Monitoring HRV or resting heart rate can be used as a tool for predicting training adaptations or preventing overtraining. Nevertheless, to analyze these indicators, we first need to know how to interpret our data, and we must be aware of many factors that must be taken into account during such an analysis.
The measurement itself is equally important as analysis in this case. Errors in carrying it out can thwart all our efforts because such data will be completely inadequate. “Garbage in – garbage out”
Although the article mainly refers to the measurement and analysis of HRV, I will mention the possibilities of using the resting heart rate, which is also a source of similar and valuable information.
Autonomic nervous system
The activity of the autonomic nervous system is largely responsible for the regulation of the current heart rate (Shaffer et al., 2014). There are two main parts to the system:
Parasympathetic activity causes a decrease and a sympathetic increase in heart rate (Singh et al., 2018a). It might therefore seem that since both parts cause such opposite reactions, the decrease in the activity of one of them will be connected with an increase in the other.
It turns out, however, that the relationships between the two parts of the system are complicated and can occur independently of each other (Shaffer & Ginsberg, 2017).
Reverse, independent or simultaneous changes in the activity of two parts of the nervous system are possible (Ernst, 2017).
This means that the drop in heart rate can be caused either by an increase in parasympathetic activity with a sympathetic one being constant or by the same level of the former while increasing the activity of the latter. Unfortunately, this can complicate the interpretation of HRV behavior and the resting heart rate.
What is HRV in cycling?
HRV, or Heart Rate Variability, is nothing more than the variability of the time intervals between successive heartbeats (Task Force, 1996).
Usually, we get the impression that our heart beats like a metronome – at exactly equal intervals. In reality, however, it’s rarely the case. Successive heartbeats differ from each other, and this variability may be a source of valuable information for us.
A high HRV actually means that our heart beats less steadily, so we can intuitively conclude that such a condition is undesirable. Usually, however, the opposite is true (but not in all cases), and a higher HRV can actually be a sign of health (Singh et al., 2018b).
Heart Rate Variability can be a non-invasive method to measure the activity of the autonomic nervous system. While using its indices we can estimate parasympathetic activity, (although this method is still not ideal, de Geus et al., 2019), it turns out that HRV can’t be used as a way of estimating sympathetic activity (Billman, 2013).
In practice, therefore, HRV can only be used for measuring parasympathetic activity. I would also like to remind you a second time that both parts of the system work in a complicated and independent way, therefore an increase in HRV may indicate an increase in parasympathetic activity, but it does not mean a decrease in sympathetic activity.
For example, Elite HRV app has something called a readiness score. It is displayed on some kind of scale. On the left side, we have sympathetic activity, and on the right we have parasympathetic, and the arrow points out where our HRV is on a particular day.
That’s a completely wrong interpretation because actually, we can conclude from HRV analysis if our parasympathetic activity is lower or higher, but we can’t say anything about the sympathetic one.
Resting heart rate and parasympathetic activity
Theoretically, a decrease in our resting heart rate should mean an increase in parasympathetic activity (Sacha, 2014). However, due to the above-described complex relationships, this may not necessarily be the case. Potentially, the decrease in resting heart rate could be caused by e.g. a decrease in sympathetic activity with the same parasympathetic level.
As noted by J. Goldberger (1999), the average RR interval (i.e. the average time between successive heartbeats; although it is measured in units other than the heart rate, in practice it gives us the same information) is the result of the interaction between both the parasympathetic and sympathetic systems.
Because of that, changes in heart rate cannot provide us with information about the activity of a given part of the nervous system. It’s rather a function of the activity of both systems together.
Nevertheless, there is an exponential relationship between resting heart rate and HRV, while when we take into account the average time between successive heartbeats, the relationship is linear (de Geus et al., 2019).
Thus generally, we can say that the higher the HRV, the lower the resting heart rate, and the lower the HRV, the higher the resting heart rate.
Despite these ambiguities, in practice, our resting heart rate appears to have a similar analytical value to the HRV. For example, in the work of D. Plews et al. (2013), the correlations between the results obtained in 10 km of running, whether in the case of averaged weekly resting heart rate or HRV, were at a similar level.
As noted by M. Buchheit (2014), the use of the resting heart rate has the most practical value. The author also suggested that both HRV and resting heart rate could be a similar source of information for us.
There are many different metrics of Heart Rate Variability. Nevertheless, as it turns out, our analyzes can be minimized to only one of them (RMSSD). Generally, HRV indicators can be divided into three main groups:
The time-domain metrics are calculated as different kinds of statistical transformations based directly on the time intervals between consecutive heartbeats. The simplest indicator is SDNN, which is the standard deviation of the time intervals between normal heartbeats. It does not take into account any abnormal beats (e.g., premature or missed beats) that are corrected (more on this later).
This indicator is influenced by both the parasympathetic and the sympathetic parts of the nervous system (Shaffer & Ginsberg, 2017). Moreover, it is inappropriate to compare its results from measurements of different lengths (Task Force, 1996).
Another metric belonging to this group is NN50, i.e. the number of beats differing by more than 50 ms, and pNN50, i.e. the percentage of such beats in the entire measurement. This indicator reflects parasympathetic activity (Shaffer & Ginsberg, 2017).
Another indicator is RMSSD, i.e. Root Mean Square of Successive Differences. It is the root of the sum of the time intervals between successive heartbeats squared divided by the number of beats minus 1. As was the case with NN50 / pNN50, it reflects the parasympathetic activity, but it is preferred over previous ones (Shaffer and Ginsberg , 2017).
The second group of HRV metrics are frequency-domain indices. They are calculated using mathematical transformations (including Fast Fourier Transform), but at least for me, they are a bit abstract.
With the help of these transformations, we can obtain four frequency bands (Sassi et al., 2015):
● ULF – ultra low frequency (below 0.003 Hz);
● VLF – very low frequency (between 0.003 and 0.04 Hz);
● LF – low frequency (between 0.04 and 0.15 Hz);
● HF – high frequency (between 0.15 and 0.4 Hz).
The power in the first two bands comes from not entirely clear sources, while understanding of power at LF and HF is more clear, and these metrics are often used in literature.
Some sources equate LF with sympathetic activity. However, as it turns out, this view is incorrect (Billman, 2013), and the power in this band is influenced by both sympathetic and parasympathetic activity (Task Force, 1996). This problem significantly hinders the usefulness of this metric in analysis.
In turn, power in HF band is a good indication of parasympathetic activity and is a non-invasive way to estimate it (Dong, 2016).
We can also distinguish one more metric related to frequency-domain indices – TP (Total Power). As the name suggests it’s just total power in all frequency bands.
The last group of HRV metrics are those based on non-linear calculation methods. They are often based on the theory of mathematical chaos. An example of an indicator belonging to this group is DFA (Detrended Fluctuation Analysis). It can distinguish the completely random behavior of a given system from a completely ordered one or a mixture of both (Gronwald, Rogers & Hoos, 2020).
Which metric to choose?
Considering that the HRV indices turned out to be only a good approximation of parasympathetic activity, the potential choice is narrowed down to only a few of them. M. Buchheit suggests, however, that we can limit ourselves to only one metric – RMSSD (Buchheit, 2014).
The author distinguishes several advantages of that metric:
● it is a parasympathetic indicator;
● the results of a calculation are more repeatable than HF;
● the influence of respiration on the results obtained
is negligible (although other authors note that its influence is unclear, Shaffer et al., 2014).
Breathing has a significant influence on the interpretation of power in HF and should therefore be controlled to obtain meaningful results. Generally, the power in HF can “jump” to the LF at low breathing rates (about 3-9 bpm) (Earnest et al., 2004, Gąsior et al., 2020).
Taking into account the above-mentioned reasons and because of the practical benefits, in our analyzes, we can limit ourselves only to this metric.
A frequently used procedure is to extract the natural logarithm of the RMSSD (lnRMSSD). This procedure is performed due to the lack of a normal distribution of this metric, which makes it impossible to carry out some statistical analyzes.
However, I cannot say whether the lnRMSSD has a greater value for analysis in relation to the raw values.
Other authors note, however, that in the context of monitoring fatigue itself, RMSSD can indicate its general level, while the analysis of frequency indicators along with the heart rate can distinguish its different types (Schmitt et al., 2015).
