Summary

Prediction of functional range of motion after ACLR

Abstract

Objective

To develop and validate an interpretable machine learning-based model that predicts the postoperative functional range of motion (ROM = 120°) after anterior cruciate ligament reconstruction (ACLR), with the aim of enhancing personalized postoperative rehabilitation strategies and improving patient outcomes.

Methods

This prospective cohort study included 43 patients who received ACLR between January 2022 and December 2022 for training and internal validation, and 19 patients between February 2023 and July 2023 for prospective validation. Data collected included demographic information, clinical characteristics, and functional assessment scores. Functional ROM was assessed at 3 months post-operation. Six different machine learning algorithms were used to develop predictive models. The best performing model were identified through the area under the receiver operating characteristic curve (AUC) and validated with a separate test dataset and multiple evaluation metrics. Shapley Additive Explanations (SHAP) was used to assess the feature importance and interpretation of the model.

Results

Among the six models, the XGBoost model demonstrated highest predictive accuracy with a test set AUC of 0.817 and training cross-validation AUC of 0.892 (Std=0.163). The global feature interpretation of SHAP showed that leg length, age, SF12-MCS score, pre-operative ROM, and LEFS score were the top five important features. We also constructed a web calculator for personalized prediction of ROM prognosis, and SHAP-based individualized feature interpretation.

Conclusions

The developed XGBoost-based predictive model provides an accurate and interpretable tool for predicting functional ROM after ACLR. This model can potentially assist clinicians and physiatrists in tailoring personalized rehabilitation programs based on individual risk profiles, expediting recovery and improving long-term postoperative joint function.