Development and validation of a machine learning-based predictive model for coronary heart disease risk in...

The incidence of coronary heart disease (CHD) continues to rise among younger populations, necessitating the development of rapid and effective risk prediction models to provide new approaches for secondary prevention of CHD.

To construct a clinical prediction model for premature coronary heart disease (PCHD) in the Chinese population based on machine learning algorithms.

A retrospective cohort study was conducted young and middle-aged patients undergoing coronary angiography at Northern Jiangsu People's Hospital (November 2018-May 2023).Feature selection was performed using Lasso regressionwith 10-fold cross-validation, followed by multivariate logistic regression. Seven supervised learning algorithms were evaluated: Logistic Regression (LR), LightGBM (LGBM), Random Forest (RF), Decision Trees (DT), Support Vector Machines (SVM), eXtreme Gradient Boosting (XGBoost), k-Nearest Neighbors (KNN), and Naïve Bayes (NB).

This study enrolled a total of 1276 participants, comprising ‌881 in the PCHD group‌ and ‌395 in the non-PCHD group. LASSO regression analysis‌ identified ‌nine potential predictors. All sevne machine learning models demonstrated good predictive performance. After excluding overfitted models, the LR model (AUC: 0.82; Sensitivity: 0.654; Specificity: 0.805; Recall: 0.654; F1: 0.749) and SVM model had higher AUC values than XGBoost (AUC: 0.794; Sensitivity: 0.858; Specificity: 0.504; Recall: 0.858; F1: 0.82) in the validation set. Therefore, we used Nomogram and SHAP summary plot to visualize and interpret the LR model and SVM model, respectively.

The LR-based nomogram and SVM-SHAP model provide clinically actionable tools for PCHD risk stratification. These models facilitate early identification of high-risk individuals for targeted preventive interventions.