Foot Posture Outweighs Habitual Footwear in Discriminating Concurrent Lower-Limb Pain in Children: A Cross-Sectional Interpretable Machine-Learning Study

Lower-limb pain is common in school-aged children, but the relative contributions of foot posture, adiposity, and footwear remain uncertain, and prediction models may be inflated by target leakage. We aimed to quantify how well foot posture, adiposity, and footwear discriminate current lower-limb pain, and whether machine learning outperforms penalised logistic regression. We analysed a cross-sectional school-based dataset of 4408 children aged 4–14 years from Türkiye. Eight classifiers were trained with five-fold cross-validation and evaluated once in a held-out 20% test set; pain-site information was excluded from the primary predictors because it encoded the outcome. Test values of the area under the receiver operating characteristic curve (AUROC) ranged from 0.945 to 0.962; extremely randomised trees (0.961; 95% CI 0.942–0.977) did not outperform penalised logistic regression (ΔAUROC 0.0003, 95% CI −0.0075 to 0.0095; p = 0.99). Foot posture alone reached an AUROC of 0.909, whereas demographics plus anthropometry reached 0.704, and adding footwear increased this only to 0.723. Therefore, concurrent lower-limb pain was discriminated mainly by foot posture and adiposity, while footwear and algorithmic complexity added little. As posture and pain were recorded concurrently, the models discriminate current rather than future pain. The findings are most relevant to first-contact assessment in general practice.