SHM Converge Scientific Abstract ePoster Gallery-Multicenter Impact of a Real-Time Early Warning System on Mortality and Hospital Length of Stay

Pdf Summary

This multicenter, non-randomized study evaluated the impact of implementing an FDA-cleared AI early warning system (eCART) across seven hospitals in one health system. The system was paired with standardized clinical workflows and weekly audit-and-feedback on performance. The study compared 12 months before and 12 months after implementation, excluding a 3-month wash-in period at each hospital.<br /><br />The primary outcome was in-hospital mortality. Secondary outcomes included observed-minus-expected hospital length of stay (O-E LOS) and estimated cost savings. The analysis focused on two groups: a target group with elevated eCART scores and a control group without elevated scores.<br /><br />Results showed that among patients with elevated eCART scores, mortality fell by 42% after implementation, and O-E LOS decreased by 0.5 days. This was associated with an estimated 677 fewer deaths and about $7.1 million in cost savings, assuming $900 per avoided hospital day. The control group did not show the same benefit. A dose-response relationship was also seen: greater compliance with the clinical workflow was associated with lower mortality, supporting the likelihood that the intervention contributed to the improved outcomes.<br /><br />In summary, real-time implementation of an AI early warning system combined with structured clinical response and feedback was associated with meaningful reductions in mortality, shorter hospital stays, and lower costs among high-risk medical-surgical patients. The study’s strengths included its multicenter design, long pre/post periods, concurrent and historical controls, and dose-response analysis, though it was limited by being non-randomized and conducted within a single health system.

Asset Subtitle

Deborah J. Rhodes

Meta Tag

Author List Dana Edelson, Deborah J. Rhodes, Jennifer Johnson, Jonathan M. Siner, Joseph Reid, Kaitie Todd

Category Research

Concept Non-Randomized Study

Concept Pre-Post Study Design

Concept Early Warning System

Concept Mortality

Concept Inpatient Mortality

Distinguished Finalist

Presenter Organization Yale New Haven Health

Presenting Author Deborah J. Rhodes

Track Outcomes Research

Keywords

AI early warning system

eCART

in-hospital mortality

hospital length of stay

cost savings

multicenter study

clinical workflow

audit and feedback

high-risk patients

dose-response relationship

Non-Randomized Study

Pre-Post Study Design

Early Warning System

Mortality

Inpatient Mortality