Key points are not available for this paper at this time.
Abstract Background Diabetes mellitus (DM) combined with coronary heart disease (CHD) significantly increases the risk of cardiovascular events with a greater mortality rate. Therefore, establishing a predictive model can help DM patients recognize their potential risk of CHD and prevent the occurrence of CHD at an early stage. Methods A total of 12124 clinical samples of DM patients were collected from two centers. Univariate and multivariate logistic regression analyses were used to preliminarily screen important factors for the risk of CHD in DM patients. We used eight kinds of machine learning (ML) algorithms (10-fold cross validation) to build different ML models for predicting the risk of CHD in DM patients, and compared their prediction performance by using various evaluation indicators. We performed external validation of the final model and utilized SHapley Additive exPlanation (SHAP) to explain it. Results 11 factors related to the risk of CHD in DM patients were ultimately selected. Among the eight ML models, the light gradient boosting machine (LGBM) model showed the best predictive performance in both the internal validation of the test set Area under curve (AUC): 0.87, 95% confidence interval (CI) (0.82–0.89) and the external validation AUC: 0.84, 95% CI (0.82–0.87). SHAP analysis identified variables that contributed to the model predictions. The ultimate predictive model was incorporated into a web-based platform. Conclusions This model serves as a valuable tool for both clinicians and DM patients, enabling early identification of CHD risk and facilitating the formulation of personalized prevention and treatment plans.
Kuang et al. (Tue,) studied this question.