63. mapping longitudinal real-world treatment and hospitalization dynamics across primary psychotic disorders and bipolar disorder

Background Primary psychotic disorders (PPD) and bipolar disorder (BD) are characterized by recurrent episodes, repeated healthcare contacts, and long-term pharmacological treatment. While clinical trials are the gold standard for estimating treatment efficacy under controlled conditions, real-world data enable assessment of treatment and outcomes over extended follow-up but require careful handling. Here, we use data from the Estonian Biobank (EstBB), comprising 212,000 individuals, to investigate treatment exposure and hospitalization in a biobank-linked real-world setting, and to assess the contribution of genetic liability to these outcomes. Methods Using electronic health records (EHR), we derived a sample of 1,625 individuals with PPD, mania, or BD. Healthcare contacts were captured through inpatient and outpatient records, and treatment periods were reconstructed from drug purchase data using multiple approaches varying in assumptions on drug supply duration. Polygenic scores (PGS) for schizophrenia (SCZ), BD, and educational attainment (EA) were evaluated in relation to healthcare contacts using negative binomial models and to rehospitalization using time-varying Cox proportional hazard models. Results EHR-derived phenotypes showed high genetic correlation with large-scale GWAS of SCZ and BD (rg > 0.88), indicating that biobank samples reflect underlying genetic susceptibility to disease. While diagnostic categories remained stable over time (median follow-up 11.2 years), with limited transition between PPD and BD, hospitalization was more frequent among individuals with PPD. All individuals had at least one psychiatric healthcare contact, and 51.6% experienced at least one hospitalization for PPD/BD. At FDR 5%, all PGSs were associated with PPD/BD-related outpatient visit counts, but not with the number of hospitalizations. Treatment patterns varied across individuals, with a median of 3 distinct drugs per individual over follow-up. Across alternative treatment period derivation approaches, being on antipsychotics or mood stabilizers was consistently linked with reduced risk of rehospitalization. Treatment was associated with reduced hazard of first rehospitalization (HR = 0.78, 95% CI 0.62–0.97), with stronger effects in recurrent event models (HR = 0.72, 95% CI 0.63–0.84). In contrast, PGSs for SCZ and BD were not associated with rehospitalization risk. Discussion Biobank-linked health data enable the investigation of clinical outcomes and genetic liability for psychotic disorders in a real-world setting. While genetic predisposition was reflected in patterns of healthcare contact, treatment showed stronger and consistent association with rehospitalization risk, with effect sizes depending on how treatment periods were defined from drug purchase data. Together, these findings highlight both the value and methodological sensitivity of real-world data for studying disease course and treatment outcomes.