is estimated between 3.1 and 56.0%, between 0.9 and 19.5%, and between 0.4 and 5.1%, respectively, in adult populations (Nock et al., 2008a). Among adults reporting SI and a plan, 56% are estimated to have made an attempt, while 15.4% have made an attempt without a plan (Nock et al., 2008b). Suicide attempts are amongst the most powerful predictors of completed suicide (World Health Organization, 2014). Regardless of completion, suicide attempts exact a considerable economic burden via medical care accrued and lost productivity (American Foundation for Suicide Prevention, 2012). Alarmingly, between 2011 and 2012, suicide attempts rose by 2.4% and remain the 10^th leading cause of mortality in the U.S. (Xu et al., 2014).

Suicidal thoughts and behaviors (STB; ideation, planning, attempt) are strongly related to substance use behaviors, including cannabis involvement (Byrne et al., 2004; Calabria et al., 2010; Johns, 2001; Moore et al., 2007; Pompili et al., 2012), especially early (Byrne et al., 2004; Lynskey et al., 2004) and heavy cannabis use or cannabis use disorders (CUD) (Fergusson et al., 2002; Johns, 2001; Lynskey et al., 2004; Pedersen, 2008; Pompili et al., 2012; Van Ours et al., 2013). For instance, Silins and colleagues (2014) recently reported that suicide attempts were substantially increased (adjusted odds-ratio > 6) in young daily cannabis users.

Other studies suggest that the relationship between cannabis involvement and STB may be explained by shared risk and protective influences (Harris and Barraclough, 1997). For instance, in a longitudinal study of Swedish conscripts, the association between cannabis use and completed suicide was entirely explained by confounders, including other substance use and psychological adjustment (Price et al., 2009). Likewise, Wilcox et al. (2010) found that the relation between CUD and SI in college students disappeared when accounting for confounding factors such as depressive symptoms and maternal depression.

Another challenge is that items querying STB are frequently embedded in diagnostic interview sections assessing major depressive or bipolar disorder, such that only individuals reporting mood-related symptoms or episodes are presented with these questions. In addition, a majority of studies have disregarded intensity and duration of ideation (Joiner and Rudd, 2000) and the distinction between planned and unplanned attempts, even though their etiology may differ. In particular, planned attempts are more common in samples that require presence of dysphoric or anhedonic mood in the assessment of suicide (Simon et al., 2002) and their relationship with substance use may also vary. For example, Borges et al. (2000) showed that using one or more substances was related to suicide attempts without planning (SANP), but not to suicide attempts that were planned (SAP). The authors attributed this difference to the disinhibition hypothesis, which proposes that, when using drugs, inhibitions to make an impulsive attempt are reduced, therefore increasing the risk of suicide attempts (Mayfield and Montgomery, 1972; Rossow and Wichström, 1994), although whether attempts were in the context of substance use was not assessed. The finding is also consistent with the notion, as stated by Conner et al. (2007) in their study on alcohol dependent men and women, that SANP are related to impulsivity, while SAP are more related to depression, and are also more likely to result in completion (Harriss et al., 2005).

Cannabis involvement and SI are both influenced by genetic factors to a similar degree (h²=40–60%; Maciejewski et al., 2014; Verweij et al. 2010) with evidence for non-additive genetic influences on SI. However, little is known of the extent to which shared genetic factors contribute to their comorbidity. One study (Lynskey et al., 2004) found that CUD was associated with SI and suicide attempts, even in identical twin pairs who shared 100% of their genetic background. The twin with cannabis dependence was at 2.9 and 2.5 greater odds of SI and suicide attempt relative to their genetically-related nondependent co-twin, suggesting that individual-specific environmental factors that are correlated across cannabis involvement and STB but are not shared by members of a twin pair were important. However, the extent of the genetic and environmental overlap between cannabis involvement and STB was not examined.

The present study expands upon this prior research by (a) studying varying levels of cannabis involvement, including use and use disorders; (b) examining both SI and suicide attempt separately; (c) expanding the definition of suicide attempts to include planning and (d) estimating the magnitude of genetic overlap between cannabis involvement and STB. We hypothesized that cannabis involvement would be associated with SI and suicidal attempts in a dose-response fashion, however, associations with the latter would only be restricted to those reporting SANP. Furthermore, we expected moderate genetic and individual-specific environmental correlations to contribute to the association between cannabis involvement and SI.

Our results confirm previous studies, which have shown support for an association between cannabis involvement and STB (e.g., Fergusson et al., 2002; Moore et al., 2007; Pedersen, 2008). In addition, unlike some other studies, associations in our sample were not explained by confounding measures (Price et al., 2009; Wilcox et al., 2010). Differences between our study and others may be related to sample characteristics, measurement of ideation/attempt, or the possible inclusion or exclusion of certain covariates. To identify potential contributions of differences in measurement we differentiated SI by duration (>1 day) but associations with cannabis involvement did not vary. In contrast, we did show a difference between suicide attempts with and without a plan: cannabis involvement was related to SANP only, confirming previous studies on number of substances used (Borges et al., 2000) and alcohol dependence (Conner et al., 2007). Even though SANPs are found to be less likely to result in completed suicide (Harriss et al., 2005), suicidal attempts, in and of themselves are amongst the leading contributors to economic burden (Palmer et al., 1995) and are powerful predictors of later completed suicide (Harris and Barraclough, 1997; Suominen et al., 2004), emphasizing the need for effective prevention strategies.

