Optimized feature selection and advanced machine learning for stroke risk prediction in revascularized coronary artery disease patients.

Background: Coronary artery disease (CAD) remains a leading cause of global mortality, with stroke constituting a significant complication among patients undergoing coronary revascularization procedures, such as percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG). Previous research has demonstrated the successful application of machine learning (ML) in predicting various postoperative outcomes, including poor prognosis following cardiac surgery and the risk of postoperative stroke. Despite these advancements, a critical gap persists in studies quantitatively linking the risk of postoperative stroke to revascularization using ML-based approaches. This study aims to address this gap by developing and validating ML models to predict the risk of stroke in CAD patients undergoing coronary revascularization, with the ultimate goal of enhancing clinical decision-making and improving patient outcomes.

Methods: We developed an ML framework to predict stroke risk in patients with CAD undergoing revascularization. A total of 5,757 patients were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Feature selection was performed using a combination of Pearson correlation analysis, least absolute shrinkage and selection operator (LASSO), ridge regression, and elastic net. Initially, 35 features were identified based on expert opinion and a comprehensive literature review; the integrated results of the feature selection methods reduced the feature set to 14. The dataset was randomly divided into training, testing, and validation subsets with proportions of 70%, 15%, and 15%, respectively. Several ML models were evaluated, including logistic regression, XGBoost, random forest, AdaBoost, Bernoulli naive Bayes, k-nearest neighbors (KNN), and CatBoost. Model performance was assessed using the area under the receiver operating characteristic curve (AUC-ROC), accuracy, and 500 bootstrapped 95% confidence intervals (CIs) to ensure robust evaluation.

Results: The CatBoost model demonstrated superior performance, achieving an AUC of 0.8486 (95% CI: 0.8124-0.8797) on the test set and 0.8511 (95% CI: 0.8203-0.8793) on the validation set. Shapley Additive Explanations (SHAP) analysis identified the Charlson Comorbidity Index (CCI), length of stay (LOS), and treatment types as the most influential predictors. Notably, compared to the best existing literature, which reported an AUC of 0.760 on the test set, our model exhibited a 9% improvement in predictive performance while utilizing a more parsimonious feature set.

Conclusion: By integrating four feature selection methods, we significantly streamlined the feature set, resulting in a more efficient and reliable predictive model. We propose the CatBoost model for the prediction of postoperative stroke in patients with CAD undergoing coronary revascularization. With its high accuracy, the proposed model offers valuable insights for medical practitioners, enabling informed decision-making and the implementation of preventive measures to mitigate stroke risk.