Cannabis and psychosis: accumulating evidence

Abstract

The report by Stefanis et al. (2004) adds to the evidence linking the heavy use of cannabis with increased risks of psychosis or psychotic symptoms. The advantages of this study include: a large and representative sample; measurement of multiple dimensions of psychotic symptoms; statistical control for reverse causality mediated via distress; and adjustment for comorbid depression. The study produces evidence of a dose/response association between cannabis use and psychotic symptoms in which: (a) rates of symptoms increased with increasing use of cannabis; (b) the early use of cannabis appeared to have a greater impact on later risks of psychotic symptoms. An interesting feature is that the study finds this association in a cultural context in which the use of cannabis by young people was low with only 6% reporting cannabis use. The authors suggest that this evidence adds credence to the hypothesis that cannabis contributes to the population expression of psychosis. There have been two recent reviews that have examined the linkages between cannabis use and the development of psychotic symptoms (Arseneault et al. 2004; Smit et al. 2004). Both have concluded that the weight of the evidence favors the hypothesis that the use of cannabis may provoke the onset of psychosis or psychotic symptoms. However, this consensus of the psycho-social consequences of cannabis use in young people has been challenged in a recently published review (Macleod et al. 2004). This review concluded that ‘Available evidence does not strongly support a causal relationship between cannabis use and psycho-social harm in young people . . .’. This review included some of the evidence on the cannabis/psychosis link. What distinguishes the conclusions of the two early reviews on cannabis and psychosis from the more recent review is not differences in reviewing methodology or the evidence but rather differences in the opinions of authors about the role of uncontrolled confounding. The reviews by Arseneault et al. (2004) and Smit et al. (2004) take the evidence at face value and conclude that since an association between cannabis use and psychosis has been found to persist after control for confounding, the weight of the evidence favors the hypothesis. Macleod et al. (2004), however, focus more critically on the issue of uncontrolled or residual confounding and suggest that because of the possibility of such confounding, strong evidence for a causal link does not exist. These different ways of describing the same evidence and issues are likely to lead to further confusion in an area already subject to controversy. For example, following the publication of the Macleod article, my local paper published a letter claiming that the review had ‘debunked the myth that cannabis causes psychosis’. In fact, it had done nothing of the sort; it had merely drawn attention to the limitations in the existing evidence and phrased conclusions in such a way that a careless reader would be prone to conclude that there was no credible evidence linking cannabis and psychosis. This debate is, of course, of direct relevance to the study by Stefanis et al. In particular, the review by Macleod et al. highlights the importance of paying attention to the control of third or confounding factors. Unfortunately, this is the very area in which the article by Stefanis et al. has limitations. Although the authors adjust their results for a number of confounding factors (other drug use, gender and school grade) it is clear that there are many other social, family and individual factors that could have confounded the association. It is, thus, entirely possible that the associations reported by Stefanis et al. could be ascribed to uncontrolled residual confounding rather than to a causal link between cannabis and psychosis. This difficulty is, in theory at least, redeemable since the longitudinal design used by Stefanis et al. has the potential for further adjustments the association between cannabis use and psychosis for wider range of prospectively assessed confounders. The tensions between the conclusions drawn by recent authoritative and peer reviewed assessments of the linkages between cannabis use and psychosis raise important issues about the types of evidence that is needed to reduce the uncertainty and controversy in this area. As I have argued in a previous letter (Fergusson 2004) on this topic, two general approaches seem promising. The first approach is through improvements in epidemiological research design and analysis to estimate the effects of-non-observed confounding factors. Here several strategies are possible. In longitudinal designs it is possible to use fixed effects regression methods to control non-observed confounders (Duncan et al. 1998; Fergusson et al. 2002). Twin designs may be used to control non-observed genetic and common environmental factors through the use of the discordant twins design (Lynskey et al. 2003). In their review Macleod et al. point to the use of genetic markers and the principle of Mendelian Randomisation as a promising approach to controlling confounding factors (Davey Smith & Ebrahim 2003). Although each of these approaches has limitations, each may add a different perspective on the issue of uncontrolled confounding. The second route to clarifying this issue is through more basic science research into the neuro-chemistry, biology and physiology of the effects of cannabis and the development of psychosis. As Stefanis et al. point out, there is suggestive but by no means conclusive evidence that the sensitization of the mesolimbic dopaminergic system may be one pathway by which the repeated use of cannabis may lead to the onset of psychotic symptoms. Better evidence on the underlying neuro-chemistry and biology of the effects of cannabis and the origins of psychosis is likely to play an important role in furthering our understanding of the extent to which statistical linkages between cannabis and psychosis reflect underlying causal processes.