Abstract
Background: Periprosthetic joint infection (PJI) remains a significant and complex complication following total hip and knee arthroplasty. This study aims to design, validate, and assess a machine learning (ML) model for predicting the likelihood of PJI in individuals undergoing revision arthroplasty procedures. Methods: A retrospective analysis was conducted on patients who underwent hip or knee revision arthroplasty between 1 January 2015 and 31 March 2021. Data were collected from preoperative clinical histories, laboratory results, and patient demographics. The final dataset was used to train multiple classification models for the preoperative prediction of PJI. Results: A total of 1360 patients were included, comprising 1141 cases in the aseptic group and 219 in the infected group. The best-performing model, a Linear Support Vector Machine (SVM), demonstrated reasonable predictive capability for PJI, achieving an area under the curve (AUC) of 0.770 ± 0.008 in the training set and 0.730 ± 0.078 in the testing set. Additionally, three key predictors of PJI were identified. Conclusions: The Linear SVM model, developed using preoperative clinical information, exhibited reasonable performance in predicting PJI. While further refinement and validation are necessary, integrating ML tools into the preoperative evaluation process has the potential to enhance personalized risk assessment, support informed decision-making, and optimize surgical preparation for patients undergoing prosthetic revision surgery.