目的 评估非计划重返重症监护室（intensive care unit,ICU）风险预测工具的构建与验证，探讨其临床应用潜力。方法 采用范围综述方法学框架，检索PubMed、CINAHL、Web of Science、Cochrane Library、Embase、中国知网、维普、万方、中国生物医学文献数据库的相关文献，检索时限为建库至2024年3月1日。结果 纳入19项研究，含18个模型，非计划重返率为2.36%~9.93%，12 个模型受试者工作特征曲线下面积（area under the receiver operating characteristic curve,AUC）超0.75。年龄、呼吸、心率、性别、疾病诊断等为非计划重返ICU常用的预测变量。结论 13 个模型以工具的形式呈现，多数模型展示了较好的区分能力。机器学习（machine learning,ML）在提高准确性方面具有优势，但模型可解释性、临床适用性及外部验证需进一步研究。未来需提升模型普适性和准确性，优化临床应用，并开发更直观、用户友好的模型。

Objective To evaluate the construction and validation of risk prediction tools for unplanned readmission to the intensive care unit(ICU),and to explore their potential for clinical application.Methods The scoping review methodology framework was used to retrieve relevant literature from PubMed, CINAHL, Web of Science, Cochrane Library, Embase, CNKI, VIP, Wanfang, and China Biology Medicine disc from the inception of each database to March 1, 2024. Results 19 studies,including 18 models,were included,with unplanned readmission rates ranging from 2.36% to 9.93%.The area under the receiver operating characteristic curve(AUC) for 12 models exceeded 0.75.Age,respiratory status,heart rate,gender,and disease diagnosis were commonly used predictive variables.Conclusions 13 models were presented as tools,and the majority demonstrated good discriminative ability.Machine Learning(ML) showed dominance in improving accuracy,but model interpretability,clinical applicability,and external validation need to be further investigated.It is necessary to enhance model generalizability and accuracy,optimize their clinical application,and develop more intuitive and user-friendly models in the future.

R473

甘肃省自然科学基金(22JR5RA920)；甘肃省教育厅优秀研究生“创新之星”项目(2023CXZX-153)