A patient is tossing and turning in her bed. She sits up, maybe moves the covers around. It might seem like a normal interruption during a night’s sleep, but it’s enough movement for an artificial intelligence system to trigger an alert that lets nurses and technicians at Bryan Medical Center know the patient might need help—and they should check out the situation, stat.
The concern for staff at the Lincoln, Neb., hospital’s inpatient rehab unit is that a patient might be trying to get out of bed. And without the proper support, they’re prone to fall.
Patient falls account for roughly 10% of problems documented in the ECRI Institute’s database of patient-safety incidents, said Robert Giannini, an ECRI patient-safety analyst—making them the second most-common event reported by providers.
That’s a serious risk to patient safety—with roughly one-third of falls among hospitalized patients resulting in an injury—and it carries a financial risk, too. The CMS doesn’t reimburse for costs attributed to patient falls, which can leave hospitals on the hook for thousands of dollars if a patient sustains an injury.
“Inpatient rehab facilities have a high concentration of patients either with a functional disability and/or a cognitive disability, so (patients) have a very high fall risk,” said Christie Bartelt, nurse manager in Bryan Medical’s inpatient rehab unit. Since all patients in the unit are at moderate to high risk for falls, they’re asked to call a nurse before getting out of their bed.
But not all patients remember to do so. For those patients, the nursing team needed to figure out how to predict when a patient was getting out of bed, so they could respond with enough time to prevent a possible fall.
About three years ago, Bryan Health began collaborating with Ocuvera, a company that uses AI to predict when a patient is about to leave their bed.
For patients deemed at risk of getting up from bed unattended, nurses mount a camera on the wall across from the bed. The AI system—trained on 200,000 hours of video—reviews the patient’s movements, watching for cues like moving a blanket or changing positions.
The AI system then pushes an alert to the smartphones of nurses and technicians on duty. The alert includes a video stream to help the staffer decide whether the patient needs help.
That video component makes systems like Ocuvera’s look “somewhat promising,” said Ismael Cordero, a senior project engineer at the ECRI Institute. An AI camera system could possibly provide more sophisticated analysis than traditional bed-exit alarms—pads or sensors spread across a patient’s bed, which alert nurses when pressure is lifted from them.
“It’s not just telling you there’s been a shift in weight, but actually giving you an image of the patient’s movements,” Cordero said, noting weight-based alarms could be fooled by a patient turning over or reaching for a book.
Ocuvera’s system uses Azure Kinect, a developer kit from Microsoft Corp., which includes a depth camera. Unlike a traditional camera, a depth camera provides a black-and-white outline of a scene based on the distance of objects—such as the patient—without identifying details. That’s important to maintain patient privacy, according to Steve Kiene, Ocuvera’s CEO. “The data is not personally identifiable,” Kiene said. “You’re not getting a high-def video of a patient.”
Bartelt said the inpatient rehab unit, which now has more than a dozen of the company’s cameras, is working on measuring improvements in patient falls over time. So far, she said the hospital has noticed a “steady decline in the last three years of our unattended falls.”
It’s difficult to tie how many of those are linked to the AI system, as Bryan Medical’s unit has continued using traditional bed-exit alarms alongside the AI system. But the AI tends to alert nurses roughly 2 minutes before a patient gets out of bed, and Bartelt said it takes nurses just 1.5 minutes to respond to an alarm.
“If I can have 30 to 45 seconds advance notice that one of my patients is trying to get up without calling for help, that’s going to make a huge difference,” she said.
Cordero noted that health systems interested in a predictive alert system should consider the additional cost, since many hospital beds come with built-in exit alarms.
Ocuvera charges for each hour that one of its cameras is used, and there’s no capital cost to purchase the system, Kiene said. He declined to share what the per-hour fee is, as the startup is still “fine-tuning” pricing for the system, but said it’s meant to be a “fraction” of what a human sitter would be paid.