Using AI to find patients who are getting sicker quicker

Hospital patients with serious illnesses, or multiple chronic conditions, can take a turn for the worse quickly.

But researchers at RWJBarnabas Health say an AI-powered early warning system is giving doctors more insight at patients who are at risk of suffering declines, allowing for earlier interventions.

RWJBarnabas teamed with researchers from Rutgers Robert Wood Johnson Medical School and outlined encouraging results in a study published by NEJM AI, a journal from the New England Journal of Medicine group.

Dr. Andy Anderson, chief medical and quality officer of RWJBarnabas Health and a co-author of the study, tells Chief Healthcare Executive® that the early warnings are improving the survival rates of patients with serious conditions.

“It really helps to identify patients that are getting sicker in the hospital so that we can intervene more quickly,” Anderson says. “We know when patients are sick that they can suddenly get worse, and by having this tool that's basically helping us watch the patient and alerting us when they're getting sicker, we can intervene more quickly.”

RWJBarnabas and Rutgers researchers examined the use of the Epic Deterioration Index, and they analyzed outcomes of more than 23,000 patients in 11 RWJBarnabas hospitals. They saw improved survival rates with Epic’s warning tool, with deaths among high-risk patients falling from 23.1% to 18.6%.

Anderson talked about the encouraging findings and how RWJBarnabas deployed staff to respond to the data on patients who may be headed for danger. He says using the alerts has helped doctors and nurses get patients into the intensive care unit sooner than they would have otherwise.

“We see that trend, and we're able to sort of step in more quickly and move them to a point of care in the ICU where they're getting much more continuous care,” Anderson says.

‘An extra set of eyes’

The Epic Deterioration Index is tied to Epic’s electronic health record, and combs through the patient’s medical record, Anderson says. The tool gathers vital signs such as blood pressure, pulse, respiratory rate, and temperature, as well as new laboratory data.

The system recalculates data every 15 minutes and alerts staff if a patient is possibly headed for trouble.

“This tool basically, every 15 to 20 minutes is refreshing data, looking at vital signs, looking at over 30 parameters and letting us know if someone's starting to get sicker,” Anderson says.

The early warning system is useful in assessing patients at risk for conditions such as sepsis, which can escalate quickly.

It’s also useful in offering more insights on patients with multiple health challenges.

“There's multiple situations when patients have chronic issues with their heart or with their lungs, for example, and they're in the hospital with pneumonia, and they're having some challenges with their breathing, and this is sort of like an extra set of eyes to continuously, you know, keep an eye on the patient,” Anderson says.

“People who have multiple chronic medical conditions can deteriorate quickly,” he adds. “And so, it’s much better to see that as soon as possible to intervene as soon as possible.”

Anderson also points to the diverse patient mix in the study as another promising sign. Nearly half of the patients studied (48.6%) were members of minority groups, and they came from a wide variety of income levels.

“We have a very diverse patient population in New Jersey, in our hospitals where the study was conducted, and we conducted the study across really all of our hospitals …. very diverse in terms of age and ethnicity and socioeconomic status, and so the results basically held up, with all those variables,” he says.

Key to success: Response team

As RWJBarnabas implemented the early warning system, Anderon says teams were organized to help intervene when patients were at risk of declining.

“It really identified the need to organize our rapid response teams and in a very standardized way to make sure that we had a full team available 24/7 to be able to step in in these situations,” Anderson says.

“It's not just how much these numbers change in terms of how sick someone is, it's how fast they're changing,” he adds. “So if someone is getting sicker quicker, it's more important to step in even more quickly.”

Doctors, nurses and the hospitalist team are engaged earlier to do more for patients who are declining and get involved sooner, he says.

Pairing AI and clinical expertise

RWJBarnabas also worked to adjust so the early warnings were triggered appropriately, to avoid the potential of “alert fatigue.”

“We don't want to have too many rapid responses,” Anderson says. “That distracts and creates work that doesn't need to happen. Nor do we want to under-alert, either. So it's really finding the right balance.”

Anderson says he sees RWJBarnabas’ use of the alert system as the “perfect example” of matching AI with clinical expertise.

“The AI tool is like a companion,” Anderson says. “It's like a helper, and you know, it's really alerting the doctors and the nurses that something is going in the wrong direction, and having that information enables our nurses and our doctors to step in and do things as soon as possible, rather than waiting for the patient to get even sicker.”

But Anderson says RWJBarnabas is using AI to help doctors and nurses make better decisions.

“In the end, I think the most important thing for patients and families is to reassure them that the ultimate decision is made by the doctor and the nurse,” Anderson says.