Medical Record Abstraction for Quality Improvement in Sepsis Care Using Artificial Intelligence: A Cluster

To test the hypothesis that AI-enabled scaling of SEP-1 measurement can improve measure performance in sepsis care.

Study Design: A prospective, cluster randomized trial conducted in the emergency departments of two academic medical centers within the UCSD health system.
Participants: 66 attending emergency medicine physicians were randomized into intervention (timely feedback using LLM) and control (standard feedback) groups.
Patient Selection: Patients with severe sepsis and/or septic shock were identified and evaluated for SEP-1 compliance using LLM at discharge.
Data Collection: Self-reported race and ethnicity data were collected to contextualize the patient mix.

AI-based abstraction showed 90% agreement with expert human reviewers.
Baseline SEP-1 compliance was 65%; a sample size of 300 patients was needed to detect a 15% increase in compliance.

The study aims to address the challenges of manual abstraction in sepsis quality measurement through AI integration.

The study was not prospectively registered due to its quality improvement design.
Limited to two academic medical centers, which may affect generalizability.

The integration of AI in sepsis management may enhance quality measurement and reporting.