Driven Digital Twin Framework for Homecare of Older Adults
Abstract
This study presents and technically evaluates a personalized, data-driven digital twin framework for interpretable monitoring of older adults receiving homecare. The framework was informed by requirements previously identified with older adults and homecare professionals and was evaluated using multimodal smart-home and physiological sensor observations collected in real residential environments. Five participants from a complete five-home deployment cohort were used for detailed personalized modeling and digital-twin evaluation. Minutelevel observations from ambient, wearable, and sleep sensors were integrated to characterize room occupancy, sleep and physiology, washroom use, mobility, and eating-related behavioral proxies. Participant-specific historical patterns were used to estimate expected behavior and identify daily statistical deviations, while contextual events captured time-specific conditions including prolonged washroom occupancy, repeated refrigerator access, and persistent sleeping heartrate or respiratory-rate deviations. During the one-week held-out evaluation, 32 contextual events were generated across 35 participant-days, comprising 25 moderate and 7 high-deviation events. These outputs were integrated into 50,400 minute-level digital-twin states and chronologically reconstructed within a virtual home environment, where future states remained concealed and contextual events became visible only when their detection criteria were satisfied. The findings demonstrate a physical-to-digital pathway that combines individualized behavioral interpretation with temporally synchronized virtual representation. The framework provides a foundation for explainable, human-in-the-loop digital twins for longitudinal homecare monitoring, while clinical effectiveness and prospective real-time performance require further validation.
