Bin Chen,1– 3,* Yaji Yang,1– 3,* Haotian Zhou,1– 3 Feilong Li,1– 3 Yidong Shen,1– 3 Qiang Cheng,1– 3 Wei Huang,1– 3 Leilei Qin1– 3 

1Department of Orthopaedic Surgery, The First Affiliated Hospital of Chongqing Medical University, Chongqing, People’s Republic of China; 2Chongqing Municipal Health Commission Key Laboratory of Musculoskeletal Regeneration and Translational Medicine, Chongqing Medical University, Chongqing, People’s Republic of China; 3Orthopaedic Research Laboratory of Chongqing Medical University, Chongqing Medical University, Chongqing, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Wei Huang, Department of Orthopaedics, The First Affiliated Hospital of Chongqing Medical University, Chongqing, 400016, People’s Republic of China, Email huangwei68@263.net Leilei Qin, Department of Orthopaedics, The First Affiliated Hospital of Chongqing Medical University, Chongqing, 400016, People’s Republic of China, Email 253505921@qq.com

Background: Preoperative diagnosis of periprosthetic joint infection (PJI) is crucial for guiding treatment strategies and improving patient outcomes. This study aims to develop a new diagnostic model for the preoperative diagnosis of PJI based on serum and synovial fluid markers and further validate its effectiveness.
Methods: We retrospectively collected data from patients admitted for joint revision surgery between January 2018 and October 2022, selecting serum and synovial fluid markers as variables for the study. The most suitable diagnostic markers were selected using LASSO regression, and eight machine learning (ML) models were constructed based on the selected markers. The diagnostic performance and clinical utility of the ML models were assessed using receiver operating characteristic curves, calibration curves, decision curve analysis, and clinical impact analysis. Finally, the best model was compared to existing diagnostic standards using an external validation cohort.
Results: A total of 376 cases were analyzed (263 in the training cohort and 113 in the validation cohort), with 111 cases (29.52%) diagnosed as PJI. The ML models included SE-IL6, SE-CRP, ESR, SF-IL6, PMN%, DD, and ALB. The eXtreme Gradient Boosting model was the optimal model, achieving an area under the curve of 0.998 (95% CI 0.993– 1) in the test set, outperforming other models. It also recorded the lowest Brier score of 0.062 and the highest F1 score of 0.985. In the external validation cohort, the accuracy, sensitivity, and specificity of the ML diagnostic model were higher than those of the MSIS 2013 and ICM 2018 diagnostic criteria.
Conclusion: Our newly developed ML diagnostic model can assist clinicians in rapidly and accurately diagnosing PJI before surgery and has potential value for timing decisions regarding two-stage revisions. It has high economic value and clinical applicability.

Keywords: periprosthetic joint infection, machine learning, diagnostic model, preoperative diagnosis, synovial fluid, inflammatory markers