What Is Behind Changes in Resting Heart Rate and Heart Rate Variability?
A large-scale analysis of longitudinal measurements acquired in free-living
Last year we published a research paper that was in the making for many years, looking at resting physiology (heart rate, heart rate variability) both between individuals or at the population level, with respect to age, sex, body mass index, and physical activity level and within individuals, with respect to stressors such as exercise intensity, alcohol intake, the menstrual cycle, and sickness.
I have published over fifty papers over the years, but this is probably my favorite, as it is the outcome of a very long process that started about 10 years ago when developing HRV4Training, the first app able to measure HRV accurately using just the phone camera (validation here).
As the adoption of the tool slowly grew over the years, and more and more people opted in to allow anonymized use of their data for research, we were able to analyze the relationships listed above on a large dataset of 9 million HRV4Training measurements collected from about 28 000 people over 5 years.
Let’s see what we learned by analyzing this data.
Population-level data
Normally, I’m not a fan of population-level analysis. I find it a bit of an outdated way of looking at physiological data (getting some people in the lab, taking a few measurements, looking at differences between people without any concern for their historical data, context, changes over time, etc.). For example, if you measure an athlete’s resting heart rate while sick, it might be 55 bpm, instead of the usual 40 bpm. According to “population averages”, nothing is wrong, and yet we know this is far from the truth.
While longitudinal measurement is the way to go forward (analyzing changes in your physiology with respect to your historical data, to highlight periods of higher stress or abnormalities), looking at population averages and stratifying such averages across different groups of individuals (men/women, based on activity level, age groups, etc.) can still provide some useful insights on the differences between resting heart rate and HRV. Our goal here is really to better grasp these differences, and therefore this analysis becomes useful even at the population level.
Some of the findings are larger-scale replications of what we knew already. Consistent results with published literature that used different data collection procedures is a good first step, which gives confidence in the quality of the data before we start digging a bit deeper.
In our paper, we looked at population-level differences in:
- Sex
- Body mass index (BMI)
- Age
- Physical activity level
Sex
In our dataset, women have higher resting heart rates than men, but very similar HRV. In fact, at a younger age, women have a slightly higher HRV (not shown here). This is of interest as a higher heart rate would normally be associated with lower HRV, but this is not the case when we look at the data grouped by sex. The discrepancy might be due to hormonal differences, as the data is often more similar between men and women after menopause.
Body Mass Index (BMI)
In our dataset, both underweight and overweight or obese categories show what we have called in the paper a suboptimal physiological profile, meaning that resting heart rate increases and HRV reduces when deviating from what is considered a normal BMI. The strongest deviation is for the obese category.
Age
Looking at age, things get a bit more insightful. In our dataset, there was no correlation between resting heart rate and age, and a moderate correlation between HRV and age. This is one of the most interesting relationships, as heart rate and HRV clearly decouple and are representative of different processes (more on this later).
Physical activity level
The association between physical activity level and resting physiology is stronger for heart rate (correlation = 0.30, moderate effect size) than for HRV (correlation = 0.21, small effect size). When we break this down by age group, things get even more interesting.
In particular, the correlation between physical activity level and HRV reduces with age, getting down to 0.13 for older individuals. Only for very young individuals (20–30 age group) there is a decent association between physical activity level and HRV.
We also built models to determine how much variance age, sex, BMI, and physical activity level could explain. In other words, are these parameters sufficient to get a good understanding of inter-individual differences in resting physiology? Well, not really, as all parameters together explain 19% of the variance in heart rate and only 15% in HRV. This means that much of the difference in HRV between people is not necessarily linked to any of these parameters.
What are the implications of these findings?
In terms of the population-level analysis, here is a summary of our findings. A low HRV in aging individuals might be associated with a deterioration of regulatory mechanisms. The weak link between physical activity and HRV as we age might similarly be associated with reduced baroreceptor sensitivity. On the contrary, increased stroke volume due to high levels of physical activity maintains resting heart rate low even for older age groups. In terms of explained variance, it is clear that genetic factors are key in explaining differences in heart rhythm between people.
An important implication here is that in our opinion, targeting improvements in HRV as intervention goals might not be realistic, given the strong heritability coupled with reductions in age and low explained variance associated with lifestyle factors such as physical activity level. There’s an important caveat here. In this work, we had a large sample. However, this sample is not representative of the whole population, but mostly of relatively healthy or health-conscious individuals. It is certainly possible to aim at improving HRV and there might be more to gain for e.g. people that do not exercise, are overweight, etc.
