Building HRV4Training Pro

Helping you making sense of physiological data

Marco Altini
6 min readJun 25, 2018

In the past few months we’ve been busy building HRV4Training Pro, a web platform for individuals and teams aiming at better understanding how different stressors affect their body, so that adjustments towards better health and performance can be made.

In this post, I’d like to cover the main approach behind our new platoform, deriving from the past 5 years of learnings. Since we launched the first and only validated camera-based Heart Rate Variability (HRV) measurement a couple of years back, we had the opportunity to learn a lot through continuous iterations and feedback from our community as well as from top scientists in the field.

From the average guy just like myself, to elite triathletes that I occasionally enjoy slowing down, HRV4Training made it extremely easy for everyone to gather meaningful data points linked to physiological stress.

So, what’s HRV4Training Pro about?

Quick step back first. What’s HRV and why do we care?

As pretty much anything affects the autonomic nervous system, collecting and analyzing longitudinal data representative of these effects, typically reflected in changes in vagal tone, can provide insights on many complex mechanisms taking place in health and disease. The best non-invasive proxy to vagal tone is HRV, as the autonomic nervous system modulates heart activity in response to stressors. Hence, the motivation to use HRV to assess recovery from workouts, as well as the influence of the many other stressors affecting our lives on a daily basis (e.g. work, family, traveling, food & alcohol intake, etc. — it all counts).

If you are new to the world HRV, you might want to read this post first, or browse through this deck.

Where are we at 5 years later?

We built the easiest and most cost-effective solution to acquire high quality physiological data, in particular HR and HRV at rest (yep, that’s HRV4Training). We built a solid community with evidence-based work at its core. We published a fair amount of work, from the validation of the camera-based measurement, to acute day to day changes in physiology (heart rate and HRV) in response to training, to methods to estimate VO2max from workout data, methods to estimate running performance and the relation between HRV, training load and injury in Crossfit. We’ve also partnered with countless universities to which we offer this platform for free, to facilitate their work. Transparency and solid scientific grounds are what we believe in, which is why we started documenting to the public and validating our work since day zero.

Alright, back to the original question: what’s HRV4Training Pro about?

Data -> Awareness -> Insights -> Actionability

From data comes awareness. However, we need to be able to interpret data correctly, read through the noise, in order to gain insights and make meaningful adjustments to our lifestyle, based on the data we collect. It’s easy to get overwhelmed when things are not put in the right context, and just add to the confusion.

HRV4Training Pro provides many insights, typically combining physiological data, subjective annotations and actual workouts data. Thanks to the many interactions with users and teams, we fine tuned the platform to highlight what we believe is the most effective way to analyze and interpret physiological data.

In particular, with our new platform, we strongly relied on the following principles:

  • everything is relative
  • going beyond day to day variability
  • multiparameter is key

Let’s look into them a little more in detail.

Everything is relative

Physiology needs to be always analyzed with respect to an individual’s historical data or normal values. In HRV4Training Pro we take this approach to the next level, allowing users and teams to build their own set of metrics and track progress over time.

For example, with the new dashboard, you can pick up to 6 parameters, including physiological data, training load and subjective annotations, and see how this combination of metrics has been evolving this week, with respect to the previous month.

We don’t stop at the past week and month, but interpret changes with respect to all your historical data, so that relative changes this week with respect to the previous month, can be put into perspective, based on what are normal, long-term variations for you (alright, it sounds a bit convoluted, but it all boils down to a nice, easy to read, radar plot).

Example of the new dashboard in HRV4Training Pro for Dan’s recent data and a set of relevant parameters. The green hexagon represents this week while the gray one his previous month. All data is normalized with respect to his historical values, for all parameters.

