The development of machine learning algorithms that can predict patients satisfaction using baseline characteristics, and preoperative and operative factors of total knee arthroplasty.

BACKGROUND: Patient satisfaction following primary total knee arthroplasty (TKA) is a crucial part to evaluate the success of the procedure. The purpose of this study was to develop to predict patient satisfaction following TKA.

METHODS: Satisfaction outcome data after 435 consecutive conventional TKAs performed between August 2020 and December 2021 were retrospectively collected. The total 26 input data were collected. The most favorable algorithm was first found using logistic regression (LR) and machine learning (ML) algorithms. To evaluate the predictive performance of the models, both area under curve (AUC) and F1-score were used as the primary metrics. The shapley additive explanations (SHAP) feature explanation in XGBoost and LR analysis were performed to interpret the model.

RESULTS: The performance of extreme gradient boosting classifier (XGBoost) was only higher than that of conventional LR in AUC (0.782 vs. 0.689). Comparing the F-1 score, only XGBoost showed better performance than LR (0.857 vs. 0.800). The most predictive feature in XGBoost was Short Form-36 physical and mental component summary scores (SF-36 MCS), followed by Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain, Bone mineral density (BMD). In the LR analysis, lumbar spine disease, WOMAC pain, and BMD were statistically significant.

CONCLUSION: XGboost showed the best performance and was superior to conventional LR in the prediction of patient satisfaction after TKA. The SF-36 MCS was the most important feature in the ML model. WOMAC pain and BMD were meaningful variables and demonstrated a linear relationship with satisfaction in both the LR and ML models.

LEVEL OF EVIDENCE: Retrospective cohort study; Level of evidence 3.