A new artificial intelligence sepsis detection system had an 89 percent adoption rate by physicians and nurses, higher than other legacy tools, which typically garner a 10 percent adoption rate, a study published July 21 shows.
The Targeted Real-Time Early Warning System, which catches sepsis symptoms earlier than traditional methods, was initially developed at Baltimore-based Johns Hopkins and was commercialized by Bayesian Health. Suchi Saria, PhD, founder and CEO of Bayesian Health, led the research on the study in collaboration with a group of researchers from Johns Hopkins University. The team also partnered with Epic and Cerner, two of the largest electronic health system providers.
More than 4,000 clinicians from five hospitals used the AI during the study in treating 590,000 patients, and the system reviewed 173,931 previous patient cases. Bayesian Health also recently announced a partnership with LifeBridge Health to help deploy the system.
One of the reasons for the increased physician adaptability, clinicians say, is a decrease in the number of false alerts for a single patient.
"One in 3 times there's a real possibility that that patient is in trouble, as opposed to all of the other automated detection problem processes … where the alerts were delivered and it was only 1 in 30 or 40," said Neri Cohen, MD, PhD, president of the Center for Healthcare Innovation and a cardiothoracic surgeon.
The system also allows for greater collaboration with clinicians and offers more visibility into why the alert was delivered, Dr. Cohen said.
"The reason that it's working is not just the AI being delivered within the physician's workflow, it's being delivered in a way that is a collaborative colleague to the clinician. … It's just like getting a consultation from someone else; you get the data that is going into what has generated that alert. So the clinician gets the opportunity to evaluate the quality of that alert," Dr. Cohen said.
The system works in the background and monitors the hospital's entire patient population. When a patient shows markers for deterioration, the system flags the patient in the EHR. The provider interacts with the application and, whether the alert is accurate or not, the system learns from each interaction.
"That's what this is all about. This is about building trust," Dr. Cohen added.
In 82 percent of sepsis cases, the system was accurate nearly 40 percent of the time, according to a press release. Previous electronic tools caught fewer cases and were only accurate 2 to 5 percent of the time.
"It's absolutely crucial for treatment to start early. But for treatment to start early, diagnosis has to be early. That's the hard part. I mean, the hard part isn't knowing what to do when someone has sepsis, it's actually recognizing that it's sepsis," said Michael Shabot, MD, founding partner of Relia Healthcare Advisors and former system chief clinical officer of Houston-based Memorial Hermann Health System.
Though sepsis is always discovered, the difficulty for physicians is being able to catch it fast enough. With the current standard of care, sepsis kills 30 percent of the people who develop it.
"Our problem is that we have a real tough time recovering patients and the death from sepsis is from failure to recover, not because we don't know what to do. … There are very few things that are left in medicine where there is an acknowledged sustained 30 percent mortality," Dr. Cohen said.
"Sepsis still carries the highest mortality for inpatients and still carries a mortality of 30 percent. Again, because of a failure to intervene soon enough," Dr. Cohen added.
The study also showed that providers were willing to trust AI and machine learning in their clinical practice. In interviews, clinicians revealed they perceive themselves as partnering with the technology even without a deep understanding of machine learning.
"This tool makes every clinician more efficient and can let you manage more patients because there's less work to do," said Lee Sacks, MD, founding CEO of Advocate Physician Partners and former system chief medical officer of Downers Grove, Ill., and Milwaukee-based Advocate Aurora Health. "So I think, going forward … [staff] are going to choose to work in organizations that have a tool like this because it makes them more effective."
Dr. Saria added that this system may show that a similar approach could generate results in other clinical areas.
"It's much easier to take a platform that shows really high-quality results in a tough use case like sepsis and now start to generate results in other clinical areas," Dr. Saria said.