Risk prediction models for delirium in ICU patients: a systematic review and critical appraisal

This systematic review aimed to critically appraise the methodological quality, predictive performance, and clinical applicability of published risk prediction models for delirium in adult intensive care unit patients.

We searched PubMed, Embase, Web of Science, and the Cochrane Library from inception to April 8, 2025. Eligible studies were observational cohorts developing or validating multivariable prediction models for ICU delirium in adults (≥ 18 years). The review protocol was registered in PROSPERO (CRD420251028221). Study selection and data extraction followed the CHARMS checklist. Risk of bias and applicability were assessed using the Prediction Model Risk of Bias Assessment Tool (PROBAST).

A total of 27 studies, encompassing 26 prediction models, were included. Most models demonstrated apparently strong discrimination (AUCs ranging from 0.67 to 0.932), with common predictors including age, sedation, APACHE-II score, mechanical ventilation, and Glasgow Coma Scale score. However, the PROBAST assessment revealed a high risk of bias across most studies, primarily due to flaws in the analysis and predictor domains. Key limitations included inadequate handling of missing data, overfitting, predominant reliance on internal validation only, and heterogeneous delirium assessment methods. Only four models underwent both internal and external validation.

Despite the promising discriminatory performance of several models, significant methodological shortcomings limit their current clinical applicability and generalizability. Future research should prioritize robust external validation, improved methodological rigor, and implementation studies to assess real-world utility before routine clinical adoption.