Predicting return to sport after multiligament knee injuries using machine learning: development and internal validation of a clinical algorithm

Carlos Suarez-Ahedo, Francisco Endara-Urresta, Carlos Peñaherrera-Carrillo, Alejandro Barros-Castro,
Predicting return to sport after multiligament knee injuries using machine learning: development and internal validation of a clinical algorithm,
The Knee,
2025,
,
ISSN 0968-0160,
https://doi.org/10.1016/j.knee.2025.10.003.
(https://www.sciencedirect.com/science/article/pii/S0968016025002558)
Abstract: Purpose
To develop and internally validate a machine learning–based predictive model to estimate the probability of return to sport (RTS) at 12 months after surgical reconstruction of multiligament knee injuries (MLKIs), and to identify the most influential clinical predictors. Methods: Patients who underwent MLKIs reconstruction between 2012 and 2022 across three tertiary care centers were included. Inclusion criteria comprised reconstruction of ≥2 ligaments, age 18–50, and ≥12 months of follow-up. The primary outcome was RTS at 12 months. Predictive variables included demographic, clinical, surgical, and functional data. Four models were compared: logistic regression, support vector machine (SVM), random forest, and XGBoost. Model performance was assessed via 10-fold cross-validation using AUC-ROC, accuracy, sensitivity, specificity, F1-score, and Brier score. SHAP analysis was used to interpret feature importance. Results: Among 220 patients, 60.9 % achieved RTS at 12 months. Machine learning models demonstrated strong predictive performance, with XGBoost yielding the highest accuracy (AUC = 0.84). Key predictors for RTS included a higher preinjury Tegner score, younger patient age, shorter time to surgery (<6 weeks), and higher baseline IKDC score. Psychological readiness and posterolateral corner reconstruction also contributed positively. Logistic regression showed inferior performance (AUC = 0.72). Conclusion: Machine learning models can accurately predict RTS following MLKIs using accessible clinical data. These tools may enhance individualized decision-making and guide postoperative rehabilitation strategies. Clinical Relevance: This model may assist clinicians in setting realistic patient expectations, personalizing care, and developing RTS-focused treatment algorithms.
Keywords: Multiligament knee injury; Machine learning; Predictive modeling; Sports medicine; Functional outcome; SHAP analysis