A Novel Methodology for Temporal Analysis of Consumer Wearable Sleep Data: Feasibility and Application to Early...

While sleep disturbances are recognized as early markers of Alzheimer's disease (AD), the practical application of gold-standard polysomnography (PSG) for long-term monitoring is limited. This study aims to establish a novel methodological framework for the time-series analysis of sleep data collected via consumer-grade wearables and to explore whether this approach can detect differentiated signals across various stages of cognitive impairment. Daily sleep patterns of thirteen participants (5 healthy controls, 4 with aMCI, and 4 with mild AD) were monitored over three months using the Fitbit Charge 2. Rather than relying on aggregate nightly averages, we implemented a time-resolved analysis across 10-minute intervals to examine the temporal dynamics of sleep architecture. The proposed analysis revealed distinct, group-dependent temporal signatures. Specifically, the aMCI and AD groups exhibited shorter deep sleep during the early phase of the night, reduced REM sleep approximately three hours after sleep onset, and consistently elevated levels of light sleep and wake after sleep onset (WASO). These findings demonstrate that time-series analysis of wearable sleep data presents the potential to identify candidate digital phenotypes associated with cognitive decline. This study supports the feasibility of using longitudinal, dynamic sleep monitoring as an exploratory analytical framework warranting further validation for the detection of pathophysiological changes in older adults, shifting the focus from simple detection to the identification of candidate temporal sleep features. All reported findings are exploratory in nature and require replication in larger, independent cohorts.