Machine Learning-Based Risk Prediction for Coronary Heart Disease Complicated by Hyperhomocysteinemia: Retrospective Study.

BACKGROUND

Hyperhomocysteinemia (HHcy) is recognized as an independent risk factor for coronary heart disease (CHD), yet accurately predicting CHD risk in patients with HHcy remains a challenge. This study aimed to develop and validate multiple machine learning models for predicting CHD risk in patients with HHcy and elucidate key predictors using Shapley Additive Explanation (SHAP) algorithms.

OBJECTIVE

This study aims to develop and validate machine learning models for predicting the risk of coronary heart disease in individuals with normal homocysteine levels, aiming to improve early risk stratification and clinical decision-making.

METHODS

This single-center retrospective study collected data from patients who were diagnosed with HHcy through electronic medical records, which were randomly divided into training (n=364, 70%), validation (n=78, 15%), and test (n=78, 15%) sets. Seven machine learning models were constructed, including logistic regression, k-nearest neighbor, decision tree, random forest, extreme gradient boost, light gradient boosting machine (LightGBM), and stacking. Six core variables (age, weight, hypertension, continuous drinking history, activated partial thromboplastin time, and carotid plaque) were utilized as inputs, with performance evaluation metrics encompassing area under the receiver operating characteristic curve, accuracy, F1-score, calibration curve, Brier score, and decision curve analysis. Additionally, SHAP interpretation was conducted on the optimal LightGBM model.

RESULTS

The LightGBM model exhibited superior performance in the test set (area under the receiver operating characteristic curve=0.807, F1-score=0.606), demonstrated good calibration (Brier score=0.2415), and yielded high clinical net benefit. SHAP analysis revealed age and activated partial thromboplastin time as the most influential predictors, followed by hypertension, weight, carotid plaque, and continuous drinking history. The correlation heat map illustrated low collinearity among variables, ensuring model stability.

CONCLUSIONS

The LightGBM model demonstrated high accuracy and interpretability in forecasting CHD risk among patients with HHcy. The integration of machine learning and interpretable artificial intelligence methods holds promise for delivering personalized early risk assessment and intervention strategies in clinical settings.