Page 34 - Modern Healthcare (January 2020)
P. 34
AI helps Nebraska
hospital reach patients
before they fall
By Jessica Kim Cohen
a patient was getting out
A PATIENT is tossing and turning in of bed, so they could re-
her bed. She sits up, maybe moves the spond with enough time
Cameras mounted on the walls of the
covers around. It might seem like a nor- to prevent a possible fall.
inpatient rehab unit allow an AI system
mal interruption during a night’s sleep, About three years ago, Bryan Health to watch for cues that a patient is getting
but it’s enough movement for an arti- began collaborating with Ocuvera, a ready to leave their bed.
ficial intelligence system to trigger an company that uses AI to predict when a
alert that lets nurses and technicians at patient is about to leave their bed.
Bryan Medical Center know the patient For patients deemed at risk of getting tient privacy, according to Steve Kiene,
might need help—and they should up from bed unattended, nurses mount Ocuvera’s CEO. “The data is not person-
check out the situation, stat. a camera on the wall across from the bed. ally identifiable,” Kiene said. “You’re not
The concern for staff at the Lincoln, The AI system—trained on 200,000 hours getting a high-def video of a patient.”
Neb., hospital’s inpatient rehab unit is of video—reviews the patient’s move- Bartelt said the inpatient rehab unit,
that a patient might be trying to get out ments, watching for cues like moving a which now has more than a dozen of the
of bed. And without the proper support, blanket or changing positions. company’s cameras, is working on mea-
they’re prone to fall. The AI system then pushes an alert to suring improvements in patient falls
Patient falls account for roughly 10% the smartphones of nurses and techni- over time. So far, she said the hospital
of problems documented in the ECRI cians on duty. The alert includes a video has noticed a “steady decline in the last
Institute’s database of patient-safety in- stream to help the staffer decide whether there years of our unattended falls.”
cidents, said Robert Giannini, an ECRI the patient needs help. It’s difficult to tie how many of those
patient-safety analyst—making them are linked to the AI system, as Bryan
the second most-common event report- That video component makes sys- Medical’s unit has continued using tra-
ed by providers. tems like Ocuvera’s look “somewhat ditional bed-exit alarms alongside the AI
That’s a serious risk to patient safety— promising,” said Ismael Cordero, a senior system. But the AI tends to alert nurses
with roughly one-third of falls among project engineer at the ECRI Institute. An roughly 2 minutes before a patient gets
hospitalized patients resulting in an in- AI camera system could possibly provide out of bed, and Bartelt said it takes nurses
jury—and it carries a financial risk, too. more sophisticated analysis than tradi- just 1.5 minutes to respond to an alarm.
The CMS doesn’t reimburse for costs at- tional bed-exit alarms—pads or sensors “If I can have 30 to 45 seconds advance
tributed to patient falls, which can leave spread across a patient’s bed, which alert notice that one of my patients is trying to
hospitals on the hook for thousands of nurses when pressure is lifted from them. get up without calling for help, that’s go-
dollars if a patient sustains an injury. “It’s not just telling you there’s been ing to make a huge difference,” she said.
“Inpatient rehab facilities have a high a shift in weight, but actually giving you Cordero noted that health systems
concentration of patients either with a an image of the patient’s movements,” interested in a predictive alert system
functional disability and/or a cognitive Cordero said, noting weight-based should consider the additional cost,
disability, so (patients) have a very high alarms could be fooled by a patient since many hospital beds come with
fall risk,” said Christie Bartelt, nurse turning over or reaching for a book. built-in exit alarms.
manager in Bryan Medical’s inpatient Ocuvera’s system uses Azure Kinect, Ocuvera charges for each hour that
rehab unit. Since all patients in the unit a developer kit from Microsoft Corp., one of its cameras is used, and there’s
are at moderate to high risk for falls, which includes a depth camera. Unlike no capital cost to purchase the system,
they’re asked to call a nurse before get- a traditional camera, a depth camera Kiene said. He declined to share what
ting out of their bed. provides a black-and-white outline of a the per-hour fee is, as the startup is still
But not all patients remember to do scene based on the distance of objects— “fine-tuning” pricing for the system, but
so. For those patients, the nursing team such as the patient—without identifying said it’s meant to be a “fraction” of what
needed to figure out how to predict when details. That’s important to maintain pa- a human sitter would be paid. l
30 Modern Healthcare | January 27, 2020

