In vivo comparison of sleep stage scoring between commercialized wearable devices and polysomnography system

Abstract

This study aims to evaluate the performance of 4 commercialized wearable devices in scoring sleep stages with the ground-truth from polysomnography (PSG) system. The comparisons were performed using data from 14 human subjects simultaneously wearing 4 wearable devices with sleep monitoring function monitored by polysomnography overnight at a Type 1 sleep lab. The compared features were categorized into 2 groups including (1) sleep–wake pattern and (2) sleep distribution. Wearable devices with sleep monitoring functions used in this study are from 4 different brandnames including Misfit, Garmin, Jawbone and Fitbit. These devicesare anonymously named as Device A, B, C and D. Using PSG system as benchmark, wearable devices earned good sensitivity, especially in detecting sleep onset and sleep period time in contrast to poor specificity, particularly in monitoring sleep stages. However, specificities in terms of wake–sleep transition features reported from wearable devices are low compared to those reported from polysomnography. Among 4 wearable devices, device C with the sensors to capture the heart rate, respiratory rate, body temperature, galvanic skin response as well as an accelerometer proved the best device in detecting not only sleep–wake transition but sleep stages as well. From the device perspective, we suggest that the usage of both actigraphy and heart rate sensors in the wearable devices and proper selection of sleep features can yield better agreement between wearable devices and the gold standards—PSG—in determining the sleep stages.