For example, L. Schmitt (2016) was able to distinguish 3 types of fatigue in a case study of an Olympic swimmer. Each type of fatigue manifested itself in the form of different trends in HF, LF, and heart rate during the Orthostatic Test (more on this later).
Nevertheless, because of the hustle of analyzing much more metrics and data, for me personally it doesn’t bring a lot of value considering the additional effort.
How to measure resting heart rate?
Proper measurement of both HRV and resting heart rate is a key factor in the success of our further analyzes. According to the “garbage in, garbage out” principle, we should pay special attention to this aspect, because the analysis of poor quality data would be inappropriate right from the start.
The procedure of measuring resting heart rate is rather simple. Nevertheless, we must remember some important rules. We should take measurements in the morning, right after waking up (and a possible visit to the toilet).
In addition, we should always perform it in the same body position (e.g. sitting or lying). We also need to make external conditions as comparable as they possibly can be (e.g. the same lighting or temperature, but this one could be difficult to control).
An important point is that our heart rate needs some time to stabilize at a constant level (Bourdillon et al., 2017). Usually, 1 minute of such stabilization is used, after which we can start the measurement.
We can measure resting heart rate using the “manual” method, sensing the pulse with our hand, and counting the beats for 60 seconds.
The second way is to use a heart rate monitor, which will allow us to perform 1 minute of measurement (after 1 minute of stabilization period).
As it will turn out, in the case of HRV, the length of the measurement is of key importance, but I don’t really know if it can be important also in the case of resting heart rate (but I suppose that’s could be the case). Nevertheless, it seems to me that 1 minute of measurement is enough.
In the past, I thought that when measuring your heart rate, you should try to spot its minimum value, but we should rather aim at obtaining the average value of the heart rate during testing, not necessarily the minimal value. The minimal heart rate during the measurement may be a source of valuable information, but the analysis which I’ll present later rather uses the average heart rate.
How to measure heart rate variability?
Measuring HRV is not as simple as it was in the case of resting heart rate. The first thing we will need to do so is a tool that will allow us to capture so-called R-R intervals, from which heart rate variability is calculated.
We have two options to choose from:
● telemetry belt;
● PPG device (photoplethysmography).
Telemetry belts identify successive heartbeats by measuring the heart’s electrical activity.
Photoplethysmography, on the other hand, is based on passing a beam of light through the skin. That light reaches the blood vessels, and thanks to this the device can detect changes in blood volume associated with each heartbeat (Singh, 2018a).
But are commercial portable devices that can measure HRV as good as a “gold standard” – ECG? The authors of one article addressing this issue (Georgiou et al., 2018) found that portable devices are by all means a good source of HRV data.
They noticed, however, that devices were working really well at rest, but during exercise, their effectiveness in relation to the ECG decreased.
Similar conclusions were made based on the results of one meta-analysis (Dobbs et al., 2019). The authors concluded that despite the small measurement error of portable devices in relation to the ECG, this error was at an acceptable level.
Of course, not all telemetry belts can record R-R intervals (i.e. the time between successive heartbeats), so to measure HRV we need the appropriate equipment.
One of the most recommended belts which capture those intervals is Polar H10. In one study, this belt proved to be very good at identifying R-R intervals during exercise (Gilgen-Ammann et al., 2019). As earlier authors noted, the effectiveness of portable devices decreases with increasing intensity, so Polar probably works great also at rest.
Nevertheless Polar isn’t the only option in the market. For example, Garmin HRM Dual or Suunto Smart Heart Rate Belt can also capture R-R intervals. It is worth getting acquainted with the manufacturer’s specification to find out whether a given belt has such a function.
PPG technology can be found in sensors, such as CorSense offered by EliteHRV, as well as in many sports watches.
Nevertheless, the HRV4Training application can use this technology using the smartphone’s camera, which is quite unique.
How can I check my HRV at home?
We must be aware that many potential factors could influence the obtained HRV measurement result. The first thing that we must be aware of are all kinds of external factors that could influence our results. They should be minimized to obtain standard conditions for each measurement.
For this reason, it is usually recommended to take measurements right after waking up (and possibly visiting the toilet), as this allows us to obtain relatively similar measurement conditions each time.
Often, measurements are also made in a darkened room. At best, we should also try to conduct measurements at a relatively constant temperature (although this can sometimes be difficult to obtain).
Another aspect that we must consider is body position. We have three options to choose from;
● lying position;
● sitting position;
● orthostatic test.
Classically, measurements are made in a lying position. Nevertheless, standing (which is part of the orthostatic test) and potentially sitting positions can compensate for the so-called HRV saturation (Buchhiet, 2014), which may skew our analyzes.
In one study (Kiviniemi et al., 2004) researchers analyzed 24 h HRV measurements. The HF index (responsible for parasympathetic activity) was compared with the mean time between successive heartbeats. The analysis was conducted in shorter segments of the 24-hour measurement.
It turned out that in some cases they didn’t observe a linear relationship between the two above-mentioned factors. Above a certain threshold, HRV began to decrease despite the still increasing mean time between beats. In simple words, heart rate was decreasing, but it didn’t result in higher, but lower HRV.
This phenomenon is called HRV saturation. In this case, it usually occurred at low heart rate levels (often seen in athletes), but they found that in the case of some people it could occur at 40 bpm, but in others at 60 bpm. That’s a quite broad range.
What does this mean for us? It may turn out that when this phenomenon occurs, a decrease in HRV won’t mean a decrease in parasympathetic activity, but it will be quite the opposite.
Of course, a decrease in heart rate (i.e. an increase in the time between successive heartbeats) does not necessarily mean an increase in parasympathetic activity, so comparing it to a simultaneous decrease in HRV does not necessarily mean an actual increase in parasympathetic activity during above mention saturation.
Nevertheless, in another paper, the level of parasympathetic activity has been manipulated pharmacologically (Goldberger et al., 2001). Because of that, we can make stronger assumptions regarding the actual level of parasympathetic activity.
It turned out that a similar relationship was observed as in a previous paper. HRV (here measured by multiple indices, including RMSSD and HF) increased with increasing parasympathetic activity, along with increasing RR intervals (decreasing heart rate). However, this happened only up to a certain point.
With very high parasympathetic activity, it was observed in some people that HRV began to decrease despite still increasing R-R intervals length. Thus, a phenomenon analogous to the previously described saturation was observed here. Based on that we can conclude that saturation indeed is something that we need to have in mind.
These examples show that HRV is only a non-invasive method of measuring parasympathetic activity. Sometimes its decline does not necessarily mean a decline in this activity, but even an increase.
Therefore, when analyzing HRV, we should also take into account the average length of R-R intervals, or resting heart rate. If we observe a decrease in HRV but combined with a simultaneous increase in R-R length, this may mean that saturation occurred, i.e. an increase, not a decrease in parasympathetic activity.
Returning, however, to the actual positions of measuring. Personally, I think that if the sitting position can compensate for this potential saturation, then perhaps it is a better choice than the lying position (in case we don’t want to do an orthostatic test).
Nevertheless, it probably can still occur even while sitting, so even in that case, it is a good idea to analyze HRV in the context of changes in R-R / heart rate.
In the case of the orthostatic test, we are dealing with a combination of a lying and sitting position. It consists of an initial measurement in the supine position, followed by a transition to a standing position.
Apart from eliminating the phenomenon of saturation, the orthostatic test can be a source of additional information for us. As noted earlier, some authors were able to distinguish between several types of fatigue based on their results.
Moreover, a frequently observed phenomenon during the orthostatic test is the low difference between HRV and heart rate in the supine position in relation to the standing position, in the case when we increase our training loads quite rapidly (e.g. as it is during the training camp) (Le Meur et al., 2013; Schneider et al., 2019). Usually, while standing heart rate is much higher and HRV lower, than while lying, but here we are talking about a situation when those parameters differ only slightly in two positions.
For example, in a study by A. Barrero et al. (2020), HRV was monitored by cyclists who participated in an event promoting women’s cycling. It consisted of completing the Tour de France 2017 route one day before the men’s race.
This event was certainly associated with cumulative fatigue. It was observed that the difference between the resting heart rate or the RMSSD in the supine position comparing it to the standing position was significantly lower during the event than before.
Therefore, perhaps observing such a phenomenon during an orthostatic test may be additional information about fatigue for us.
