Can Digital Behavioral Phenotypes Predict Long-Term Community Participation After Stroke? Evidence from Wearable Data and Smartphone-Based Behavioral Research

Abstract

Long-term stroke rehabilitation requires attention to the situation of patients' participation in social and community activities. Traditional follow-up assessments can only provide clinical information, but they are difficult to comprehensively reflect the daily behaviors of patients outside the medical setting. This study conducted a longitudinal secondary analysis to explore whether digital behavioral phenotypes obtained through wearable devices can predict the community participation of elderly individuals with a history of stroke one year later. The study used the publicly available registered data from the National Health and Aging Trends Study, with the 11th round wrist acceleration data as the baseline, and the 12th round data was used to assess subsequent community participation. Digital behavioral phenotypes included activity level, inactivity behavior, behavior fragmentation, time rhythm, and daily volatility. Community participation was measured using a breadth score, which combined indicators such as social visits, religious activities, organizational activities, and leisure outings. Regression analysis and prediction models were used to compare the predictive effects of demographic characteristics, health status, functional indicators, and behavioral characteristics obtained through wearable devices when used alone or in combination. The empirical analysis further evaluated the application value of digital behavioral phenotypes obtained through wearable devices as potential biomarkers for neurorehabilitation, and the results showed that they have the potential to complement clinical assessment and provide support for long-term community rehabilitation.