Antidepressant composition change before bipolar disorder diagnosis among patients initially diagnosed with major depressive disorder.

BACKGROUND: Some patients initially diagnosed with major depressive disorder (MDD) later receive a bipolar disorder (BD) diagnosis. We examined whether antidepressant composition shifted toward specific agents before BD diagnosis and whether identified signals were confined to the final 180 days. METHODS: We conducted a matched case-control study using Taiwan's National Health Insurance Research Database. Cases were individuals diagnosed with MDD at ages 10-30 years who received a BD diagnosis at least 1 year later. Controls with MDD but no BD by the matched reference date were matched 1:4. Staged drug-level analyses addressed same-class burden, co-prescribing, composition change, and non-antidepressant treatment escalation. Temporal sensitivity analyses evaluated the final 180 days and the lagged days 181-365 window before the reference date. RESULTS: The sample included 8552 cases and 34,208 controls. Of 26 medications, 20 met leave-one-out screening criteria and 11 passed both post-leave-one-out checks. Paroxetine, duloxetine, and escitalopram showed composition-change signals in the main model. All three were associated with BD case status in the final-180-day analysis. After excluding the final 180 days, duloxetine remained associated (odds ratio, 1.04; 95% confidence interval, 1.01-1.07; false discovery rate-adjusted p = 0.043), whereas paroxetine and escitalopram did not. CONCLUSIONS: The duloxetine composition-change signal was not confined to the immediate 180-day prediagnostic period, whereas paroxetine and escitalopram signals were observed only in that period. These diagnosis-preceding prescribing patterns remain susceptible to protopathic bias and should not be interpreted as causal medication effects.