目的 系统评价ICU机械通气患者呼吸机相关性肺炎（ventilator-associated pneumonia，VAP）发生风险的预测模型，为临床医护人员开发或选择合适的风险预测模型提供参考。方法 系统检索中国知网、万方数据库、维普期刊库、中国生物医学文献数据库、PubMed、EMbase、Web of Science、CINAHL和Cochrane Library中截至2025年4月30日发表的相关文献。2名研究者独立筛选文献与提取数据，并进行质量评价。结果 共纳入21篇文献，涉及27个预测模型，受试者工作特征曲线特征下面积为0.722~1.000。质量评价结果显示，21篇文献总体偏倚风险较高，13篇文献适用性较好，偏倚风险主要源于未选择合适的数据来源、样本量不足、自变量和缺失数据处理不当、仅采用单因素分析筛选预测因子、模型性能评估不全。重复报告出现频次较高的预测因子是ICU住院时间、呼吸机使用天数、急性生理学和慢性健康状况评分系统Ⅱ评分、气管切开、联合使用抗生素。结论 现有的ICU机械通气患者VAP发生风险预测模型的预测性能较好，但偏倚风险较高。未来研究者应重点关注研究设计的方法学细节及报告的规范性，以提高模型性能，便于临床推广。

Objective To systematically evaluate risk prediction models for ventilator-associated pneumonia （VAP） in ICU mechanically ventilated patients，so as to provide references for clinical doctors and nurses to develop or select appropriate risk prediction model. Methods We systematically searched CNKI，Wanfang，VIP，CBM，PubMed，EMbase，Web of Science，CINAHL，and Cochrane Library for relevant studies published before April 30，2025. Two investigators independently screened the literature and extracted data，and evaluated the quality. Results 21 studies were included，involving 27 models with an area under the subject working characteristic curve of 0.722~1.00. The overall risk of bias in 21 studies were high，and the applicability in 13 studies was good. The bias risks were mainly due to failure to select appropriate data sources，insufficient sample sizes，improper handling of independent variables and missing data，screening predictors based on univariate analysis，and incomplete model performance evaluation. The most frequently reported predictors were ICU length of stay，duration of mechanical ventilation，APACHE Ⅱ score，tracheostomy，and combined antibiotic use. Conclusion Existing VAP risk prediction models demonstrate satisfactory predictive performance but exhibit high bias risks. Future research should prioritize methodological rigor in study design and standardized reporting to enhance model validity for clinical implementation.