So we can see that the orthostatic test can be a source of additional information, but we must also be aware of the downsides related to this solution. Unfortunately, this test is much less practical than measuring HRV in a sitting or lying position.
It requires more data to be interpreted, and that means, we will need more time to carry it out. Nevertheless, if this is not a problem for us, we can certainly use it.
Length of HRV measurement
Classically, it is recommended that the HRV measurement should last for at least 5 minutes (Buchheit, 2014).
Nevertheless, some authors have used measurements as short as 1 min (Esco and Flatt, 2014). They found that they were decent comparing it to the standard 5 min of measurement.
Nevertheless, N. Bourdillon and his team (2017), verified the minimum duration of HRV measurement against a longer standard. Orthostatic tests consisting of 7 minutes in the supine position and 6 minutes in the standing position were analyzed.
However, the reference point was 4 minutes of the total duration of a test. (i.e. slightly shorter than the recommended 5 minutes of measurement) for both the lying position (they analyzed 3-7 minutes) and the standing position (3-6 minutes).
Then the HRV criterion data was compared to the shorter time windows included in the 4 min measurement. It turned out that for the results to be comparable with criterion measurement, in the case of RMSSD we can reduce the measurement time to 2 minutes, while to calculate LF and TP we need at least 4 minutes.
Before taking the measurement, we should also use a stabilization period, which in this case was 1 min. A. Flatt and M. Esco (2016a) researched the issue of the length of the stabilization period and, they concluded that 1 min is sufficient.
In summary, the authors of the previously cited study suggested that the measurement time can be shortened to:
● 3 min in the case of RMSSD measured in a lying or sitting position (1 min of stabilization + 2 min of measurement);
● 6 min for RMSSD in orthostatic test (1 min + 2 mins lying down, 1 min + 2 mins standing);
● 5 min in a lying or sitting position, when, in addition to the RMSSD, we want to use frequency-domain metrics (1 min + 4 min);
● 10 min for the orthostatic test in which, apart from the RMSSD, we measure the frequency-domain indices (1 min + 4 min lying down, 1 m and + 4 min standing).
Even though shorter measurements turned out to be sufficiently accurate compared to longer solutions, when we have more time, it is worth using the classic 5-minute length, but also remembering about a 1-minute stabilization period. In total, such a measurement will take us 6 minutes.
One of the issues that should be mentioned in the case of HRV measurement is the so-called artifacts. They consist of some kind of erroneous heartbeats, either premature or missed.
They appear for both purely technical and biological reasons. Regardless of the reason, actions should be taken to correct them, because if left uncorrected, they can change the values of the obtained HRV metrics by up to 50% (Alcantara et al., 2020).
They are a common phenomenon, so we should not be surprised if they appear in our measurements.
The “gold standard” of editing them is manual verification of a given measurement, but this method is not practical under normal conditions. However, most of the available software or apps automatically detect and correct any artifacts.
It seems, however, that although everyone agrees that we need to correct them, at least for the moment, there is no standard method of doing so. We deal with many different approaches.
It often happens that because of different correction algorithms used, we can obtain different HRV values using different apps or software (despite the same raw data).
In addition, we need to remember that if the measurement contains too many artifacts, we should repeat it as the results may be skewed.
However, how much is too much is rather a subjective opinion, but for example, the Kubios software assumes an acceptable threshold of 5% artifacts for the entire measurement.
Which software/app to choose?
While it is possible to calculate the HRV metrics in Google Sheets or Excel (at least for RMSSD it is not overly complicated), we will rather need some software that will do it for us.
There are several options on the market, but it turns out not all of them are equally effective.
The most frequently used software for HRV analysis is Kubios. It is (at least for non-commercial use) completely free.
Nevertheless, because of a need for frequent HRV measurements for obtaining reliable analysis results, this solution would prove to be very impractical in the long run.
This is due to the need to record the measurement on a device (e.g. a bicycle computer) and then export the data to the computer each time we carry out the measurement.
Nevertheless, the Kubios app for Android or iOS smartphones has recently been created. Since the Kubios desktop application can be considered a “gold standard”, it seems to me that an app using the same technology should be just as good.
However, personally, I haven’t come across any publications that would specifically check the effectiveness of a smartphone app
The nice thing about the app is that we can regulate both the heart rate stabilization period and the length of the measurement in the settings. As I described earlier, it is of great importance.
However, I am not sure if we will be able to perform an orthostatic test with it.
To carry out the measurement using the Kubios app of software, we will need a telemetry belt that supports Bluetooth technology.
Another solution is the ithlete app. In that case, HRV measurement can be conducted both using a telemetry belt and a dedicated PPG sensor worn on the finger.
In 2013, A. Flatt and M. Esco compared the effectiveness of the application in relation to the ECG in the RMSSD measurement (only this metric is available). The results turned out to be very similar.
Later in 2017 (Esco, Flatt, and Nakamura), the PPG sensor offered by the manufacturer was also verified, but unlike the previous paper where measurements were made while lying down, sitting and standing positions were also verified.
It turned out that in the supine position the compliance with the ECG was high, but some discrepancies were observed in sitting and standing positions. However, the authors provided solid arguments that they did not necessarily have to be of practical importance.
Based on this, it seems that the application is doing well in measuring HRV against the “gold standard”. However, there is one point that, at least for me, precludes the use of ithlete.
It is because ithlete doesn’t allow us to change both measurement length and the stabilization period. Measurement lasts 55 s and a 1 min stabilization period is used. Unfortunately, based on Bourdillon et al. data (2017), such short measurements turned out to be different from the standard longer solutions.
Additionally one of the disadvantages is that the app isn’t free.
An orthostatic test theoretically is not possible here (unless we take two measurements in a row if possible).
The application that I personally use is Elite HRV (this does not prove its effectiveness, of course). It is completely free (although there is a paid version), and it contains all the features we need.
We can freely choose the measurement time and the stabilization period.
To conduct measurement we will need either a telemetry belt or a PPG finger sensor offered by the manufacturer.
A. Perrotta and his team (2017) compared RMSSD calculated with Elite HRV and Kubios software. This was possible due to the raw data being exported from the application.
It turned out that despite the high correlation between the HRV metrics values (r = 0.92), the results of another statistical analysis showed that the two calculation methods were not always compatible with each other.
The authors noted that one of the reasons behind some of the discrepancies was the use of different artifact correction methods.
Nevertheless, they pointed out that if we constantly use a solution that is associated with some kind of error, but we always compare the results obtained in this way, it will still have a practical value for us.
It is worth noting, however, that the work was published in 2017, and the app has since been updated.
Indeed, various artifact detection and correction algorithms may have contributed to such observations. For example, another publication (Gambassi et al, 2020) compared the R-R intervals (intervals between consecutive heartbeats) recorded by Elite HRV with the ECG.
Only the raw results of both measurement methods were compared here (without correcting the artifacts) and the recorded RR intervals were very similar to each other.
So it turns out that the application records heart rate data well, but the problem may lie with how it calculates HRV or how it corrects artifacts.
From my own experience, I will add that one thing intrigues me about the app. After every measure, the number of detected and corrected artifacts is presented. I don’t understand why the algorithm detects artifacts and does not correct them (so why even bother with detecting?). Also, I am not entirely sure if the number of uncorrected artifacts represents actual artifacts.
Nevertheless, if there is an obvious premature beat (which can be seen on the heart rate chart as a sharp spike, e.g. 120 bpm, with an average heart rate of 60 bpm), it is often corrected by the app.
However sometimes even in such an obvious case, algorithms don’t correct an artifact.
Nevertheless, I continue to use the app to have comparable data with the previous measurements, because I started using this app, and then after some while, I found the above-mentioned issues.
Elite HRV also doesn’t allow us to directly perform an orthostatic test. We could perform it by taking a longer measurement, but the app would not be able to distinguish between the two body positions.
Therefore, we would have to export the measurement to the Kubios software, which would be highly impractical.
However, we could take two measurements in a row, one while lying down and one standing.
One unique feature of Elite HRV is the ability to conduct HRV biofeedback sessions. It involves manipulating our breathing in such a way as to obtain the appropriate level of sinus arrhythmia (Shaffer, McCraty & Zerr, 2014).
Sinus arrhythmia is the phenomenon characterized by an increase in heart rate when we breathe in and a decrease during expiration.