The entire platform is built around relative changes over different time scales, see for example a multi-parameter analysis of your physiological trends, in which significant changes in the past two weeks, are derived based on what are normal variations for you in the previous two months:

Looking at baseline changes on multiple parameters relevant to your physical condition (e.g. HRV, HRV, coefficient of variation, training load, etc), the web app can automatically determine if a recent HR or HRV trend is changing in a trivial way, or if the change is something to take more seriously, based on historical data. Based on a combination of physiological parameters and training load, the app will also estimate your current condition, if you are for example adapting well to a specific training block, or having issues and risking to accumulate fatigue or to go into overtraining

Going beyond day to day variability

The second most important point is the ability to abstract and go beyond day to day variability and acute changes, so that we can focus on baseline changes and the big picture. This is true not only for physiological measurements, but also other modeling techniques used for example to estimate freshness or injury risk.

HRV4Training Pro builds on our previous work on physiological trends to easily highlight how your baseline is changing with respect to your historical data and allow you to understand if variations are just normal or are consistently outside of your normal ranges, at a glance:

An example of normal ranges (greenish bar) and baseline (blue bar) changes over time. Periods of significantly higher stress can be spotted easily as they end up below the lower bound of the normal ranges, while variations within the green band are most likely just due to normal variability in physiology on a day to day basis. p.s. Nothing says true love like matching physiology.

Similarly, you can see below how acute and chronic load as well as freshness and injury risk can be abstracted at the week level and in relative terms, to provide more meaningful feedback.

If you are interested in learning more about these specific analyses, please refer to the user guide that you can find here.

In the top panel, we show two percentages to provide more intuitive meanings to the current acute and chronic load. In particular, acute training load is expressed as a percentage of chronic training load, so that you can see how much you are training in the past week, with respect to the past month and a half. This way you can easily keep an eye on relative increases in load and make sure you don’t overdo it.

Multiparameter is key

Granted that HRV is a strong marker of physiological stress and that it can be extremely valuable per se to track it. However, it is obvious that the ability to collect and analyze multiple physiological (and not only physiological) parameters is key to aid interpretation and provide additional context around our measurements.

HRV4Training Pro makes it very easy to look at the big picture, for example in the dashboard view shown above, or with the resting physiology analysis and correlations analysis shown below:

The Coefficient of Variation has been shown in research to be linked to trouble in adapting to a particular training block (or life stressor) when increasing over time. We use it in combination with resting heart rate and HRV, as well as training load, to automatically determine how you are responding to your training program, as shown above in the detected trend plot.
The Correlations page lets you explore the relationship between physiological variables (e.g. HR and HRV) and your annotated Tags. Don’t forget that these are just correlations, and don’t imply a causal relationship. Other variables might be key in explaining the correlations you are seeing (or not seeing).

Similarly, it should be no surprise that HRV data should be properly contextualized, and not analyzed alone. In particular, only by looking at objective data representative of physiological stress (HR and HRV), subjective data (your own or your athlete’s feeling, as well as aspects more difficult to quantify such as muscle soreness), and external / training load, we can make well-informed decisions.

In the image above we can see 1) a positive physiological response to increasing training load, highlighted by stable or increasing HRV. 2) A negative response to increased lifestyle stress, shown by a reduction in HRV despite constant training load. Integrating physiological data (HRV), training load, and subjective metrics is key to the proper interpretation of an individual’s response to training and lifestyle stressors and can help us make meaningful adjustments to improve health and performance.

That’s all for now.

I hope you found this read somewhat useful and it will help you make use of the new platform. For a more comprehensive overview of HRV4Training Pro, check out the user guide here.

There is a lot to learn from being a little more aware of our physiology and of how we respond to different life stressors.

Our goal is to help you build a little more on top of the daily measurements so that you can read through the noise and make meaningful changes based on relative, significant variations in physiology and associated parameters.

Enjoy HRV4Training Pro

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 the co-founder of HRV4Training and loves running.

Twitter: @altini_marco



Marco Altini

Founder, Data Science @ouraring Lecturer @VUamsterdam. PhD in Machine Learning, 2x MSc: Sport Science, Computer Science Engineering. Runner