0326 Sleep Stage Prediction With Raw Acceleration And Photoplethysmography Heart Rate Data Derived From A Consumer Wearable Device
Year: 2019
Type: article
Abstract: Wearable, multisensor, consumer marketed devices that estimate sleep are now commonplace, but the algorithms used by these devices to score sleep are not disclosed. Additionally, the raw acceleration and heart rate data collected by the device sensors is rarely accessible. Moreover, approaches to distinguish sleep from wake typically consider small windows of time in isolation and neglect previously established, validated mathematical models describing sleep-wake control such as the well-known contribution of the circadian oscillator. We collected raw acceleration data (in meters per second squared) and heart rate from the Apple Watch worn by participants undergoing polysomnography using code of our own creation. Using this data, we compared the contributions of multiple features (motion, heart rate, and "clock proxy"-- a stand-in for circadian drive) to classifier performance across several algorithms. Best performance was achieved using a neural net trained on motion, heart rate, and "clock proxy" features. To assess generalizability of our results, we tested the models trained on Apple Watch data with data from the Multi-ethnic Study of Atherosclerosis (MESA) and were able to meet or exceed past sleep staging algorithm performance. This study demonstrates, for the first time, the ability to analyze raw acceleration and heart rate data from a ubiquitous wearable device with various methods to improve accuracy of sleep and sleep stage prediction. Additionally, we demonstrate the relevance of including estimated circadian phase as an input to improve sleep and sleep stage estimation. All code used in this project is being made open source. This work is supported by an Exercise Sports Science Initiative grant from the University of Michigan.
Source: SLEEP
Institutions: University of Michigan, Dartmouth College
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $4,388
OpenAlex ID: W4254234402