1. Taylor and colleagues evaluated the implementation of an AI-informed triage decision-support tool.
2. With the AI tool, nurses more accurately identified patients requiring critical care, and the department experienced decreased flow times.
Evidence Rating Level: 2 (Good)
Study Rundown: Emergency department (ED) triage often utilizes subjective criteria, causing bias. AI-based triage tools embedded in electronic health records (EHRs) have the potential to improve triage performance and patient flow. Taylor and colleagues assessed an AI-informed clinical decision support tool’s ability to assist ED nurses with triage. The tool was implemented across three EDs, and outcomes from 180-day periods before and after implementation were compared. The primary outcomes included triage acuity distribution, identification of critical care and emergency surgery, hospitalization among low-acuity patients, and time to treatment and departure. The tool was required to provide an individualized rationale for its decisions, and nurses retained final decision-making authority. The study found that sensitivity in identifying critical care as high-acuity improved from 78.8% to 83.1% following AI-tool implementation. Additionally, after AI intervention, there was a median decrease of 4 minutes from patient arrival to moving to the initial care area. Overall, AI-supported triage was associated with improved critical-care identification and patient flow.
Click here to read the study in NEJM AI
Relevant Reading: Clinical Impact of Artificial Intelligence-Based Triage Systems in Emergency Departments: A Systematic Review
In-Depth [cross-sectional study]: This quality improvement study included 174,648 visits across three EDs in Connecticut. There were 83,404 visits during the 180-day pre-implementation period and 91,244 during the 180-day post-implementation period. The AI triage tool collected data, such as demographics, arrival mode, vital signs, chief complaint, and active issues, to predict outcomes, such as need for critical care or emergency surgery. Then, an Emergency Severity Index recommendation with an individualized explanation was generated, but triage nurses could accept or override it. Study outcomes included sensitivity and specificity for classifying patients requiring critical care or emergency surgery as high acuity; hospitalization among low-acuity patients; and time from arrival to initial care area, disposition, and departure. The study found that following implementation, low-acuity assignments increased from 23.9% to 35.4%, while mid-acuity and high-acuity assignments decreased. Sensitivity for identifying critical-care patients as high acuity increased from 78.8% (95% confidence interval [CI], 76.4%-80.8%) to 83.1% (95% confidence interval [CI], 80.9%-85.0%; p<0.001). Emergency surgery triage sensitivity did not significantly change. Median time to initial care decreased from 12 to 8 minutes, disposition from 190 to 182 minutes, and departure from 311 to 292 minutes. However, this study was limited by temporal confounding from the observational design, possible cross-site contamination, and the absence of long-term patient outcomes. Nonetheless, this study provided evidence that AI has the potential to improve ED triage accuracy and patient flow.
Image: PD
©2026 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.