The relationship between cannabis involvement and STB could be explained in multiple ways. For example, the self-medication hypothesis states that individuals use psychoactive substances to reduce negative affective states, such as STB (Khantzian, 1985, 1997). However, longitudinal studies have found little support for this hypothesis (Harris and Barraclough, 1997; Van Ours et al., 2013). Alternatively, the impaired functioning theory hypothesizes that cannabis involvement causes STB (Newcomb et al., 1999). Physical, psychological, or emotional functioning is impaired by the repeated and, often, early onset, use of cannabis, resulting in a higher risk of suicidal behaviors. Some longitudinal studies indeed showed a significant relationship between early cannabis use and later risk of suicidal behaviors (e.g., Lynskey et al., 2004; Van Ours et al., 2013; Wilcox and Anthony, 2004). In the present study, while we did not have data on age at first CUD symptom, median age at first cannabis use coincided with the median age at first SI (18 years) in cannabis users and only preceded first suicide attempt by a year. Hence, cannabis involvement likely occurred contemporaneously. Finally, as stated by the disinhibition theory, cannabis intoxication prior to an attempt could also directly lead to an increased risk of suicide attempt. Related studies (Borges et al., 2000; Conner et al., 2007) allude to the possible role of facets of disinhibited behavior as mediators. While we did not assess substance-induced disinhibition or trait level aggression in our study, the inclusion of conduct disorder (which encompasses both facets to some degree) did not attenuate the association. In addition, only 56 individuals (13.7% of attempters) indicated that their suicide attempt was related to co-occurring drug use. A more likely alternative is that both cannabis involvement and STBs are influenced by and related to a common set of risk and protective influences and that any observed association is due to these confounding effects (Price et al., 2009).

We also found that the link between cannabis involvement and SI was strongly influenced by shared genetic influences. This is consistent with pre-clinical evidence regarding the role of the endocannabinoid system in both the behavioral effects of cannabis and the regulation of depressive and anhedonic mood states (Gorzalka and Hill, 2011). One prior study of 277 discordant twins (Lynskey et al., 2004; utilizing data from Sample 1) found that CUD was associated with a 2.5–2.9 increased OR of SI and attempt. In other words, even when individuals were matched for their segregating genes and early family environment, a history of CUD significantly increased the likelihood of STB, suggesting the possibility of causal/individual-specific factors in this association. We found support for such individual-specific factors influencing both cannabis involvement and SI in male and female twins (rE=0.21). Furthermore, due to the smaller sample size, twin models with SANP could not be fit to the data.

Finally, due to the relatively small number of individuals reporting SAP and SANP, we did not further probe whether intent to die, an aspect of self injury that is associated with risk influences similar to suicide completion, including a greater than 2-fold increase in drug use disorders, modified these associations (Nock and Kessler, 2006). Of those reporting SAP and SANP, 60% and 40% reported intent respectively. Stratifying by intent alone (no attempt, attempt without intent and attempt with intent) did not yield significant associations with any level of cannabis involvement (OR = 0.78–1.17, p > 0.05). Similarly, associations with SAP remained non-significant when stratified by intent. For SANP, while the point estimates were no longer statistically significant, the magnitude of the ORs suggested that while intent may enhance the association, SANP without intent was also associated with cannabis involvement. For instance, the OR for SANP with intent was 2.89 [95% C.I. 0.94–8.88] relative to an OR of 1.98 [95% C.I. 0.88–4.47] for SANP without intent.

The present study has some limitations. First, as age of onset of CUD symptoms was not queried and age of onset of STBs was limited to SI and attempt, it is not possible to make causal inferences. Second, with self-report data, participants could be unwilling to report or unable to recall cannabis use and CUD symptoms or STB. Third, the most severely affected (i.e., those completing suicide) are not represented. Fourth, an abbreviated version of the CUD assessment was used. When analyses were conducted with the full CUD assessment in sample 2 and the abbreviated assessment in sample 1, results remained largely unchanged (available upon request). Fifth, we utilized a sample of twins for epidemiological analyses. However, it is important to note that twins, after statistical accommodation of the clustered nature of their data (as done in this study), are representative of the general population (Kendler et al., 1995; Pulkkinen et al., 2003). Finally, despite the statistically significant association with suicide ideation as well as SANP, cannabis involvement was not the strongest correlate of STBs. In addition to cannabis involvement, substance use and misuse, as a whole, different psychopathologies such as MDD and conduct disorder, and childhood sexual abuse should be considered in the etiology of STBs. Relatedly, despite the inclusion of several important confounders, the possibility that an unmeasured covariate was responsible for this association cannot be excluded.

As SI was reported by 26.2% in the present study and SANP are particularly difficult to prevent (Conner, 2004), research on possible correlated factors is important. In the present study, cannabis involvement was related to SI regardless of duration and to SANP, even when controlling for possible confounders. However, the odds ratio was modest and results from some previous studies differed from ours, emphasizing the need for further replication. If cannabis use and endorsing CUD symptoms exacerbate the likelihood of these behaviors, then more well-designed studies are required to unpack the etiological mechanisms underpinning this association. Such research is particularly important in light of current developments regarding the legalization of cannabis use in the U.S. in the states of Washington and Colorado and also in other countries. As 3.9% of first-time cannabis users have been found to develop cannabis dependence within 24 months (Chen et al., 2005), with risk increasing with chronic use, the study of potential harmful correlates and consequences of cannabis use and escalating cannabis involvement should be made a priority.