Machine Learning Based Diagnostic Models for Hip and Knee Prosthetic Joint Infection: A Systematic Review.

Abstract

Prosthetic joint infection (PJI) is a devastating complication that affects up to 1.7% of patients within 2 years following total hip arthroplasty (THA) or total knee arthroplasty (TKA). PJI is associated with significant patient morbidity, reduction in quality of life, prolonged hospitalisation, and healthcare expenditure. With a 5-year mortality reported as high as 21%, PJI is one of the most feared complications of joint arthroplasty. Identification of PJI is of critical importance to ensure successful and definitive treatment. However, diagnosis remains challenging due to the lack of a gold standard, culture-negative infection, and varying sensitivity and specificity of diagnostic tests. The rapid expansion of machine learning (ML) in the literature has led to the emergence of models that utilise patient demographics, clinical features, serological studies, synovial fluid biomarkers, and imaging to improve PJI diagnostics. The purpose of this study was to describe the literature on using ML to diagnose PJI. A systematic review of the literature for original studies describing ML use in PJI diagnostics following THA or TKA was conducted. This review identified 12 studies applying ML to diagnose or predict PJI, through patient demographics, clinical features, and imaging. Most models demonstrated good predictive performance, with Area Under the Curve (AUC) from 0.68 to 0.993. However, few studies validated their models externally. In conclusion, ML presents a promising approach to enhance PJI diagnostic accuracy, which may reduce diagnostic delays and ensure appropriate treatment. Further studies are needed to assess the model's generalisability and validate these models in external cohorts. Statement of Clinical Significance: The diagnosis of PJI remains a challenge due to limitations in current diagnostic criteria. ML offers a data-driven approach to improve diagnostic accuracy, potentially allowing earlier and more accurate identification to ensure appropriate treatment in a timely manner.