Predicting return to sport after multiligament knee injuries using machine learning: development and internal...

Area of Science:

• Orthopedic surgery
• Sports medicine
• Machine learning in healthcare

Background:

• Multiligament knee injuries (MLKIs) require complex surgical reconstruction.
• Predicting return to sport (RTS) is crucial for patient outcomes and rehabilitation planning.
• Existing prediction methods may lack accuracy and comprehensive data integration.

Purpose of the Study:

• Develop and validate a machine learning (ML) model to predict RTS at 12 months post-MLKI surgery.
• Identify key clinical predictors influencing RTS after MLKI.
• Compare the performance of different ML algorithms for RTS prediction.

Main Methods:

• Retrospective analysis of 220 patients undergoing MLKI reconstruction (2012-2022).
• Utilized demographic, clinical, surgical, and functional data for prediction.
• Compared logistic regression, SVM, random forest, and XGBoost models, validated with 10-fold cross-validation.
• Employed SHAP analysis for feature importance interpretation.

Main Results:

• 60.9% of patients achieved RTS at 12 months.
• XGBoost model demonstrated superior predictive performance (AUC=0.84) compared to logistic regression (AUC=0.72).
• Significant predictors for RTS included higher preinjury Tegner score, younger age, shorter time to surgery (<6 weeks), and higher baseline IKDC score.

Conclusions:

• Machine learning models offer accurate RTS prediction post-MLKI using readily available clinical data.
• These predictive tools can personalize patient expectations and guide postoperative rehabilitation strategies.
• The developed model supports informed clinical decision-making for MLKI recovery.