The lifetime risk of stroke is 1 in 6 with an estimated 33 million stroke survivors worldwide. Ideally acute stroke patients would receive an accurate and rapid prognosis regarding return of motor function, followed by application of those therapies most able to improve it. Yet decisions regarding post-acute treatment of stroke patients are made on short-term assessments of function that may be influenced by concurrent treatment, time-of-day, motivation, and other factors. Those assessments are often delayed, with resultant delays in rehabilitation treatments. There are important decisions that need to be made about the setting where rehabilitation occurs, if it is needed, and where the stroke patient will best live in the long-term. This research project aims to significantly add to the current understanding of biomarkers that can be used to provide better diagnosis, rehabilitative treatment, and long-term disposition advice for veterans who experience upper-extremity impairments from stroke. The gaps in knowledge we aim to address are the unknown relationships between 1. immediate post-stroke movement and functional ability, and 2. between sympathetic tone and psychological response to disability. Clinicians do not yet know how to use the data from wearable technologies that measure these factors – a problem caused by the volume of data generated and lack of reliable biomarkers derived from it. Our central hypothesis is that application of machine learning techniques to data from a multimodal sensor array worn by a patient for multiple hours can provide better evidence of motor ability, assess latent psychological factors, and predict recovery trajectory better than conventional short-term assessments. It may also allow more rapid personalization of therapy plans based on real-world deficits discovered through sensor-based data. We will test our central hypothesis by pursuing the two following specific aims with associated working hypotheses: 1. Collect functionally relevant data from a wearable inertial, electromyographic, and electrodermal sensor array. Working Hypothesis: A few strategically placed sensors can capture functional movement and state of the autonomic nervous system. Kinematic and physiological measures taken during task performance will be correlated with motor impairment and functional status. Completion of this aim will lead to the identification of functional variables derived from multimodal sensor measurements and demonstrate the feasibility of, and challenges to, inpatient use of a sensor array. 2. Predict key clinical outcomes from sensor array-derived variables in acute stroke inpatients being evaluated for post-discharge therapies. Working Hypothesis: Machine learning techniques, including Bayesian fusion, will predict deficits and discharge disposition from the multimodal variables collected. The electrodermal response to challenging movement is an unexplored area that may provide insight into motivation and affecti...