Objective To systematically evaluate risk prediction models for hypothermia in patients with severe trauma, in order to provide an evidence base for the selection and refinement of predictive tools in clinical practice. Methods A comprehensive search was conducted in electronic databases including PubMed, Web of Science, Embase, Cochrane Library, CINAHL, China National Knowledge Infrastructure, Wanfang Data Knowledge Service Platform, VIP Network, and SinoMedfrom from inception to July 31, 2025, to identify studies on risk prediction models for hypothermia in severe trauma patients. Two reviewers independently screened the literature, extracted data, and assessed the risk of bias and applicability using the Prediction Model Risk of Bias Assessment Tool. Results A total of 8 studies detailing the development of prediction models were included, yielding 8 models. The single sample sizes ranged from 91 to 489, with the number of outcome events ranging from 61 to 231. The area under the receiver operating characteristic curve for the model ranged from 0.704 to 0.990. Calibration assessment was conducted in all 8 studies, with 6 of them employing classic calibration testing methods such as the Hosmer-Lemeshow test. The most frequently identified predictors included the modified Glasgow Coma Scale score, Revised Trauma Score, Injury Severity Score, wet clothing on admission, and ambient temperature at the time of injury. PredictionModel Risk of Bias Assessment Tool indicated that while all models demonstrated good applicability, they exhibited a high risk of bias, primarily attributable to insufficient sample sizes and suboptimal handling of predictive variables, inadequate standardization of hypothermia measurement, and lack of effective validation. Conclusions Current prediction models show good discriminative potential, but due to the high risk of bias, there is insufficient evidence to recommend their routine use in clinical decision-making. Clinicians may refer to core predictors for empirical risk assessment. Future research should prioritize large-scale, multicenter external validation of existing models with high discriminative power and strictly standardize methodological aspects.