Developing and validating a machine learning model to predict chronic pain following total knee arthroplasty

This study aims to compare the performance of various machine learning algorithms in predicting chronic pain after total knee arthroplasty (CPSP).

Patients with CPSP after total knee arthroplasty at the same medical center between January 1, 2021, and January 1, 2023, were selected for this study. A retrospective cohort design was employed to collect samples, which were then randomly divided into a training set and a test set in a 7:3 ratio. Valid high-risk factors were identified using the Least Absolute Shrinkage and Selection Operator (LASSO) method. Subsequently, five predictive models were constructed and evaluated based on machine learning (ML) algorithms, including Decision Tree (DT), Light Gradient Boosting Machine (LGBM), Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). In the test dataset, the model's performance was evaluated using metrics including accuracy, precision, recall, specificity, F1-score, Brier score, and area under the curve (AUC). The Brier score helped identify the most suitable model, and SHAP values were explained to analyze the key factors affecting the predictions.

This study enrolled 785 patients who underwent total knee arthroplasty, with 549 in the training - set and 236 in the test - set. The overall CPSP incidence was 39.6%. Nine high - risk factors were identified: hospital stay length, albumin levels, acute postoperative pain status (APSP), non-operative pain status, pain catastrophizing, osteoporosis, preoperative operative-area pain score, education level, and rehabilitation site. The AUC values were: DT(0.877), LGBM(0.914), SVM(0.890), RF(0.918), and XGBoost(0.898). The Brier scores were: DT (0.123), LGBM (0.119), SVM (0.126), RF (0.111), and XGBoost (0.124). These findings suggest that the RF model had the best performance.

The incidence of CPSP in TKA patients is high, which has a significant adverse effect on body function and needs to be paid attention to. Nine risk factors have been identified. RF model can effectively identify CPSP patients, which is helpful for clinical medical staff to early identify and intervene in high-risk patients.