Use of Genetically Informed Methods to Clarify the Nature of the Association Between Cannabis Use and Risk for Schizophrenia
Nathan A. Gillespie, Kenneth S. Kendler
JAMA Psychiatry, Published online November 4, 2020
Cohort studies and meta-analyses have documented a robust association between cannabis use, heavy use, and misuse with future risk of schizophrenia.1-3 Despite adjusting for covariates, including current psychotic symptoms, other psychopathology, and social integration,(1) the ability of these models to determine the degree to which cannabis causes schizophrenia is limited and dependent on their ability to capture all relevant confounders. When evaluating efforts to reduce cannabis use as a means of preventing schizophrenia, the proportion of this association that is causal is critical.
Given the high heritability of schizophrenia,we reviewed reports that relied on4geneticmethods (Table)
capable of addressing the nature of the cannabis schizophrenia association. We evaluated 3 hypotheses :
(1) it is entirely causal,
(2) it is partly causalandpartly confounded by genetic/familial effects and/or reverse causation,
or (3) it is entirely noncausal.
(We are unable to review the literature regarding short-term psychiatric effects of cannabis administration.(9)) Confounding here refers to genetic risks that increase the probability of both using/misusing cannabis and schizophrenia, thereby explaining at least part of the association. In this example, reverse causality refers to a theoretical underlying mechanism in which schizophrenia liability or symptoms increase the risk of using cannabis.The first 2 methods are natural experiments, discordant relative designs, and mendelian randomization, that directly evaluate each hypothesis. The 2 other methods, linkage disequilibrium score regression (LDSR)andpolygenic risk scores (PRSs), measure genetic associations, which, although less definitive, provide evidence of correlated genetic risks that undermine the plausibility of hypothesis 1.Each hypothesis generates distinct expectations under each
method (Table). Therefore, our approach determines which of the hypothetical expectations best fit the aggregate results.By examiningmultiplemethods with different strengths and limitations,we attempt to triangulate the observed results hoping thereby to increase confidence in our conclusions.
Discordant relative designs examine declines in exposure- outcome associations while systematically increasing control over confounding genetic or environmental risks shared between relatives of increasing closeness.(4) If hypothesis 1 is correct, then the expected odds ratio (OR) for the cannabis schizophrenia association should be the same for unrelated pairs in the general population and in relative pairs including full sibling pairs and monozygotic twin pairs discordant for cannabis exposure,
ie, OR for unrelated pairs = full sibling pairs OR = monozygotic twin pairs OR (Table). In contrast,
when ORs decline as relatedness increases but remain significantly greater than unity in close relatives (OR for unrelated pairs > full sibling pairs OR > monozygotic twin pairs OR > 1), hypothesis 2 is supported. Under the entirely noncausal hypothesis, ORs decrease rapidly in discordant pairs of relatives and are not significantly different from unity in monozygotic twin pairs (OR for unrelated pairs > full sibling pairs OR 1; monozygotic twin pairs OR = 1). We are aware of only 1 such analysis examining the cannabis-schizophrenia association. Giordano et al (4) reported a significant increase in schizophrenia among individuals with cannabis misuse. However, as genetic relatedness increased, the ORs declined. The change in ORs from 10.44 in the general population to 3.52 for discordant monozygotic twin pairs, a 66% decline, is consistent with hypothesis 2 and suggests substantial confounding.
Mendelian randomization is a type of instrumental variable analysis for approximating randomization to experimental and control conditions. Using single nucleotide variants as instrumental variables, Vaucher
et al (1) found that the genetic liability to lifetime cannabis use was associated with a significant increase in the risk of schizophrenia by 37%. In contrast,Gage et al (5) and Pasman et al (6) found much stronger evidence that genetic liability to schizophrenia was associated with an increase in the risk of lifetime cannabis use, ie, reverse causality. These mendelian randomization studies rely on single-nucleotide variant instrumental variables associated with lifetime cannabis use, not heavy use or misuse. However, in aggregate, results from available mendelian randomization studies are most consistent with hypothesis 2.