Emory expanded AI fall alerts to five hospitals after a pilot reported fewer falls
A Monaco Stage discussion put Emory’s fall-prevention rollout alongside robotic vascular ultrasound and home medication dispensing. Each system faces a different test of how well a new signal leads to care.
Emory Healthcare expanded an AI-enabled fall-prevention system to five additional hospitals after a pilot unit recorded fewer total falls and falls with injury, according to a quality-improvement report by its nursing and digital teams. VirtuSense founder and chief technology officer Deepak Gaddipati discussed the deployment Wednesday during a panel on moving physical AI prototypes into patient care.
Moderator Tamara Moores Todd, chief health informatics officer at Intermountain Health, asked Gaddipati, Vexev co-founder and co-CEO John Carroll, and HiDO Health co-founder and CEO Charles Gellman how their products work in a clinical setting or a home. Summit host Ian Khan introduced the group during The Futurist Physical AI Summit on the Monaco Stage at HLTH USA 2026. The discussion covered three different ways of collecting a signal and three different people who may need to act on it.
Fall alerts depend on a nurse getting there
Gaddipati said the company started with a system intended to assess fall risk and later tackled the timing of a potential fall. He described a sensor designed to recognize movements that can precede a patient getting out of bed. “We’re able to figure that out ahead of time,” he said, estimating that the system can detect signs 30 to 60 seconds before some patients rise. He described the device telling a patient that “help is on the way” while alerting clinical staff.
The useful interval in that example is the time between the alert and the patient doing something that may require help. A nursing team needs a way to decide who receives the alert, who can get to the room and what happens when several patients need help at once. A prediction can give staff more time to respond. The result depends on how reliably staff can act within that window and what the team learns from alerts that did not lead to an intervention.
In its published report, Emory described evaluating a light detection and ranging sensor platform in one unit and spreading its implementation process to five additional hospitals. Local nursing leadership worked with system leaders and the digital team to refine how the platform was used. The report described the work as a quality-improvement project, with the initial unit’s fall outcomes supporting the decision to expand. Its design leaves open how much of the observed change came from the sensor and how much came from the workflow around it.
For a second hospital evaluating the approach, a fall rate alone leaves questions that an implementation report can help answer. The unit’s staffing, its baseline fall rate, the beds where sensors were used and the way nurses escalated alerts all affect how another team would reproduce the work. Following response measures alongside rates of falls and injuries gives leaders a way to see whether a change in outcomes coincides with the workflow they intended to put in place.
Gaddipati also described what happened when a fall-risk result entered a clinic workflow. Giving the clinician and patient a result generated questions about who should speak with the patient, what action should follow and who would be responsible if nothing happened. The example gave Moores Todd’s questions about caregivers and staff a concrete setting. It illustrated why a site has to decide what the result is for and assign the next step before adding a new alert to a visit.
Putting vascular imaging closer to dialysis care
Carroll framed Vexev’s first use case around patients who already have frequent contact with a care team. “We want to start by focusing on clinics,” he said, pointing to dialysis sites where patients can need assessment of the vascular access used for treatment. In the workflow he described, a clinic that cannot provide its own scan has to refer the patient to another site. The proposed robotic system aims to move some of the work of acquiring images into the clinic.
“We want to reverse that paradigm,” Carroll said, referring to imaging that is often obtained after a problem presents. He described an ambition to make more proactive imaging possible and, over time, use the resulting information to study how vascular disease develops. Vexev describes its VxWave system as investigational and says it is unavailable for sale pending regulatory clearance. The potential to shorten the gap between a clinical concern and a scan depends on what uses are supported by evidence and authorized.
Moving a scan changes where some of the work happens. It also creates questions about responsibility for reviewing results and following up on abnormal or uncertain images. A dialysis program would need to know which patients should have scans, how findings should reach the clinician responsible for their access and how quickly that team can arrange additional evaluation. Those questions help define what an evaluation should measure alongside the quality of the images.
Medication dispensing shifts which work caregivers do
Gellman described a home dispenser that uses facial recognition and records a dispensing event, with notifications intended to tell a caregiver when a routine deviates. “You’re only notified if something happens,” he said of the design. Moores Todd asked how the information would reach a family member. Gellman pointed to texting, saying that “a simple text message can change everything” when a caregiver needs to know that a dose may have been missed.
Gellman cited a feasibility study at Rush University Medical Center involving 35 people living with dementia and their caregivers. The in-clinic study combined two cohorts and assessed setup and dispensing usability. Participants were shown how the device worked, and the dispensing exercise used vitamin C. Researchers logged two critical errors and 17 noncritical errors across both cohorts. Gellman was a co-author and is affiliated with the company.
The test shows what researchers observed when patients tried the technology in a clinic. Experience in the home may depend on changes in cognition, how a caregiver supports setup and how a family responds when a notification arrives. Over time, an alert for a prescribed regimen may also have to account for a prescription change or a dose that a patient took away from the device. Those situations help explain why measuring usability is one step in assessing medication safety.
The products discussed onstage address different clinical problems, each with an operational decision at the end. A nurse has to reach the room, a clinician needs to follow up on a scan, and a caregiver has to decide what to do when a phone buzzes. Emory’s report offers one example of evaluating technology alongside the response it requires. For the other products, evidence about that response will help determine where they can make a difference.