Joint-specific measures improve risk adjustment in total knee arthroplasty: A machine learning approach

Area of Science:

• Orthopedic Surgery
• Medical Machine Learning
• Arthroplasty Outcomes

Background:

• Accurate risk adjustment is crucial for assessing outcomes and quality in total knee arthroplasty (TKA).
• Existing prediction models often overlook joint-specific pathology, relying primarily on patient demographics and comorbidities.
• Machine learning (ML) offers potential for improved risk prediction in TKA.

Purpose of the Study:

• To evaluate if incorporating radiographic and clinical joint-specific parameters enhances ML-based risk adjustment for TKA.
• To determine if joint-specific features improve the prediction of postoperative complications and residual pain after TKA.
• To compare the performance of ML models with and without joint-specific variables.

Main Methods:

• Retrospective analysis of 1207 primary TKA procedures (2018-2022).
• Predictor variables included patient factors and joint-specific parameters (radiographic and clinical).
• Stacked gradient-boosting ensemble models (XGBoost, CatBoost) were developed and evaluated using AUC, accuracy, sensitivity, and specificity.

Main Results:

• Incorporating joint-specific parameters significantly improved the prediction of postoperative complications.
• Area Under the Curve (AUC) for major complications increased from 0.66 to 0.74; for any complications, it rose from 0.64 to 0.72.
• No improvement in prediction accuracy was observed for residual pain (Visual Analogue Scale ≥ 4).
• Key joint-specific predictors included prior septic surgery, large bone defects, and preoperative knee flexion < 70°.

Conclusions:

• Joint-specific features enhance ML-based prediction of postoperative complications in TKA.
• These parameters do not improve the prediction of residual pain following TKA.
• Joint-specific data may refine risk adjustment strategies for TKA complications.