Machine learning-based mortality risk assessment in first-episode bipolar disorder: a transdiagnostic external validation study

Background: Accurate mortality risk prediction could enhance treatment planning in bipolar disorder, where mortality rates rival those of many cancers. Such prognostic tools are lacking in psychiatry, where assessments typically emphasize immediate suicidality while neglecting long-term mortality risks, and their clinical use is debated. We evaluated the recently developed machine learning model MIRACLE-FEP, initially developed for first-episode psychosis, in predicting all-cause mortality in patients with first-episode bipolar disorder (FEBD), hypothesizing that it would provide accurate risk prediction and guide pharmacotherapy decisions. Methods: We utilized national register-based cohorts of FEBD patients from Sweden (N = 31,013, followed 2006–2021) and Finland (N = 13,956, followed 1996–2018). We assessed the MIRACLE-FEP model's performance in predicting all-cause mortality using the area under the receiver operating characteristic curve (AUROC), calibration, and decision curve analysis. Additionally, we conducted a pharmacoepidemiologic analysis to examine the relationship between predicted mortality risk and pharmacotherapy effectiveness. Findings: MIRACLE-FEP achieved an AUROC = 0.77 (95%CI = 0.73–0.80) for 2-year mortality prediction in Sweden and 0.71 (95%CI = 0.67–0.75) in Finland. For 10-year all-cause mortality prediction, the model demonstrated an AUROC of 0.71 in both cohorts. The model demonstrated relatively good calibration and indicated potential clinical utility in decision curve analysis. Among patients with predicted risk exceeding the observed two-year mortality rate in FEBD, the lowest mortality risk was observed with polytherapy regimens (compared to non-use of antipsychotics or mood stabilizers), including quetiapine and lamotrigine (HR = 0.42, 95%CI = 0.23–0.80) or mood stabilizer polytherapy (HR = 0.47, 95%CI = 0.27–0.82). Conversely, in patients with predicted risk below this threshold, complex pharmacotherapy was not associated with a significant reduction in mortality risk. Interpretation: MIRACLE-FEP offers a promising approach to predicting long-term mortality risk and could guide proactive treatment decisions, such as targeting combination pharmacotherapy, in FEBD. Funding: The Swedish Research Council for Health, Working Life and Welfare, FORTE (2021-01079).