Assessing evidence for a causal link between cannabis and psychosis: a review of cohort studies.

Over the past five years, the release of cohort studies assessing the link between cannabis and psychosis has increased attention on this relationship. Existing reviews generally conclude that these cohort studies show cannabis has a causal relationship to psychosis, or at least that one cannot be excluded. Few studies have evaluated the relative strengths and limitations of these methodologically heterogeneous cohort studies, and how their relative merits and weaknesses might influence the way the link between cannabis use and psychosis is interpreted. This paper reviews the methodological strengths and limitations of major cohort studies which have looked at the link between cannabis and psychosis, and considers research findings against criteria for causal inference. Cohort studies that assessed the link between cannabis and psychosis were identified through literature searches using relevant search terms and MEDline, PsycINFO and EMBASE. Reference lists of reviews and key studies were hand searched. Only prospective studies of general population cohorts were included. Findings were synthesised narratively. A total of 10 key studies from seven general population cohorts were identified by the search. Limitations were evident in the measurement of psychosis, consideration of the short-term effects of cannabis intoxication, control of potential confounders and the measurement of drug use during the follow-up period. Pre-existing vulnerability to psychosis emerged as an important factor that influences the link between cannabis use and psychosis. Whilst the criteria for causal association between cannabis and psychosis are supported by the studies reviewed, the contentious issue of whether cannabis use can cause serious psychotic disorders that would not otherwise have occurred cannot be answered from the existing data. Further methodologically robust cohort research is proposed and the implications of how evidence informs policy in the case of uncertainty is discussed.