Data-Driven Approach to Development of a Risk Score for Periprosthetic Joint Infections in Total Joint Arthroplasty...

Area of Science:

• Orthopedic Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Periprosthetic joint infection (PJI) is a significant complication following total joint arthroplasty.
• Personalized risk prediction and management are crucial for improving patient outcomes.
• Electronic health records (EHRs) offer large-scale data for developing predictive models.

Purpose of the Study:

• To develop and evaluate data-driven, surgery-specific models for predicting PJI risk.
• To assess the utility of augmented EHR data and machine learning approaches for PJI prediction.

Main Methods:

• Utilized a large arthroplasty registry (58,574 procedures, 41,844 patients) from 2000-2019.
• Augmented registry data with EHR clinical, procedural, and laboratory variables (>100 predictors).
• Implemented traditional and machine learning models (e.g., lasso, random forest, XGBoost, neural networks) with 10-fold cross-validation.

Main Results:

• All models demonstrated similar discrimination in PJI risk prediction (c-statistic differences < 0.08).
• Relaxed lasso models showed the highest concordance: 0.787 (primary hip), 0.722 (revision hip), 0.681 (primary knee), 0.699 (revision knee).
• Predictors varied significantly across the four surgical groups (primary/revision hip/knee).

Conclusions:

• Augmenting data with EHRs provided only limited improvement in PJI risk stratification.
• Machine learning approaches yielded modest improvements in PJI risk prediction, potentially not justifying the added complexity.