I see assumptions and claims, but i dont find them practical.
Can we get a links to papers? Or it’s just subjective experience and unsystemical analysis?
Sleep is complex process, even in a free running conditions. It doesnt looks simple in my opinion.
Statistical data analysis is used to model data and explain that chaos. Sometimes it cannot be explained because of lack of data or other limitations. This method have some limitations, but it seems like best tool we have for now. I dont think someone can build statistical model with multiple predictors / outcomes just by thinking - but that’s not required. We have machines to do that for us
I’m tracking all of that and even more. Why i need it to be manageble? If i have a big data set with different combinations of factors - i can use statistical models to explain part of sleep variability. Nobody talking about explain all variability, but some part of it seems to be explainable.
This agrument is empty. How it prevent sleep to be explained in n=1 even for particular teenager? This is just highlight of social problems, not about our ability to analyze sleep data.
I dont see proofs. Which papers and real evidence stated that sleep is ONLY simple to analyze during free run?
Another questions is how can we make free run sleep possible in modern environment? Why would we need to achieve that? Circadian rhytm seems to be pretty stable and a lot of sleep papers recommends keeping constant sleep schedule (go to bed everyday at same time) which is opposite to free run (sleep when you feel sleepy even at random times)
Looks weird. Evidense of absence = absence of evidence? Getting insignificant results doesnt mean there is no associations/connections. There is a big chance of not being able detect them. Why? small dataset, lack of statistical analysis understanding, high error rate and variance, just small subset of factors (predictors) were measured etc
Doesnt looks healthy. I’m pretty sceptical on recommendations when person who recommend doesnt follow by himself or follow in unhealthy way (writing posts at 4 am?).
Anyway, that’s interesting to read I’m not trying to overcriticize, just some thoughts
It maybe worth to assess sleep stages with eeg wearable / psg lab. To make sure deep/rem is enough.
Accuracy doesnt equal to agreement. So if you will compare oura with other non-precise wearable, results are generally agreement and not measure of accuracy. To call it accuracy we might need to compare it vs gold standard or something not far from it. In case of oura 3 i thinks something like hypnodyne zmax, dreem 2 or psg should be used to assess accuracy. Quantified Scientist on youtube will do that and i’m going to benchmark oura gen3 vs dreem 2. I’ll post results when i have them. I bought x3 dreem 2 headbands to be covered for a few years of experiments
Blood pressure seems to be a response to some internal and enviromental conditions. In that case i’m not sure how it can affect sleep. Even if there is connection, i’m not sure if single blood pressure measurement before/after sleep will reveal it. Something like continous one might be, but not easy to achieve
How do you assessing your cognitive functions in objective way? I cant find a fast and simple way for daily assessment of different measures (working, short and long term memory, fluid intelligence etc)
@Max_Eastwood This is highly-relevant to our discussions in this thread regarding Oura ring accuracy. Posting it here for anyone discovering this thread in future.
Accuracy for 2-stage detection (sleep, wake) was 94% for a simple accelerometer-based model and 96% for a full model that included ANS-derived and circadian features. Accuracy for 4-stage detection was 57% for the accelerometer-based model and 79% when including ANS-derived and circadian features. Combining the compact form factor of a finger ring, multidimensional biometric sensory streams, and machine learning, high accuracy wake-sleep detection and sleep staging can be accomplished.
I’m neither Rob nor Max, but I suspect that despite the relatively mediocre sleep tracking performance of the Oura 3 thus far, it will improve significantly with more data and Oura’s tuning of their algorithms for the 3rd gen tracker. More data from the general population, plus more data from any individual we wish to track performance for, and more time to improve the algorithm will hopefully yield good results.
i’m pretty sceptical on that. Their strategy looks greedy - post paper with exceptional results just before release of a new ring. They selling new ring with old algorithm and still nobody can verify new sleep algorithm.
Their paper shows new sleep algo is better than any consumer EEG device which looks unrealistic. I think quality of sleep will improve but just a bit, maybe they will be somewhere around fitbit devices, but i dont expect it anywhere around EEG. For years their algo showed poor performance, so my priors on it are low.
I dont think adding circadian modelling and more precise HRV / temperature can provide same volume of information as brain waves. But it may improve accuracy. A lot of people have social jet lag, some people have circadian rhythm disorders, shift work etc - i dont understand how they will model this correctly for everyone. Incorrect circadian model for someone may lead to poor sleep prediction results…
Paper results might be biased, they may be played with train/test datasets. They may tuned neural networks to show good results for that particular dataset etc. Anyway lets wait for real world results, i think we will have enough data to judge in the mid/end of 2022.
I’ve ordered 4 new oura rings for me and family, and my wife also uses dreem 2. I’ll share oura/dreem our data when its available. Also i’ve bought one more dreem 2 headband from ebay and plan to give it to my mom, so i might have data for 3 persons. Merging it with Rob’s data will make our dataset even more powerful.
Yeah, I think your scepticism is well-founded. The results seem both conveniently-timed and surprisingly good. I’m very interested to see what you find from first-hand experience with your family and from comparisons to Dreem. Keep us posted!
@gianlucatruda I have some new results which you might be interested to see Now i have a personal home sleep research PSG “lab” with OpenBCI multichannel EEG approach (500Hz with frontal, occipital and temporal brain areas covered) and Hypnodyne ZMax 2-channel EEG (256Hz, frontal area).
I did some analysis on newest Oura ring 3 sleep staging algo and posted all details in my blog.
Here is a confusion matrix for all nights in analysis:
I can see an improvement from 60% to 75% of accuracy for 4-stage sleep classification compared with previous generation ring which is a good news.
Bad news is that deep sleep detection doesnt improve that much, 66% accuracy is imho too low to detect day-to-day changes in deep sleep.
I’m not trying to overcriticize, just some thoughts
Now i have a personal 