Machine learning models for predicting severe clinical events in hospitalized patients with coronary artery disease

Background and purpose: Coronary artery disease (CAD) represents the leading cause of mortality on a global scale, with severe clinical events such as resuscitation or death occurring frequently during the course of hospitalisation. The utility of existing predictive models may be constrained by their incomplete utilisation of the depth of electronic medical records (EMRs), which could limit their effectiveness and scope. This study aims to develop and validate interpretable risk prediction models to predict severe clinical events in hospitalized patients with coronary artery disease, enhancing clinical decision-making and patient management. Methods: We conducted a retrospective study using EMRs from CAD patients admitted to Xiyuan Hospital between 2016 and 2024. The dataset includes structured and unstructured data extracted via natural language processing (NLP) from EMRs. We developed five machine learning (ML), including Logistic Regression (LR), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Gaussian Naive Bayes (GNB), and Deep Neural Network (DNN). The discrimination ability was comprehensively evaluated by the area under the curve (AUC); sensitivity, specificity, and F1 score. SHapley Additive exPlanations (SHAP) were used to interpret model predictions. Results: Of the 6,971 patients included, 268 (3.84%) experienced severe clinical events during hospitalization. The DNN model demonstrated the best performance, with an AUC of 0.995 (95% CI: 0.985–0.999). The SHAP analysis demonstrated that the most significant predictors were the admission principal diagnoses of acute CAD, followed by the presence of urinary occult blood and the mental state of the patient. Conclusion: Using NLP and ML models to integrate data from EMRs enables early warning of severe clinical events in hospitalized CAD patients. The interpretable prediction models developed in this study can assist clinicians in more accurately predicting severe clinical events, thereby enhancing clinical decision-making and patient management.