Risk predictive models for delirium in the intensive care unit: a systematic review and meta-analysis

Background: An emerging approach to prevent delirium in an intensive care unit is the use of risk prediction models. At present, there is no scientific comparison of the predictive effect of the prediction model. This systematic review and meta-analysis aimed to compare the performance of available delirium risk prediction models for intensive care units. Methods: As of June 1st, 2019, articles on delirium prediction models of the intensive care patients were searched in the Cochrane Library, PubMed, Embase, Web of Science, CINAHL, ProQuest, and four Chinese databases. Studies describing the development or validation of risk prediction models for predicting delirium in ICU patients were included. The Prediction model Risk of Bias Assessment Tool (PROBAST) was used to assess the quality of included studies. A meta-analysis of the predictive performance was performed using the forest plot package in R3.6.1. Results: A total of 21 studies with 14 models were included in this article. PRE-DELIRIC, E-PRE-DELIRIC, and recalibrated PRE-DELIRIC model were the most popular models, which had been externally validated in at least two studies. The pooled area under the receiver operator characteristic curve (AUC) were 0.844 (95% CI: 0.793-0.896), 0.763 (95% CI: 0.680-0.846) and 0.776 (95% CI: 0.738-0.813) respectively. Most of the other models were with C-statistics above 0.7. Conclusions: The E-PRE-DELIRIC model, PRE-DELIRIC model, or both are recommended to predict ICU delirium risk. However, the recommendation should be considered with caution because of substantial heterogeneity. The protocol was registered with PROSPERO (CRD42019130802).