Outcomes & Risk Prediction

XGBoost model for TKA complications: moderate discrimination for major complications (AUC 0.68), no better than chance for residual pain (AUC 0.53)

A retrospective single-center study of 783 primary total knee arthroplasties at Technical University of Munich reports that an XGBoost model trained on AAHKS-defined risk factors achieved moderate accuracy for predicting major complications and any complication, and performed at chance level for predicting residual pain at one year.

Journal of Orthopaedics · Apr 16, 2026