| Objective To investigate the impact of soluble suppression of tumorigenicity 2 (sST2) and triglyceride-glucose (TyG) index on the severity of coronary artery stenosis in patients with coronary heart disease (CHD), and to develop and validate an optimal machine learning prediction model. Methods Data from 193 patients diagnosed with CHD at the First Affiliated Hospital of Bengbu Medical College from June 2021 to February 2023 were collected. Based on the Gensini score, patients were categorized into a mild stenosis group (n=92) and a severe stenosis group (n=101). Differences in characteristics between the two groups were compared, and multivariate logistic regression analysis was performed to assess the associations of sST2 and TyG with the severity of coronary stenosis. Additionally, various machine learning algo-rithms, including random forest, support vector machine, XGBoost (eXtreme Gradient Boosting), and backpropagation neural network, were em-ployed to compare their predictive capabilities for coronary stenosis. The optimal model was identified by calculating the area under the receiver operating characteristic curve (AUC). Results Compared with the mild stenosis group, the severe stenosis group had significantly higher levels of sST2 and TyG index (P<0.001). Multivariate logistic regression analysis revealed that sST2 (OR=1.082, 95%CI: 1.008~1.161) and TyG index (OR=3.834, 95%CI: 1.856~7.921) were risk factors for severe coronary stenosis in CHD patients. The predictive performance of different machine learning models surpassed that of the traditional logistic model (AUC: 0.708, 95%CI: 0.576~0.841), with the XGBoost model exhibiting the highest predictive accuracy (AUC: 0.848, 95%CI: 0.740~0.955). Furthermore, variable importance rankings across all machine learning models consistently placed sST2 and TyG index among the top five most important predictors. Conclusion sST2 and TyG index are closely associated with the severity of coronary stenosis in CHD patients. Combining these biomarkers with machine learning algorithms can significantly enhance the predictive accuracy for coronary stenosis risk. |
