However, observational studies entail a risk of selection effects. This risk is even more pronounced in pharmacoepidemiological studies, where confounding by indication is a major hurdle. Treated patients are likely to differ from nontreated patients in symptom severity. This is well illustrated in our study, where the rate of mania in the group of patients on a concurrent mood stabilizer was higher overall than in the antidepressant monotherapy group (2.4–6.8 times, depending on the period compared). This is not surprising, as the patients previously treated with a mood stabilizer received this treatment precisely because they weremore prone to manic episodes than the patients with no prior mood stabilizer treatment. We therefore employed a within-individual design, in which the rate of mania is compared between a non-treatment period and a treatment period in the same patient. This reduces the confounding due to disorder severity, genetic makeup, and early environmental factors that can be expected between those who receive a prescription for a mood stabilizer and those who do not.