Enhancing one-year mortality prediction in STEMI patients post-PCI: an interpretable machine learning model with risk stratification.

Background: ST-elevation myocardial infarction (STEMI) poses a significant threat to global mortality and disability. Advances in percutaneous coronary intervention (PCI) have reduced in-hospital mortality, highlighting the importance of post-discharge management. Machine learning (ML) models have shown promise in predicting adverse clinical outcomes. However, a systematic approach that combines high predictive accuracy with model simplicity is still lacking. Methods: This retrospective study applied three data processing and ML algorithms to address class imbalance and support model development. ML models were trained to predict one-year mortality in STEMI patients post-PCI, with performance evaluated using accuracy, sensitivity, precision, F1-score, area under the receiver operating characteristic curve (AUROC), and the area under the precision-recall curve (AUPRC). Results: We analyzed data from 1,274 patients, incorporating 46 clinical and laboratory features. Using the Random Forest (RF) algorithm, we achieved an AUROC of 0.94 (95% confidence interval (CI): 0.90-0.98), an AUPRC of 0.44 (95% CI:0.15-0.76) in the internal validation set, identifying five key predictors: cardiogenic shock, creatinine, NT-proBNP, diastolic blood pressure, and left ventricular ejection fraction. By integrating risk stratification, the model's performance improved, achieving an AUROC of 0.97 (95% CI: 0.96-0.99) and an AUPRC of 0.74 (95% CI: 0.60-0.84). Conclusion: This study highlights the feasibility of constructing accurate and interpretable ML models using a minimal set of predictors, supplemented by risk stratification, to improve long-term outcome prediction in STEMI patients.