Cannabis & depression: Demystifying propensity score techniques

INTRODUCTION: Increased cannabis use among adolescents and reports linking cannabis use to other psychiatric disorders has prompted research on the relationship between adolescent cannabis use and later depression. Using propensity score (PS) adjustment techniques we test the potential causal link between adolescent cannabis problem use and young adult major depression. METHODS: A cohort of 2,311 first-graders was interviewed over a 15-year period. We used parametric (multivariable logistic regression) and non-parametric (generalized boosted modeling) PS estimation techniques and five alternative applications of the PS, including 1:1 matching, full matching, subclassification, Inverse Probability of Treatment Weighting, and weighting by the Odds. The PS adjusted the final logistic regression of major depression on cannabis problem use by balancing all observed potential confounding covariates. Decision criteria based on minimizing the standardized bias (effect size) of the potential confounding covariates were created to aid in the selection of the best average treatment effect model. RESULTS: Unadjusted, 24% of individuals with cannabis problems reported depression whereas only 13% of individuals without cannabis problems reported depression. The PS-adjusted analyses all resulted in increased odds of depression among those with cannabis problems. DISCUSSION: The variation in magnitude of the effect and statistical significance does not directly suggest one of the PS techniques to be superior. This work suggests: based on the decision criteria, non-parametric estimation of the PS may be an improvement over parametric estimation for these data, and adolescents with cannabis problems are at increased odds for young adult major depression, but the causal link is modest.