Related Experiment Videos
Development and validation of a machine learning model for predicting coronary artery disease risk: a retrospective
Yingying Zhou1,2, Yuxin Chen3, Shaoren Rao1
1Department of Cardiology, The First Affiliated Hospital of Hainan Medical University, Haikou, Hainan, China.
Background:
The mortality rate from coronary atherosclerotic heart disease is rising year by year, affecting an increasingly younger population, and is projected to continue rising in the future. Although medical technology is advancing, coronary angiography cannot be widely adopted for screening due to factors such as economic constraints, disparities in regional healthcare standards, and patient compliance. This study aimed to develop and validate a machine learning-based predictive model for assessing the severity of coronary artery lesions and to explore related risk factors.
Methods:
This retrospective study included 400 patients who underwent their first complete coronary angiography at the First Affiliated Hospital of Hainan Medical University between 2023 and 2025. The patient dataset from July 2023 to July 2025 (n = 300) was randomly divided into an internal training set and a test set in a 7:3 ratio. The temporal validation set consisted of patients (n = 100) from the same center between August and December 2025. Key feature variables were used to construct predictive models using eight machine learning algorithms, including Extreme Gradient Boosting (XGBoost), Logistic Regression (LR), Random Forest (RF), Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), Naive Bayes (NB), and Gradient Boosting Machines (GBM). Model performance was analyzed using receiver operating characteristic (ROC) curves, 10-fold cross-validation, and decision curve analysis (DCA). The optimal model was validated on the temporal dataset, and its interpretability was assessed using SHapley additive explanation (SHAP). The model was further successfully implemented as a web application using the Shiny package (version 1.13.0) in R 4.5.2.
Results:
Ten predictors were retained. The XGBoost model exhibited the best discrimination, with AUCs of 0.889 (95% CI: 0.819-0.945) in the internal test set and 0.803 (95% CI: 0.718-0.888) in temporal validation. The model also showed good calibration and low Brier scores, with stable performance in ten-fold cross-validation. DeLong's test confirmed its superiority over competing models. DCA and SHAP analysis further supported its clinical utility and interpretability.
Conclusion:
This study developed a machine learning model for predicting the severity of coronary lesions, demonstrating satisfactory performance in a single-center setting.
