Abstract:

Objective To conduct a scoping review of studies on machine learning-based prediction models for intraoperative hypothermia in adults，in order to provide a reference for optimizing the management of intraoperative hypothermia. Methods Following scoping review guidelines，we systematically searched original studies in databases including Embase，Web of Science，CINAHL，Cochrane Library，Elsevier ScienceDirect，Scopus，PubMed，IEEE Electronic Library Electronic Library，CNKI，CBM，Wanfang Database，and VIP Database from their inception to September 26th，2025. The included literature was summarized and analyzed. Results A total of 15 articles were included，encompassing 42 prediction models. The area under the receiver operating characteristic curve（AUC） of the models ranged from 0.717 to 0.974，with an average value of 0.830. The quality of data requires improvement. Commonly used machine learning algorithms included Random Forest，XGBoost，Logistic Regression，Decision Tree，and Support Vector Machine. Frequent predictive factors were anesthesia duration，surgery duration，intraoperative fluid volume，BMI，baseline body temperature，operating room temperature，and age. The studies only conducted internal validation，lacking external validation；only a few studies performed hyperparameter optimization or interpretability analysis. Conclusion In the past 5 years，the number of machine learning-based prediction models for intraoperative hypothermia in adults has grown rapidly. Machine learning demonstrates significant potential in predicting intraoperative hypothermia，with tree-based models（e.g.，Random Forest，XGBoost） showing particularly optimal performance. Future efforts should focus on improving data quality，promoting multi-center prospective studies，employing hyperparameter tuning methods，strengthening external validation，incorporating explainable artificial intelligence techniques to enhance model interpretability and clinical translation，and evaluating their cost-effectiveness and clinical utility to achieve precise temperature management.

Key words: Machine Learning Algorithms, Intraoperative Hypothermia, Nursing Care, Prediction Model, Scoping Review

摘要：

目的 对基于机器学习算法构建成人术中低体温预测模型的相关研究进行范围综述，为优化术中低体温管理提供参考。 方法 依据范围界定审查指南，系统检索Embase、Web of Science、CINAHL、Cochrane Library、Elsevier ScienceDirect、Scopus、PubMed、IEEE Electronic Library、中国知网、中国生物医学文献数据库、万方数据库、维普数据库中的相关研究，检索时限为建库至2025年9月26日，对纳入文献进行归纳和分析。 结果 共纳入15篇文献，共42个预测模型，模型受试者工作特征曲线下面积为0.717~0.974，平均值为0.830。数据的质量有待提高。常用机器学习算法包括随机森林、XGBoost、Logistic回归、决策树及支持向量机等。常见预测因子为麻醉时间、手术时间、术中输液量、BMI、基础体温、手术室温度和年龄。研究仅进行内部验证，缺乏外部验证；仅少数研究进行了超参数优化或可解释性分析。 结论 近5年来基于机器学习算法成人术中低体温预测模型数量快速增长，机器学习在术中低体温预测中展现出良好潜力，尤其以树模型（如随机森林、XGBoost）性能较优。未来，应提高数据质量、推进多中心、前瞻性研究，采用超参数调优方法，加强外部验证，引入可解释人工智能技术，促进模型解释及临床转化，并评估其成本效益与临床效用，以实现精准的体温管理。

关键词: 机器学习算法, 术中低体温, 护理, 预测模型, 范围综述