Epidemiological studies have repeatedly shown that individuals who use cannabis are more likely to develop psychotic disorders than individuals who do not. It has been suggested that these associations represent a causal effect of cannabis use on psychosis, and that psychosis risk may be particularly elevated when use occurs in adolescence or in the context of genetic vulnerability. This study, however, does not support these hypotheses, suggesting instead that observed associations are more likely due to confounding by common vulnerability factors.
Observational studies have repeatedly linked cannabis use and increased risk of psychosis. We sought to clarify whether this association reflects a causal effect of cannabis exposure or residual confounding.
We analyzed data from 2 cohorts of twins who completed repeated, prospective measures of cannabis use (n = 1544) and cannabis use disorder symptoms (n = 1458) in adolescence and a dimensional measure of psychosis-proneness (the Personality Inventory for DSM-5 Psychoticism scale) in adulthood. Twins also provided molecular genetic data, which were used to estimate polygenic risk of schizophrenia.
Both cumulative adolescent cannabis use and use disorder were associated with higher Psychoticism scores in adulthood. However, we found no evidence of an effect of cannabis on Psychoticism or any of its facets in co-twin control models that compared the greater-cannabis-using twin to the lesser-using co-twin. We also observed no evidence of a differential effect of cannabis on Psychoticism by polygenic risk of schizophrenia.
Although cannabis use and disorder are consistently associated with increased risk of psychosis, the present results suggest this association is likely attributable to familial confounds rather than a causal effect of cannabis exposure. Efforts to reduce the prevalence and burden of psychotic illnesses thus may benefit from greater focus on other therapeutic targets.
…One powerful approach that can be helpful in testing for residual confounding involves comparing both monozygotic (MZ) and dizygotic (DZ) twins who differ in their cannabis exposure. Termed “discordant twin” or “co-twin control” analyses, this approach allows for examination of the effects of cannabis use while simultaneously controlling for all measured and unmeasured genetic and environmental factors shared between twins (McGue et al 2010). If cannabis is a causal contributor to long-term psychotic illness, both MZ and DZ twins who use more cannabis in adolescence than their co-twins should be more likely to experience psychosis. If this “twin difference” is not observed, it suggests the association between cannabis and psychosis is likely driven by confounding familial factors.
To date, only 2 studies have tested links between cannabis and psychotic symptoms using co-twin comparisons. Both reported that these associations were largely attributable to shared familial factors, but also that they observed evidence consistent with a small, independent, and potentially causal effect (Karcher et al 2019; Nesvåg et al 2017). However, these studies are also characterized by a shared set of limitations, which constrains the implications of their findings.
One limitation is that both studies used data from cross-sectional surveys of adult twins, which precluded tests focusing specifically on cannabis use occurring during the sensitive period of adolescence. A second limitation is that both studies used single, lifetime assessments of cannabis use and use disorder, which are subject to the many well-documented sources of bias that reduce the accuracy of retrospective measures (e.g., normal forgetting, revisionist recall). Methodological research suggests that this reduction in accuracy may be particularly problematic in a twin study context, as exposure measurement error tends to bias within-twin-pair estimates more dramatically than corresponding unpaired associations (Frisell et al 2012). Finally, the relatively coarse, binary measures of cannabis exposure used by these studies (including current use [yes/no], frequent use [>100×/not], and lifetime use disorder) are characterized by reduced variability relative to more continuous measures of cannabis use, and thus a reduced power to detect effects. Co-twin control studies of cannabis and psychosis that employ repeated, dimensional measures of cannabis use over time are thus needed to address these concerns and establish more accurate estimates of cannabis’s true causal effects.
The present study aimed to address these needs by examining associations between adolescent cannabis exposure and psychosis in a twin sample that combines data from 2 longitudinal cohort studies at the Minnesota Center for Twin and Family Research (MCTFR). In contrast to the few previous co-twin control studies, twins in these cohorts were assessed repeatedly using gold-standard, self-report and interview measures of cannabis use administered prospectively throughout adolescence. Using these measures, we created a continuous index measuring cumulative cannabis use prior to and during adolescence (“adolescent cannabis use index”) and a binary variable indicating presence or absence of a diagnosable cannabis use disorder (i.e., abuse or dependence) prior to and during adolescence.
…Does Adolescent Cannabis Exposure Predict Greater Adult Psychoticism Independent of Shared Environmental and Genetic Factors, Consistent With a Causal Effect? Results from co-twin control analyses are also presented in Table 4. Co-twin control models capitalize on twin differences to examine effects of cannabis exposure accounting for familial liability. In contrast to our individual-level analyses, these models indicated predominantly statistically-significant between-pair effects (estimates ranging from 0.14 to 0.20 for cannabis use index and from 0.43 to 0.59 for cannabis use disorder), suggesting an effect of preexisting, shared familial liability. They also indicated consistently small, nonsignificant within-pair effects (estimates ranging from −0.01 to 0.01 for cannabis use index and from −0.04 to 0.06 for cannabis use disorder), suggesting no effect of cannabis exposure (for full model results, see Supplemental Table 6).
…Do We Find Evidence Suggesting a Potential Causal Effect of Cannabis on Psychoticism in Genetically Vulnerable Individuals? Although we observed no statistically-significant within-pair associations suggesting a causal effect of cannabis exposure on psychoticism in the full analytic sample, this does not rule out the possibility that cannabis may increase psychoticism in subsets of particularly vulnerable individuals. Consequently, we next conducted analyses examining this possibility using one of the most obvious indicators of potential vulnerability: polygenic risk of schizophrenia. Results from these analyses are presented in Table 5. Our first set of models showed that, consistent with our expectations, twins with higher schizophrenia polygenic risk scores tended to score higher on our measure of adult psychoticism as well as its facet scales. Higher polygenic risk of schizophrenia was also associated with higher scores on our adolescent cannabis use index (β [95% CI] = 0.08 [0.02, 0.14], p = 0.014), and higher likelihood of meeting criteria for an adolescent cannabis use disorder (OR [95% CI] = 1.53 [1.11, 2.20], p = 0.010). Our second set of models indicated that schizophrenia polygenic risk and each measure of cannabis exposure both generally made incremental contributions to the prediction of scores on the adult psychoticism scale and its facets. Our third set of models tested the hypothesis that cannabis and polygenic risk interact such that individuals with higher levels of genetic risk are more affected by adolescent cannabis exposure. Interactions between the cannabis use index and polygenic risk in these models were all nonsignificant (βs [95% CIs] ranging from −0.04 [−0.10, 0.02] to 0.04 [−0.02, 0.10], all ps ≥ 0.213). Similarly, all interactions between cannabis use disorder and polygenic risk in corresponding models were also nonsignificant (βs [95% CIs] ranging from −0.04 [−0.20, 0.12] to 0.12 [−0.04, 0.28], all ps ≥ 0.141), except in the model predicting Perceptual Dysregulation (β [95% CI] = 0.17 [0.01, 0.34], p = 0.038). Nevertheless, because this single statistically-significant result would not survive correction for multiple testing, we conclude that results suggest little to no moderation of the effects of cannabis on Psychoticism by polygenic risk of schizophrenia overall.