Sample bias may obscure results

Di Forti et al ¹ present their paper as further evidence of the link between high-potency cannabis and psychosis. Obviously, a major issue in case–control studies is the sampling, and any difference between case and control groups needs to be carefully considered. The authors state that ‘there was no significant difference between the cases and control groups in age, gender, ethnicity, educational qualifications or employment status at time of assessment’. However, I would raise concerns about the employment status of the participants and I would respectfully highlight that this statement does not seem consistent with the information provided in the accompanying table of sample characteristics. This table states that 58.4% of cases and 43.2% of controls were unemployed. The percentages in this table have some inaccurate rounding but more worryingly, contrary to the authors' report, there is a clear statistically significant difference (P = 0.001 using a z-test for proportions).

This also seems to be a highly relevant and clinically significant difference that may have introduced considerable bias into this study and merited the attention of the 14 authors. In the discussion the authors state ‘the increased availability of skunk cannot alone explain why our control group members are less likely to prefer higher-potency types than the cases group across time’. The requirement to hold down a job may be a highly significant reason why controls smoked cannabis of lesser potency less often than the unemployed. Moreover, individuals who are unemployed are highly likely to have poorer social and health status, which further serves to obscure the true role of cannabis in this study.