Risk prediction models for post-stroke cognitive impairment:a systematic review

ZHANG Jie; XI Chongcheng; KONG Yun; ZHONG Kelong; AN Xuemei

doi:10.12102/j.issn.1009-6493.2024.10.005

Abstract：Objective:To systematically evaluate the risk prediction models for post-stroke cognitive impairment.Methods:Research related to risk prediction models for post-stroke cognitive impairment was retrieved from China National Knowledge Infrastructure,Wanfang Data,China Biology Medicine database,PubMed,EMbase,the Cochrane Library,Web of Science,and EBSCO.The retrieval period was from establishment of databases to January 30,2023.2 researchers independently screened the literature,extracted data,and evaluated the risk of bias and applicability for inclusion in the study.Results:A total of 16 studies were included,including 19 risk prediction models for post-stroke cognitive impairment.Among them,16 models used Logistic regression analysis,2 models used random forest method,and 1 model used LASSO regression method.The area under the curve(AUC)of receiver operator characteristic during modeling were ranged from 0.773 to 0.940.4 models were subjected to the Hosmer-Lemeshow(H-L)test,with 2 models reported P-values and their P≥0.05.11 models underwent internal validation,5 models underwent external validation,and 4 models underwent both internal and external validation simultaneously.The 16 studies had good applicability,but there was a high bias,and the main problem was concentrated in the analysis field.Conclusions:The overall performance of the risk prediction models for post-stroke cognitive impairment is good,but the quality of the models need to be improved.In future research,it is necessary to optimize the research design,expand the sample size,select appropriate predictive factors according to clinical needs,improve statistical analysis methods.It also should focus on external validation of the model to verify its generalization ability.