ATLANTA – Emory University researchers will soon begin testing whether the use of artificial intelligence (AI) can reduce mortality in a subgroup of patients with sepsis, a life-threatening disorder that leads to organ dysfunction following an infection. Sepsis contributes to at least 1.7 million adult hospitalizations and at least 350,000 deaths annually in the U.S., according to the Centers for Disease Control and Prevention (CDC).

Emory has received up to $750,000 in funding over two years from the Kaiser Permanente Augmented Intelligence in Medicine and Healthcare Initiative (AIM-HI) Coordinating Center to demonstrate the value of AI in diverse, real-world settings to enhance diagnostic decision-making in health care. Emory’s AI and sepsis research proposal was one of five selected out of 120 applications to receive funding from AIM-HI.

Through the Precision REsuscitation with Crystalloids In SEpsis (PRECISE) trial at Emory, researchers hope to determine if their AI algorithm can identify a subgroup of patients with sepsis who might benefit from a simple intervention — a switch in the type of intravenous (IV) fluids given.

“IV fluids are one of the most common treatments given to hospitalized patients with suspected sepsis, but we don’t know which class of fluids — normal saline or balanced crystalloids — is better for our patients,” says Sivasubramanium (Siva) Bhavani, MD, assistant professor in the Division of Pulmonology, Allergy, Critical Care and Sleep Medicine at Emory University School of Medicine, and principal investigator of the trial.

The need for precision medicine to determine best outcomes in these patients is a critical next step, according to the researchers, as even a tiny mortality reduction from sepsis could have a huge public health impact.

“We know that the one-size-fits-all approach has not worked in fluid resuscitation in these patients,” Bhavani explains. “We have previously developed and validated the AI algorithm that we are deploying in this trial using retrospective data from Emory hospitals. The algorithm uses bedside vital signs to identify a sepsis subgroup known as Group D. In a secondary analysis of clinical trial data, we have found that Group D has a 15% absolute reduction in death with balanced crystalloids compared to normal saline. Now, we will have the opportunity to test our algorithm prospectively in a randomized controlled trial.”

The researchers note that in other subgroups (Groups A, B and C) [*see descriptions below], fluid choice may not matter. But in Group D (older patients described as having low blood pressure and relatively lower heart rates, respiratory rates and temperature), the fluid choice may be critical.

The PRECISE trial is expected to get underway in summer 2024. The single-blinded, randomized controlled trial will be conducted systemwide in emergency departments and intensive care units (ICUs) at six Emory Healthcare hospitals. During the trial, Emory’s AI algorithm will be applied to identify Group D patients and randomize them to usual care or intervention, with intervention being an electronic health record (EHR)-alert that nudges clinicians to use balanced crystalloids instead of normal saline. The end goal will be to evaluate a potential reduction in death after up to 30 days of care in the hospital.

Emory hopes to enroll more than 1,500 patients in the trial over the course of a year. For more information on the PRECISE trial, visit the trial’s ClinicalTrials.gov site.

Past published research from Emory applied unsupervised learning algorithms to bedside vital signs to identify sepsis subgroups.

Group A had elevated temperature, heart rate and respiratory rate, and relatively lower blood pressure.

Group B also had elevated temperature, heart rate and respiratory rate, although not as elevated as Group A, but with high blood pressure.

Group C has lower temperature, heart rate, respiratory rate with normal blood pressure.

Group D is defined above.

Groups A and B were younger patients, with Groups C and D consisting of older patients. Groups A and D had the worst outcomes, with the highest risk of requiring life support and the highest risk of death.