Insights From Inputs: Enhancing Revision Total Joint Arthroplasty Resource Allocation With Machine Learning...

Area of Science:

• Orthopaedic Surgery
• Health Informatics
• Machine Learning in Healthcare

Background:

• Revision total knee arthroplasty (rTKA) and revision total hip arthroplasty (rTHA) are complex, resource-intensive orthopaedic procedures.
• Accurate prediction of surgical duration, length of stay, and readmission is crucial for resource management and patient outcomes.
• Comparing predictive model performance across different data sources is essential for optimizing healthcare resource allocation.

Purpose of the Study:

• To compare the accuracy of machine learning models (Artificial Neural Networks - ANNs) using administrative versus institutional datasets for predicting key outcomes in rTKA and rTHA.
• To identify significant preoperative predictive features for these outcomes.
• To evaluate the performance of ANNs against multivariable regression models.

Main Methods:

• Utilized national (2014-2019) and institutional (2012-2022) arthroplasty databases for rTKA and rTHA cases.
• Developed Artificial Neural Networks (ANNs) and multivariable regression models for predicting surgery duration, length of stay, and 30-day readmission.
• Compared model performance between datasets using buffer accuracy (BA) and identified feature importance with Shapley Additive exPlanations (SHAP) values.

Main Results:

• Institutional ANNs showed higher accuracy for predicting surgery duration (76.2% vs 45.4% for rTKA, 55.4% vs 43.1% for rTHA).
• National ANNs demonstrated superior accuracy for predicting length of stay (81.8% vs 67.7% for rTKA, 71.4% vs 48.2% for rTHA).
• National database ANNs achieved AUC scores of 0.593 (rTKA) and 0.590 (rTHA) for 30-day readmission prediction.

Conclusions:

• ANN model performance significantly differed based on the dataset used, highlighting the importance of data source selection.
• All developed models outperformed predictions based on historic averages.
• Future research should leverage accurate, arthroplasty-specific datasets and identified significant predictive features for improved outcomes.