【摘要】 目的 系统评价脑卒中患者发生医院感染风险预测模型的特征，为选择适合的风险预测模型辅助临床决策提供参考。方法 检索Pubmed、Cochrane Library、CINAHL、Web of science、Embase、中国知网、万方、维普、中国生物医学文献数据库，将有关脑卒中患者发生医院感染风险预测模型的文献纳入，检索时限为建库时间至2025年4月。采用预测模型偏倚风险评估工具PROBAST对纳入文献的风险预测模型偏倚风险和适用性进行评估，并对纳入模型的共性预测因子采用Stata 17软件进行Meta分析。结果 共纳入10项研究，共23个预测模型。纳入研究的模型受试者工作特征曲线下面积(area under the curve,AUC)均>0.7，表明预测模型的整体预测性能良好。纳入文献所构建的模型适用性较好，但偏倚风险较高。Meta分析显示23个模型中具有预测作用的共同预测因子为出血性脑卒中、合并糖尿病、存在吞咽困难、静脉血栓、NIHSS评分和留置胃管。结论 目前已构建的脑卒中患者发生医院感染风险预测模型的偏倚风险较高，建议未来严格遵照PROBAST报告条目开展多中心、大样本的风险预测模型构建研究，同时对模型进行外部验证，以提升构建模型的普适性与外推性。更多还原

【Abstract】 Objective To systematically evaluate the characteristics of the risk prediction model for nosocomial infections in stroke patients,and to provide a reference for choosing an appropriate risk prediction model to assist clinical decision-making.Methods Pubmed,Cochrane Library,CINAHL,Web of science,Embase,CNKI,Wanfangdata,VIP,and Sino Med were retrieved.The literatures related to the risk prediction model of nosocomial infections in stroke patients were included.The search period is from the establishment time of the database to April 2025.The risk of bias and applicability of the risk prediction models in the included literature were evaluated by using the PROBAST risk assessment tool for prediction model bias,and the common predictors of the included models were analyzed by meta-analysis using Stata 17 software.Results A total of 10 studies and 23 prediction models were included.The area under the receiver operating characteristic curve( AUC) of the models included in the study was all > 0.7,indicating that the overall predictive performance of the prediction model was good.The models constructed by the included literature have good applicability,but the risk of bias is relatively high.Meta-analysis showed that the common predictors with predictive effects in the 23 models were hemorrhagic stroke,combined diabetes,presence of dysphagia,venous thrombosis,NIHSS score and indwelling gastric tube. Conclusion The bias risk of the currently constructed risk prediction models for nosocomial infections in stroke patients is relatively high.It is recommended that in the future,the construction research of multi-center and large-sample risk prediction models be carried out strictly in accordance with the PROBAST report items,and the models be externally verified at the same time to improve the universality and extrapolation of the constructed models.更多还原