Fall Detection and Prevention for Memory Care Through Real-time Artificial Intelligence Applied to Video: A Randomized Control Trial

This process enables staff to know about falls without requiring residents wear a device and to see how falls occur for residents that cannot advocate for themselves while still protecting resident privacy by only uploading video when safety critical events are detected. Seeing how the resident went to the ground (1) prevents the need for emergency room visits when residents intentionally moved to the ground without risk and (2) allows the care team to determine what caused an event like a fall and what changes can be made to reduce risk.

PRELIMINARY EVIDENCE. The proposed study follows a series of pilots. In pilot 1, we showed the technical feasibility of detecting falls from video with 200 falls acted out by healthy subjects. In pilot 2, in a 40-resident facility, we demonstrated the acceptance of privacy-safety tradeoffs and showed a reduction of total facility falls by 80% by providing the system for 10 repeat fallers. In pilot 3, we addressed repeatability of fall reduction in a cohort of 87 residents with ADRD in 11 facilities of three partner networks. In pilot 4 (NIH SBIR Phase I), we demonstrated that falls can be detected reliably in real-time within the partner facilities. We detected 93% of the falls; reduced the time on the ground by 42%; showed that when video was available, the likelihood of EMS visit was reduced by 50%; and reduced total facility falls by 38%.