Any accurate watch that measures HRV (Heart rate variability) with public API?
It just needs to be accurate enough for me to see trends and make experiments 
This video seems to show that Oura is pretty accurate
Is there any studies showing that any watches are also accurate for HRV? (I’m only interested if they have public API)
Any company can claim their product measures HRV but the question is accuracy. I haven’t found one yet. The issue is that the watches, rings and bands all use PPG (photoplethysmography) which is optical technology to determine blood volume changes and estimate when the heart beat. Unlike a heart rate monitor chest strap which measures the crisp electrical signals of your heart. Some PPG devices will work well for short periods of time if you hold perfectly still(really still). Motion has been the great challenge for PPG accuracy. PPG is great for an average heart rate but falls short for clinical grade HRV. I am hoping to use PPG technology for HRV measurement in the future but I have been waiting a very long time already.
But doesn’t this suggest that at least oura is pretty accurate?
Most accurate is EKG and this is easy with a chest sensor. HRV really needs low little information, just the EKG peaks (a clear point) and a time stamp. The issues with HRV is how you do the computation and what time frame you use. Many methods. The underlying data is easy.
EKG = ECG - Look at the movesense sensor from Suunto.
What do you mean?
I just want something that is accurate enough to make experiments and show trends. Maybe the Oura is fine for that?
Hi Lars, I’m not an expert here but I think @JBDow’s comment is worth considering carefully. She has many years of experience looking at different inputs for HRV analysis. My guess is some important issues may be hidden by the Oura/Polar graph you shared. Oura may give a measure that correlates closely with Polar over time, but still have enough additional noise second to second that make the data problematic for some use cases. As always, the devil is in the situational details. If you want to do something with average HRV across nights, it seems like Oura would be a good choice, but doing things that require comparing HRV over seconds or minutes might fail.
Second by second or minute by minute HRV is not important for me 
It’s only being able to tell if I’ve been having to much stress or overtraining Also I would like to make experiments where I compare the average HRV in one week/month to another week/month
Oura seems like a good product for me.
My only questions left is whether the emfit or a watch could be just as accurate.
If the emfit also can give good average HRV across nights, then it’s more practical, since I can just install it and leave it. But I’m afraid about whether its imprecise and whether it will read wrong if I sleep together with somebody else.
If the watch can give good average HRV across night, then that would be 2-in-1, but I just haven’t tried to sleeping with a watch, and I don’t know if they are as accurate as the Oura ring, because I can’t find any studies on them.
I know that @azure has found the Oura to be reliable for a wide variety of measurements, not sure she’s looked at HRV specifically, but I know that overall for her studies she’s found it to be a good instrument.
Do you know if the watches are as accurate as Oura?
Thanks Lars for posting the graphs. I would want to see the raw data and see the specifics of how the data was captured. With HRV measurements you want precision to the microsecond.
There are many factors that will improve your overall analysis when using HRV. (1) To be able to determine your state of recovery you would want to be performing a min 3 - 5 minute session, daily, each morning at approximately the same time and in the same position. As with most biometrics they are dynamic and fluid. They will be changing through out the day depending on a multitude of factors so test at the same time each day. (2)The measurement time needs to be long enough. You can get a good enough rMSSD read in 3 minutes but in the frequency domain (LF HF) it needs a 5 minute reading. It uses a 5 minute moving window of data. It takes 5 minutes for the data to settle especially in the frequency domain. (3) The signal quality from the BLE device is important. Poor contact with the device (contact pads) can cause poor quality data. The filtering algorithms are important for removing electrical artifacts and ectopic beats from the data. Even if advanced filtering is used, too much bad data can cause a false high HRV result. (4) There are 32 different HRV metrics that can be derived from the RR data in the time domain, frequency domain and non-linear. Each metric gives different a different insight. When the phrase HRV is used it is important to be specific as to which metric. They are all HRV.
With all of that being said, maybe an average HR metric, daily (make sure it is measured at the same time of day) is sufficient for what you are trying to discover. Average HR is an excellent metric and frequently undervalued. It’s easy to capture and PPG devices are very good since it is an average.
Good luck!
Okay, so just to be sure that I understood you:
Both the Oura and other PPG devices like watches should be accurate enough for me, if all I want is average HRV. I’m going to measure HRV in the night by wearing the watch or Oura ring, so that should be long enough and consistent enough.
What do you mean about 32 different HRV metrics? Where can I read more about that?
No, average heart rate(HR) not average HRV. I was just proposing the possibility that maybe you don’t really need HRV measurements for your analysis and an average heart rate measurement would suffice. I am not clear on what your end objective is for this analysis. The PPG devices are very good measuring average heart rate.
I would suggest looking at the Kubios documentation. Kubios is an application originally developed by the University of Finland(I believe) is the gold standard for HRV research and analysis.
But I don’t understand… If the coefficient of determination (r^2) was equal to 0.943 (from this Oura study) doesn’t that mean it is very accurate?
Doesn’t that mean that 94.3% of the variation from the professional ECG can be “explained” by the Oura ring? Wouldn’t that be accurate enough to see trends and make experiments?
For example I would like to be able to see the following:
So it just needs to be accurate enough to make these kinds of conclusions.
Do you know if the watches are as accurate as the Oura ring?
Relative to Oura accuracy, this thread may be helpful in starting to assess that.
Thanks @mmschladen. I haven’t listened yet but this looks very good. I think that friction in measurement is a huge barrier, even for those of us who enjoy these projects. I’ve found that even small glitches can fatally interrupt a self-research project. The continuous background measurement that Oura provides is probably worth designing around, for projects that want to use HR data.
It seems like the guy in the podcast is very satisfied with his Oura and that he thinks it is accurate enough.
Doesn’t hurt to try it See if the results make sense to you. If you are incorporating other biometrics you can also discover patterns and interesting insights. This is all about discovery.
Tell me if you find anything, because the only thing I’ve managed to do meanwhile is buy a PulseSensor for Arduino and take manual measurements, which is clearly suboptimal 
regards, printsbery
@Agaricus the night time rmssd, reported every 5 minutes from the ring, is GOOD and available. Their inter-beat-interval (BI) data, from which one could derive any type of HRV (not just rmssd) is not publicly available yet to my knowledge. I hope they decide to open it up!