A new study published in npj Digital Medicine suggests that walking data collected from hip-worn accelerometers is viable for assessing the risk of falls in older women.
While this specific investigation employed a medical-grade triaxial accelerometer (the ActiGraph GT3X+), the researchers noted in their writeup that the success of these sensors bode well for population-level analyses of fall risk that would use accelerometers more common among consumers, such as those already built into most smartphones.
“The least expensive low-end smartphones (e.g., the LG Optimus Zone 3 which now costs $30) can measure gait as accurately as the most expensive high-end medical accelerometers, while also being more accurate than fitness devices,” the researchers wrote in the study. “Thus, there is a potential path towards screening and prevention of falls at population scale for the aging population, by leveraging already carried personal phones.”
In the study, researchers enrolled women who were originally recruited for the Women’s Health Initiative in the 1990s and had consented to subsequent extensions of the project. Among the 67 participants who met the inclusion criteria, the average age was 77.5 years.
Each participant’s fall risk was classified as “high” or “low” based on their scores on the Short Physical Performance Battery assessment and self-reported history of falls within the previous year. The participants were then asked to wear the hip triaxial accelerometer and perform a timed 400 meter walk. Gait data generated by the sensor were subjected to 11 different models designed to generate similar fall risk classifications, which were compared to those established using the more traditional assessments.
Through this comparison, researchers found the models created with accelerometer data to be, on average, 73.7 percent accurate, 81.1 percent precise, and have an area under curve (AUC) of .706. Among these, the model that performed the best included a combination of triaxial data, cross-correlations, and traditional measures of gait. This particular model scored a 78.9 percent accuracy rate, an 84.4 percent precision rate, and an .846 AUC.
"Our prediction showed that we could very accurately tell the difference between people that were really stable and people that were unstable in some way," Bruce Schatz, head of the Department of Medical Information Science at the University of Illinois College of Medicine at Urbana-Champaign and an author on the study, said in a statement.
These findings are generally consistent with those of prior investigations, and generally lend credence to the concept of wearable accelerometers for fall prediction, the researchers wrote. In a statement, Schatz said that he could someday see an implementation of the technology based on older patients’ phone accelerometers and automatic doctor notifications aiding in fall prevention.
"The question is: is it known how to take the signal, how to take whatever comes out of (a device), and predict something that's useful?" Schatz said. "I believe strongly the answer is yes."