Estimating the per-protocol effect of lithium on suicidality in a randomized trial of individuals with depression or bipolar disorder

Background: The CSP590 randomized trial was designed to estimate the effect of lithium on suicidality. After a third of the intended number of participants were enrolled, the hazard ratio of suicidality was 1.10 (95% CI: 0.77, 1.55). Based on this, the trial was stopped for futility. However, only 17% of patients adhered to the specified protocol. Aims: The objective was to estimate the per-protocol effect of lithium on suicidality, that is, the effect of adhering to the treatment strategies as specified in the protocol. Methods: We stopped individuals’ follow-up if/when they showed evidence of nonadherence. We then conducted the analysis in the restricted sample, adjusting for prognostic factors that predict adherence via inverse probability weighting. The primary outcome was the 12-month risk of suicidality (including death from suicide, suicide attempt, interrupted attempt, hospitalization specifically to prevent suicide). Results: The estimated 12-month risk of suicidality was 18.8% for lithium, and 24.3% for placebo. The risk ratio was 0.78 (95% CI: 0.43, 1.37) and the risk difference −5.5 percentage points (95% CI: −17.5, 5.5). Results were consistent across sensitivity analyses. Conclusions: With one-third of the targeted sample size, lithium effects (compared with placebo) ranging between a 17.5% reduction and a 5.5% increase in the risk of suicidality were highly compatible with the data. Thus, a protective effect of lithium on suicidality among patients with bipolar disorder or major depressive disorder cannot be ruled out. Trials should incorporate adequate per-protocol analyses into the decision-making processes for stopping trials for futility.