Detecting sleep in free-living conditions without sleep-diaries: a device-agnostic, wearable heart rate sensing approach

Abstract

AbstractThe rise of multisensor wearable devices offers a unique opportunity for the objective inference of sleep outside laboratories, enabling longitudinal monitoring in large populations. To enhance objectivity and facilitate cross-cohort comparisons, sleep detection algorithms in free-living conditions should rely on personalized but device-agnostic features, which can be applied without laborious human annotations or sleep diaries. We developed and tested a heart rate-based algorithm that captures inter- and intra-individual sleep differences, does not require human input and can be applied in free-living conditions. The algorithm was evaluated across four study cohorts using different research- and consumer-grade devices for over 2,000 nights. Recording periods included both 24-hour free-living and conventional lab-based night-only data. Our method was systematically optimized and evaluated against polysomnography (PSG) and sleep diaries and compared to sleep periods produced by accelerometry-based angular change algorithms. Against sleep diaries, the algorithm yielded a mean squared error (MSE) of 0.04 to 0.06 and a total sleep time deviation of -2.70 (±5.74) and 12.80 (±3.89) minutes, respectively. When evaluated with PSG lab studies, the MSE ranged between 0.06 and 0.11 yielding a time deviation between -29.07 and -55.04 minutes. Our findings suggest that the heart rate-based algorithm can reliably and objectively infer sleep under longitudinal, free-living conditions, independent of the wearable device used. This represents the first open-source algorithm that can infer sleep using heart rate signals without actigraphy or diary annotations.