How UC San Diego Health used AI to reduce its sepsis mortality rate

Artificial intelligence (AI) helped UC San Diego Health reduce its sepsis mortality rate by 17%, according to a study the health system published in January.

Sepsis is a leading cause of death in US hospitals—at least 1.7 million people develop the condition annually and of that group, about 350,000 patients die in the hospital or are discharged to hospice, CDC data shows. Sepsis also costs the US healthcare system roughly $62 billion annually, according to an article published in journal Critical Care Medicine.

Clinicians at UC San Diego Health developed an AI model called Composer that can detect sepsis in patients roughly four to six hours before a clinician would be able to diagnose it, Gabriel Wardi, the medical director of hospital sepsis at the health system, told Healthcare Brew.

“If you can predict when sepsis is going to happen ahead of time, that might be able to improve patient care and thereby have better outcomes for patients,” Wardi said.

Sepsis treatments, like antibiotics, “have greater benefits” when given earlier, the study authors noted.

How it works: Composer is a deep-learning model, a type of AI in which computers are taught to process data similarly to humans.

The model is trained to look through roughly 150 variables—like demographics, vital signs, and medications—to determine if a patient is at risk of developing sepsis, according to Wardi. Composer pulls this data from electronic health records (EHR).

If a patient is flagged as high risk, clinicians are notified via an alert in the EHR.

“[Composer] works silently and safely behind the scenes, continuously surveilling every patient for signs of possible sepsis,” Wardi told UC San Diego Health.

The health system tested Composer in two of its emergency rooms as part of a study comprising more than 6,000 adult patients with sepsis from January 2021 to April 2023.

Based on the study’s results, the health system is expanding Composer’s use beyond the two emergency rooms to detect postoperative sepsis, or sepsis that sets in after a surgery, Wardi said.

“It is because of this AI model that our teams can provide life-saving therapy for patients quicker,” Wardi told UC San Diego Health.