A Machine Learning Approach to Predict Mortality in Patients Enrolled in Cardiac Rehabilitation After Percutaneous Coronary Intervention for Acute Coronary Syndrome

Abstract Background and aims: Patients with acute coronary syndrome (ACS) treated with percutaneous coronary intervention (PCI) have a residual risk of adverse events and all-cause mortality. Traditional static risk assessment models are not designed to incorporate the complex and dynamic nature of post-intervention conditions which may be overcome by machine learning (ML) approaches. This study aimed to develop and validate a model for the prediction of all-cause mortality of ACS patients post-PCI using advanced ML. Methods and Results: Data from 1,261 CAD patients (mean age 61 ± 11.1 years, 16.6 % women) who underwent PCI during an episode of ACS was analysed using ML techniques. External validation was performed using the Eu-CARE dataset of 1,386 CAD patients and predictions were explained using the model-agnostic SHapley Additive exPlanation (SHAP) methodology. Three models, LightGBM, Random Forest and SVM, outperformed logistic regression in terms of accuracy (mean AUC > 0.8). LightGBM was identified as the most potent prediction model in terms of balanced accuracy (0.87), with a sensitivity of 0.86 and a specificity of 0.88 indicating strong predictive power. External validation confirmed reliable detection of true positives, although specificity dropped to 0.7. The highest SHAP values for impact on the model were detected for the variables “diabetes”, “physical exercise capacity (METs achieved during stress test)”, “average daily sitting duration”, “statin use”, “history of family heart disease”, “number of affected vessels”, “revascularization history”, “serum creatinine levels”, and “chronic hepatopathy”. The model was rated as “recommended” using the independent validation screening tool. Conclusion: We present a ML-based model with high predictive capacity to identify CAD patients post PCI with higher mortality risk. The model could serve in clinical practice as a decision support tool to identify high-risk patients who could benefit from intensified personalized treatment and lifestyle interventions.