Systematic Review of the Intraoperative Hypothermia Risk Prediction Models in Total Joint Arthroplasty Patients

Introduction: Machine learning (ML) identifies risk factors for intraoperative hypothermia (IH) more comprehensively than traditional scoring systems, offering effective guidance for nursing care. Despite promising results in total joint arthroplasty (TJA) patients—a high-incidence group—the quality of existing ML models requires systematic evaluation. This study reviews IH risk prediction models in TJA, focusing on their development quality and predictive performance. Purpose: This study aims systematically review and evaluate intraoperative hypothermia risk prediction models in TJA patients. Patients and Methods: A systematic search was conducted across nine databases (including PubMed, Embase, Cochrane Library, Web of Science, CINAHL, Wan fang database, CNKI, VIP database, and SinoMed) from inception to October 2025. Two independent reviewers performed the literature screening and data extraction, utilizing the PROBAST tool to assess study quality. Results: Eight studies were included, all involving model development and internal validation; four also performed external validation. Algorithms used were primarily Logistic regression (7 studies) and Random Forest (1 study). All models demonstrated good calibration and strong discriminatory ability, with the Area Under the Curve (AUC) values rangng from 0.791 to 0.938. Key predictors identified across studies include patient factors (age, BMI, hemoglobin level, ASA classification), surgical factors (duration, fluid/irrigation volume, blood loss, operating room temperature), and anesthesia factors (duration, active warming). Conclusion: IH risk prediction models for TJA patients demonstrate high performance and clinical applicability, with consistent predictors identified across the literature. However, the included studies exhibited a relatively high risk of bias. Future research should ensure high-quality data handling and standardization of validation processes. Prospective, multicenter studies are needed to refine these models, thereby providing clearer guidance for clinical decision-making. With the advancement of artificial intelligence, integrating current predictive models into visualized clinical tools will facilitate nursing decisions and reduce the incidence of intraoperative hypothermia in TJA patients. Prospero Registration Number: CRD420251134154. Keywords: total joint arthroplasty, intraoperative hypothermia, prediction model, systematic review, operating room nursing