Machine Learning in Predicting Cardiac Events for ESRD Patients: A Framework for Clinical Decision Support

<b>Background/Objectives:</b> Patients with end-stage renal disease (ESRD) are at an increased risk of major adverse cardiac events (MACEs), highlighting the need for accurate risk prediction and personalized interventions. This study aims to develop and evaluate machine learning (ML) models to identify key predictive features and enhance clinical decision-making in MACE risk assessment. <b>Methods:</b> A dataset comprising 84 variables, including patient demographics, laboratory findings, and comorbidities, was analyzed using CatBoost, XGBoost, and LightGBM. Feature selection, cross-validation, and SHAP (SHapley Additive exPlanations) analyses were employed to improve model interpretability and clinical relevance. <b>Results:</b> CatBoost exhibited the highest predictive performance among the models tested, achieving an AUC of 0.745 (0.605-0.83) with balanced sensitivity and specificity. Key predictors of MACEs included antiplatelet use, the grade of left ventricular hypertrophy, and serum albumin levels. SHAP analysis enhanced the interpretability of model outputs, supporting clinician-led risk stratification. <b>Conclusions:</b> This study highlights the potential of ML-based predictive modeling to improve MACE risk assessment in patients with ESRD. The findings support the adoption of ML models in clinical workflows by integrating explainable AI methods to enable individualized treatment planning. Future integration with electronic health record systems may facilitate real-time decision-making and enhance patient outcomes.