An Automated Digital Biomarker of Mobility

Abstract

The Timed Up and Go (TUG) test is a common clinical endpoint, but is limited by the need to conduct it in the presence of a trained evaluator, usually a clinician. Herein, we propose a sensor-agnostic automated pipeline based on machine learning to predict TUG scores using day-to-day walks captured using commonly used wearable sensors by generating a passive and continual stream of mobility biomarkers without the need of conducting scripted TUG tests. We validated our pipeline against data from 303 participants in three cohort datasets, each with a different primary focus population of healthy elderly adults, Parkinson's disease patients, and patients with mild cognitive impairment or dementia. In addition to TUG scores, the three datasets include walking data collected from different wearable sensors, i.e., a lower-back-worn accelerometer, wrist-worn accelerometer, and in-sole pressure gait sensor, respectively. Our leave-one-subject-out validation using participants from all cohorts showed that a random-forest predictive model achieved an accuracy of 1.7 ± 1.7s (mean absolute error ± standard deviation) and 84.8% predictions within the minimal detectable change (± 3s) with reasonable generalization across cohorts. Through the validation on data collected using he three types of commonly used wearable sensors, we demonstrated the ability of our proposed pipeline to leverage heterogeneous inputs for predicting TUG scores from walking data, suggesting the feasibility to generate a continual stream of TUG predictions as a novel digital biomarker of mobility by leveraging naturally occurring walks in free-living scenario. Our investigation also suggests that, for certain cohorts (e.g., Parkinson's disease population), applying a cohort-specific model instead of using a model trained with mixed cohorts might further improve performance.