From prescription to prognosis: longitudinal associations between different antidepressant classes on medical comorbidity in those with bipolar disorder

Individuals with bipolar disorder experience high rates of cardiometabolic and other chronic medical comorbidities, yet there is limited bipolar-specific evidence on whether antidepressant class is associated with differential long-term medical risk. Using longitudinal electronic health record (EHR) data from a tertiary academic health system, we examined whether antidepressant class (selective serotonin reuptake inhibitors [SSRIs], serotonin–norepinephrine reuptake inhibitors [SNRIs], tricyclic antidepressants [TCAs], monoamine oxidase inhibitors [MAOIs], and atypical antidepressants) was associated with incident medical comorbidity among patients with bipolar disorder. This retrospective cohort included 50,706 antidepressant treatment instances among 15,613 unique patients with bipolar disorder. Comorbidities were derived from harmonized Charlson and Elixhauser indices. Baseline comorbidity status, after January 1, 2019, was taken from the most recent encounter prior to antidepressant initiation; follow-up was the first encounter after antidepressant discontinuation. Incident comorbidity was defined as absence at baseline and presence at follow-up within an instance. Generalized linear mixed-effects models with patient-level random intercepts, adjusted for age, sex, baseline comorbidity count, and log-transformed observation duration, with inverse probability of treatment weighting (IPTW), estimated odds of incident comorbidity by antidepressant class, with pairwise comparisons using a contrast matrix of log-odds differences with propagated standard errors. SSRIs were most commonly prescribed (50.6% of instances), followed by SNRIs (22.9%), atypical antidepressants (22.1%), TCAs (4.3%), and MAOIs (< 0.1%). In adjusted models, TCAs showed the highest overall odds of incident comorbidity (OR = 1.315, 95% CI 1.161–1.489) and were also higher than atypical antidepressants (OR = 1.065, 95% CI 0.999–1.137) and SNRIs (OR = 1.102, 95% CI 1.027–1.182). Atypical antidepressants (OR = 1.26, 95% CI 1.16–1.38 vs. SSRIs) and SNRIs (OR = 1.31, 95% CI 1.19–1.43 vs. SSRIs) formed an intermediate risk tier with largely similar profiles (atypical vs. SNRI OR = 1.03, 95% CI 0.94–1.14). Across individual conditions, the largest elevations for TCAs were observed for cardiac arrhythmias (OR = 2.55, 95% CI 1.98–3.28), renal disease (OR = 1.91, 95% CI 1.32–2.77), and uncomplicated hypertension (OR = 1.63, 95% CI 1.19–2.24). Sensitivity analyses separating bupropion and mirtazapine within the atypical class revealed divergent but non-significant effects (bupropion OR = 1.048, 95% CI 0.977–1.125, p = 0.190; mirtazapine OR = 1.122, 95% CI 0.975–1.291, p = 0.107), supporting retention of the combined atypical class in the primary analysis. In a large real-world bipolar cohort, antidepressant class was associated with distinct patterns of incident medical comorbidity, with a largely consistent hierarchy of risk (TCAs highest; SSRIs lowest). Although exposure duration and adherence are incompletely captured in EHR prescribing data and residual confounding is likely, these findings provide clinically relevant comparative safety signals to inform prescribing decisions and motivate future studies incorporating pharmacy fills, longitudinal exposure trajectories, and stronger causal inference designs.