Abstract
Study objectives
Consumer sleep technologies (CSTs) enable low-burden longitudinal sleep monitoring, and their measures are often interpreted as equivalent to polysomnography (PSG) measures. We applied measurement reliability-aware approaches to determine whether CST-derived ‘sleep’ measures (1) are interchangeable or device-specific, (2) can reliably assess participant-level sleep characteristics and (3) converge onto interpretable principal components representing essential aspects of sleep.
Methods
Data from 74 older adults (20 people living with dementia) were collected at-home (total=752 paired nights) using four modalities simultaneously: research-grade actigraphy (Axivity), a wearable (Withings Watch), a nearable (Withings Sleep Analyser) and Sleep Diary, followed by one in-lab PSG assessment. We applied Bland-Altman analyses, repeated-measures correlation analyses, intraclass correlation coefficients (ICC) and principal component analysis (PCA).
Results
Single-night between-device correlations and correlations with PSG were weak (r<0.3) for most measures but moderate (0.3≤r<0.7) for duration- and timing measures. Bland-Altman analyses revealed systematic disagreement with PSG, with total sleep time and sleep efficiency overestimated, wake after sleep onset underestimated, and poor agreement for sleep-stage durations. The number of nights required to obtain acceptable reliability (ICC≥0.7) varied across measures, modalities and participant groups, with 71% of measures reaching an ICC ≥0.7 within seven nights. Reliability-filtered PCA yielded interpretable principal components, with Duration being the only sleep aspect showing consistent moderate between-device associations.
Conclusions
Correlation of most sleep measures across assessment modalities is poor, suggesting that the consumer sleep technologies tested in our study may not measure the same essential aspects of sleep. Duration was the only sleep aspect transferable between these devices.