Some publications suggest that HRV biofeedback may have some effect on athletic performance (Jiménez Morgan and Molina Mora, 2017), although much more research is certainly needed to draw clearer conclusions.
Interestingly, in one study, HRV biofeedback turned out to be an effective method of reducing stress during exams in students practicing sports. (Deschodt-Arsac et al., 2018).
Polar Flow is theoretically a free training analysis service (not only HRV). However, we must remember that to use it, we need to poses one of the manufacturer’s devices.
It is probably necessary to have one of the Polar’s GPS devices, while the telemetry belt alone may not be enough to use software (but maybe I am wrong).
Nevertheless, something that distinguishes it from the other solutions is the dedicated orthostatic test protocol. In the case of previously described apps, this test was theoretically possible to conduct, but it was not a built-in function.
According to the manufacturer, the test consists of 4 minutes (2 minutes lying down, 2 minutes standing), and the change of position is signaled by a sound.
Based on its results, we will obtain information about the heart rate in the standing and lying positions and its maximum value. Concerning HRV, we will gain information about RMSSD in a standing and lying position.
The duration of the test is consistent with the recommendations of N. Bourdillon and others (2017), but from the materials provided by the manufacturer, I am not able to determine whether the stabilization period is used here (I do not use Polar Flow myself, nor do I have a device from this manufacturer).
Unfortunately, I couldn’t come across a publication that would verify the effectiveness of HRV calculations by the Polar Flow application. It often happened that the authors exported raw data from Polar Flow to Kubios, but I did not find a direct comparison.
Another app on the list is HRV4Training. This app is based on the RMSSD measurement and uses among others D. Plews’ method, which I’ll describe later in the post.
Although by default app calculates its own HRV metric (which is based on some transformation of lnRMSSD), we can change it to the raw RMSSD
Considering the number of different functions, this app has by far the most of them among the previously presented.
This is at least the case with HRV, as Polar Flow also has several other functions, perhaps not directly related to HRV.
One unique thing about it is that we can measure HRV using a smartphone camera. Nevertheless, it is also possible to use a telemetry belt.
However, is the smartphone camera a sufficiently accurate tool for measuring HRV? It seems so. This is evidenced by the results of a study by D. Plews et al. (2017). They compared HRV measurement using ECG, Polar H7 telemetry belt, and the HRV4Training app. It turned out that both the telemetry belt and the smartphone camera allowed obtaining results comparable to the “gold standard”.
The app also allows us to use different measurement times: 1, 2, 3, and 5 minutes (although in EliteHRV or the Kubios app we can manipulate them to a greater extent).
Since I do not use the app personally, I am not sure if it uses the stabilization period (I also didn’t find such information on the website, but maybe I missed something).
Although shorter HRV measurements may not always be the most accurate solution, M. Altini (founder of HRV4Training) (2015) noted in one blog post that using a smartphone camera for longer than 1 minute may increase the number of artifacts caused by the movements of a finger placed on it.
In addition, HRV4Training is the second app, next to Polar Flow, that has a built-in ability to perform an orthostatic test.
The downside is that the app is paid, but it contains many additional analyzes that go far beyond the HRV measurement itself. Moreover, at least for the moment, it is not possible to make several measurements in one day.
There is also a desktop version of the software, HRV4Training Pro, which is really advanced. However, it should be remembered that for an average user, additional information obtained by using it might not always be necessary.
As was the case with EliteHRV, there is also the possibility of using HRV biofeedback here, but for this, we need to download the other app HRV4 Biofeedback. It is only available in the Apple Store.
In addition to the app itself, HRV4Training has an excellent HRV blog. There is a lot of interesting information there, so I think that if you want to broaden your knowledge on this topic, it is worth checking it out (https://www.hrv4training.com/blog)
Comparison of the Elite HRV, HRV4Training, and other apps
One very interesting publication (Stone et al., 2021) compared several HRV measurement options with the “gold standard”, i.e. ECG.
What’s unique about this paper is the fact that RMSSD values calculated by the software/app of a given manufacturer (and not the exported raw results, analyzed in other software) were compared with the values derived from the ECG, calculated in Kubios.
So the authors didn’t examine whether a particular app collects R-R intervals data properly, but rather they compared calculated by the app HRV metrics with ECG obtained ones and calculated in Kubios.
Several combinations of devices and software were compared in this way:
● HRV4Training using a telemetry belt (Polar H10) and a smartphone camera;
● Oura Ring (manufacturer’s software);
● Elite HRV with telemetry belt (Polar H10) and PPG CorSense sensor;
● Firstbeat using the manufacturer’s belt;
● Camera HRV app with iPhone 8.
Although only 5 people participated in the study, 148 different combinations of devices and software were compared. 3 different statistical methods were also used to determine the effectiveness of a given solution.
Summarizing all the analyzes, the authors noticed that HRV4Training with a telemetry belt and Oura Ring were the solutions closest to the “gold standard”. The Elite HRV with a telemetry belt and HRV4Training with a smartphone camera were a bit worse.
When it comes to comparing the CorSense sensor and the telemetry belt in the case of the Elite HRV, some analyzes showed that they were comparably effective, while others showed that CorSense was even slightly more accurate (which is a bit surprising).
Nevertheless, personally, it would not make me buy it.
Interestingly, the Firstbeat belt turned out to be the most accurate device for measuring average heart rate, while the RMSSD results were significantly skewed. It’s hard to say what caused this, but perhaps it was the way RMSSD was calculated (and artifact corrected) by the software.
Based on the above paper, Camera HRV does not seem to be a suitable tool for either heart rate measurement or RMSSD as the results, in this case, were complete garbage.
Summary of the measurement section
The first element that we need to take care of while conducting a measurement is obtaining the most standardized measurement conditions possible.
For this reason, we should conduct it after waking up and visiting the toilet, preferably in a darkened room and the same air temperature (but as I wrote earlier with this, it can be quite difficult).
Another very important aspect is the body position. It should be always the same because the relationships between parasympathetic and sympathetic in each of the three previously mentioned items will differ from each other (Martinmäki et al., 2006).
Among other things, for this reason, it is completely inappropriate to compare measurements taken in different body positions. We don’t compare apples to apples then.
We could choose either the lying or sitting position, or use the orthostatic test. However, we must be consistent with the position and use the same one each time.
It is also inappropriate to compare measurements of different duration (Mourot, 2018). Nevertheless, the authors who were verifying the minimum length of the HRV measurement did just that but they were checking the strength of the relationship between the shorter and longer measurements (Bourdillon et al., 2017).
Nevertheless, we should choose one measurement length and stick to it all the time.
In this regard, we can use the proposed shortened measurement variant lasting 3 minutes (1 minute of stabilization + 2 minutes of measurement), or the classic 6-minute one (1 minute of stabilization + 5 minutes of measurement).
It is also worth noting that HRV measurements calculated by different apps or software could differ. This is probably due to the use of different algorithms for detecting and correcting artifacts. For that reason, we should also perform measurements using one app or software.
In summary, we can distinguish three principles of HRV measurement:
● always under the same conditions;
● always in the same position;
● always of the same duration.
Heart Rate Variability Analysis
As important as proper acquiring data is proper analysis of it. Otherwise, the obtained values would be another meaningless number.
HRV metrics are characterized by quite substantial day-to-day variability. Because of that only frequent measurements can give us a complete picture of HRV trends (Flatt and Esco, 2016b). Analyzing only isolated values taken out of context can lead us to wrong conclusions.
For example, we could test our HRV only once a week on Thursday. After a couple of weeks, we could see what HRV values are typical for us. Then it could happen that for some reason our HRV is much lower on one Thursday.
We would think that something is wrong, but actually, on all other days in a week our heart rate variability could be normal, but that particular Thursday was just an exception.
So we could easily come to the wrong conclusion about our training state, for example, we could think that we are a little bit tired from training, but without having in mind a broader picture, we could be completely wrong.
What is good heart rate variability HRV?
We need to realize that there is no good or bad HRV. Of course in a clinical setting really low HRV sometimes can be used as a marker for predicting the future health of a patient, but it doesn’t necessarily apply to healthy athletes.
Many different factors affect the level of our HRV. These include age, gender, genetic factors, and physical activity (Fatisson et al., 2016). We can distinguish some kind of norms for a given metric (Shaffer and Ginsberg, 2017), but they can be only an indication for us whether our HRV is slightly lower or higher than in the case of the whole population.
