Risk prediction models for pneumonia in hospitalized stroke patients: A systematic review

ObjectiveTo systematically evaluate risk prediction models for pneumonia occurrence during hospitalization in stroke patients. MethodsComputer searches were conducted in the PubMed, Embase, Web of Science, Cochrane Library, and EBSCO databases for literature related to risk prediction models for pneumonia in hospitalized stroke patients, with search dates ranging from database inception to June 13, 2024. Two researchers independently screened the literature, extracted the data, and evaluated the risk of bias and applicability of the included studies via the Prediction Model Risk of Bias ASsessment Tool (PROBAST). ResultsA total of 43 studies were included, among which 33 studies developed a total of 56 new models, and 25 studies externally validated 19 models. Among the 56 new models, 29 used a logistic regression model (LR), 25 used a machine learning model (ML), 1 used a classification and regression tree model (CART), and 1 used a linear regression model. The reported area under the curve (AUC) ranged from 0.565 to 0.960. The number of predictors explicitly reported for one model was 1,046, with the top three predictors most commonly used being age, the National Institutes of Health Stroke Scale (NIHSS) score, and dysphagia. The PROBAST results revealed that all 43 studies had a high risk of bias, and 27 studies had poor applicability. ConclusionAlthough the pneumonia risk prediction models for hospitalized stroke patients in the included studies achieved good predictive performance, the overall quality needs improvement. Future research should follow stricter study designs, standardized reporting practices, and multicenter large-sample external validation.