Efficient triage of emergency patients is crucial for the immediate identification of critically ill individuals and enables rapid interventions to improve patient outcomes. Systematic review and meta-analysis of the diagnostic test accuracy (DTA) of the Emergency Severity Index (ESI) for identifying critically ill adult patients in the emergency department (ED). We considered all studies (case-control and cohort studies) that evaluated the DTA of the ESI in adult patients attending an ED. The outcome of a triage system is the high urgency of treatment, commonly used reference standards are short-term mortality or admission to an ICU. We searched four bibliographic databases up to 13 February 2025. Screening, inclusion, data extraction, and assessment of methodological quality followed standard Cochrane methodology. We calculated measures of DTA for all studies against the reference standards and calculated pooled estimates using a bivariate random effects model. We included 27 studies, representing 510 777 patients. Methodological quality according to the QUADAS-2 tool was high, except for risk of bias in patient selection, which was high for 12 (44%) studies. A total of 18 studies provided data for the reference standard short-term mortality, with an estimated pooled sensitivity of 81.8 95% confidence interval (CI): 71.8-88.9, specificity of 70.5 (60.5-78.8), diagnostic odds ratio (DOR) of 10.8 (5.4-21.4), positive likelihood ratio of 2.77 (2.02-3.81), and negative likelihood ratio of 0.26 (0.16-0.41). For the reference standard ICU admission, based on 10 studies, pooled estimates were sensitivity of 81.5 (65.2-91.2), specificity of 81.7 (71.9-88.6), DOR of 19.7 (5.5-70.7), positive likelihood ratio of 4.45 (2.58-7.84), and negative likelihood ratio of 0.23 (0.11-0.49). Those results remained stable in the sensitivity analysis. ESI showed a moderate-to-high diagnostic accuracy for identifying critically ill patients at the ED. These findings support the role of the ESI guided by a principal understanding of the limitations inherent to any triage tool.
Wandl et al. (Fri,) studied this question.