Related Experiment Video
Updated: Jun 29, 2026
14:56
An Experimental Paradigm for the Prediction of Post-Operative Pain (PPOP)
Published on: January 27, 2010
Predicting persistent pain after total knee arthroplasty using different machine learning algorithms
Anni Rajamäki1, Aleksi Reito2, Mari Karsikas2
1Coxa Hospital for Joint Replacement and Faculty of Medicine and Health Technology, Tampere University, Tampere, Finland. anni.rajamaki@tuni.fi.
Acta Orthopaedica
|June 22, 2026
Summary
Predicting outcomes after total knee arthroplasty (TKA) is challenging. Machine learning models using extensive patient data showed poor accuracy in identifying patients with persistent pain or low functional scores one year post-TKA.
Area of Science:
- Orthopedic surgery
- Biostatistics
- Machine learning in healthcare
Background:
- Total knee arthroplasty (TKA) has a 10-20% dissatisfaction rate.
- Predictive models can improve patient selection and counseling.
- Identifying patients at risk for poor outcomes is crucial.
Purpose of the Study:
- Develop a precise machine learning model to predict TKA outcomes.
- Identify patients with residual pain or low Oxford Knee Score (OKS) one year post-TKA.
- Stratify patients based on predicted functional outcomes.
Main Methods:
- Retrospective cohort study of 11,755 primary TKA patients.
- Utilized 751 patient-related variables.
- Employed Extreme Gradient Boosting (XGBoost) machine learning algorithm.
- Assessed model performance using Area Under the Curve (AUC).
Main Results:
- 850 patients (7.2%) experienced persistent pain one year post-TKA.
- The prediction model achieved an AUC of 0.67.
- Key predictors included lower preoperative OKS, younger age, valgus malalignment, and specific medication use.
Conclusions:
- Current machine learning models demonstrate poor predictive accuracy for TKA outcomes.
- Predicting pain and functional outcomes post-TKA remains a significant challenge.
- Even with extensive data and advanced algorithms, precise prediction is difficult.
