Abstract
BACKGROUND: Acute myocardial infarction (AMI) remains a leading cause of mortality and disability globally. Although percutaneous coronary intervention (PCI) has significantly reduced in-hospital mortality (IHM), the resultant class imbalance complicates accurate risk prediction. While machine learning (ML) demonstrates potential in predicting IHM, there is a lack of models that provide both high accuracy and personalized risk assessment. METHODS: This retrospective study was conducted at the First Hospital of Lanzhou University from January 1, 2019, to December 31, 2020. We employed three data processing methods: synthetic minority over-sampling technique (SMOTE), Boruta, and grid search cross-validation (GSCV). Subsequently, six ML algorithms were implemented. Model performance was evaluated using accuracy, sensitivity, precision, F1-score, area under the receiver operating characteristic curve (AUROC), and area under the precision-recall curve (AUPRC). RESULTS: The study cohort consisted of 1693 patients diagnosed with AMI, of whom 34 (2.0%) experienced IHM following PCI. After employing SMOTE to balance the dataset, 32 independent risk factors were identified using the Boruta feature selection method. Among the evaluated ML models, ensemble algorithms demonstrated superior performance. For instance, the Light Gradient-Boosting Machine (LightGBM) framework achieved a predictive accuracy with an AUROC of 0.93 (95% confidence interval (CI): 0.82-1.00) and an AUPRC of 0.62 (95% CI: 0.17-0.96). Additional performance metrics included an accuracy of 0.988, a precision of 0.625, a sensitivity of 0.625, a specificity of 0.994, and an F1-score of 0.625. CONCLUSION: Utilizing SMOTE for class balancing, Boruta for feature selection, GSCV for optimal hyperparameter tuning, and LightGBM for model development achieved strong predictive performance for IHM following AMI. These findings underscore the significance of robust processing and careful algorithm selection.