Study record · validation · 2019

Sleep stage prediction with raw acceleration and photoplethysmography heart rate data derived from a consumer wearable device

Walch O, Huang Y, Forger D, and Goldstein C

Sleep, 42(12), zsz180 · 2019

Why this study matters to CircaTest

Important because it demonstrates what an open, peer-reviewed Apple Watch sleep classifier can achieve from raw sensor data, independent of Apple's proprietary algorithm. The accompanying PhysioNet dataset is the most-used open dataset for wearable sleep classification research. Useful counterpoint to Apple's closed-algorithm white papers.

Abstract

Wearable, multisensor, consumer devices that estimate sleep are now commonplace, but the algorithms used by these devices to score sleep are not open source, and the raw sensor data is rarely accessible for external use.…

Read the full abstract on the source →

Source: PUBMED · Excerpt for fair-use commentary; full abstract via the source link

Population

Age

adult (specific range not stated in PubMed abstract)

Reference standard

psg

Participants undergoing polysomnography at the University of Michigan with simultaneous Apple Watch raw sensor collection. The PubMed abstract does not state the sample size or age range; the n value is left at 0 to make this absence explicit. Secondary sources report n=31; verify in the full paper before relying on this number. Models were cross-validated against the Multi-Ethnic Study of Atherosclerosis (MESA) dataset, which is mentioned in the abstract.

Devices and metrics

Apple Watch (Series 2/3 era; raw sensor data via custom app)

All studies for this device →

Metric	Value	95% CI	Note
Accuracy	90%	—	Sleep-wake classification accuracy from a research-built classifier on raw Apple Watch data.
Sensitivity	93%	—	Sleep epoch sensitivity.
Specificity	59.6%	—	Wake epoch specificity.
Accuracy	72%	—	Wake / NREM / REM 3-stage classification with all features.

Cite this study

Walch O, Huang Y, Forger D, and Goldstein C (2019). Sleep stage prediction with raw acceleration and photoplethysmography heart rate data derived from a consumer wearable device. Sleep, 42(12), zsz180. https://doi.org/10.1093/sleep/zsz180

AI-assembled record · errors are possible

This page was built and is maintained by AI working from the published abstract and PubMed metadata for the cited study. CircaTest does not have access to the full text of every paper in its corpus. Categories of error that have happened on this site and could happen again include:

Per-device metric values inferred when the abstract reports only an aggregate range
Stage-specific accuracy figures pulled from secondary sources rather than the primary paper
Sample size or population characteristics missing from the abstract
Editorial framing in the curatorial note that overstates the study's actual scope
Year typos and author-name transliteration errors

Where a metric value reads see source, the published abstract did not contain that value and CircaTest deliberately left it null rather than guess. For any decision that depends on the underlying numbers, please consult the linked PubMed, PMC, or DOI source directly. Corrections are welcome via the about page.

Source links

Added to the CircaTest meta-analysis on 2026-04-06. How CircaTest evaluates studies →