I just devoted an evening to see if my understanding of accuracy of sleep detection was out of date, or if my conclusions from looking into the research a year ago were still ok. It remains a popular topic for empirical study, perhaps because it’s bound to provoke curiosity and there are both some obviously good methods and some room for cleverness.
The quick answer is: the situation has not changed much. Where devices have heart rate as well as actigraphy, they are mostly good at detecting total sleep time. They are still unreliable at sleep staging. The commonly complained about problem (“does not catch my wakeups”) still exists, though the details are a bit different for the different devices.
Recently purchased an Ultrahuman Ring Air and it is catching my wakeups. The survey paper does not include it among the five devices tested; probably as the device was not released at the time. So far the sleep data it reports has proven useful although this was not the primary purpose of my purchase.
I find my Fitbit does catch most of my wakeups. Any time I remember being awake, and I later check the night’s record, I do see wakefulness at that time. The exact number of wakeups and times are sort of messy. I tend to wake up at least once, between 2 am and 4 am, and sometimes I’m awake for a while. During these times, my record shows me bouncing between awake and light sleep. So, perhaps not precise, but good enough for my purposes.
Too bad they didn’t evaluate Autosleep (iOS app) + Apple Watch. This is able to track disrupted sleep, which I’ve confirmed by repeated comparison with my own observations. It’s tracking sensitivity is adjustable. It is also able to track my daytime naps. It is the most data-rich of all the sleep tracking apps I’ve tried. It has a history export to CSV. The download of my entire 5 years worth of daily tracking data took a few minutes.
My sleep last night was interrupted by a medical issue which woke me up and kept me out of bed for 20 minutes. The Ultrahuman Ring app recognised that I have woken and offered to save my sleep up to that moment as “a nap” stupidly I clicked the yes button which resulted in my total sleep time being just shy of 60 minutes short. Won’t click that yes again in future but return to my bed and go back to sleep.
I’ve been using AutoSleep plus Apple Watch for years but long ago gave up trying to get much in the way of actionable insights. That said, i find it more accurate than other apps including Apple’s own.
I’ve been using Garmin, Oura and Apple for AutoSleep for a qualitative composite measure of sleep for 3+ years. Neither one by itself is accurate enough but together I appear to get the most accurate and reliable data. For example, Oura overestimates sleep awakening but is great with sleep onset and offset while the other two can show true awakenings.
Hi Mark, that’s interesting that Oura overestimates awakening. Do you use that measure in any way? I like total sleep time from my Fitbit, and trust it. I also see my main wakeups clearly. (When I get out of bed or check the time on my wrist.) Restless sleep, “half-awake” states seem to show as a mix between light and awake, which is reasonable. When you say Oura overestimates wakeups, what are you noticing?