AI can reduce sepsis-related deaths, study shows


Using artificial intelligence and a machine learning system, providers were able to diagnose and treat cases of sepsis earlier than with traditional methods, decreasing sepsis patient mortality by nearly 20%, according to new research.

The Targeted Real-Time Early Warning System, developed by Johns Hopkins University and Bayesian Health, tracks patients from their arrival in the hospital to their discharge, reviewing their medical records, symptoms and lab results to alert doctors about individuals at risk of life-threatening complications.

To date, the vast majority of predictive analysis sepsis studies have been theoretical, based on retrospective data in a lab setting, said Suchi Saria, founding research director of the Malone Center for Engineering in Healthcare at Johns Hopkins.

Saria, who is also the founder and CEO of New York-based Bayesian Health, said the research she led focused on the clinical deployment and adoption of the system to see if it could reliably and earlier identify patients at risk for deteriorating from infection.

Annually, around 1.7 million U.S. adults develop sepsis and nearly 270,000 die, according to the Centers for Disease Control and Prevention.

The studies looked at the system’s use by more than 4,000 clinicians at five academic and community-based hospital settings over two years, reviewing data from more than 700,000 encounters with patients, 17,538 of whom had sepsis.

Researchers found the system was able to identify and flag 82% of all sepsis cases included in the study, meaning the A.I. software missed less than one-fifth of septic patients. Providers confirmed that out of all the alerts they received from the system, 38% were accurate, actually flagging septic patients rather than those with other conditions that have symptoms similar to the ailment.
Previously, electronic tools identified less than half as many cases of sepsis and were accurate 2% to 5% of the time, Saria said.

“Health systems are inundated with tools and technology,” Saria said. “They’re only going to adopt something if, one, it’s easy to implement and adopt, two, there’s an evidence base supporting it, and three, it can help them solve more than one problem.”

When the software flags a patient it believes to be at risk of developing sepsis, it immediately alerts physicians and nurses through the patient’s electronic medical records system and suggests treatment protocols such as requesting blood cultures or prescribing antibiotics.

Clinicians receive messages specifically tailored to their different positions and roles in patient care.

If a provider disagrees when a patient is flagged for sepsis, their feedback is used to adapt the system’s performance and how it monitors the rest of that patient’s hospital stay, Saria said.

Because the symptoms of sepsis—such as fever and confusion—are common in other conditions, it is often easy to overlook the infection until it’s too late.



Source link

Leave a Reply

Your email address will not be published.