Development and Validation of a Machine Learning-Based Prediction of Chronic Postsurgical Pain Following Total Hip and Knee Arthroplasty: An Exploratory Study.

INTRODUCTION

Chronic postsurgical pain (CPSP) is a concern following total hip arthroplasty (THA) and total knee arthroplasty (TKA), substantially diminishing surgical benefits. We aimed to develop a machine learning (ML) prediction model for risk of developing CPSP.

METHODS

This cohort study included 383 patients recruited from a tertiary medical center between March 2022 and July 2024. Dataset was divided into a training set and a testing set. We used Lasso regression for feature selection and developed prediction models using six ML algorithms: logistic regression (LR), random forest (RF), extreme gradient boosting (XGBOOST), support vector machine (SVM), multilayer perceptron (MLP), and k-nearest neighbor (KNN). Model performance was evaluated using metrics such as receiver operator characteristic (ROC) curve and precision-recall (PR) curves, accuracy, sensitivity, and specificity. The optimal model was deployed on a web application using Shiny, and Shapley additive explanations (SHAP) analysis was used to interpret model predictions.

RESULTS

Among the 383 eligible patients, the median age was 67.4 years, and 58.0% were women. The incidence of CPSP in TKA and THA was 26.0% and 12.7% respectively. Four predictors were selected: preoperative pain catastrophizing scale (PRO-PCS), preoperative Tampa Scale for Kinesiophobia (PRO-TSK), preoperative numerical rating scale (PRO-NRS) at rest, and time to first ambulation. In the testing set, the RF model demonstrated the best performance with ROC-AUC of 0.872 and PR-AUC of 0.773. The accuracy, sensitivity, and specificity were 0.90, 0.77, and 0.94, respectively.

CONCLUSION

This study developed a clinically applicable ML model for CPSP prediction for THA, while the CPSP prediction for TKA was only conducted for exploration and still requires further verification.