Heart rate variability normal values

Blog post by Marco Altini.

Check out this blog post for a more recent update on HR and HRV population values based on HRV4Training data.


In this post I will be looking at heart rate variability (HRV) normal values in relation to other parameters (e.g. age and gender), with the aim of trying to provide some more perspective around HRV values, at least for HRV4Training users or people with similar characteristics (i.e. trained individuals).

What are normal values and what can we use them for?

For many other parameters we are often dealing with, such as body weight, temperature, blood pressure, we typically have in mind some range reference values that are supposedly ok or indicative of a good health condition. For example, a healthy heart rate is somewhat between 40 and 70, systolic and diastolic blood pressure should be around 90-120 and 60-80 respectively, etc. Typically, we know these values because they were derived from some sample of the population in past epidemiological research.

Simply put, normal values are summary statistics from a big number of people (our sample). If we collect data properly, they will be representative of the values in our population. However, a simple average HRV of the entire population might not be the most insightful summary to look at. 

What we can do, is to start stratifying and creating subsets of the population with different characteristics (e.g. age, gender, etc.) and try to understand what factors influence HRV, and what are normal values for people similar to us. 

​For example, being a $30$ years old male with a BMI of $24.6$ and training about $4$ times per week, I’d like to see my values with respect to people with characteristics similar to mine, more than the entire population. By looking at how much I deviate from people similar to me, I can determine if I am more or less the average person with my characteristics (i.e. I don’t deviate much from the mean) or if my HRV is particularly low or high with respect to other people similar to me (i.e. my values deviate much from the mean of the sample, see later). Note that this is far from obvious given current medical practice where we often look at values with respect to everyone else, see for example blood pressure or heart rate.

In HRV we always stress the importance of looking at your own baseline and monitor deviations from the baseline, however the baseline must also come from somewhere. Our baseline HRV is probably affected by some factors that we cannot easily measure (genetics?), other factors that we cannot change (age), and factors we can probably influence (lifestyle). 

Unfortunately, when it comes to HRV and normal values, things get a bit difficult to extrapolate. This is due to many reasons, one being that there is no single HRV number, and therefore before we start talking about normal values we should be talking about what feature we want to consider and why. Secondly, the protocol used to acquire HRV data differs between studies, with supine tests, tilt tests, nocturnal HRV recordings, all frequently used. Finally, acquiring HRV data was not that easy until a couple of years ago, therefore making it difficult just to acquire representative data on a decent amount of people.

HRV4Training sample

The lack of a clear, standardized protocol and HRV feature to use, make deriving normal values a bit more complicated. However, since the protocol and feature to be used are clearly defined in HRV4Training (i.e. typically a 60 seconds measurement from which we will extract rMSSD), we can use this data to create some reference values.

A note on bias
Without going too much into statistical terminology, normal values come from a sample of the population. As such, our values will depend on what criteria we used to select the sample. If I select only patients with cardiovascular disease as my sample, the mean HRV will most likely be very low. If I select only professional athletes, the mean HRV will most likely be very high. As a result, any comparison we do later on, will be biased by what our sample is. 

This is what happens in HRV4Training as well. The user base is people interested in tracking fitness, optimizing performance, measuring HRV and as such, this is a very biased sample of fit individuals, with respect to the entire population (i.e. everyone on this planet). So this post is really about normal values for these individuals, i.e. trained and on the fit side of the spectrum. If you are a user, you can probably find this quite useful, since you will be able to understand where you stand with respect to other people similar to you, i.e. HRV4Training users.

Still, I like also to put things into perspective in a broader sense, and therefore I did a small literature search for normal values or simply HRV values (rMSSD) in published studies. This way we can see also where HRV4Training users stand with respect to other samples of the population, before we dig into our data. 

HRV Normal values in published literature

In literature, values are often reported without any stratification, i.e. we get the mean and standard deviation of the HRV for all participants in the study together. Sometimes this is ok because the sample included in the study is very uniform (e.g. all male participants between 20 and 25 years old, a classic for studies done at university). Some studies do stratify participants, typically by age, and sometimes by gender or physical activity level. A very common distinction is between sedentary individuals and trained ones or athletes. 

For my review, I focused on studies reporting rMSSD during a supine test or in some limited cases nocturnal values, since rMSSD is our metric of interest and is invariant of the time window used to collect the data. This is important because all studies use different protocols. 

