Decision tree models highlight lower social and physical function in individuals with recurrent plantar diabetes-related foot ulcers: A 1-year prospective observational cohort study

AIM: We investigated whether decision tree (DT) models and predictor selection strategies can improve stratification of individuals with diabetes at high risk of plantar foot ulcer recurrence. METHODS: In a 12-month prospective cohort of 58 participants (mean age 65 years, 83% men) at high risk of diabetes-related plantar ulcer recurrence, three DT models were developed: a predefined model with predictors linked to cumulative plantar tissue stress, and two data-driven models with predictors selected using univariate or elastic net-regularized logistic regression. Performance evaluation included sensitivity, specificity, accuracy and area under the receiver operating characteristic curve (AUC). RESULTS: Twenty-two participants (38%) developed a recurrent plantar ulcer. The predefined model showed low sensitivity (57%) and specificity (64%). The univariate selection model showed higher performance estimates (sensitivity 71% and specificity 82%). This DT was based on social and physical functioning (36-Item Short Form Health Survey) and self-selected walking speed, with lower social and physical functioning scores indicating higher risk of recurrence. CONCLUSION: In this exploratory study, data-driven DT models identified quality-of-life domains as candidate predictors of plantar ulcer recurrence. If validated in larger cohorts, quality-of-life assessments could represent a low-cost addition to current risk stratification in diabetes-related foot ulcer prevention.