Optimized Feature Selection and Advanced Machine Learning for Stroke Risk Prediction in Revascularized Coronary Artery Disease Patients
Preprint
Abstract
Background: Coronary artery disease (CAD) is a leading cause of mortality, with stroke being a major complication following coronary revascularization procedures such as percutaneous coronary intervention (PCI) and coronary artery bypass grafting (CABG). While machine learning (ML) has been used to predict postoperative outcomes, a gap remains in quantifying stroke risk in revascularized CAD patients. This study aims to develop and validate ML models for stroke risk prediction, enhancing clinical decision-making. Methods: We extracted 5,757 patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and applied Pearson correlation analysis, least absolute shrinkage and selection operator (LASSO), ridge regression, and elastic net for feature selection. The initial 35 features were reduced to 14. The dataset was split into training (70%), testing (15%), and validation (15%). We evaluated multiple ML models, 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) and 500 bootstrapped 95% confidence intervals (CIs). Results: CatBoost achieved the highest performance with an AUC of 0.8486 (95% CI: 0.81240.8797) on the test set and 0.8511 (95% CI: 0.82030.8793) on validation. Shapley Additive Explanations (SHAP) identified Charlson Comorbidity Index (CCI), length of stay (LOS), and treatment types as key predictors. Our model outperformed existing studies, improving AUC by 9% while using a more refined feature set. Conclusions: Integrating multiple feature selection methods, our streamlined model improves efficiency and reliability. The proposed CatBoost model offers a high-accuracy approach to predicting postoperative stroke in CAD patients undergoing revascularization, supporting clinical decision-making and preventive strategies.Keywords
All Related Versions
- Published version: BMC Medical Informatics and Decision Making, 25 (1), 1.
This publication has 20 references indexed in Scilit:
- Artificial intelligence in cardiovascular medicine: clinical applicationsEuropean Heart Journal, 2024
- Merging machine learning and patient preference: a novel tool for risk prediction of percutaneous coronary interventionsEuropean Heart Journal, 2024
- MIMIC-IV, a freely accessible electronic health record datasetScientific Data, 2023
- Explanation of machine learning models using shapley additive explanation and application for real data in hospitalComputer Methods and Programs in Biomedicine, 2022
- Murine models of clonal haematopoiesis to assess mechanisms of cardiovascular diseaseCardiovascular Research, 2021
- Applications of artificial intelligence/machine learning approaches in cardiovascular medicine: a systematic review with recommendationsEuropean Heart Journal - Digital Health, 2021
- Stroke risk prediction using machine learning: a prospective cohort study of 0.5 million Chinese adultsJournal of the American Medical Informatics Association, 2021
- Cardiac Amyloidosis and the Risks of CardioversionJACC, 2019
- The Risk of Stroke Following Coronary Revascularization - a Population-based Long-term Follow-up StudyScandinavian Cardiovascular Journal, 2002
- Established Risk Factors and Coronary Artery Disease: The Framingham StudyAmerican Journal of Hypertension, 1994