A machine-learning–based risk model for predicting wound healing after revascularization in patients with chronic limb-threatening ischemia

Abstract Objective Achieving complete healing of ischemic wounds represents a central therapeutic goal in the clinical management of patients with chronic limb-threatening ischemia (CLTI). This study aimed to develop a machine-learning model to predict 1-year wound healing and to identify its predictors in patients with CLTI. Methods This multicenter retrospective study analyzed 772 patients with CLTI and ischemic wounds who underwent infrainguinal revascularization between 2015 and 2022. Patients were randomly divided into the training (n = 579) and test (n = 193) cohorts. A random forest model was developed using 15 variables encompassing demographics, clinical, and limb-related factors, such as wound, ischemia, foot infection (WIfI) and Global Limb Anatomic Staging System (GLASS) stages. Model performance was evaluated using the area under the receiver operating characteristic curve (AUCROC), accuracy. Feature importance was assessed and compared with the results of the multivariable logistic regression analysis. Results The model exhibited excellent performance in the training subset (AUCROC = 1.00, true-positive rate [TPR] = 1.00, false-positive rate [FPR] = 0.12) and maintained acceptable discrimination in the internal validation (AUCROC = 0.72, TPR=0.89, FPR=0.63) and independent test cohorts (AUCROC = 0.78, TPR=0.93, FPR=0.70). The corresponding accuracies were 0.96, 0.72, and 0.77, respectively, indicating robust generalizability across datasets. In the ML model, serum albumin, hemoglobin, and age were identified as the most influential feature importance, followed by treatment modality, GLASS inframalleolar (IM) stage, and WIfI wound grade. In contrast, multivariable logistic regression analysis indicated that low serum albumin levels, advanced GLASS IM stage, and endovascular therapy were independently associated with poorer 1-year wound healing outcomes. Conclusion The ML-based model showed acceptable predictive performance for 1-year wound healing in patients with CLTI. Systemic factors, including low albumin, anemia, and advanced age, were identified as key determinants, underscoring the importance of comprehensive systemic management in addition to revascularization.