Predicting lower limb amputation risk based on clinical factors and wound characteristics using machine learning

Abstract This study investigates the multi dimensional risk factors driving lower limb amputation (LLA) in diabetic patients, aiming to delineate the distinct predictive weights of systemic metabolic markers, inflammatory flux, and localized wound characteristics. Utilizing a comparative framework of six machine learning architectures, we identified Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) as superior tools for identifying high-risk profiles, achieving a peak clinical sensitivity (Recall) of 0.800 and a discriminative stability (ROC-AUC) of 0.787. Our results demonstrate that systemic markers, specifically blood glucose and C-reactive protein (CRP), serve as the primary drivers of surgical risk, providing a critical biological lead time that significantly outperforms traditional localized wound morphology in predictive precision. A novel contribution of this work is the identification of divergent, gender specific risk architectures, males exhibit a Metabolic Dominant profile triggered by acute physiological flux, while females demonstrate a Cumulative Multifactorial risk pattern influenced by disease chronicity and nutritional status (Prealbumin). Furthermore, we identify a sharp one-year vulnerability window for amputation risk that varies across age specific cohorts. These findings advocate for a paradigm shift toward personalized, biomarker driven risk stratification, prioritizing acute metabolic stabilization in men and comprehensive longitudinal monitoring in women to optimize limb salvage outcomes.