Assessing Risk of Early ICU Admission or Death at Emergency Department Triage: Clinical Judgment Versus Early Warning Scores.

Abstract

Background: Early Warning Scores (EWS) are widely used to standardize the identification of clinical deterioration, yet their comparative performance against clinical judgment in Emergency Department (ED) triage remains uncertain. We aimed to evaluate whether commonly used EWS match or outperform clinical judgment in predicting early adverse outcomes.

Methods: We conducted a retrospective observational study including 361,927 adult ED presentations at a tertiary-care academic center from 2015 to 2024. Clinical judgment was operationalized as the triage category assigned at the initial evaluation. Five EWS (NEWS, NEWS2, MEWS, REMS, and ViEWS) were computed using vital signs recorded at ED presentation. The primary outcome was a composite of 24-h mortality or intensive care unit (ICU) admission. Discrimination, calibration, decision curve analysis, and reclassification metrics were used to compare models.

Results: The primary outcome occurred in 1.17% of patients. Discriminatory performance was highest for ViEWS (AUC 0.875) and clinical judgment (AUC 0.872), with no significant difference between them. Other EWS demonstrated significantly lower AUCs. In precision-recall and threshold-based analyses, clinical judgment maintained higher specificity at higher-risk thresholds while preserving adequate sensitivity. Decision curve analysis showed comparable or greater net benefit for clinical judgment than for all tested EWS across clinically relevant thresholds. Reclassification metrics showed no improvement with EWS over clinical judgment.

Conclusion: Commonly used EWS did not clearly outperform triage clinical judgment in predicting early ICU admission or death. These findings support further investigation of how structured scores and clinical judgment may provide complementary information, particularly in identifying patients whose risk may be modified by timely escalation of care.