Validation of real-world actigraphy to capture post-stroke motor recovery.

Stroke is a leading cause of long-term disability, but advances for rehabilitation have lagged those for acute treatment. Large biological studies (e.g., "omics"-based approaches) may offer mechanistic insights for recovery, but to enable those studies, researchers need to collect detailed recovery phenotypes at scale, e.g., in thousands of people with minimal burden for participants and researchers. This study investigates the concurrent validity between remotely collected wearable sensor data and clinical assessments of motor recovery post-stroke. We specifically focus on the "use ratio", which is the activity level of the paretic arm relative to the non-paretic arm, measured via bilateral wrist-worn accelerometers. Utilizing a large, harmonized multi-site dataset of adults with stoke, we analyzed cross-sectional (N=198) and longitudinal (N=98) changes in use ratio, the Action Research Arm Test (ARAT) and the Fugl-Meyer Assessment upper extremity subscale (FM-UE). Our findings indicate strong concurrent validity of the use ratio and the ARAT, and the use ratio and the FM-UE both cross-sectionally (i.e., differences between people) and longitudinally (i.e., changes within a person). Notably, while the use ratio strongly correlated with FM-UE and ARAT initially, the strength of these correlations reduced over time. This decreasing correlation might be explained by the increasing influence that personal and environmental factors play as recovery progresses. Additionally, these correlations were also stronger for the use ratio than for hours of activity for the paretic/nonparetic arm alone, suggesting that it is specifically asymmetry of activity that correlates with clinical measures. Thus, the use ratio is an efficient and clinically valid measure of motor recovery post-stroke that can be deployed at scale to collect biologically meaningful phenotypes.