At-home Wearables and Machine Learning Capture Motor Impairment and Progression in Adult Ataxias

A significant barrier to developing disease-modifying therapies for spinocerebellar ataxias (SCAs) and multiple system atrophy of the cerebellar type (MSA-C) is the scarcity of tools to sensitively measure disease progression in clinical trials. Wearable sensors worn continuously during natural behavior at home have the potential to produce ecologically valid and precise measures of motor function by leveraging frequent and numerous high-resolution samples of behavior. Here we test whether movement-building block characteristics (i.e., submovements), obtained from the wrist and ankle during natural behavior at home, can sensitively capture disease progression in SCAs and MSA-C, as recently shown in amyotrophic lateral sclerosis (ALS) and ataxia telangiectasia (A-T). Remotely collected cross-sectional ( = 76) and longitudinal data ( = 27) were analyzed from individuals with ataxia (SCAs 1, 2, 3, and 6, MSA-C) and controls. Machine learning models were trained to produce composite outcome measures based on submovement properties. Two models were trained on data from individuals with ataxia to estimate ataxia rating scale scores. Two additional models, previously trained entirely on longitudinal ALS data to optimize sensitivity to change, were also evaluated. All composite outcomes from both wrist and ankle sensor data had moderate to strong correlations with ataxia rating scales and self-reported function, strongly separated ataxia and control populations, and had high within-week reliability. The composite outcomes trained on longitudinal ALS data most strongly captured disease progression over time. These data demonstrate that outcome measures based on accelerometers worn at home can accurately capture the ataxia phenotype and sensitively measure disease progression. This assessment approach is scalable and can be used in clinical or research settings with relatively low individual burden.