AI speeds hospital sepsis care reviews in clinical study
UC San Diego researchers said language models reviewed sepsis care measures in seconds and delivered feedback to emergency physicians sooner.
By Priya Raghavan · Science Reporter
3 min read
AI tools helped UC San Diego clinicians review care for severe sepsis patients faster and send feedback to emergency teams while care was still underway, according to a clinical study published in JAMA Network Open. The findings matter because sepsis treatment is time-sensitive, while national quality reporting can take months, according to the researchers.
The study was conducted by researchers and physicians at the University of California San Diego School of Medicine and UC San Diego Health. UC San Diego said the team used artificial intelligence and large language models to assess complex care measures and provide targeted feedback to hospital clinicians.
The work focused on the Centers for Medicare & Medicaid Services SEP-1 measure for severe sepsis and septic shock, according to UC San Diego. The measure is difficult to review because it requires detailed analysis of medical records, and UC San Diego said the usual clinical review process involves a 63-step evaluation of large patient charts.
That manual process can require months of work by multiple reviewers for a small number of cases, UC San Diego said. In the study, the researchers reported that large language models could scan hundreds of patient charts and produce contextual insights in seconds.
After the AI system reviewed charts, UC San Diego said it sent a notice to emergency department clinical leaders for further review. The information was then shared with the medical teams treating patients with sepsis, giving physicians feedback close to the time care was being delivered.
Gabriel Wardi, a co-corresponding author and emergency and critical care physician at UC San Diego Health, said physicians rarely receive rapid individualized performance feedback, even for urgent conditions such as sepsis. Wardi, who also leads the Division of Critical Care in the Department of Emergency Medicine at UC San Diego School of Medicine, said near-real-time performance measurement could make quality reporting more useful for improving care.
Karandeep Singh, a study co-author and chief health artificial intelligence officer at UC San Diego Health, said the system allowed teams to receive guidance during a point in care when feedback could be applied. Singh said the effort improved compliance with national sepsis quality measures and helped care teams improve their work.
According to the U.S. Centers for Disease Control and Prevention, at least 1.7 million adults in the United States develop sepsis each year. The CDC says about 350,000 adults die from the serious blood infection annually.
The study also found that large language models could improve efficiency by correcting errors and reducing administrative costs through automation, according to UC San Diego. The researchers said the approach could be scaled across different health care settings.
Aaron Boussina, the paper’s first author and affiliate faculty member at the Joan & Irwin Jacobs Center for Health Innovation at UC San Diego School of Medicine, said smaller privacy-preserving language models can turn large volumes of chart documentation into rapid, usable insights. Boussina said the approach can build best practices into the care delivery process.
This story draws on original reporting from Medical Xpress.