Risk Prediction Models for Perioperative Hypothermia: A Systematic Review.

Background: Perioperative hypothermia is a frequent complication causing patient discomfort and increasing risks like surgical site infection, coagulation dysfunction, slow drug metabolism, cardiovascular events, and prolonged hospitalization, which severely affect prognosis. Due to its significant impact, this study systematically reviews and evaluates existing risk prediction models for perioperative hypothermia. The aim is to provide clinical staff with a reference for selecting or developing an appropriate prediction model.

Methods: A systematic search was carried out in PubMed, Embase, Web of Science, the Cochrane Library, and CINAHL to find relevant studies on perioperative hypothermia risk prediction models from the inception of databases to May 23, 2024. Two reviewers independently screened abstracts and full texts for eligibility. Data collection followed the checklist for critical appraisal and data extraction for systematic reviews of prediction modelling studies (CHARMS). The prediction model risk of bias assessment tool (PROBAST) checklist assessed the risk of bias and applicability of the data.

Results: This study included 11 papers (14 risk prediction models). Models showed good predictive performance (the area under the curve (AUC) range: 0.700-0.870). Nine studies reported calibration; validation involved internal (n=3), external (n=3), or both (n=3). PROBAST indicated high risk of bias in all 11 papers, primarily due to insufficient model validation. The most common predictors were age, baseline temperature, BMI, fluid/infusion/rehydration volume, operating room temperature, anesthetic time, and operative time.

Conclusion: The overall discrimination and applicability of perioperative hypothermia risk prediction models are good, but the risk of bias is high and the quality of studies needs to be further improved. In the future, a more standardized approach should be used to optimize existing models, develop more targeted prediction models with a low risk of bias, and conduct internal and external validation to improve their predictive accuracy in clinical application.