Artificial intelligence (AI)-triggered rapid response notifications were associated with lower mortality among inpatients

1. Nahass and colleagues evaluated the implementation of an AI-powered electronic alert system that automatically activated the rapid response team.

2. Automated rapid response team notifications were associated with lower inpatient deaths and more rapid response activations.

Evidence Rating Level: 2 (Good)

Study Rundown: Newer machine-learning models, such as the Epic Deterioration Index (EDI), may reduce in-hospital mortality. However, there is a lack of adequate prospective studies demonstrating their mortality benefits. Nahass and colleagues assessed whether combining EDI with electronic alerts and automated rapid response team (RRT) notifications was associated with reduced inpatient mortality. Researchers compared high-risk adult admissions before and after implementing a system that automatically prompted timely bedside assessment. The primary outcome was risk-adjusted in-hospital mortality. Additional outcomes included RRT activation and escalation to higher-level care. The study found that unadjusted mortality decreased from 23.1% before implementation to 18.6% afterward, with an adjusted odds ratio of 0.82. RRT activations increased from 25.3% to 37.5%, while escalation of care did not increase significantly. This study demonstrated that integrating AI-powered deterioration predictions into clinical response workflows may improve outcomes among higher-risk patients.

Click here to read the study in NEJM AI

Relevant Reading: Detecting Patient Deterioration Using Artificial Intelligence in a Rapid Response System

In-Depth [prospective cohort]: Adult admissions with an EDI score of at least 60 were enrolled across 11 hospital sites. 10,803 patients were included in the pre-implementation cohort, and 12,329 were included in the post-implementation cohort. A modified EDI excluding the Glasgow Coma Scale generated predictions every 15 minutes, and scores of at least 60 triggered RRT mobile notifications unless suppression criteria were applied (e.g., existing intensive or comfort care, recent alerts, or recent RRT events). Before implementation, scores were calculated but not used. Escalation was defined as transfer from acute to intermediate or intensive care, or from intermediate to intensive care. Deaths occurred in 2,500 pre-intervention patients versus 2,297 post-intervention patients, an absolute difference of −4.5 percentage points (95% confidence interval [CI], −5.6 to −3.5). The adjusted mortality odds ratio was 0.82 (95% CI, 0.74-0.91), while the adjusted RRT activation odds ratio was 1.74 (95% CI, 1.61–1.88). The mortality rate for the group just below the threshold (EDI 55–59) decreased from 2.9% to 2.3%, with an adjusted odds ratio of 0.99 (95% CI, 0.70-1.39). Meanwhile, the mortality rate for the group just above the threshold decreased from 5.5% to 3.8%, with an adjusted odds ratio of 0.86 (95% CI, 0.63 to 1.17). Only 46.6% of post-intervention encounters generated notifications. This study was limited by unequal seasonal distributions of the pre- and post-intervention phases, incomplete implementation at two hospitals, and uncertain generalizability beyond the sites studied. Nonetheless, this study provided evidence that connecting an AI-enabled prediction model in the EHR with real- time alerts to the clinical team could reduce mortality among admitted patients.

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.