Most studies have really tiny sample size (i.e. 7-8 participants per group), which makes the distribution of the data very broad, since HRV changes so much between individuals. Here is what I could extract, summarized in a few points:

• For young highly trained individuals or athletes, values are higher, in the range $70-120$, still with great variability ($40-50$) [3, 4, 6, 7, 10].
• For young ($20-40$ years old), sedentary individuals, values range from the middle $30$s to the $50$s, with standard deviation $20-30$ [4, 5, 6, 9, 10].
• Healthy, physically active and young (age $20-30$ years old) individuals have rMSSD values around $60$ with broad standard deviation (20-30) [1, 2], which could be raised in certain studies following training programs of $12-16$ weeks [2, 5], up to about $80 \pm 40$. Note the huge variability. This category is defined more or less in between the two previous ones, but there is no clear criteria. 
• In patients populations, e.g. chronic heart failure patients, rMSSD is in the $20$s, with much lower variability ($10-15$) [8, 10].

Note that most of these studies include very young individuals and things change quite a bit with age. Given this data, I'm gonna keep it simple and assume a normal distribution and simulate some data (which is wrong but good enough for our purpose):

Now, we include HRV4Training data as well:

As expected, HRV4Training data have higher HRV than the general population (i.e. sedentary individuals) and are typically on the trained/athlete side of the spectrum. We are basically overlapping with these last two groups. The real distribution is right skewed. 

The good news for HRV4Training users is that even if your values are lower than other HRV4Training users similar to you, you are most likely still "better" than the average population. 

All about HRV4Training users

Now it’s time to look in more detail at HRV4Training users. In this analysis I included about 848 users, i.e. the ones with at least one month of data. The sample size is pretty good with respect to what is typically available in clinical studies. One of the advantages of outsourcing data collection to users. The distribution of the data is very spread out ($\mu = 79.3$, $\sigma = 31.7$), as we could see from the plot above. Let’s look at some of the variables that have a more or less known relation with HRV, to start stratifying. More specifically, I will look at:

• Age: HRV is known to be decreasing with age.
  
• Gender: the relation is unclear from research. In some studies there are no differences, in other men have higher values, in yet other studies women have higher values.
  
• Body Mass Index (BMI): higher BMI is often linked to being less active, (in the general population, see later) which in turn can be linked to lower HRV values (see summary from literature above)
  
• Trainings per week: more activity and intense trainings have been linked to higher HRV.

Age
HRV decreases with age, we can capture this easily in our data, regardless of other factors, since this is by far the strongest parameter influencing the cross sectional HRV values of our dataset (and results in literature as well). On the left I plotted rMSSD vs Age, while in the figure on the right I first clustered age into groups ($20 - 30$, $30 - 40$, etc.) and then plotted boxplots per age group. This last plot makes the relation a bit more clear, with rMSSD reducing with age:

The biggest reductions in rMSSD for HRV4Training users are between $20$ and $50$ years old, with values progressively reducing from $110$ ($15-20$ years old), to $90$ ($20-30$ years old), $83$ ($30-40$ years old) and $70$ ($40-50$) years old. Then we have pretty much the same values until the last age group.

Gender
In literature, it is sometimes reported that there is a difference between men and women in HRV. However, some studies report higher HRV for women (until a certain age, then things level off), while other studies report higher HRV for men. Let's see what is the relation between rMSSD and gender for HRV4Training users. 

First, I'll compute summary statistics over the entire dataset:

• Male: $\mu = 79.2$, $\sigma = 32.2$
  
• Female: $\mu = 79.8$, $\sigma = 28.9$
  

Pretty much the same. However, given the strong influence of age, looking at gender differences without stratifying by age might not make much sense. So let's combine these two parameters and look at rMSSD differences between gender across age groups:

We have way more men ($n = 716$) than women ($n = 132$). For some age groups there is really little data for women, so I kept only age groups with at least $20$ users. From these data, it seems HRV reduces with age for men, but values are pretty much the same for women. Given the small sample size per age group in women, I suspect there is a confounding effect of fitness / training level here. Difficult to derive any sound conclusion.

BMI
BMI is typically a decent parameter if we look at the average population, meaning that most people with high BMI will be overweight, and most people with low BMI will be underweight. However this changes completely when looking at athletes and people with a lot of muscle mass (think about Lebron James). Therefore, I don't expect any strong links between HRV and BMI in HRV4Training users. Here are the plots:

As BMI categories I used the standard definition of the world health organization. $90\%$ of the users are between the normal and overweight categories. Almost nobody is underweight ($0.9\%$). From these data we see no consistent differences across ages for BMI categories, even when looking only at normal or overweight groups, that are the ones with the most data. This validates our initial hypothesis that standard relations between BMI and HR/HRV or health in general, fall short when looking at trained individuals and athletes.

Trainings
Finally, since physical activity categories (sedentary, trained, etc.) are often based on how much studies participants train, it is interesting to look at HRV values with respect to weekly number of trainings. HRV4Training users input how much they train during registration, choosing between occasionally, 1-2 times, 3-4 times and everyday. Let's look at the data:

Here I excluded the extremes of age (<$20$ and >$60$) again, since there are very little data when we start stratifying in many subgroups (as Andrew Gelman says, there is never enough data, as soon as we have more, we'll just stratify more). 

