Introduction and Objective: Cardiovascular disease (CVD) is a major complication in diabetes, requiring effective risk prediction tools. This study developed an AI-driven scoring system for CVD prediction, comparing machine learning models with the Framingham risk score (FRS).Methods: We analyzed two health datasets, the China Health and Retirement Longitudinal Study (CHARLS) and Shanghai Diabetes Studies (SDS), each with a 5-year follow-up. CVD was defined based on clinical histories and self-reported diagnoses of heart disease or stroke. Predictors included anthropometric measures, baseline labs, and disease status. Non-collinear datasets were created using LASSO-based feature selection. Models predicted CVD and were evaluated using stratified cross-validation and external validation, employing AUC, F1, and ROC curve metrics.Results: Logistic regression outperformed other models with the highest AUC (0.66). Compared to FRS (AUC: 0.61), the AI model demonstrated superior predictive performance. External validation (SDS) confirmed these findings, with consistent metrics across datasets, supporting the robustness and generalizability of the AI-driven scoring system.Conclusion: An AI-driven scoring system outperformed the Framingham risk score, providing a robust and generalizable tool for CVD prediction in diabetes.
T. Julaiti: None. J. Pan: None.
Hainan Provincial Natural Science Foundation of China (821MS159)
Source link
Leave a Reply