300 Section - Translating Machine Learning into Clinical Practice: Predicting Functional Recovery After Total Knee Arthroplasty

Total knee arthroplasty (TKA) is an effective treatment for end-stage knee osteoarthritis (OA). Although most patients show improved function, 10–30% continue to experience limitations beyond one year. This study aimed to develop predictive models using preoperative variables to forecast 3-month functional outcomes after TKA with machine learning algorithms.

Design: We retrospectively analyzed 313 patients who underwent TKA between September 2013 and January 2022. Data included demographic and anthropometric variables, clinical characteristics, and functional assessments: Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC), visual analogue scale (VAS), EQ-5D, Timed Up and Go (TUG), 6-minute walk test (6MWT), stair climbing test (SCT), and peak torque of knee extensor and flexor strength, measured preoperatively and at 3 months. Five algorithms—Linear Regression, Poisson Regressor, Tweedie Regressor, Passive Aggressive Regressor, and ARD Regression—were used. External validation employed data from 7 additional patients. Feature importance and SHAP (SHapley Additive exPlanations) values were applied to interpret predictors. The 10 most influential variables in each model, along with significant factors from multivariate analysis, were comprehensively evaluated.

Results: Among 313 patients (mean age 71.8 ± 5.9 years; 15.7% male), predictive performance varied by algorithm. The most significant preoperative predictors of 3-month outcomes were TUG, SCT ascending and descending, and peak torque of knee flexor and extensor strength. These consistently ranked highly across feature importance analyses and SHAP values. Models incorporating these predictors demonstrated robust explanatory power and reliable external validation performance.

Conclusion: Machine learning models using preoperative clinical and functional variables effectively predicted short-term recovery after TKA. TUG, stair climbing, and muscle strength were key predictors. Incorporating these measures into preoperative evaluation may support prognosis estimation, patient counseling, and individualized rehabilitation planning.