However, when analyzing the HRV in our specific case, we will rather check whether the HRV on a given day or week is lower, higher, or the same as usual. We don’t necessarily need to compare our HRV with others, but rather analyze our specific case.
HRV data cannot be correctly interpreted by us without taking other factors into account. One of them is the training phase.
HRV cannot be analyzed in isolation
In the periodization of cycling training, we deal with the concept of a mesocycle. It consists of a few training weeks (usually three) and the so-called recovery week. During the training weeks, our training loads will increase, while during the recovery period, they will be lowered.
A decrease in the average weekly RMSSD values in a training week may be a sign of fatigue, while in the recovery week such behavior of HRV metric is actually positive and rather indicates recovery (Plews, 2014) (I’ll describe it in more detail in the section on analysis methods).
This example shows that the same HRV behavior, but analyzed in a different training context, will have a completely opposite meaning for us.
Another element that we must pay attention to is the saturation phenomenon described previously. For this reason, we need to analyze HRV along the length of R-R intervals or resting heart rate changes. Sometimes a decrease in the RMSSD may not be caused by a decrease in parasympathetic activity, but it could be higher in the case of saturation.
M. Buchheit (2014) also notes that the metrics related to the measurement of a heart activity can inform us about the current state of the cardiovascular system, but they don’t necessarily reflect the condition of other systems, e.g. muscles.
Based on HRV or heart rate we can conclude for example that we are already fresh and well regenerated, but maybe there are other factors that we need to take into account (e.g. muscle damage or glycogen resynthesis). That’s why the author recommends analyzing (along with HRV analysis) training diaries, questionnaires, or performing jumping tests (although I’m not sure if the jumping tests could be useful in the cycling context).
It is a really important point. It shows that HRV can’t be used as a global marker of fatigue, readiness to perform, etc. because it is related only to the activity of one system of the whole body. That’s why HRV measurement and analysis isn’t a panacea for all our problems in training.
We need also to take into account such basic things as our well-being. We should be asking ourselves if we are feeling good or a little bit fatigued. That will give a much broader picture of our training state.
In the past, I blindly believed in HRV metrics. Even when I was feeling good, but HRV was telling me something different, I trusted it more than myself.
I’ve changed my view a little bit because I did a couple of times interval training when HRV was telling me not to do so, but surprisingly the training went very well. In the other cases, HRV was telling me that I’m ready for intense work, but actually, I struggled to finish a session.
That opened my eyes and showed me the limitations of using HRV in training. Of course, it is very useful, but we need to have in mind that is not a perfect solution, and we need to take into consideration also other aspects – not analyze HRV in isolation.
In the literature, I have come across two great methods that allow us to analyze the HRV. One was developed by the Finnish researcher A. Kiviniemi, while the other comes from New Zealand and its author is D. Plews.
A. Kiviniemi conducted two interesting studies on the HRV-guided training (ie. based on HRV we decide what training we are doing). In the first of them, carried out in 2007, one group of runners followed a classic predefined training plan, while in the case of the second group, the morning HF level determined the training on a given day.
It turned out that the group training in the traditional way increased the maximum speed obtained in the step test, while the VO2max did not change. The HRV-based group, on the other hand, improved both VO2max and maximum speed. Moreover, the improvement in relation to the “traditional” group was higher and statistically significant.
In another study from 2010, the team conducted a similar experiment. This time, the SD1 metric was used, obtained from the Poincare chart, which is in fact analogous to the RMSSD (Michael et al., 2017).
In this case, we were again dealing with standard training compared to the HRV-guided approach. However, for women, two methods of determining training based on HRV were compared. One of them was analogous to that used in men, while the other one allowed for less training load to be achieved.
In the case of men, there was no difference in the increase in VO2max between the groups, but training based on HRV resulted in a greater increase in maximum power in the step test on a bicycle ergometer.
In the case of women, no significant differences were observed between the training effects of the methods. Nevertheless, the second method allowed to obtain the same effect, but at a lower cost, as the training loads were lower in this case.
The above-described HRV analysis method is based on one-day records of our HRV. In general, the idea behind it was to perform high-intensity training only when the HRV was the same or higher than the reference value. In the case where HRV is low, only low-intensity training is performed.
In the first study, the HF index was used. However, as I mentioned before, there are many reasons why our analysis can be limited to RMSSD. Moreover, in later work, SD1 was used, which is very similar to RMSSD, therefore the choice of this metric shouldn’t reduce the effectiveness of the method itself.
We can distinguish here three states in which our HRV can be found:
● it can be unchanged;
To determine all of these ranges, the first step is to take daily HRV measurements for 10 days. Then we need to calculate the average value of the RMSSD from this period, which, however, will be updated with each subsequent day, so it will become a moving average.
Then we should calculate a reference value based on this, which is the 10-day average minus the one standard deviation of RMSSD for the same period. Like the average, the value will also be updated every day.
Then we need to place the 10-day RMSSD average, the reference value, and the value for a particular day on one chart. On this basis, we will be able to determine where the RMSSD value on a given day is in relation to the established ranges.
If the RMSSD is below the mean minus the standard deviation, we may consider the HRV to be low on that day. If the value on a given day falls between the mean and the mean minus the standard deviation, we can assume that the HRV is constant. On the other hand, when the RMSSD on a given day is higher than the average, we consider HRV to be high.
In the case of the previously described method for men in one of the studies, when HRV was constant or high, they performed intensive training (i.e. above the mean minus the standard deviation), while when it was low, they performed low-intensity training (i.e. below the mean minus the deviation).
On the other hand, in the second method used in women, participants performed high-intensity training only when their HRV was high (i.e. only when the HRV value was above the 10-day average).
We can conclude that maybe the second method in the case of women is better, but actually, I don’t think that we can extrapolate those findings to a whole population of women (for example among others because of a rather low level of endurance performance of participants).
In addition, several additional rules were introduced:
● maximum 9 training days in a row;
● up to 2 intensive workouts in a row;
● HRV’s two-day downward trend was also considered low HRV.
A pattern similar to the one below was used:
Training based on HRV actually allows you to get slightly better results compared to the traditional approach. Nevertheless, the gains are by no means relatively large (Granero-Gallegos et al., 2020; Medellín Ruiz et al., 2020).
Nevertheless, as J. Medellín Ruiz noted in his meta-analysis, training based on HRV can increase the number of people who respond positively to a given training (ie. responders), and eliminate cases when a given method will not allow someone to improve results, or worse, cause them to drop (ie. non-responders).
Therefore even if such an approach may not allow us to obtain much better results than the classic one it may potentially increase the probability that our training program will be effective at all.
Nevertheless, HRV-guided training can be difficult to implement in many cases. Using it will mean that we don’t really know what type of training we’ll be doing on each day of the week.
When we have a job and other responsibilities during the week, it could be unsustainable for us, because we can’t plan training sessions in advance.
For this reason, it seems to me that for most people slightly different methods would be suitable. We could use an HRV as a warning sign, rather than for a prescription of training sessions on a particular day.
Therefore, we can adopt the rule that if our RMSSD is low (i.e. below the average minus standard deviation), and we had an interval training planned on that day, perhaps we can change it a little bit (but considering other aspects such as wellbeing as well), and do a low-intensity session.
If our HRV is lower than usual, it can be a kind of warning sign for us. Perhaps we haven’t recovered from our previous training yet, or maybe the day before we had a stressful day at work.
However, if everything is fine and our HRV is not low, we do not necessarily have to modify anything in our plan and we can stick to it.
It is also worth mentioning that the authors of the above-described method also considered the case when the daily HRV value decreased for two consecutive days as low HRV.
The use of the Kiviniemi method in the context of the resting heart rate
We could utilize the same methodology, but use the resting heart rate for the calculation instead of HRV. Nevertheless, I am not entirely sure if this approach would provide the same information as it was for HRV.
To do this, we would have to “reverse” all the calculations made for the HRV. This is because the decrease in resting heart rate may indicate an increase in parasympathetic activity (although, as we already know, not always), and in the case of HRV, such a change means its increase.
As with HRV, we need to calculate our 10-day average for resting heart rate (which will later be updated daily as a moving average). Then we should add (and not subtract as was the case with HRV) to the 10-day average resting heart rate, a standard deviation for the same period, which will also be updated daily.
