Fall risk analysis using machine learning, the Timed Up and Go test, and inexpensive wearable IMU sensors
Venous Roshdibenam, Stephen Baek, Gerald Jogerst, Priyadarshini R. Pennathur, Daniel V. McGehee, Geb W Thomas, Rajan Bhatt · 2021
Falls are a traumatic and possibly life-threatening incident experienced by an increasing population of elderly adults. Starting at the age of 65, the risk of falls starts to grow exponentially. Even if they are not fatal, they cause physical injuries such as broken bones, brain damage, or mental traumas such as fears of falls. These detrimental consequences lead to reduced mobility, increased risk of losing independence and reliance on caregivers or being placed in a nursing home, and overall reduced quality of life. Although some falls are not preventable such as falls due to age, some can be prevented by detecting the contributing risk factors. The clinicians investigate the risk factors by reviewing the patients’ fall history, monitoring the potential contributing health problems, and evaluating patients’ gait, strength and balance during some functional tests such as the Timed Up and Go (TUG) test. Often these assessments are subject to a clinician’s judgment and not a precise measurement of risk factors. In addition, conducting all the assessments would be time-consuming and cumbersome for the clinicians to accomplish during the clinical visits. Therefore, it is critical to build an objective fall-risk detection model that can efficiently measure biometric risk factors with minimal means and costs.Wearable sensor data were introduced as an inexpensive alternative to precisely quantify ambulatory kinematics during the TUG test to predict prospective falls. However, they require a long-term evaluation of large samples of subjects’ locomotion to predict actual falls. Our goal was to enhance geriatrician’s assessment using a clinically practical assessment tool. Hence, we studied an easy deployable quantitative approach, using three non-intrusive wearable sensors to measure participants’ gait kinematics of their neck, right, and left feet during the TUG test. The sample in this study is representative of older patients with multiple co-morbidity seen in daily medical practice. This data collection ensured convenient capture of various gait impairment aspects at different body locations. We built a sensor data-driven Machine Learning (ML) fall-risk detection algorithm that could closely align with an experienced geriatrician’s fall screening of older adults.