Designing machine learning-based models to predict residual pain after functional treatment of acute lateral ankle ligamentous sprains

BACKGROUND: Machine learning (ML) has emerged as a transformative approach for outcome prediction in diverse disease contexts. This study sought to construct a prognostic prediction model leveraging ML algorithms and to elucidate risk factors linked to residual pain in patients with acute lateral ankle ligamentous sprains (ALALS) after functional treatment. METHODS: A total of 695 ALALS patients who received functional treatment were included in the analysis. Using a Visual Analog Scale (VAS) score of ≥ 4 at the one-year post-treatment mark as the criterion, participants were grouped into pain-free (PF) and residual pain (RP) categories. The optimal feature set was determined with backward elimination and exhaustive search. The performance of the model was assessed using key metrics, including the area under the curve (AUC), classification accuracy, sensitivity, and specificity. RESULTS: The study yielded two key findings: (1) ML models outperformed traditional statistical models in predictive performance, with Support Vector Machines (SVM) achieving the highest accuracy (AUC = 0.90); (2) Both ML and statistical models identified several significant predictors of residual pain, including Ankle Function Score (AFS), injury grade, spontaneous pain, inability to bear weight, and active range of motion (AROM). CONCLUSION: This study highlights the potential of ML classifiers, particularly SVM, in effectively predicting residual pain in ALALS patients following functional treatment. Additionally, it identifies key multivariate predictors, providing critical perspectives to support clinical decision-making and optimize patient management.