Physiological correlates of training load (a.k.a. relation between heart rate, heart rate variability, and fitness)

Blog post by Marco Altini

Following up on our previous posts on user generated science, HRV population values, and HR & HRV by sport, this week will be exploring another interesting aspect; the relation between physiological variables such as HR & HRV and fitness, as derived from training load.

This is a particularly interesting aspect as research has been rather inconclusive, at least on the relation between HRV & fitness. Many studies tried to estimate fitness (e.g. VO2max) using HRV features. At the individual level, during an intervention that aimed at increasing physical activity and fitness, HRV did not appear to be increased in a dose-dependent manner with increasing levels of physical activity [1] (meaning that training increased VO2 max but not HRV). Similarly, at the group level, most studies typically reported changes in resting heart rate, but no changes in HRV [2, 3]. Resting heart rate changes are slower and less variable on a day to day basis, thus might be a better proxy to overall fitness level, leaving HRV for the day to day variability and recovery assessment. This should be no surprise as most sub-maximal tests to estimate VO2max typically involve some sort of pre-defined exercise and measuring HR at specific intensities. The lower the HR, the higher the fitness. In my own PhD research [4, 5, 6] I found similar relations, with submaximal HR being a decent predictor of VO2max when measured both at rest (e.g. while sleeping) and while performing light activities of daily living, such as walking, automatically detected using accelerometer data and pattern recognition methods (more on this research can be found here). On a personal level, I discussed some of these aspects a couple of years ago, during the early days of HRV4Training, full blog post here, in which I show how HRV during three months of increased training load and improved aerobic fitness was very representative of the previous day training and recovery, but not representative of fitness, while HR was trending down, as we would expect with increased fitness.

This being said, others [7, 11] found correlations between VO2max and HRV, sometimes only for certain participants, while still reporting consistent reductions in resting HR [12]. 

Cardiovascular responses to physical exercise can explain reductions in HR associated with aerobic training, as the heart muscle improves over time (left ventricular diastolic cavity dimensions, wall thickness and mass increase [8]), increasing stroke volume and reducing heart rate. However, other aspects are much less understood. For example, endurance training reduces resting and submaximal exercise blood pressures, and the mechanism of reduced blood pressure at rest is not known [9]. 

Given many of the complexities related to cardivascular control mechanisms and the many, small, inconclusive studies, looking at the relation between HRV & training load, several questions remain unanswered.

As we keep gathering data from thousand of users, ranging from inactive individuals to recreational and elite athletes, together with additional tags related to training load, we have the opportunity to observe patterns at the population level in our userbase.  In this post, we try to shed some light on the following questions:

• What is the association between resting HR, HRV and training load across individuals? 
• Does the relation between resting HR, HRV and training load differ based on the type of sport? (e.g. running, biking, etc.)

Dataset & assumptions

For this analysis, we included all users that used HRV4Training between October 2015 and April 2016, for a period of at least 1 month. We considered one month the minimum amount of time necessary to get an understanding of an individual's training load. We used the same timeframe to determine an individual's baseline resting HR and HRV (i.e. the average score). We included only users that reported being either runners, cyclists or triathletes. As these sports include a high aerobic component, and the relations between HR, HRV & fitness we are interested in are typically resulting from aerobic exercise, it makes more sense to start exploring these relations on a limited dataset of aerobically trained individuals. We also excluded all users not reporting how many times per week they train at registration. Eventually, we ended up with a bit less than 1100 users (approx 500 runners, 350 cyclists and 250 triathletes).

The main simplification/assumption we make in this analysis is that training load is representative of fitness. Training load and fitness (or VO2max) are not the same thing. This is an analysis of data acquired in uncontrolled free-living settings, and we did not design a study or had participants performing a VO2max test to assess fitness. However, training load has long being used as a measure of fitness, as intuitively, the more training you can take, the more fit you are (for a more comprehensive discussion on this topic, check out this great post by Andy Coggan on TrainingPeaks).

As we are looking at a broad set of individuals of different training levels and we are interested in macro changes between individuals, I believe making these simplifications is less problematic with respect to similar analysis over a small set of individuals with similar characteristics. Finally, I believe VO2max is a questionable measure as the results of a VO2max test are affected by several limitations (even just results being dependent on the type of test and normalization procedures). Better understanding relations between cardiac activity, training load and VO2max is an interesting area of research, however, it might not necessarily be relevant in the context of optimizing performance, as genetic factors might be behind both individual variations in HRV & VO2max [12]. Given the assumptions and considerations above, we will be using fitness & training load interchangeably in this post.

