Diagnostic accuracy of delirium screening tools in the emergency department: A network meta-analysis

Background: Delirium in the emergency department (ED) is common, frequently missed, and assessed using multiple screening tools with uncertain comparative performance. We compared diagnostic accuracy and administration time and explored efficient screening strategies. Methods: Embase, MEDLINE, and CENTRAL were searched through June 6, 2026, with citation searching. Eligible studies evaluated structured delirium screening tools against standardized diagnostic criteria and provided 2 × 2 data. Two reviewers screened records, extracted data, and assessed risk of bias using the QUADAS-2 tool. A Bayesian bivariate random-effects network meta-analysis was performed. Primary analyses compared six functional categories; secondary analyses evaluated individual instruments. Results: Twenty studies comprising 9664 examinations evaluated 16 tools. The Delirium Triage Screen had the highest sensitivity [0.96; 95% credible interval (CrI) = (0.84, 1.00)], whereas composite delirium diagnostic algorithms had the highest specificity [0.95; 95% CrI = (0.92, 0.97)] and Youden Index [0.74; 95% CrI = (0.64, 0.81)]. Instrument-level estimates varied within categories; among higher-performing tools, 4AT combined favorable performance [Youden Index = 0.74; 95% CrI = (0.58, 0.83)] with the largest evidence base (7 datasets). Administration time was not associated with diagnostic performance (Spearman ρ = −0.31; P =. 54). In simulation at 10% delirium prevalence, sequential arousal/sedation followed by attention/working-memory assessment yielded a positive predictive value of 73.8% versus 63.4% for composite algorithms, with estimated administration times of 1.0 versus 3.0 min. Conclusion: Category-level and instrument-level analyses provided complementary information on screening approaches and tool selection. Sequential arousal- and attention-based screening showed greater modeled efficiency and warrants prospective evaluation before clinical implementation.