Abstract
OBJECTIVE: To develop and validate a machine learning (ML) model for predicting long-term depression risk in ACS patients following percutaneous coronary intervention (PCI). METHODS: This retrospective cohort study included 1951 ACS patients who underwent PCI in 2023. Feature selection was conducted using the Boruta algorithm, and restricted cubic spline (RCS) analysis was applied to assess non-linear associations. Six ML models were trained and tested using a 70:30 train-validation split. Model performance was evaluated using Area under the curve(AUC), sensitivity, specificity, F1-score, calibration curves, and decision curve analysis. SHapley Additive exPlanations (SHAP) were used to interpret feature contributions. RESULTS: Among the 1951 patients, 382 (19.6%) developed long-term depression. After feature selection via the Boruta algorithm, ten key predictors were identified, including NYHA classification, diabetes, thyroid-stimulating hormone (TSH), and left ventricular ejection fraction (LVEF). The LGBM and XGBoost models achieved the highest discrimination, with AUCs of 0.849 (training) and 0.652 (validation) for LGBM, and 0.814 (training) and 0.699 (validation) for XGBoost. Calibration curves showed good alignment between predicted and observed outcomes. SHAP analysis confirmed NYHA classification, TSH, and diabetes as the most influential features. Decision curve analysis demonstrated the clinical benefit of both models across a range of thresholds. CONCLUSION: The models demonstrated potential for early risk stratification of post-PCI depression and may inform targeted clinical interventions.