​Enough with the explanations, time to look at some data. At registration, each user specifies how frequently they train. Here is what we have for the users included in this analysis:

When we look at user generated data, we always ask ourselves a question first. Does these data make sense? We need to make sure what we look at is meaningful data, as there is no oversight while collecting data. While we can never be sure, each time we can confirm previous findings or gain small insights, we increase confidence on the reliability of the data. We can gain a small insight also from the plot above. For example, we can see how most users train either 3-4 times per week or every day, and we have runners and cyclists that also report training less (either occasionally or 1-2 times per week). However, we basically have no triathletes training that little, and most triathletes report training every day. This makes sense as doing three sports obviously takes much more time, and gives us additional confidence on the reliability of the data. There is no doubt some users have no interest filling in the tags, using all the features in the app, or might even be misusing the app for reasons unknown to us. However, as we included only users that used the app for more than a month, the likelihood that these users are included in this dataset is extremely low.

Heart rate, heart rate variability, and training load

Time to look at physiological data. We will be looking at training load in two ways:

1. Self-reported number of trainings at registration. Similarly to what is shown above, users report training "occasionally", "1-2 times per week", "3-4 times per week", or "every day". We can assume individuals that train every day have a higher fitness with respect to individuals that train much less, and see what are the respective HR & HRV values.
2. As habits change, a better way to gather training load information is probably to use the tags users can fill in in the app. Basically every time you measure, the app asks if you worked out the day before, and you can annotate a series of training-related parameters. We will look at the relation between trainings per week as self-reported as users used the app, and physiological parameters as well.

To keep things simple, for this analysis we will not look at other metrics that might be more representative of training load, such as intensity, RPE, TSS, training distance, etc. Something we might be exploring in future posts.

1. Trainings per week as self-reported at registration

Let's start with trainings per week as self-reported at registration. We will be looking at the three selected sports, starting with heart rate data:

As triathlete have basically no data for the ''1-2" and "occasionally" category, I don't think the data is particularly relevant there, However, for runners and cyclists we can see a clear reduction in HR from the "occasionally" to the "everyday" category. Sorry for not ordering them by number of trainings, would have been easier to read, but we will do that with the following analysis on the actual number of trainings recorded.

​What about HRV?

HRV data shows much smaller changes, with no clear trends. Runners training occasionally have pretty much the same HRV than runners training every day (last two boxplots). These data seems to confirm that HR at rest is a physiological parameter highly representative of training load, while HRV is not.

However, we can do better. Let's look at this data with respect of all trainings annotations during the period of 1 to 5 months in which the app has been used by each one of the over 1000 users included in this analysis. 

2. Tags annotated after the measurements

As I mentioned above, we can capture training load using trainings per week as reported during registration or compute trainings per week using the tags annotated after each measurement. If our training habits don't change much, and we spend a few seconds per day filling in our tags, these two metrics should be rather similar. We can also run a quick check, and look at these variables with respect to each other:

Self-reported annotations are consistent, as users that reported training a certain number of times per week at registration, ended up annotating a very similar number of trainings using the tags. We then used the annotated tags to compute the number of actual trainings per week.

Let's see what we get when plotting the relation between training load as derived from daily tags and heart rate, this time on a continuous scale:

Quite clearly, regardless of sport type, we have a very strong linear relation showing that HR is much lower in more fit individuals, or at least in individuals able to withstand a much higher training load.

These are expected results. However, it is very important to show them and to show how consistent they are with 1) our understanding of human physiology and the effects of aerobic training on the heart [9] 2) previous literature on the topic [2-6]. Why? Because once again we collected these data in uncontrolled free-living settings using a phone camera. These data shows clearly how the morning routine (i.e. our protocol) and the high quality of the data collected (an example here) can be used for more explorative analysis, such as the one that follows on HRV & training load.

So, what about HRV?

For HRV data, there seem to be no relation with training load. While the plot for runners seems to trend upwards, I think this is simply the effect of a few outliers, as very few individuals belong to the extreme right and one is showing particularly high rMSSD. The central part of the data, where we have most individuals, shows no trend between HRV and training load. On the contrary, we had a clear decreasing trend in the HR data above.

In these lats two plots I've also highlighted age (dot size) as we've previously reported a very strong link between HRV and age, and we need to make sure results are not confounded by age. As you can see, individuals of all ages are spread across the entire training load range, so we don't seem to have a problem there.

Summary

In this post we explored the relation between fitness (as expressed by training load) and physiological parameters such as heart rate and heart rate variability on a sample of more than 1000 HRV4Training users. 

​We have not found any relation between HRV and training load, while we could confirm a reduced HR for individuals able to take up a much higher training load.

