A few months ago I had a nice chat with Sameer and Wikus at Biostrap, where we discussed ways to make wearables more useful.
As someone developing apps and tools for heart rate variability (HRV) measurement and analysis, I feel like many sensors could be a lot more useful if they implemented standard communication protocols.
It makes sense that companies making wearables want to develop their own analytics and solutions (scores, etc.), but a lot more could be done with these sensors if they also implemented standard communication protocols, so that any HRV app could talk to the sensor.
The team at Biostrap seemed to agree and just developed their new sensor, called Kairos, which can indeed be paired with any third-party app and used for data collection via standard protocols.
Note that this device at the moment is mostly meant for a different market (B2B, not B2C), hence as far as I know it is not readily available to the consumer. However, if you are doing research using wearables, this might be a great device to look at, as it offers a number of additional features, including also access to the raw PPG data (for more information, please contact Biostrap directly, as I do not have all the details).
In this blog, I just want to look at data quality for our application of interest, which is HRV analysis.
Measurement setup
I'm using the HRV Logger, our research app for data collection. The HRV Logger can pair to any sensor that implements the standard heart rate profile over Bluetooth, for example, Polar straps, but also the Biostrap Kairos.
In this case, I linked the Kairos and a Polar H10 to two different phones running the HRV Logger and recorded simultaneously.
RR intervals and rMSSD
I then collected a few RR intervals while resting and while deep breathing, to elicit different oscillations in instantaneous heart rate, which can be easily seen in the figures below.
In the figures below, we can see that the data looks really good already, even without any processing, as we can spot only one artifact during the second minute of the recording:
As always, the measurement was taken without any movement, as PPG is completely unreliable unless we are very still (and do not contract any muscle, as I’ve shown here and here for other sensors such as the Scosche Rhythm24 and the Whoop band).
I then applied the artifact removal we use in HRV4Training, which seems to be one of the most effective out there, as we can outperform other apps in terms of accuracy for HRV analysis, even when paired to the same sensor (e.g. a Polar H10).
After this step, the data looks basically identical between the Polar H10 and Biostrap sensor:
Finally, we can compute rMSSD for each minute, obtaining the following:
That’s all. I was very happy with the data quality of this sensor.
This is yet another demonstration that PPG can be used to estimate HRV and that pulse rate variability (i.e. HRV from PPG), when measured at rest in healthy individuals, provides all we need.
Thank you, Sameer, Wikus, and Biostrap for your work and for sending me this sensor, which I think I will be using quite a bit for biofeedback sessions.
I hope this was informative, and thank you for reading!
Marco holds a PhD cum laude in applied machine learning, a M.Sc. cum laude in computer science engineering, and a M.Sc. cum laude in human movement sciences and high-performance coaching.
He has published more than 50 papers and patents at the intersection between physiology, health, technology, and human performance.
He is co-founder of HRV4Training, advisor at Oura, guest lecturer at VU Amsterdam, and editor for IEEE Pervasive Computing Magazine. He loves running.
Twitter: @altini_marco