Truth or dogma

a journey in the world of heart rate variability

I will start this one with a bit of background about my journey.

I had no master plan to work on heart rate variability (HRV), or even with anything related to it, or in sports science. Many years ago, while studying computer science engineering, I found myself with little interest in the various topics that were part of my field (computer networks, computer graphics, programming in its various forms, cryptography, etc.).

Nothing really clicked until during the last year, I took a course on what was called “embedded systems”. Basically, sensors placed on the body or in the environment, which would sense stuff, to then be processed. Through this course, I had the opportunity to go abroad and ended up at IMEC, the largest R&D institute on nanotechnologies in Europe. In particular, IMEC had a branch that worked on the development of what we called “body area networks”. There, I started building hardware, firmware, and software for sensors that could measure brain activity, heart activity, movement, muscle contractions, and more. These were the early prototypes of today’s wearables.

For me, as a computer scientist, it was extremely fascinating to be able to use my skills to measure, analyze or estimate the body’s response to different aspects of daily life. This was about 15 years ago. After a few years of working on the development of these sensors, I went back to my computer science roots and did a Ph.D. in applied machine learning, working on the estimation of energy expenditure and cardiorespiratory fitness from data collected with these sensors. During these years, thanks to my colleagues and the experiments they were doing in the context of work-related stress, I learned about HRV.

It was good timing, as the first Android phones came out, and it was possible to code apps and build tools that for the first time, potentially many more people could use. I always loved building tools, hence I started coding on Android and later on iOS (when iPhone started supporting the Bluetooth low energy protocol, which allowed for communication with third-party sensors like Polar straps).

I studied, I built tools, I self-experimented, I learned from the literature, and I learned from other people doing similar work. I then moved to the US to start a company, focused on the development of a new sensor to measure uterine activity during pregnancy, and potentially unveil relationships between changes in physiology and pregnancy complications. I had already developed the first versions of HRV4Training at that point, which I was running as a side business, mostly because I found it interesting and useful for myself. Eventually, I figured this HRV thing was really where my interest was and decided to try to focus all my energies on this. I moved back to Europe and went back to University to study sports science.

During these 15 years, I got 4 degrees, started 2 companies, and published about 50 papers in peer-reviewed journals, plus a few patents. However, during this entire time, I have often been doubting what I have been doing and changed my mind many times when it comes to HRV. Sometimes I changed my mind because I understood something that I didn’t understand before. Other times because thanks to the technology we have developed, we could look at the data in ways that were not possible before, and the data changed my thinking. Other times working with people, and users of our tools, was a key part of changing my mind. Other times it was my own data and experience to change my mind.

Where am I going with this story?

This is a field that keeps me interested for various reasons: there is a solid scientific background in terms of assessing autonomic modulation of heart rhythm, but its application and use in real life, outside of controlled lab settings, is far from obvious. The human body responds to stressors in ways that can be unpredictable. The technological and physiological challenges associated with the measurement and analysis of HRV data are something that remains extremely fascinating to me, and non-obvious to a point that I didn’t get bored yet.

This is why I am still here, building these tools, trying to understand the data or the limits of what the data can tell us.

Collaborating and discussing with others (scientists, companies, etc.) is a key part of the process for me. This is why I have developed good relationships with many “competitors” (to the point that I even work for them at times). I have no interest in selling more, having more users, or whatever game others love to play in their sad quest for world domination.

I love building tools and I want them to be useful and grounded in science.

As I keep doing my work, I think it is essential that I try to be a decent scientist, and seek truth instead of dogma. Seeking truth can be painful, or at least difficult, as it forces us to face our bias, challenge our assumptions, and assess that we were probably wrong and might need to update our beliefs.

As HRV is more widespread these days, it is also challenged more frequently, which makes me think more and more about my own assumptions and work.

Since you are probably here because of HRV and not because of any particular interest in my ramblings, below are a few aspects that came to mind in which I changed my mind, and some others where I am more and more pondering what the truth might be.

• Breathing rate: I once thought that breathing rate had to be controlled for HRV measurements to be meaningful. I now think that this is not the case because of: 1) scientific literature showing that longitudinal data, collected within-individual, shows the same trends when breathing self-paced 2) interaction with users that felt that breathing using a pacer is unnatural, further impacting their measurement, 3) my own data analysis of repeated differences in HRV measurements under conditions of paced vs non-paced breathing, showing no differences between the methods (and therefore no improved reliability for paced breathing). You can learn more about this topic in this blog.

