Machine learning for high-risk multimorbid antipsychotic user identification and cardiovascular event prediction

Abstract

<p>Introduction:<br>Pre-existing multimorbidity is a known risk factor for major adverse cardiovascular events (MACE)<br>among antipsychotic users. However, the association between multimorbidity and polypharmacy<br>patterns and MACE remains to be clarified.<br>Aims:<br>Identify specific multimorbidity patterns and antipsychotic use associated with increased MACE risks<br>and develop and validate a time-to-event prediction model.<br>Methods:<br>This retrospective cohort study utilized electronic health records from the Hong Kong CDARS<br>database. Patients aged 18-65 years, with records of 2 or more chronic health conditions within three<br>years prior to the initial date of antipsychotic intake, were enrolled. Baseline characteristics,<br>including age, sex, chronic disease history, antipsychotic usage history, and previous one-year drug<br>intake history, were collected. The dataset was randomly divided into training and validation subsets<br>based on the initial year of antipsychotic prescription. A Conditional Inference Survival Tree (CISTree)<br>was employed to classify MACE risk groups. Eight machine learning models were trained using 5-fold<br>cross-validation for hyperparameter optimization and validated on the validation set.<br>Results:<br>27,466 patients were included. The CISTree model identified older patients (>44 years) with chronic<br>kidney disease (CKD), cancer, hypertension, and using antianginal and antiplatelet drugs but not<br>taking antidepressants as having the highest MACE incidence rate (173.065 per 1,000 person-years;<br>95% CI: [125.023, 230.990]). The Random Survival Forest (RSF) model outperformed the other seven<br>models, identifying age, antidepressant usage, and chronic kidney disease (CKD) as the top three<br>significant predictors. Furthermore, factors associated with a lower risk of MACE included younger<br>age (<44 years), the use of antianginal, antibacterial, or antidepressant drugs, no usage of<br>antiplatelet drugs or haloperidol, and the absence of CKD.<br>Discussion:<br>We identified highly specific high-risk groups in multimorbidity people using antipsychotics;<br>prediction based on the same features demonstrates excellent power and potential in aiding clinical<br>decisions.<br></p>