Abstract
Objective Polysomnography (PSG) is the gold standard for sleep assessment but is intrusive and unsuitable for long-term, real-world monitoring. This study compared a wrist-worn wearable device and a radar-based nearable as potential alternatives for naturalistic, home-based sleep monitoring in a working adult population. Method Thirty-four working adults (mean age 45.3 years) used both devices for up to six weeks, yielding 620 matched sessions and over 500,000 30-second sleep epochs. Agreement between devices was assessed for total sleep time (TST), wake after sleep onset (WASO), and device-derived sleep stage classification using intraclass correlation (ICC), Bland-Altman analysis and kappa statistics. Results Agreement was excellent for TST (ICC = 0.91), with wearable underestimating sleep duration by a mean of 3.8 minutes (95% limits of agreement: −74 to +67 minutes, a band of approximately 140 minutes). Agreement was low for WASO (ICC = 0.34), with systematic magnitude-dependent discrepancies. Epoch-level analyses showed substantial agreement for 2-state sleep–wake classification (PABAK = 0.78) but only moderate agreement for 4-state classification (Cohen’s κ ≈ 0.49), with the wearable classifying more epochs as light sleep, particularly when the nearable identified wake or REM. Discussion While the two device classes provide compatible population-level estimates of total sleep duration, agreement for wake detection and algorithmic sleep stage classification was considerably lower. These findings underscore the need for caution against interchangeable use of wearable and nearable devices for individual-level monitoring or research outcomes dependent on wake or sleep stage metrics.