Predicting 1-year outcomes after anterior cruciate ligament reconstruction from preoperative physical performance using machine learning: an exploratory study

This study aimed to develop machine learning (ML) models to predict 1-year outcomes after primary anterior cruciate ligament reconstruction (ACLR) using preoperative variables. A retrospective cohort of 75 patients who underwent ACLR between 2016 and 2023 was analyzed. Preoperative features included demographic characteristics, treatment-related variables, isokinetic muscle performance, and patient-reported outcome measures. One-year outcomes were defined as return to pivoting sports (RTS; Tegner activity level ≥ 6) and achievement of the patient acceptable symptom state (PASS; International Knee Documentation Committee score ≥ 75.9). Variables identified in univariable logistic regression were used for ML model development. Multiple algorithms were trained and evaluated using 3-fold cross-validation. Model performance was assessed using area under the receiver operating characteristic curve (AUC), calibration plots, and Brier scores. Among the models, the random forest algorithm demonstrated the best performance, achieving an AUC of 0.832 for RTS and 0.815 for PASS. SHapley Additive exPlanations analysis identified younger age and greater preoperative flexor strength of the uninvolved limb as the most important predictors. In this exploratory study, ML models incorporating preoperative clinical and physical performance variables showed promising potential for predicting 1-year outcomes after ACLR. However, these findings should be interpreted cautiously given the small sample size and lack of external validation. The models should therefore be considered exploratory and hypothesis-generating rather than ready for clinical implementation, and require further validation in larger, independent cohorts. Key Terms : anterior cruciate ligament reconstruction, machine learning, RTS, patient acceptable symptom state.