Multimorbidity and major adverse cardiovascular events in antipsychotic users: Time-to-event prediction by explainable machine learning

Antipsychotic treatment is associated with higher risk of major adverse cardiovascular events (MACEs), and risk may vary by multimorbidity and concomitant medications. Using Hong Kong electronic health records, we followed 26,274 MACE-free adults (18-65 years) with multimorbidity who initiated antipsychotics, capturing demographics, chronic conditions, and prior medication use. We applied a conditional inference survival tree to define clinically interpretable risk profiles and compared ten time-to-event machine learning models using time-dependent ROC, calibration, and decision curve analyses. The highest-risk profile was age >48 years with chronic kidney disease, antibacterial/antiplatelet use, no antidepressant use, and no metastatic cancer (171.3 per 1,000 person-years). A random survival forest model showed the best discrimination (C-statistics 0.841, 0.835, and 0.824 at 1, 3, and 5 years, respectively), with age, antidepressant use, and chronic kidney disease as key predictors. These results support practical cardiovascular risk stratification for antipsychotic initiators with multimorbidity.