Wearable Activity Streams and Relapse Prediction across Cardiac Rehabilitation Programs

Abstract

The integration of digital health technologies into cardiovascular care presents an unprecedented opportunity to monitor patient trajectories continuously outside clinical settings. This prospective cohort study investigates the utility of high-resolution wearable activity streams for predicting relapse in patients enrolled in secondary cardiac rehabilitation programs. By equipping a diverse cohort of post-acute cardiac patients with continuous multi-sensor wearable devices, this study collected longitudinal data encompassing step counts, heart rate dynamics, and sleep architecture over a twelve-month follow-up period. Traditional rehabilitation models rely heavily on episodic clinic visits, which often fail to capture the granular behavioral and physiological fluctuations that precede clinical deterioration or program dropout. In this study, we propose a comprehensive data processing and machine learning framework designed to extract subtle temporal patterns from continuous wearable streams. Through rigorous survival analysis and the deployment of advanced predictive algorithms, we identify distinct activity signatures that strongly correlate with an increased risk of relapse, defined as either hospital readmission or the complete cessation of physical activity protocols. The findings demonstrate that specific features, such as increased sedentary bout duration and blunted diurnal heart rate variability, serve as early warning indicators of rehabilitation failure. The results highlight the potential for wearable sensors to facilitate highly targeted, preemptive interventions, thereby optimizing resource allocation and improving long-term cardiovascular outcomes.