Interpretable machine learning model for predicting recurrence in patients with diabetic foot ulcers.

Area of Science:

• Medical Informatics
• Computational Biology
• Diabetes Mellitus Research

Background:

• Diabetic foot ulcer (DFU) presents a significant challenge due to chronic disease course and high recurrence rates.
• Accurate prediction of DFU recurrence is crucial for timely interventions and improved patient outcomes.
• This study focuses on developing a machine learning (ML) model for predicting 3-year DFU recurrence risk.

Purpose of the Study:

• To develop and validate a machine learning model for predicting the 3-year recurrence risk of diabetic foot ulcers.
• To identify key risk factors associated with DFU recurrence using ML techniques.
• To enhance clinical decision-making for patients with diabetic foot ulcers.

Main Methods:

• A cohort of 494 DFU patients was divided into training and testing sets.
• Feature selection involved LASSO, mRMR, Fisher score, and RFE; seven ML algorithms were evaluated.
• Model performance was assessed using AUROC, with calibration via Platt scaling and interpretability through SHAP analysis.

Main Results:

• The XGBoost model achieved a high predictive performance with an AUROC of 0.924.
• The calibrated XGBoost model showed good calibration with a Brier score of 0.096.
• SHAP analysis confirmed the clinical relevance of identified risk factors, aligning with existing knowledge.

Conclusions:

• The developed XGBoost model demonstrates strong predictive accuracy and clinical relevance for DFU recurrence.
• Further multicenter validation with larger sample sizes is recommended to enhance generalizability.
• The model shows promise for improving patient management and outcomes in DFU care.