A dynamic weighted ensemble learning framework for cardiovascular risk prediction in type 2 diabetes: a comparative...

Area of Science:

• Cardiovascular Health
• Diabetes Mellitus Research
• Integrative Medicine

Background:

• Cardiovascular complications are the leading cause of death in diabetes mellitus patients.
• Traditional risk assessment models have limitations in capturing complex, non-linear relationships and generalizing across diverse populations.
• Existing models often fail to integrate diverse diagnostic indicators effectively.

Purpose of the Study:

• To develop and evaluate a novel multi-index dynamic weighted ensemble model for predicting cardiovascular complications in Type 2 diabetes.
• To integrate Traditional Chinese Medicine (TCM) tongue diagnosis indexes with modern medical biomarkers.
• To enhance the accuracy and generalizability of diabetes risk assessment.

Main Methods:

• A cross-sectional study enrolled 3,111 Type 2 diabetes patients, with 2,895 included in the final analysis.
• Base models were built using Random Forest (RF), Gradient Boosting Decision Tree (GBDT), K-nearest Neighbors (KNN), and eXtreme Gradient Boosting (XGBoost) algorithms.
• A dynamic weighted ensemble framework was developed to optimize the integration of diverse diagnostic indexes.

Main Results:

• The ensemble model achieved high performance metrics: 95.68% accuracy, 94.92% sensitivity, and 96.21% specificity.
• SHAP analysis identified significant non-linear influences of indexes like chest tightness, ESR, and tongue purple on diabetes risk.
• Ablation tests confirmed the superiority of the ensemble framework over single-algorithm approaches.

Conclusions:

• The integrated ensemble model significantly outperforms existing risk assessment models for diabetic cardiovascular complications.
• This approach provides a reliable tool for early screening and personalized intervention strategies.
• The study highlights the clinical value of integrating TCM and Western medical diagnostic indexes.