Context

Bipolar disorder is most often severe and recurrent, requiring pharmacological maintenance treatment in combination with psychoeducation and/or other interventions.1 Evidence for the efficacy and tolerability of pharmacological treatment is primarily based on randomised controlled trials (RCTs). However, RCTs are usually characterised by a relatively short duration of follow-up and by relatively small sample sizes. Therefore, for detecting and evaluating rare and long-term side effects, large observational cohort studies have been conducted. Even though selection bias and confounding cannot be avoided in non-randomised studies, various approaches in terms of design and analysis are available for balancing the comparative groups and/or for confounder control, for example, case–control design, use of propensity score models or use of multivariate regression analysis.