An interpretable machine learning model for diabetic foot risk classification in patients with diabetes

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Diabetology

Background:

• Diabetic foot is a severe complication of diabetes, often leading to ulcers, infections, and amputation.
• Early identification of high-risk patients is crucial for effective preventive strategies.
• Current clinical evaluations require enhanced tools for accurate diabetic foot risk stratification.

Purpose of the Study:

• To develop an interpretable machine learning-based platform for classifying current diabetic foot risk status.
• To support clinical risk stratification and personalized care for diabetic patients.
• To create a tool for early identification of patients needing preventive interventions.

Main Methods:

• Retrospective analysis of data from 1938 diabetic patients (internal cohort) and 695 (external validation cohort).
• Development and comparison of ten machine learning models, with feature selection using Recursive Feature Elimination (RFE).
• Interpretation of model predictions using SHapley Additive exPlanations (SHAP) to build a risk classification platform.

Main Results:

• The CatBoost model, using sixteen selected features, achieved an AUC of 0.935 ± 0.016 and 0.88 accuracy in the internal cohort.
• The model demonstrated stable generalizability in the external validation cohort with an AUC of 0.922 ± 0.006 and 0.88 for key metrics.
• The developed model showed robustness against data noise and class imbalance.

Conclusions:

• A reliable CatBoost-based diabetic foot risk classification model incorporating sixteen clinically accessible features was developed.
• The model exhibits stable and dependable performance across diverse patient cohorts, indicating strong generalizability.
• The platform holds potential for real-world clinical application in risk classification and early preventive care for diabetic foot complications.