Then, the resting heart rate on a particular day, its average over 10 days, the average increased by the standard deviation, can be summarized on one graph.
Following the methodology outlined by Kiviniemi et al (2007, 2010), if your resting heart rate on a given day is higher than the reference value, it may be better not to do intensive training on that day. However, if your resting heart rate is below this threshold, we can definitely perform this type of training.
I decided to compare (based on my data) whether the Kiviniemi method will be able to give us the same information depending on whether we use resting heart rate or HRV.
I verified both approaches based on the number of days identified as those with low HRV or high resting heart rate. In theory, during these days, my parasympathetic activity was lower than usual.
It turned out that during the selected period for analysis I observed many more cases (8) of low HRV than I did with my resting heart rate (5).
It was different only on March 16, where HRV was low despite constant resting heart rate. On the other hand, on March 27, HRV fell beyond the lower limit, while the heart rate did not rise until the following day (March 28).
One interesting observation is the fact that at that time, I had two periods in which HRV was clearly higher (03/20 – 23/03 and 02/04 – 05/04) than usual. When we look at the heart rate we can see the opposite trend, which could indicate an increase in parasympathetic activity.
Nevertheless, on April 6, the heart rate turned out to be low, despite the low HRV (possibly a saturation phenomenon, although all measurements were taken while sitting).
We can see that we certainly can apply the Kiviniemi methodology in the case of the resting heart rate.
Of course, there were some discrepancies with HRV here, but it’s hard to say what they might have resulted from. I think it might in part be due to the fact that the decrease/increase in resting heart rate does not always mean a decrease/increase in parasympathetic activity. The other thing could be just measurement errors related especially to HRV metrics.
Perhaps the phenomenon of saturation played a role here, although as I mentioned I was taking the measurements while sitting, so at least theoretically it should be reduced.
The D. Plews method was originally not used for HRV-guided training, although it was also successfully used in this regard (Vesterinen et al., 2016), but it seems to me that originally it was supposed to monitor training adaptations and prevent overtraining.
Unlike the Kiviniemi method, it is not based on one-day RMSSD values, but rather on averaged values over one week. This approach probably stems from, among others, the fact that, based on the results of one publication, in which the one-day values of lnRMSSD (i.e. the natural logarithm of RMSSD) were not correlated with the MAS (Maximal Aerobic Speed) and the time obtained in a 10 km run, while in the case of averaged values over 1 week, this relationship was much stronger (Plews et al., 2013).
In addition, in the triathlon case study, an athlete who was overtrained and didn’t perform well in the target competition showed a steady decrease in 7-day lnRMSSD values up to race day (Plews et al., 2012). So maybe this approach can be potentially effective in the early identification of overtraining.
Nevertheless, some authors believe that averaging the RMSSD values may lead to the loss of some information (Schneider et al., 2019). But, as I will explain below, both approaches (ie. one-day vs averaged HRV values) are a valuable source of information for us, although they can be used for completely different purposes.
Let’s get to the technical aspects related to the using Plews method. It uses SWC (Smallest Worthwhile Change), which is the smallest significant change in the RMSSD.
There is always some kind of error related to different measurements. It is no different in the case of HRV. Perhaps a 1 ms change in the RMSSD weekly average value has no practical significance for us, as it could only be a random change.
Therefore, the SWC helps us to determine the smallest change in a given parameter that will be of practical importance to us.
The problem is that there is no agreement on how to specify such a change for the RMSSD. There are for example approaches based on using a coefficient of variance or standard deviation for its calculation. Although maybe there shouldn’t be any difference in using each of them, who knows.
In some cases, SWC is calculated by multiplying RMSSD average value over a longer period of time (but how long?) by 0,5 of the standard deviation. On the other hand, we can come across a different approaches, e.g. 0.3, 0.5, 1 x coefficient of variance.
Moreover, M. Buchheit (2014), noted that the SWC was originally used to determine the minimal improvement in a person’s performance that could increase the probability of winning a competition.
However, the application of this concept to studying changes in the body’s physiological response (e.g., changes in the resting heart rate or HRV) is somewhat less clear.
Another issue is the determination of the time range based on which SWC is calculated. D. Plews (2014) in his doctoral dissertation indicated that the SWC determined as 0.5 of the coefficient of variance of the RMSSD from 14 days turned out to be a good method for tracking changes in HRV.
On the other hand, other authors used 28 days time frame to determine the SWC and calculated it as 0.5 x standard deviation (Vesterinen et al., 2016).
On the other hand, in another publication, authors used the SWC determined based on 5-week HRV data as 0.5 x the coefficient of variance (Le Meur et al., 2013).
Besides the ambiguity regarding the length of the period based on which SWC is calculated, another problem is when in a training phase we’ll determine it.
Usually, authors talk about the “normal” training period, without a lot of training load (Vesterinen et al., 2016).
However, depending on what we consider to be a “normal” period, we will obtain different results of calculations. We could assume that the “normal” period is the time when we do not perform intensive training and the training volume is not so high (similar to the recovery weeks). But our interpretation of this could be also totally different.
Based on SWC we could conclude whether our HRV is higher lower or the same as in this “normal” baseline period. But depending on when we conducted our baseline measurement, we could potentially get different results from our analysis. We could also ask a question – what training phase is the best for SWC determination?
Another problem is that the SWC must be updated after a certain training period. We certainly cannot assume that its value will be constant over time, given the commonly observed relationship between HRV and performance level.
Endurance training generally leads to an increase in HRV and a decrease in resting heart rate (Sandercock et al., 2005). Nevertheless, the relationship between performance level and HRV resembles an inverted U.
Both in people not undertaking physical activity and in the best athletes, HRV may be lower compared to people somewhere in the middle of the performance spectrum (Plews, 2014). Nevertheless, in the case of very well-trained individuals, lower values may be due to the previously described saturation phenomenon.
For this reason, we would have to update our SWC after some time. The only question is after what time? For example, in one publication this was done after 4 weeks of training (Vesterinen et al., 2016), but it is difficult to say whether it was optimal or not.
Another problem arises here. When we update our SWC after a fixed time range (eg. 4 weeks) new SWC would be based not on the baseline “normal” training period, but on the different one. So it seems that to establish a new SWC we would rather repeat the “normal” training period in the future to be able to compare apples to apples.
But the unanswered question remains – whether that “normal” training period is optimal for the SWC determination (maybe it would be better to determine it along with our training blocks), and what exactly means the keyword “normal”? I don’t really know the answers.
There is one more problem. Calculation of SWC is based on the coefficient of variation or standard deviations, but we could calculate these indexes from one-day HRV values or from 7 days rolling average. But since we are trying to establish SWC for 7 day rolling average I think that we should perform our calculations based on this average .
We can see, therefore, that there is absolutely no agreement as to how to specify this smallest worthwhile change.
Due to the above-described uncertainties, I decided to run some simulations on my own data to check how different approaches affect the obtained SWC result.
At the beginning, I checked if there was any difference when we used the coefficient of variance (CV) or standard deviation (SD) for the calculations. It turned out that the result was identical, which is due to the method of calculating the coefficient of variance (deviation/mean x 100%) itself.
In this case, I determined the SWC as 0.5 x CV / SD, obtained from the 7-day moving average RMSSD (not from each day’s values).
The next step was to calculate the SWC based on data from 14 and 28 days. It turned out that the obtained results were different (2.1 vs 3.45).
With the SWC We define some kind of area that defines when our HRV should be considered normal or unchanged.
To find it, you just need to subtract and then add the SWC to the 7-day moving average RMSSD average to get two thresholds. It may seem strange that we need to calculate the mean of the mean, but when we define this range for this particular indicator, we should do so.
I checked how the observed differences in the SWC affected the determination of this range.
It turned out that the range obtained from 14 days was higher than that determined from 28. Although both overlapped in part, they nevertheless differed significantly from each other. So which one is correct? I have no idea.
Such differences could affect the interpretation of our analyzes. Taking into account the fact that different methods of determining SWC can affect this “base rage” (ie. the range of HRV values that we consider normal or unchanged) in my mind using SWC in the HRV context can be a little bit confusing for us.
For this reason, it seems to me that we could limit our analysis only to tracking the trends in the 7-day RMSSD moving average, as the application of the SWC, in this case, is very unclear.
