Clinical Geriatric Functional Assessment Using Wearable Sensing and Machine Learning

Physical function is critically important for older adults as it directly impacts their independence, quality of life, and overall health. To assess the physical functions of older adults, primary health provider are optimally placed, but they do not complete the assessments due to time and space constraints. In this paper, we introduce a geriatric functional assessment (GFA) system using a wearable device and machine learning methods to assess older adults' physical function during their clinical visits and enable clinicians to monitor physical function results. Specifically, the GFA system includes a badge-sized wearable prototype with motion and visual sensors that continuously collect sensor data during ~15-minute clinical visits. Then, the GFA system integrates three different physical function inference methods using a support vector classifier, random forest, and deep neural network to infer actionable clinician-familiar measures, including gait speed, sit-to-stand, and balance metrics. We tested the GFA among 21 older adults who visited the same clinic. The evaluation of the inference methods shows that the random forest classifier performs better than the Support vector classifier and Dense neural network; the accuracy score is increased from 68.80 to 74.10 for gait speed, from 60.70 to 66.90 for sit-to-stand and from 69.80 to 76.2 for balance. Moreover, we showed that the GFA's usability score is promising, 79 out of 100, and we identified critical design issues that should be addressed in future designs.