Northwell Health research arm develops AI tool to help hospital patients avoid sleepless nights

Female doctor talking to male patient in hospital bed
Patients’ sleep gets interrupted every four to five hours in a hospital to check for vital signs, according to the Feinstein Institutes. Using deep learning, a type of artificial intelligence, the research team saved about half of patients’ overnight sleep. (Getty/monkeybusinessimages)

The Feinstein Institutes for Medical Research, a division of the Northwell Health system, has developed an artificial intelligence tool to gauge when hospitalized patients need to be woken up during the night.

By letting patients sleep, the health system can speed up patients’ recovery and discharge them faster. It can also help avoid causing delirium in patients, noted Theodoros Zanos, Ph.D., assistant professor at the Feinstein Institutes’ Institute of Bioelectronic Medicine.

“It actually is a critical piece of a person getting better in the hospital,” Zanos told Fierce Healthcare. “We know that sleep disruption can essentially slow down recovery.”

Zanos and Jamie Hirsch, M.D., a nephrologist and internal medicine physician at Northwell Health, along with several colleagues published a report on the project in Nature Partner Journal Digital Medicine on Nov. 13. For the study, the Feinstein Institutes took 24.3 million vital sign measurements over about 2 million patient nights, according to Zanos. The team collected data from 2.13 million patient visits at multiple Northwell Health hospitals between 2012 and 2019.

The Feinstein Institutes is located in Manhasset, New York, and houses 50 research labs, 3,000 clinical research studies and 5,000 researchers and staff. It will conduct a pilot of the clinical tool, called Let Sleeping Patients Lie, at Huntington Hospital, according to Zanos, who runs the Neural and Data Science Lab at the Feinstein Institutes. It will then roll out the tool to several other hospitals in the Northwell Health system, the largest health system in New York state.

“What we're trying to do is to use an actual algorithm to identify the people that are safe to be left asleep,” Zanos said. “Clearly the reason to wake them up and measure vitals is out of an abundance of caution, so we’ve got to make sure that we don't have people deteriorating or decompensating overnight when we're not watching them inside our hospitals.”

RELATED: Early AI adopters beginning to see success predicting readmissions, ED visits

Despite the need to monitor many patients at night, the majority could be safely left sleeping, Zanos said.

“The question becomes, how do you identify them? And that's what we set out to do,” Zanos said. “I think we have something that could definitely save a lot of sleepless nights.”

Patients’ sleep gets interrupted every four to five hours in a hospital to check for vital signs, according to the Feinstein Institutes. Using deep learning, a type of AI, the research team saved about half of patients’ overnight sleep.

“Studies have shown that this is one of the more major complaints of patients, the fact that they can't sleep at night in hospitals, and a lot of people have been kind of trying to get at this in different ways,” Zanos said.

Easing clinicians’ workload

In addition to improving the sleep of patients, the clinical tool will reduce the workload of clinicians by 20% to 25% per overnight shift and address employee burnout. Patients that are less sick will just get a visual look-over by clinicians during the night, and more acutely ill patients will be awakened for a vital sign check. After spending 10% of a shift collecting vital sign checks, nurses spend 20% to 35% of their time documenting the vital signs in patients' electronic health records. Therefore, the Let Sleeping Patients Lie app can improve efficiency for clinicians, researchers said.

The machine learning algorithm uses a modified early warning score (MEWS) along with the previous night’s vital sign data and information on heart rate and elevated blood pressure to inform a clinician’s decision on whether to disturb a patient’s sleep, Zanos explained. They wake up patients with a MEWS score of 7 out of 14.

RELATED: The unexpected ways AI is impacting the delivery of care, including for COVID-19

Research plans during the COVID-19 pandemic

Zanos told Fierce Healthcare that Northwell Health will play it safe when it comes to using the clinical tool during the COVID-19 pandemic. Some of the research for the clinical tool was scheduled for March but had been put on hold due to the health crisis. In fact, the research may get delayed further until after a possible second wave of COVID-19. The team has a couple of floors in the Huntington Hospital in mind to begin work. Zanos does not plan to use the tool with COVID-19 patients since they will need their vital signs monitored closely.

“This is not a COVID algorithm,” Zanos explained. “It's not tested with COVID patients. Now that does not mean that it wouldn't work with COVID patients. Since we haven't looked into how this would work with a COVID patient, the suggestion is that you shouldn't use it in COVID units or on COVID floors.”

Although the AI algorithms in the Let Sleeping Patients Lie app will not be used with COVID-19 patients at Northwell Health, Zanos did publish a paper in the journal Bioelectronic Medicine about how AI and ML tools could help with COVID-19 clinical decision-making. These decisions were formed based on “Emergency ML” models, according to the Bioelectronic Medicine study.

In fact, Northwell Health says that COVID-19 cost the hospitals $1.1 billion in operating revenue during the first six months of 2020.