Sleep 5.4 Challenges in Using Wearable Devices for Accurate Sleep Monitoring

How Garmin, Fitbit, and Apple Watch estimate sleep stages, the conditions that affect their accuracy, and their limitations compared to EEG.

Video Breakdown:

00:00 Introduction to Wearable Devices and Sleep Tracking

01:03 Challenges in Differentiating REM Sleep

02:10 PPG Signals and Heart Rate Variability

02:57 Conditions Affecting Wearable Accuracy

04:37 EEG vs. Wearable Data for Sleep Stages

05:45 Factors Influencing Wearable Accuracy

06:22 Conclusion and Final Thoughts

So if you are a wearable device, and you're relying on heart rate, heart rate variability and movement like Garmin and Fitbit and Apple, what do you look for?
For non-REM sleep, you are focusing on reduction of heart rate. You're seeing a reduction in SDNN or total power, an increase in the normalized high frequency or the RMSSD. And then the other thing I just wanted to point out is, the reason why RMSSD can potentially even go down during non-REM sleep is because I had mentioned to you during non-REM sleep, the slow wave sleep, you have a reduction in the tidal volume.
Because you have a reduction in the tidal volume, your respiratory sinus arrhythmia actually could potentially decrease. And this could lead to the RMSSD to stabilize or even go down as opposed to go up. So sometimes RMSSD is not, the best way to differentiate non-REM three and non-REM two
Then if you wanna differentiate in terms of, REM sleep, ideally you want to focus on, the increase in heart rate variability, reduction in normalized heart frequency, and then the heart rate. But again, this is. It's hard to differentiate from wake period because wake also shares all these characteristics and so you ideally want to rely on reduced activity count, that you ideally should see, during REM sleep.
But as I had shown in the previous graph, you can have people acting out their dreams. And this is why REM sleep, when it comes to wearables, it does a pretty poor job of differentiating REM sleep from wake. That's, one of the things that you have to be aware of. Now, if I were to create a device, I would realize that the tidal volume is significantly reduced. And so how can you incorporate that information? Also, because during REM sleep, you have poikilothermia. If there's a matching between the internal and the external temperature, that might also indicate that you're in REM sleep.
The other thing I wanna point out is, what we found out is at nighttime, the PPG signals that we get are really good, more than during the day. The heart rate and heart rate variability signals is really good. It's because, as I had mentioned before, in the nighttime, your vessels in your skin are dilating, so your PPG signals are actually gonna be pretty good at night.
That's why the heart rate variability and heart rate data is so good at night. And so if you, you know, if your focus is the heart rate variability during the night, then I think you'll be reassured that the PPG data will be better. During the day, again, you're gonna be sensitive to the vasal constriction.
If you're cold, you're gonna see less of a PPG signal because the vessels are constrict. So these are things that you need to consider. Knowing all these things will tell you which conditions will not do so well when it comes to these wearable distinctions of sleep stages. When you have sleep related movement disorders like REM, sleep, behavior disorders, PLMD restless legs, sleepwalking, these sort of destroy the movement component of your algorithm. The cardiac conditions, again, the cardiac conditions will rely on heart rate variability, which assumes a normal sinus rhythm. If you have cardiac arrhythmias, autonomic nervous systems, or even just have a pacemaker, you're just not gonna have the accurate response using these wearable devices. Respiratory conditions like sleep apnea can cause significant amount of arousals change the heart rate variability, cause can cause people to cough or wheeze. All these things can also change their algorithms. Neurological conditions like Parkinson's.
I mean, Parkinson's has multiple things. I mean, you have tREMors, and muscle rigidity, which can affect the movement effects. You have also increased incidents of PLMD, so you have movement, During non-REM sleep, and also you have autonomic nervous system changes as well.
So a number of reasons why Parkinson's may not be working well, but all these other conditions also can change things. You have pain conditions, we can change heart rate variability and how a person moves. And then psychiatric conditions. So these are all sort of the clinical conditions where these wearable devices may not perform well because of the way the algorithm works.
The other thing I just want you to note is that sleep stages are EEG defined, whereas most wearables infer sleep stages via heart rate, heart rate, variability movement, and respiration. And so the correlation between the peripheral signals, which are the signals that you get in the body. Are not necessarily going to agree with what's defined, what happens in the brain.
So there's that disconnect there. I might even argue that these wearable based autonomic nervous system data actually could be more important because it has long-term physiological consequences. But either way, I just want you to know that these wearable devices may never be as sort of concord with EEG simply because you have this brain and body divide.
The other thing is that PSG bases their sleep changes using certain consensus rules , for example, a 30-second epoch, and many wearable devices may not follow these consensus rules. So as a result you may have some disagreement between the two.
Accuracy depends on the algorithm. Complexity, sensor data quality, diversity and size. You always have to consider the sleep context from which you're getting this data, using wearables. Obviously a lot of personal physiological differences can affect accuracy. Age, fitness level, health conditions, disorders that can alter sleep related physiological patterns, and then you have room temperature, noise, sleep schedule inconsistencies, all which can affect these autonomic physiological changes that could then change the accuracy of the sleep stages.
And most of the data that we have to validate wearable devices against PSG have been, done on healthy individuals and it may not do well with people with atypical sleep physiology.

Course episodes

Sleep 5.1 How Consumer Wearables Measure Sleep Quality

Sleep 5.2 Heart Rate, HRV, and Muscle Activity During Sleep

Sleep 5.3 Respiration & Temperature Monitoring with Wearables

Sleep 5.4 Challenges in Using Wearable Devices for Accurate Sleep Monitoring