UC San Diego Risk Stratification AI Predicts Sepsis, Reduces Mortality

Researchers from the University of California (UC) San Diego School of Medicine have found that an artificial intelligence (AI) model deployed in emergency departments to forecast patients’ sepsis risk has significantly reduced mortality, according to a study published last week in npj Digital Medicine.  

Sepsis is a major cause of morbidity and mortality in inpatient settings, and developing models to predict which patients are at risk of the infection has been a priority for medical researchers in recent years.

Many of these models are driven by AI, which can ingest and analyze a large number of data points to inform predictions.

However, the success of these models has been mixed, as a study last year revealed that the Epic Sepsis Prediction Model (SPM) achieved better accuracy at higher prediction thresholds than three existing sepsis prediction tools: Systemic Inflammatory Response Syndrome (SIRS), Sequential Organ Failure Assessment (SOFA), and quick Sepsis-Related Organ Failure Assessment (qSOFA). Despite this, SPM missed more sepsis cases and was less timely with its outputs than its counterparts, limiting its clinical application.

To address some of the pitfalls associated with existing models, UC San Diego researchers built their own model, known as COMPOSER.

COMPOSER is a deep learning tool designed to continuously monitor patients for over 150 variables associated with sepsis infection, including demographics, medical history, vital signs, current medications, and lab results.

The tool begins working in the background as soon as a patient is checked into the emergency department to catch sepsis risk as early as possible.

“Our COMPOSER model uses real-time data in order to predict sepsis before obvious clinical manifestations,” explained study co-author Gabriel Wardi, MD, chief of the Division of Critical Care in the Department of Emergency Medicine at UC San Diego School of Medicine, in a news release. “It works silently and safely behind the scenes, continuously surveilling every patient for signs of possible sepsis.”

If a patient presents with multiple relevant variables, and may be at high risk for sepsis as a result, the tool will then notify the patient’s care team via the electronic health record (EHR). From there, the nursing team can review the patient's risk with a clinician to develop a treatment plan.

“These advanced AI algorithms can detect patterns that are not initially obvious to the human eye,” said study co-author Shamim Nemati, PhD, associate professor of biomedical informatics and director of predictive analytics at UC San Diego School of Medicine. “The system can look at these risk factors and come up with a highly accurate prediction of sepsis. Conversely, if the risk patterns can be explained by other conditions with higher confidence, then no alerts will be sent.”

To assess COMPOSER’s impact on patient outcomes, the health system deployed it initially in the emergency departments of UC San Diego Medical Center, Hillcrest and Jacobs Medical Center. The research team evaluated outcomes for over 6,000 patients admitted before and after COMPOSER’s deployment in December 2022.

During the study period, the model was shown to have reduced sepsis mortality rates by 17 percent.

The researchers emphasized that the tool allows for patient risk to be identified more efficiently than traditional approaches, helping care teams provide potentially life-saving care more quickly.

Because of these significant improvements in patient outcomes, the model has since been deployed in hospital inpatient units throughout the UC San Diego Health network and is set to be implemented at UC San Diego Health East Campus, the health system’s newest location.

The researchers also underscored the potential of EHR-integrated AI tools to improve patient care more broadly.

“Integration of AI technology in the electronic health record is helping to deliver on the promise of digital health, and UC San Diego Health has been a leader in this space to ensure AI-powered solutions support high reliability in patient safety and quality health care,” stated study co-author Christopher Longhurst, MD, executive director of the Jacobs Center for Health Innovation, and chief medical officer and chief digital officer at UC San Diego Health.

In a recent interview with HealthITAnalytics, Longhurst expanded on what some of these AI efforts look like, joining leadership from Sentara Healthcare to shed light on how health systems interested in pursuing AI development and deployment can navigate concerns about automation bias in healthcare.