Submovements Derived from Wearable Sensors Capture Ataxia Severity and Differ Across Motor Tasks and Directions of Motion.

Digital measures derived from wearable sensors are a promising approach for assessing motor impairment in clinical trials. Submovements, which are velocity curves extracted from time series data, have been successful in characterizing impaired movement during specific motor tasks as well as from natural behavior. In this study, we evaluate the influence of different limb movements on submovement kinematic properties. Individuals with ataxia (n = 70) and healthy controls (n = 27) wore inertial sensors on their wrists and ankles and performed five neurologically-relevant tasks-finger-nose, fast alternating hand movements (AHM), finger-chase, heel-stomping, and heel-shin. A common framework was applied to extract submovements from each task and eight submovement kinematic features were analyzed. Though submovement kinematic properties changed in response to disease severity, they were primarily influenced by motor task and direction of motion. Modeling experiments revealed that accounting for task and direction of motion improved estimation of ataxia severity; the best performing model accurately estimated clinician-administered ataxia ratings (r = 0.82, 95%CI: 0.77-0.86), and found the finger-chase task to be most informative of severity. Although there were differences across tasks, in general, individuals with ataxia had submovements with lower peak accelerations and more variable kinematics. Relationships between ataxia severity and submovement durations, distances, and peak velocities were more task dependent. These results demonstrate that a common submovement analysis approach can be used to estimate ataxia severity across a wide range of motor tasks and that estimation of severity can be improved by accounting for movement type and direction of motion.