At Eight Sleep, we strive to provide our members with not only an optimal sleep experience, but also highly accurate biometric data during sleep. Our sensing technology, embedded in the Pod’s Active Grid Cover, is designed to monitor a variety of biometrics without any wearable and for up to two people on the same bed. This design presents a set of technical challenges, but it is also a better experience for our members who don’t have to wear any device to bed in order to monitor their sleep and health.
Today, we are launching key updates to our heart rate (HR) and heart rate variability (HRV) algorithms, live in the Eight Sleep App, to all Pod members.
The performance of these HR and HRV algorithms were validated against gold standard electrocardiogram (ECG) on more than 100 nights of data, comprising over 46,000 minutes of sleep.
Eight Sleep’s sleeping HR algorithm has a mean absolute error of less than one beat per minute (bpm), a mean absolute percent error of 1.5%, and mean error of -0.3 bpm.
One of the best ways to visualize this excellent accuracy is to look at the correlation value (i.e. correlation coefficient) between the ECG HR and the Pod HR. A correlation value of zero means there is no relationship between the two devices, and a correlation value of 1.0 is the highest correlation value and means that the HR values from the Pod perfectly match that of the ECG.
The correlation value between the Pod HR and ECG HR is 0.995 (r² = 0.99) – a near perfect match (Fig. 1).
Figure 1. Correlation between the Eight Sleep Pod HR and the gold standard ECG HR (r² = 0.99). Each dot represents a single night of sleep (n=108). The black dashed line represents a correlation value of 1.0 or the best-fit line (1:1 match between Pod and ECG HR).
There are several ways to calculate HRV. At Eight Sleep we present your HRV values as the RMSSD (Root Mean Square of Successive Differences between heartbeats). This is the HRV value presented by most wearable devices (e.g. Oura, WHOOP, Fitbit). HRV gives you a measurement of how much the time between your heartbeats varies, in milliseconds. Generally, a higher HRV is better, and changes in your HRV can indicate how well-rested you are, both physically and mentally. Note that there is a wide range in “normal HRV,” so we recommend looking at your own HRV trends across weeks and months to evaluate changes in your sleep and recovery.
Eight Sleep’s HRV algorithm has a mean error of -2.5 milliseconds, a mean percent error of 6.8%, and correlation value of 0.90 compared to gold standard ECG.
Figure 2. The bars show the mean Pod HRV (blue bar) vs. the mean ECG HRV (orange bar) in milliseconds for all nights (n = 108). The black line indicates the 95% confidence intervals.
Note to current Eight Sleep members about our HRV update: With this latest HRV update, you will likely notice a decrease in your nightly HRV compared to previous measurements on the Pod. On average, members will see a 20 ms decrease. This is due to a change in our HRV computation method and does not reflect a change in your overall health.
Participants: 42 participants were recruited for this study for a total of 108 nights of data. ~22% of participants were female, and the average age of participants was ~40 years old, with an age range of 23 to 73 years. ~33% of the participants had a pre-existing condition, including atrial fibrillation, sleep apnea, restless leg syndrome, or self-identified as a snorer.
Methods: Participants slept on the Pod for 1-3 nights while wearing an ECG device on their chest (VitalConnect RTM). This ECG device was used as the gold standard (i.e. ground truth) for sleeping HR and HRV to compare to the Pod’s sleeping HR and HRV. Participants self-selected their Pod temperature settings.
Statistical Analysis: After the study was complete, each participant’s ECG data were compared to their Pod HR and HRV data for each night of sleep.
HR analyses: The Pod calculates HR each minute, and then a 5 min median of Pod HR and ECG HR were compared across each night. With these 5 min values throughout the night, the following metrics were calculated: mean absolute error (the absolute difference between the Pod and ECG HR), mean error (the difference between the Pod and ECG HR), and mean absolute percent error (the mean absolute error divided by the ECG HR for that minute). The correlation value (Pearson r correlation coefficient) was calculated based on the Pod’s single nightly HR value displayed in the app vs. the ECG nightly value.
HRV analyses: Eight Sleep reports a nightly HRV value (RMSSD). This nightly value was compared to the nightly ECG HRV value, where RMSSD was calculated for both devices every minute throughout the night and then aggregated at the five-minute and 15-minute levels. Lastly, all 15-minute values were used to generate a single nightly HRV value. With these nightly values, we calculated the mean error, mean percent error, and Pearson r correlation coefficient.
Authors: Dr. Nicole Moyen and Brian Schiffer