Detecting sleep outside the clinic using wearable heart rate devices

Area of Science:

• Biomedical Engineering
• Sleep Science
• Wearable Technology

Background:

• Multisensor wearables offer potential for large-scale, longitudinal sleep monitoring.
• Current methods often rely on manual annotations, limiting scalability and objectivity.
• Device-specific algorithms hinder cross-cohort comparisons and broad adoption.

Purpose of the Study:

• To develop and validate a device-agnostic, heart rate-based algorithm for objective sleep monitoring.
• To assess the algorithm's performance in free-living conditions across diverse populations and devices.
• To provide a scalable solution for sleep analysis without human annotation.

Main Methods:

• Developed a heart rate-based algorithm for sleep inference, independent of specific wearable devices.
• Evaluated the algorithm on four study cohorts, encompassing over 2000 nights of data.
• Compared algorithm outputs against polysomnography, sleep diaries, and an acceleration-based method.

Main Results:

• The algorithm demonstrated low mean squared error (0.04-0.06) and minimal total sleep time deviation ([Formula: see text]2.70 ± 5.74 min) compared to sleep diaries.
• In lab-based polysomnography studies, mean squared error ranged from 0.06 to 0.11, with time deviations between [Formula: see text]29.07 and [Formula: see text]55.04 minutes.
• The algorithm proved effective in capturing inter- and intra-individual sleep variations in free-living settings.

Conclusions:

• The developed open-source algorithm reliably infers sleep in free-living conditions using heart rate data.
• Its device-agnostic nature and lack of need for annotations facilitate large-scale, objective sleep research.
• This approach enhances the utility of wearables for continuous sleep monitoring and analysis.