AI System Enables Real-Time Sepsis Quality Assessment and Improves Adherence

Complex chart reviews tied to national quality measures can take months, delaying improvement efforts while sepsis continues to cause major morbidity and mortality in U.S. hospitals. The Centers for Disease Control and Prevention estimates that at least 1.7 million adults develop sepsis annually and about 350,000 die. To help address this challenge, researchers have developed an artificial intelligence approach that accelerates sepsis care assessment and enables targeted clinician feedback.

Investigators at the University of California San Diego School of Medicine, in collaboration with UC San Diego Health, evaluated large language models (LLMs) for automating medical-record abstraction in sepsis quality measurement in the emergency department. The work focused on the Centers for Medicare & Medicaid Services (CMS) Severe Sepsis and Septic Shock Early Management Bundle (SEP-1). By structuring and interpreting free-text clinical documentation, the LLM-based system supports near-real-time evaluation of complex care processes. The study was conducted with the Joan & Irwin Jacobs Center for Health Innovation at UC San Diego Health.

The system reproduces the traditionally manual SEP-1 review, which spans 63 steps and typically requires months of effort by multiple abstractors for only a few cases. Instead, the LLM scans hundreds of charts and generates contextual insights within seconds, often while care is ongoing. After each automated review, notifications go to emergency department leadership for validation and are then shared with treating teams to provide case-specific feedback and recommendations aligned with SEP-1.

According to the study, published in JAMA Network Open on June 25, 2026, this workflow improved compliance with national sepsis quality measures at the institution. The team also reported efficiency gains from automated error correction and reduced administrative costs, with a design that can scale across care settings. The investigators noted that small, privacy-preserving language models can distill actionable insights from extensive chart documentation, embedding best practices into care delivery.

“Medicine can learn a lot from professional athletics, where every player knows their performance statistics almost immediately, and that feedback changes how they train and perform. Rarely do physicians get that same type of quick, individualized feedback, even for conditions as time-sensitive as sepsis. By measuring performance in near-real time, we can turn quality reporting from a retrospective administrative exercise into something that actually helps physicians improve care,” said Gabriel Wardi, MD, emergency and critical care medicine physician at UC San Diego Health and chief of the Division of Critical Care in the Department of Emergency Medicine at UC San Diego School of Medicine.

“By using AI to quickly assess sepsis quality care measures, we are able to provide guidance to our care teams in the most teachable moment. In turn, this has resulted in improved compliance with national sepsis quality measures and helps our teams consistently improve the care they provide to the communities we are proud to serve each day,” stated Karandeep Singh, MD, chief health artificial intelligence officer at UC San Diego Health and Joan & Irwin Jacobs Endowed Chair in Digital Health Innovation at UC San Diego School of Medicine.

Related Links
UC San Diego