This is why HRV as an absolute value is often of less interest in our work. On the other hand, HRV was able to capture day-to-day stressors within individuals with high sensitivity, as I cover below.
Individual-level data
In the second part of our paper, we analyzed individual stress responses to:
- Training
- Menstrual cycle
- Sickness
- Alcohol intake
For this analysis, we used on average 1 year of data per person, from 28 000 people. This is in my view the most interesting part of the paper. Why? This type of analysis allows us to answer important questions: can a morning measurement capture individual stress responses effectively? Is it worth the trouble to look at HRV, or is HR enough? What is the difference between the two, when it comes to stress responses?
Let’s try to answer all of these questions.
Data collection
Before getting into the analysis, I’d like to cover a bit our procedure used for data collection and analysis. Measurements and annotations (training intensities, sickness, etc.) were collected using HRV4Training, first thing in the morning. Most measurements were taken with the phone camera (validation here).
How do we analyze individual stress responses? For each person, any given day there will be many stressors. However, if we take hundreds of days of data per person, and look at one stressor at a time, we can isolate the stressor and better understand its impact on resting physiology.
Training intensity
Let’s start by looking at the impact of training intensity on resting physiology. Here we split training intensity into low vs high-intensity days. The change in HRV is 4.6% between high and low-intensity days, while for heart rate, the difference is only 1.3%. HRV is therefore more sensitive to this stressor.
Interestingly, the change in HRV does not reduce across age groups, indicating how HRV captures training stress equally well for older individuals, while the change in heart rate reduces, making heart rate even less useful as we age. Additionally, women tend to have a less marked response (more about this later).
We also split training intensity into four categories, as shown below. Once again we can see how HRV is more sensitive to changes in training intensity, but also how these measurements capture very well self-reported training intensities.
Menstrual cycle
We know from previous studies that there should be a group-level difference in heart rate and HRV between the first and second phases of the menstrual cycle, with a typical suppression in HRV during the latter (luteal phase). In our dataset, the change in heart rate was 1.6% between the follicular and the luteal phases, while the change in HRV was 3.2%. Once again, HRV is more sensitive. These differences might also be the reason why other stressors show somewhat less marked responses in women, as the responses might be confounded by underlying changes in physiology occurring with hormonal changes associated with the menstrual cycle.
Alcohol intake
Looking at changes in resting physiology in response to alcohol intake we can see that they are 3–4 times larger than changes due to training or the menstrual cycle (6% change in heart rate and 12% change in HRV). This highlights how we need to remember that if we want to use HRV to capture stress or guide training, we need a holistic approach, in which a certain lifestyle is key.
Sickness
Not surprisingly, sickness is also a very strong stressor, similarly to alcohol intake (6% change in heart rate and 10% change in HRV), as shown in the figure below.
What are the implications of these findings?
When using HRV for training guidance, lifestyle is key, and poor lifestyle or health issues will take over.
Changes due to training intensity and the menstrual cycle are typically 3–4 times smaller than changes due to sickness or alcohol intake. A holistic approach to health and performance is needed.
Additionally, changes in HRV are 2–4 times larger than changes in heart rate in response to the same stressors, making HRV a more sensitive metric.
Interpretability
When we contextualize the percentage changes reported in this paper with what we know from literature, e.g. that the smallest practical or meaningful change in heart rate is 2% and in HRV is 3%, we can see how changes in heart rate are below this threshold, and therefore smaller than normal day to day variability. This means that heart rate is not sensitive enough unless we have very strong stressors (e.g. alcohol intake or sickness). This also means that while HRV is more sensitive, it is also less specific, as shown by the typically smaller effect sizes.
In other words: changes in heart rate are often of no practical utility (smaller than daily variability). On the other hand, higher stress will be reflected in HRV data no matter where it comes from and it might be difficult to get to the source (context is key).
We speculate that these findings might lead to new forms of HRV-guided training, where rest days are prescribed based on large changes in HR (as these capture only very strong stressors), while training intensity is modulated based on more subtle HRV responses (e.g. reducing intensity).
You can find the full text of the paper, here.
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.
Marco is the founder of HRV4Training, a data science advisor at Oura, an Editor at IEEE Pervasive Computing (Wearables), and a guest lecturer at VU Amsterdam.
He loves running.
Twitter: @altini_marco
