Probability Score for the Diagnosis of Periprosthetic Joint Infection: Development and Validation of a Practical...

Area of Science:

• Orthopedic Surgery
• Infectious Diseases
• Machine Learning in Medicine
• Biomarker Discovery

Background:

• Diagnosing periprosthetic joint infection (PJI) relies on complex criteria-based systems often inconsistently applied in clinical practice.
• Existing diagnostic methods can lead to suboptimal adoption and implementation, even among experts, highlighting a need for improved accuracy and accessibility.
• Accurate and timely PJI diagnosis is crucial for effective patient management and preventing treatment complications.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for preoperative PJI diagnosis.
• To generate a PJI probability score using only synovial fluid (SF) biomarkers within a 24-hour timeframe.
• To address the limitations of current diagnostic criteria by creating a more accessible and accurate tool.

Main Methods:

• A two-stage ML model was built using 104,090 SF samples from 2,923 institutions (2018-2024).
• Unsupervised learning identified sample clusters, followed by supervised logistic regression to generate PJI scores (0-100) based on 10 SF biomarkers (excluding culture results).
• Model performance was validated against modified 2018 International Consensus Meeting criteria, including probabilistic reclassification of inconclusive cases.

Main Results:

• The ML model achieved high diagnostic accuracy in the validation cohort: 99.3% sensitivity and 99.5% specificity (pre-reclassification).
• The model successfully resolved 95% of samples deemed inconclusive by clinical standards.
• Key influential biomarkers included alpha-defensin, neutrophil percentage, and white blood cell count; the model performed well in culture-negative infections.

Conclusions:

• The ML model provides exceptional diagnostic accuracy for PJI using SF biomarkers, significantly reducing diagnostic uncertainty.
• The algorithm enables definitive diagnostic information within 24 hours, independent of SF culture results.
• This ML approach matches clinical standard accuracy while simplifying implementation, bridging the gap hindering current criteria-based diagnostics.