While HRV values per age groups are similar across number of trainings per week, we can see consistent differences between for example, everyday and occasionally, the two extremes in our categories. These differences highlight that more trainings can be representative of higher rMSSD, as reported in literature.

Sport
Finally, something purely anecdotal on the relation between sport practiced, HRV and age in the HRV4Training userbase. This can also be indicative of training / physical activity level, since some sports clearly require more (and more intense) trainings, e.g. triathlon, while others require much less effort (e.g. walking).

Quick summary of what we can see from this plot:

• Walking is practiced by older users, and shows the lowest HRV values (makes sense).
  
• Crossfit and triathlon show the highest HRV values. For crossfit, this is also associated to the youngest users, however this is not the case for triathlon users, therefore highlighting how HRV is higher in triathlon users regardless of age.

• Body building users are also on the younger side, however their HRV seems quite average (for this very fit set of individuals). 
  

All parameters together
We were exploring the relation between the different parameters and HRV. Similarly to what was reported in literature, values for different stratifications (as well as the entire dataset) have very broad distributions. Thus, normal values cover a very broad range. Something else we can do at this point is to build a simple regression model to map the relation between all these parameters together and rMSSD. Once we do so, we obtain a model explaining less than $10\%$ of the variability in rMSSD ($R^2 = 0.09$), when including age, gender, trainings, BMI and even heart rate in our model. What does this mean? It means that from these parameters only, we cannot really predict accurately HRV values for an individual with specific characteristics. Why? Because we are missing out on many other factors that might have a bigger effect on HRV baseline, for example genetics (accounting up to $26\%$ according to literature [11]). As a matter of fact, of all the parameters analyzed, only age shows a strong effect. In literature, clear differences are always reported between sedentary and trained individuals. However, within trained individuals / athletes, differences are much less clear. Since HRV4Training users are basically all trained, we cannot really discriminate much at this level, except for some small differences we could highlight between users reporting training occasionally, and users reporting training everyday. 

Bringing your data back to you

Since last spring, HRV4Training included some extra features, one of them being population comparisons. You can simply go to Menu - population comparisons in the app and see how you compare with respect to others. This is an example of how data gathered from many users, you included, can be made more useful to you once it has been aggregated among many users. Another example is the analysis of orthostatic data I published a few weeks ago.

Anyways, to conclude, here is a brief explanation of how to read the plots under population comparisons. Let's break it down in two parts. Upper plot:

And the bottom plot, showing user-specific information:

That's all. We've seen how HRV values change (or don't change) across different parameters and how this data was fed back to HRV4Training to provide users with a bit more context on their HRV values. If you liked the post, don't miss the next ones, by registering to the HRV4Training mailing list. Thanks!  

Follow @marco_alt

References

[1] Hynynen, E., et al. "Effects of moderate and heavy endurance exercise on nocturnal HRV." Int J Sports Med 31.6 (2010): 428-432.
[2] Gamelin, F. X., et al. "Effect of training and detraining on heart rate variability in healthy young men." International journal of sports medicine 28.7 (2007): 564-570.
[3] Seiler, Stephen, Olav Haugen, and Erin Kuffel. "Autonomic recovery after exercise in trained athletes: intensity and duration effects." Medicine and science in sports and exercise 39.8 (2007): 1366.
[4] Mourot, Laurent, et al. "Decrease in heart rate variability with overtraining: assessment by the Poincaré plot analysis." Clinical physiology and functional imaging 24.1 (2004): 10-18.
[5] van Rensburg, DC Janse. "HEART RATE VARIABILITY (HRV) ASSESSMENT OF PHYSICAL TRAINING EFFECTS ON AUTONOMIC CARDIAC CONTROL."
[6] Aubert, André E., Bert Seps, and Frank Beckers. "Heart rate variability in athletes." Sports medicine 33.12 (2003): 889-919.
[7] Esco, Michael R., and Andrew A. Flatt. "Ultra-short-term heart rate variability indexes at rest and post-exercise in athletes: evaluating the agreement with accepted recommendations." Journal of sports science & medicine 13.3 (2014): 535.
[8] Nolan, James, et al. "Prospective study of heart rate variability and mortality in chronic heart failure results of the United Kingdom heart failure evaluation and assessment of risk trial (UK-Heart)." Circulation 98.15 (1998): 1510-1516.
[9] Sandercock, Gavin, et al. "Heart rate variability measures: a fresh look at reliability." Clinical science 113.3-4 (2007).
[10] Melanson, EDWARD L. "Resting heart rate variability in men varying in habitual physical activity." Medicine and science in sports and exercise 32.11 (2000): 1894-1901.
[11] Singh, Jagmeet P., et al. "Heritability of Heart Rate Variability The Framingham Heart Study." Circulation 99.17 (1999): 2251-2254.