• Body position: for almost a decade, I thought that there was no difference in measuring while lying down or while sitting. I did not understand well the orthostatic measurement and how changing body position causes a response that becomes more representative of our body’s capacity to handle stress (i.e. of what we could call readiness). You can learn more about this topic, in this blog.

• Continuous measurement of HRV: I thought that measuring HRV continuously could be a useful way to track stress in real-time. Many years ago I built a tool that would even compute your normal range in a given location and hour of the day to account for differences in the circadian rhythm and home vs office environments, and then highlight deviations from your normal. I now understand that this is not a meaningful use of HRV if we look at the physiological limitations of what we are measuring with HRV. I now think that less is more, and only measuring HRV with a morning spot check or during the night, we collect actionable data on resting physiology. Additional applications do exist, but require a very specific context and controlled conditions (e.g. before and after exercise or during deep breathing), not a continuous or random sampling, despite their popularity. I discuss issues with continuous measurements of HRV in more detail here.

• Exercise HRV: I thought that a method based on exercise HRV data was effective in distinguishing different exercise intensity domains based on universal thresholds. Given the properties of the metric used (i.e. an autocorrelation measure where absolute thresholds do have a meaning even between individuals), this was potentially possible. However, the data has shown that the method is not reliable at the individual level. Please note that I do believe this is a really interesting tool in different contexts (e.g. tracking fatigue, or even exercise intensity), just not based on the same thresholds for every person. I discuss my view in more detail here.

• PPG vs ECG: I thought that PPG was equivalent to ECG given the data I have collected. However, this is the case only under certain circumstances, typically in healthy people and for resting measurements, outside of exercise / physical activity or other protocols that would alter e.g. blood pressure. It took me a while to better grasp all the elements involved, and it is important that we use PRV (pulse rate variability or HRV captured via PPG as we do with a phone camera or optical methods in other wearables) as a surrogate of HRV only under certain circumstances. You can learn more about this topic, in this other blog.

• User-generated data: one of the main motivations of my early work as a data scientist was the idea that by deploying a clinical grade tool to the consumer, and collecting data longitudinally with the right reference points, we could understand a lot more about the relationship between physiology, stress, and other factors. I am now quite disillusioned about this. I think that we can do it for obvious acute stressors that have a long-lasting impact (training, alcohol, sickness, menstrual cycle, etc. - which I published here), but I think that we really have no idea and no way to understand long-term changes. There’s seasonality, there are changes in the environment, in our diets, in our relationships, in our training, in our mental state, etc. - so many changes all the time that I think the best I can do is not a large scale analysis, but to build a good tool, so that you, the only person with all the required context, can possibly understand what is going on, and make meaningful adjustments. I am not saying that there isn’t anything to learn from large-scale data analysis, but given the data and the complexities I have learned over the years, I am not driven by this anymore. I am more interested in building tools that allow the user to do their own math (e.g. providing the required context such as a normal range for your physiology, a questionnaire to annotate subjective feel, etc.).

• HRV-guided training: the principle of HRV-guided training is to reduce training intensity when HRV is suppressed. This makes sense because the timing of the stressors matters, and it is probably a bad idea to put high stress on the body when it is already responding poorly (note that a low HRV does not mean that there is high stress, it means that there is a poor response to stress, hence you are in a danger zone since a good response to very high stress is a stable profile). I have not changed my mind about this but I do question it much more now than I used to in the past. Is this really the case? Or can I just ignore this and go hard and my body will come around? What if my body does need some shaking, so to speak? These are some of the questions that I’d like to see answered but are difficult to study systematically, at the individual level. There are certainly cases where it is a good idea to reduce stressors, and even the current research has shown that this approach does lead to improved performance. However, I wonder, are these the cases in which HRV and feel are aligned (e.g. you are feeling poorly and it is also reflected in your data), or is this a general rule (i.e. the physiology knows better)? Maybe if you feel great the physiology is less of a concern?

There are probably quite a few more, but these are the main ones that came to mind recently and prompted this blog.

I don’t have all the answers, but I am interested in seeking the truth when it comes to this application, which I think can be very useful in the right context.

Let’s see how far we get.

Marco holds a PhD cum laude in applied machine learning, an M.Sc. cum laude in computer science engineering, and an 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

New to HRV4Training?