How to use the Plew’s method in practice
Although, in my opinion, the use of SWC in the context of HRV may sometimes mislead us, the second component of the above-described method, i.e. the 7-day RMSSD moving average, can be a source of valuable information.
One-day HRV values were used in the earlier Kiviniemi method. In turn, in this case, we are dealing with values averaged over a longer period of time. So we can ask ourselves which approach is better?
As we already know, averaged values, on the one hand, were better in predicting training adaptations in runners, and on the other hand, allowed for the identification of overtraining (Plews et al., 2012, 2013). So we can take this as an argument for the use of averaged values.
However, as aptly noted by V. Vesterinen and his team (2016), one training session has a negligible effect on the 7-day RMSSD value, but we can spot that change in behavior of the one-day values.
Therefore, both of the above-presented approaches are perfectly correct, but we can use them for different purposes.
The Kiviniemi method works rather on a micro-scale. It can be used to monitor HRV within one training week and introduce some modifications on this basis (for example decide what type of training we’ll be doing on a particular day).
In turn, the Plews method will work well on the macro scale. It is better suited for analysis of the behavior of HRV over the entire training mesocycle.
So how does the 7-day RMSSD moving average behave during training? The best answer is it depends. It depends mainly on the training phase. I have already pointed out that the interpretation of HRV behavior depends on whether we are analyzing training or recovery weeks.
The characteristic behavior of the 7-day mean RMSSD is very common and usually leads to optimal results (Plews, 2014; Buchheit, 2014).
The behavior is as follows:
● in the training weeks we see an increase in the 7-day moving average of the RMSSD;
● it declines in recovery weeks.
This is well illustrated by the example provided by M. Buchhiet (2014). The chart below shows data on the various parameters measured in the runner during the training period.
Starting from the top, the first graph shows the time per 10 km and the times achieved during the 5 x 1600 m interval training. This data is expressed as a percentage of the baseline time before the training period, so the lower the percentage, the better the time, and therefore better performance.
The second graph shows the behavior of the resting heart rate (blue 7-day moving average, gray values for a given day). The third shows the 7-day moving average of lnRMSSD (in red) and the one-day values (in gray).
The next graph shows the ratio of lnRMSSD to the average R-R interval length. This indicator can help detect the phenomenon of saturation, which I will describe a little later.
The graph at the bottom shows the training load, calculated using the method developed by C. Foster, based on the global RPE after a given training session.
In this case, training load is calculated as a score on the Borg CR-10 times exercise duration.
In blue, I marked the period in which the runner increased his training loads, so we can define it as the training phase. If we pay attention to the behavior of the 7-day average lnRMSSD, we notice that its value increases with increasing loads.
We observe a simultaneous decrease in the 7-day moving average of the resting heart rate. Nevertheless, at one point towards the end of this phase, we can still observe a decrease in HRV and an increase in heart rate.
When you look at the last week, HRV rises initially while heart rate drops. Nevertheless, at the end of it, both indicators return to the values observed in the training period.
If we pay attention to the results obtained during the 10 km run (black dots in the first graph), we’ll notice that at the end of the entire mesocycle, the time over this distance improved significantly for this athlete.
Although this is only one example, it shows that the “undulating” (ie. HRV lower in a training phase and higher in the recovery period) behavior of HRV and the resting heart rate in different training phases can indeed be an optimal response, indicative of a positive training adaptation.
RMSSD behavior versus functional overreaching
Y. Le Meur and his team (2013) conducted an interesting experiment. 24 triathletes were divided into 2 groups. The first training phase was identical for both of them.
It consisted of typical training performed by athletes for 3 weeks followed by a recovery week. After that first group repeated the same training schedule, but the other one increased their training loads by 40%.
This increase wasn’t related to weekly training load metrics such as TSS or TRIMP, but rather to the volume of a given training session. The authors gave an example that a training session lasting one hour and containing six 400 m repetitions at MAS (Maximal Aerobic Speed), turned into 85 min training with 10 such repetitions.
This increase was done to cause functional overreaching (FOR) in triathletes. This phenomenon is characterized by an initial drop in performance which is reversed after the recovery period, but performance, in this case, is higher than before the training block.
On the other hand, non-functional overreaching (NFOR) is characterized by the fact that after the rest period, our performance returns only to the level before the training block but without any improvement.
Whether or not the athletes performance was declining was verified by applying weekly maximum tests to the exhaustion on the treadmill in the second part of the experiment.
It turned out that people who entered the state of functional overreaching showed identical reactions in the context of resting heart rate and HRV as those described above.
During the period of increasing training loads, a decrease in the resting heart rate in the supine position was observed in 11 out of 13 triathletes in this group, and in all of them standing in the standing position (the orthostatic test was used).
Taking into account weekly HRV values (no rolling average used here), lnRMSSD increased in both positions, while lnHF (similar to RMSSD, the natural logarithm of HF power was taken here) increased in the standing position.
These changes were reversed after the application of a recovery week. HRV decreased and the resting heart rate increased, but they were slightly higher than before the overload.
In addition, the athletes who entered the state of functional overload significantly improved their performance after taking a rest, despite the initial decrease in performance during the overload period.
In contrast, in the control group, both HRV and resting heart rates were relatively constant. Although this group improved results compared to the beginning of the experiment, it was to a lesser extent than the functionally overreached athletes.
The results of this experiment show that an increase in weekly HRV values or a decrease in resting heart rate may indicate a functional overreaching. So, such behavior of the HRV metrics may indicate positive training adaptations, which, however, are revealed only after the use of rest.
On the other hand, the example of the control group shows us that an increase in HRV / decrease in resting heart rate is not a necessary condition for positive training adaptations to occur.
Perhaps these individuals improved their performance to a lesser extent, but we can conclude from this example that the stable values of weekly HRV/resting heart rate don’t have to be a negative sign during training weeks.
Is functional overreaching the best training strategy?
In the previous example, functional overreaching turned out to be a better approach than using the more stable amount of training load. However, is this training tactic also the most optimal?
This was verified in a 2014 study (Aubry et al.). In many aspects, it was analogous to the above-mentioned one. We are dealing with triathletes again, divided into a group with increased and constant training loads.
The same system of 3 training weeks and 1 recovery week was used, but after the end of the training phase, the recovery period was slightly longer here and lasted 4 weeks.
The first group performed the same training program in two consecutive mesocycles (3 + 1), while the second group increased their load concerning the first mesocycle by 30%, while as before, it was related to the time of the training sessions rather than different measures of training load (e.g. TRIMP).
Here, the performance level was determined by a test to exhaustion on a bicycle ergometer. Both groups performed it at the beginning and end of the second training mesocycle and each week during the recovery period.
It turned out that two more subgroups could be distinguished in the group increasing the load. Some people maintained the level of their performance, despite the increase in training loads, therefore the criterion of functional overreaching was not met here. The authors described these people as very tired (AF – Acutely Fatigued).
In the remaining participants of this group, a decrease in performance was observed after increasing the load, therefore functional overreaching could be diagnosed in their case.
All groups were able to improve their results after applying a recovery period (i.e. reduce volume while maintaining intensity). Nevertheless, it turned out that the best results were obtained in the group of people who were acutely fatigued (AF), and not those with functional overreaching (FOR).
In addition, the authors initially hypothesized that functional overreaching would be more effective, but after more time of recovery.
It turned out, however, that both in the case of acute fatigue and functional overreaching, the best results in a performance test were obtained in the second week after the overload period.
These results show that increasing training loads is obviously beneficial for performance development, but functional overreaching does not necessarily maximize training adaptations.
Nevertheless drawing conclusions based on group averages, can be often misleading, especially in the case of endurance sports (but others too).
This is due to the widely observed phenomenon of heterogeneity of training responses to the same training program. (Bouchard et al., 1999). In simple words – the same training has completely different effects on different people.
Here, however, the individual responses of specific people are also shown in the graph. In the control group where participants didn’t increase their loads, 2 people were not able to improve the maximal power obtained in the test to exhaustion.
However, in both the functionally overreached and the acutely fatigued groups, all participants improved their power during the test, although it did appear that the acutely fatigued group had more cases of significant improvement.
