目的 系统评价脑卒中患者吞咽障碍风险预测模型，以期为临床实践提供参考依据。 方法 检索中国知网、万方、PubMed、Cochrane和Embase数据库中脑卒中患者吞咽障碍风险预测模型相关文献，对文献特征、研究类型、预测因子、模型构建方法和结果等进行分析和比较。 结果 共纳入11项研究，7项为模型的开发研究，4项为模型的开发和验证研究。受试者工作特征曲线下面积(the area under curve，AUC)均＞0.7，模型区分度较好；5项研究报告了校准度，拟合优度检验结果均P＞0.05，提示模型有较好的校准能力。模型适用性较好，但偏倚风险较高，主要原因为样本量不合理、直接剔除缺失数据、基于单因素分析法筛选预测因子、缺乏模型性能评估等。 结论 脑卒中患者吞咽障碍风险预测模型的研究尚处于发展阶段，未来可结合机器学习算法，开发本土化、预测性能良好、使用简便的预测模型。

Objective To systematically evaluate the risk prediction models of dysphagia in stroke patients，so as to provide a reference for clinical practice. Methods CNKI，Wanfang，PubMed，Cochrane and Embase databases were searched for literature related to risk prediction models of dysphagia in stroke patients.The literature characteristics，study types，predictors，model construction methods and results were analyzed and compared. Results A total of 11 studies were included，with 7 studies on model development and 4 studies on model development and validation.The area under the area under curve（AUC） was greater than 0.7，indicating that the model had good discrimination.Five studies reported the calibration degree，and the goodness-of-fit test results were all P＞ 0.05，indicating that the model had good calibration ability.The model had good applicability，but had a high risk of bias，mainly due to unreasonable sample size，direct exclusion of missing data，screening of predictors based on univariate analysis，and lack of model performance evaluation. Conclusion The research on risk prediction models of dysphagia in stroke patients is still in the development stage.In the future，machine learning algorithms can be combined to develop localized prediction models with good prediction performance and user-friendly features.

R47-05