Related Experiment Video
Updated: Jun 26, 2026
07:42
A Data-Driven Approach to Quantifying Immune States in Sepsis
Published on: February 7, 2025
Medical Record Abstraction for Quality Improvement in Sepsis Care Using Artificial Intelligence: A Cluster Randomized
Aaron Boussina1,2, Claire Allison3, Kimberly Quintero4
1Division of Biomedical Informatics, University of California, San Diego, San Diego, California.
JAMA Network Open
|June 25, 2026
Summary
Large language models (LLMs) improved sepsis care quality reporting by providing real-time feedback, enhancing compliance with the Severe Sepsis and Septic Shock Management Bundle (SEP-1) metric. This AI-driven approach offers a more efficient alternative to manual processes.
Area of Science:
- Healthcare Quality Improvement
- Artificial Intelligence in Medicine
- Clinical Informatics
Background:
- Traditional hospital quality reporting is manual, costly, and has limitations in improving patient care outcomes.
- The Centers for Medicare & Medicaid Services (CMS) Severe Sepsis and Septic Shock Management Bundle (SEP-1) is a key quality metric for sepsis care.
Purpose of the Study:
- To evaluate if near-real-time quality measurement using large language models (LLMs) can enhance performance on the CMS SEP-1 quality metric.
- To assess the impact of AI-enabled feedback on sepsis care compliance.
Main Methods:
- A single-blind, cluster randomized trial involving 66 physicians across 2 academic emergency departments.
- Physicians were randomized to receive LLM-determined SEP-1 compliance feedback at discharge or standard care.
- Primary outcome was overall SEP-1 compliance; secondary outcomes included expert agreement, 30-day mortality, and ICU admissions.
Main Results:
- The intervention group showed a significant improvement in SEP-1 compliance (82.9%) compared to the control group (70.1%), an absolute improvement of 13.0%.
- LLM determination showed high agreement with expert review (92%).
- No significant differences were observed in 30-day mortality or ICU admissions between groups.
Conclusions:
- AI-enabled abstraction and feedback for sepsis care significantly improved compliance with the SEP-1 quality measure.
- AI-driven quality integration can address limitations in current hospital reporting and support learning health systems.