Use of Hearing Aids Embedded with Inertial Sensors and Artificial Intelligence to Identify Patients at Risk for Falling
Kristen K. Steenerson, Bryn Griswold, Donald P. Keating, Majd Srour, Justin R. Burwinkel, Erin Isanhart, Yifei Ma, David A. Fabry, Achintya K. Bhowmik, Robert K. Jackler, Matthew B. FitzgeraldObjective
To compare fall risk scores of hearing aids embedded with inertial measurement units (IMU-HAs) and powered by artificial intelligence (AI) algorithms with scores by trained observers.
Study Design
Prospective, double-blinded, observational study of fall risk scores between trained observers and those of IMU-HAs.
Setting
Tertiary referral center.
Patients
Two hundred fifty participants aged 55–100 years who were at risk for falls.
Interventions
Fall risk was categorized using the Stopping Elderly Accidents, Deaths, and Injuries (STEADI) test battery consisting of the 4-Stage Balance, Timed Up and Go (TUG), and 30-Second Chair Stand tests. Performance was scored using bilateral IMU-HAs and compared to scores by clinicians blinded to the hearing aid measures.
Main Outcome Measures
Fall risk categorizations based on 4-Stage Balance, Timed Up and Go (TUG), and 30-Second Chair Stand tests obtained from IMU-HAs and clinicians.
Results
Interrater reliability was excellent across all clinicians. The 4-Stage Balance and TUG showed no statistically significant differences between clinician and HAs. However, the IMU-HAs failed to record a response in 12% of TUG trials. For the 30-Second Chair Stand test, there was a significant difference of nearly one stand count, which would have altered fall risk classification in 21% of participants.
Conclusions
These results suggest that fall risk as determined by the STEADI tests was in most instances similar for IMU-HAs and trained observers; however, differences were observed in certain situations, suggesting improvements are needed in the algorithm to maximize accurate fall risk categorization.