​What does this mean for you or anyone monitoring these physiological measures?

• HRV should be used as a measure of recovery, as shown many times in literature [11], in our own research [13] and in this blog. Understanding how training as well as other stressors are affecting our physiology on a day to day basis as well as in the longer term, can help in optimizing performance but not necessarily in quantifying fitness.
• The population comparison in HRV4Training has the only purpose of trying to give you some perspective of what are HR and HRV values for other users with characteristics similar to yours. Such population comparisons seem to have little to say about who is more fit, or at least about who is able to train at higher training load, as the plots above show that HRV is pretty much all over the place regardless of training load. 
• Resting HR consistently shows reductions with increased levels of both training load and fitness, highlighting how it can be used more reliably to quantify or estimate fitness level. If you are looking for an easy way to quantify your progress or compare across people, resting HR seems a better starting point than HRV.

Last few points on what this post is not about:

• Each individual analyzed in this post is a single data point representative of a certain training load, HR and HRV, over a period of at least 1 month. This means we did not look at changes over time within an individual and at how HR and HRV relates to training load / fitness within an individual, but only at the population level. Does HR and HRV changes within an individual when increasing training load? Something for another blog post.
• While individuals that are able to take up a much higher training load, might actually be more fit (higher VO2max or whatever being more fit means) and perform better (e.g. run a 3 to 10km faster, as performance is often quantified in other studies this way), we did not look at performance & aerobic capacity (VO2max). Thus, it could still be that regardless of training load, higher HRV might be associated to individuals with higher VO2max, due to genetics or other reasons, as reported by others [12], when controlled by training load.​

What's next?
While we found some interesting relations in this post, we can definitely improve our analysis in at least two ways, and we will be doing so in future posts. First, we can quantify training load using other parameters, instead of simply the number of trainings per week. Secondly, we can start exploring not only the relation between physiological parameters (HR & HRV) and training load, but also the associations with performance, as we started gathering objective data on trainings for at least certain sports, due to the recent link to Strava. By adding parameters related to training distance, pace/speed, elevation gain, & time, we can start exploring more of the relation between physiological variables and performance. Stay tuned.

Follow @marco_alt

References

1. Melanson, E. L. (2000). Resting heart rate variability in men varying in habitual physical activity. Medicine and science in sports and exercise, 32(11), 1894-1901.
2. Lee, C. Matthew, and Albert Mendoza. "Dissociation of heart rate variability and heart rate recovery in well-trained athletes." European journal of applied physiology 112.7 (2012): 2757-2766.
3. Grant, Catharina C., et al. "Relationship between exercise capacity and heart rate variability: Supine and in response to an orthostatic stressor." Autonomic Neuroscience 151.2 (2009): 186-188.
4. M. Altini, PhD thesis: "Personalization of energy expenditure and cardiorespiratory fitness estimation using wearable sensors in supervised and unsupervised free-living conditions". download pdf.
5. M. Altini, P. Casale, J. Penders, O. Amft, "Cardiorespiratory fitness estimation in free-living using wearable sensors" accepted for publication in Artificial Intelligence in Medicine. download pdf.
6. M. Altini, P. Casale, J. Penders, O. Amft, "Cardiorespiratory fitness estimation using wearable sensors: laboratory and free-living analysis of context-specific submaximal heart rates". Accepted for publication in the Journal of Applied Physiology. download pdf.
7. https://www.google.com/patents/US6026320 
8. Fagard, R., Aubert, A., Staessen, J., Eynde, E. V., Vanhees, L., & Amery, A. N. T. O. O. N. (1984). Cardiac structure and function in cyclists and runners. Comparative echocardiographic study. British heart journal, 52(2), 124-129.
9. Aubert, A. E., Seps, B., & Beckers, F. (2003). Heart rate variability in athletes.Sports medicine, 33(12), 889-919.
10. Buchheit, M. (2014). Monitoring training status with HR measures: do all roads lead to Rome. Frontiers in physiology, 27(5), 73.
11. Buchheit, M., & Gindre, C. (2006). Cardiac parasympathetic regulation: respective associations with cardiorespiratory fitness and training load.American Journal of Physiology-Heart and Circulatory Physiology, 291(1), H451-H458.
12. ​Buchheit, Martin, et al. "Monitoring endurance running performance using cardiac parasympathetic function." European journal of applied physiology108.6 (2010): 1153-1167.​
13. Altini, M., & Amft, O. HRV4Training: Large-Scale Longitudinal Training Load Analysis in Unconstrained Free-Living Settings Using a Smartphone Application.