Predicting Diabetic Foot Ulcer Outcomes: Machine Learning-Based Refinement of IWGDF-Approved Classifications for Outpatient Services.

INTRODUCTION

This study aimed to validate and compare six IWGDF-approved classification systems for predicting poor prognostic outcomes in diabetic foot ulcers (DFUs), using the same sample in Iran. We also proposed modifications to enhance the performance and feasibility of these tools, in outpatient setting.

METHODS

A prospective cohort study was conducted involving 616 DFUs from 400 patients over a six-month period. We assessed the performance of six wound classification systems: Wagner, UTWCS, PEDIS/IDSA, SINBAD, WIFI, and DiaFORA. We adjusted for the key variables associated with poor outcomes that could affect the performance of these systems, employing ten machine learning techniques along with the Least Absolute Shrinkage and Selection Operator (LASSO) and random forest methods for feature selection methods.

RESULTS

Our findings indicated that both the SINBAD and UTWCS systems exhibited comparable effectiveness in predicting outcomes, significantly surpassing other systems. Notably, modifications to the WIFI system-specifically, redefining the wound depth classification to a clearer category-yielded improved predictive capabilities, outperforming the existing systems like SINBAD and UTWCS in predicting poor prognostic outcomes in outpatient setting.

CONCLUSION

SINBAD and UTWCS systems yield the best performance in our samples. Our proposed modification on the WIFI system can enhance its applicability for outpatient services according to reported performances.