Temporal machine learning framework for diabetic foot ulcer healing trajectory prediction.

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Diabetic Foot Ulcer Management

Background:

• Diabetic foot ulcer (DFU) management is often reactive, relying on current wound status.
• Proactive treatment planning could significantly improve healing trajectories and patient outcomes.

Purpose of the Study:

• To develop a machine learning (ML) framework for predicting DFU healing phase transitions.
• To enable proactive DFU treatment planning using routinely collected clinical metadata.
• To integrate a recommendation system for treatment adjustments.

Main Methods:

• Analyzed longitudinal data from 268 patients with 329 DFUs across 890 appointments.
• Engineered 103 features, including temporal measurements normalized by inter-appointment intervals.
• Optimized an Extra Trees classifier using Bayesian tuning and feature selection to predict healing transitions (favorable, acceptable, unfavorable).

Main Results:

• Identified 30 essential predictors, reducing dimensionality by 70.9%.
• Achieved 78% ± 4% accuracy with an average AUC of 0.90.
• The integrated recommendation system showed 88.7% agreement for offloading prescriptions and chronicity-stratified performance for dressings.

Conclusions:

• The ML framework shows potential for predicting DFU trajectory using accessible clinical metadata.
• The system can guide proactive treatment planning, distinguishing wounds for standardized protocols versus those needing iterative experimentation.
• Prospective validation is pending to confirm the framework's clinical utility.