In our review of the wearables landscape some time ago (now outdated by some dozen incumbents and newcomers, from Jawbone and Garmin to Huawei and Xiaomi), we focused on Heart Rate Variability as the apex of analytics for wellbeing. Back then, only the Basis band offered HRV analysis (Fitbit and Nike+ didn’t and still don’t, and neither does Apple). Since, lots of devices have started to support HRV analysis, like the Polar H7 and the Mio Alpha, and the (now being discountinued) Microsoft Band, which we used in our pilot at BNPP with AXA in order to track over 500 users’ stress, fatigue and burnout.
What is HRV?
Heart rate variability is the variation of the time interval between heartbeats. It is measured by recording the beat-to-beat interval over a period of time, typically 2m or longer. It is also referred to as “RR interval variability” (where R is a point at the peak of the heart wave and RR is the interval between successive Rs).
HRV is related (and can be used to measure) emotional arousal or elevated state anxiety, and lower HRV (as in, less variety in the interval between heartbeats) can be indicative of clinical conditions such as congestive heart failure, diabetes, but also PTSD and high levels of worry.
This is why we at BioBeats use it: to measure anxiety and stress at the physiological level, to correlate it to the psychological (which our users self-report).
Methods used to detect beats include: ECG and the pulse wave signal derived from a photoplethysmograph (PPG), which is the typical green laser you can see on Fitbits, the Apple Watch and others.
Other devices like the Apple Watch of the Fitbit range are (surely) tracking HRV, but they do not expose it in the SDKs to developers, making it impossible for third-party developers to create applications that use this vital biometric marker in their work. One core reason for this is battery life: keeping a PPG sensor activated and capturing cardiovascular data for longer than a minute takes a toll on battery life. Capturing it 24/7, or even 2 minutes out of every 10, needs extraordinary battery consumption magic.
But the battery life issue isn’t the biggest impediment to capturing HRV and using it in applications
The biggest problem with HRV taken from PPG sensors commonly found in consumer wearables is that the noise level (a disruption in the signal captured from the sensor) is horribly high: in the the wearables we have evaluated so far at BioBeats, even gentle movement of the wrist (moving to grab a glass of water whilst sitting down, for example) created enough noise that it became almost impossible to capture HRV from the PPG signal.
When the wearer stood up, or ran, the signal became almost entirely noise, and the heartbeats were simply not visible. This is why in most wearable data dashboards, manufacturers offer only Heart Rate over time (essentially, they capture your Heart Rate when they can, i.e. when you’re very still) and ignore the rest of the data.
This is a big problem in terms of the lost opportunity to create applications that look at wellbeing from more than a fitness and general health perspective. In order to look at wellbeing from a holistic perspective, combining both mental health and physical health, HRV is essential, but capturing it from typical consumer wearables is near to impossible.
At BioBeats, we’ve worked hard at solving this problem. The use of PPG instead of ECG for HRV analysis had not been extensively explored, and we found we had to push the boundaries in order to get the data we needed. With the exception of some variables, we showed that HRV analysis from PPG is generally as reliable, once we take care of the noise problem, as ECG. But first, we had to explore the reliability of the heart rate sensor in a number of devices, gaining a deep understanding of the problem motion introduces in trying to capture heartbeats from wearables, and find a way to filter unreliable portions of the data using the accelerometer sensors as a counterbalance.
In order to do this, we conducted a small experiment using the Microsoft Band 2, capturing 15m of data at a time, and comparing these to a Polar H7 chest strap from a single 41-year-old male subject. After removing outliers and filtering out unreliable sections of data, we found that some aspects of the data available were not reliable (such as sympathovagal balance/imbalance), but in the end we were able to show that is possible to automatically discard unreliable data to estimate the quality of the wearable sensor signal. We believe this could be used to selectively turn off the PPG sensor, affording significant battery life savings for the wearable. It will also allow us to avoid performing HRV analysis on unreliable data. In this way, we will be able to look only at data that has significant value, and to keep the wearable working for longer.
Why does this matter?
The wearables industry is just waking up. By now, the idea that we should be measuring HRV is well understood, and there are good posts out there on the importance of measuring it, entire blogs on its relationship to fitness training, and very good writeups on its relationship to stress overall. But, right now, we’re getting higher fidelity data from capturing heart rate from a smartphone’s camera than we are from wearables. Soon, companies like MC10 and others will give us on-skin and intradermal sensors that link to our smartphones without the need for the house arrest bracelet look. High end luxury watchmakers are already realising the value of adding biometri data to their lines. And obviously nanoparticle IoT computing will go wet and within our bodies, these sensor networks will give us data that will make the above problem go away for good.
But in the meantime, dealing with the noise will be a real battle. To read more about our experiment, and see our results, download our white paper on Profiling the propagation of error from PPG to HRV features in consumer wearable devices.