Unfortunately, in the case of this paper, HRV measurements were not performed among the participants. Therefore, on this basis, we can’t conclude whether average RMSSD values tend to increase both in the state of acute fatigue and functional overreaching. It would be quite valuable information because maybe trends in HRV can differentiate those two states.
The control group in this study did not increase training loads at all during the experiment. In practice, we more often deal with planning the increases in training load, but maybe not as substantial as in the case of an overload group (but of course not always).
It would be interesting to see if the less substantial increase in training load (for example 10% not 30%), could be even more effective (or it could be the other way around) than the acute fatigue state. Unfortunately, on the basis of this paper, we won’t get the answer.
Saturation phenomenon and the 7-day rolling average of the RMSSD
I have already mentioned the term saturation in this text many times. However how we can deal with it in practice?
D. Plews (2014) in his doctoral dissertation presented the possibility of using the ratio of RMSSD to the average length of the R-R interval (RMSSD:R-R ratio).
This index is calculated by dividing the RMSSD value obtained on a given day by the average length of the R-R interval during that measurement:
RMSSD:R-R ratio = RMSSD / average R-R
This metric could be helpful in the interpretation of trends related to the RMSSD 7 day rolling average.
When we deal with the saturation, there will be a simultaneous decline in both the ratio and the RMSSD. This will generally mean a decrease in RMSSD coupled with a decrease in resting heart rate.
However, should we also average this ratio across for example 7 days? I don’t know the exact answer, but M. Buchheit in the previously presented example of a runner used both one-day and averaged values. Unfortunately, I have no idea which approach is better.
D. Plews points out that a simultaneous decrease in RMSSD and an increase in the ratio may indicate fatigue and undesirable training adaptations. This means a simultaneous decrease in HRV and an increase in resting heart rate.
It is worth noting, however, that this is only reflected in the context of training weeks, not recovery weeks. In that case, such a relationship would be completely normal.
In addition to the two above-mentioned relationships, M. Buchheit (2014) distinguished two more situations:
● an increase in RMSSD with a decrease in RMSSD: R-R ratio or the same level of it;
● an increase in the RMSSD with a simultaneous increase in the RMSSD:R-R ratio.
The author interprets the first case as a sign that we are coping well with the training load.
However, as he notes, the second situation may occur at the beginning of the training block, probably in the case of people with saturation.
The author points out that if such a phenomenon occurs in the case of short training blocks, it could indicate readiness for competitions. However, if it is prolonged, it may on the other hand indicate cumulative fatigue.
Plew’s method for resting heart rate
There’s no doubt that this analysis method is suitable for resting heart rate. This is shown by the previous examples already cited. In the publication of Plews et al. (2013), averaged resting heart rate over a week turned out to be only slightly less correlated with MAS (Maximal Aerobic Speed) and the results of a 10 km run than HRV.
It was similar to the case of the chart prepared by M. Buchheit. There, the resting heart rate response coincided with those observed with lnRMSSD.
For this reason, there is nothing to prevent us from using this method in the case of resting heart rate. The principles of the analysis are almost identical.
Here we should also calculate the 7-day rolling average (but for resting heart rate). We can also calculate the SWC for this parameter, although as I pointed out, there is a lot of confusion in this regard.
The only difference is that the changes in your resting heart rate will be opposite to those seen with HRV. This is due to the fact that a decrease, not an increase in heart rate, may (but, as I have repeatedly pointed out, doesn’t have to be) evidence of an increase in parasympathetic activity.
Therefore, all the rules for HRV will be quite the opposite for heart rate.
Summary: Behaviour of your resting heart rate and HRV depending on the training phase
We already know that HRV should not be interpreted in isolation from other important factors, such as the training phase or saturation. Therefore, below I have decided to collect all information regarding the behavior of the RMSSD 7-day moving average as well as for the case of resting heart rate.
As the training phase, I mean the period when we increase our maintaining higher training loads. The most common duration of this phase is 3 training weeks.
When the moving average for heart rate decreases and HRV increases, we’ll probably get a performance increase soon.
However, it is worth remembering that this behavior may indicate functional overreaching (FOR). In his case, our performance in the training phase will drop, but after the recovery period, we should get a performance boost.
However, it is difficult to say whether an increase in HRV/decrease in resting heart rate is always a sign of FOR.
I didn’t come across any data that could dispel these doubts, although I did observe such HRV reaction in my own case, and at the same time, there was no indication that functional overreaching was the case.
The second positive trend in heart rate/HRV moving average is their stable value. As shown in the example of a study on FOR among triathletes (Le Meur et al., 2013), an increase in those metrics is not necessary to obtain an improvement.
What’s less desirable during the training phase is probably a significant decrease in the HRV moving average or an increase in your resting heart rate.
Perhaps even in that situation, after the recovery period, we could also get an increase in performance, but if that trend is prolonged it could be a sign of overtraining. If this is the case rest, not the training will improve our results.
However, we need to have the saturation effect in mind here. If we observe that the RMSSD rolling average is indeed going down but at the same time the resting heart rate or RMSSD: R-R ratio is going down, we shouldn’t be worried about that.
Despite the decrease in HRV, it may even mean an increase in parasympathetic activity, so in this particular case, such a trend could be interpreted as a positive thing.
The recovery phase
As the recovery phase, I mean the period when we reduce (e.g. by 40, 50%) our training load. This will include recovery weeks, where we usually reduce the volume intensity and tapers before races where the intensity remains the same despite the decrease in volume.
In this context, something that was negative during the training phase will be considered normal here.
When we observe an increase in the moving average of the RMSSD or a decrease in the resting heart rate during the training phase, it is the complete opposite for the recovery phase.
Here we consider the drop in HRV and increase in resting heart rate normal.
However, other reactions are probably possible as well. When both HRV and heart rate have been stable during the training phase, you may find that so will also in the recovery one.
Probably during the recovery week, it may also happen that the RMSSD will increase and the heart rate will drop. This was the case of the runner from M. Buchheit’s chart.
Monitoring your individual case
The reactions outlined above are typical, but we cannot rule out that there may be deviations from this time to time. Therefore, it makes sense to monitor the behavior of HRV or resting heart rate in our individual cases.
I think that we can analyze HRV trends in relation to how our performance changes over time. That way we could possibly spot individual HRV trends, that lead us to performance gains.
Perhaps on this basis, we will be able to make interesting observations.
Shaffer, F., McCraty, R., & Zerr, CL (2014). A healthy heart is not a metronome: an integrative review of the heart’s anatomy and heart rate variability. Frontiers in psychology, 5, 1040. https://doi.org/10.3389/fpsyg.2014.01040
Billman, GE (2011). Heart rate variability – a historical perspective. Frontiers in physiology, 2, 86.
Singh, N., Moneghetti, KJ, Christle, JW, Hadley, D., Plews, D., & Froelicher, V. (2018a). Heart Rate Variability: An Old Metric with New Meaning in the Era of using mHealth Technologies for Health and Exercise Training Guidance. Part One: Physiology and Methods. Arrhythmia & electrophysiology review, 7 (3), 193–198.
Singh, N., Moneghetti, KJ, Christle, JW, Hadley, D., Froelicher, V., & Plews, D. (2018b). Heart rate variability: an old metric with new meaning in the era of using mhealth technologies for health and exercise training guidance. part two: prognosis and training. Arrhythmia & electrophysiology review, 7 (4), 247.
Shaffer, F., & Ginsberg, JP (2017). An overview of heart rate variability metrics and norms. Frontiers in public health, 5, 258.
Ernst G. (2017). Heart-Rate Variability-More than Heart Beats ?. Frontiers in public health, 5, 240. https://doi.org/10.3389/fpubh.2017.00240
Electrophysiology, Task Force of the European Society of Cardiology the North American Society of Pacing. “Heart rate variability: standards of measurement, physiological interpretation, and clinical use.” Circulation 93.5 (1996): 1043-1065.
Sacha, J. (2014), Interaction between Heart Rate and Heart Rate Variability. Annals of Noninvasive Electrocardiology, 19: 207-216. https://doi.org/10.1111/anec.12148
de Geus, EJ, Gianaros, PJ, Brindle, RC, Jennings, JR, & Berntson, GG (2019). Should heart rate variability be “corrected” for heart rate? Biological, quantitative, and interpretive considerations. Psychophysiology, 56 (2), e13287.
Billman, GE (2013). The LF / HF ratio does not accurately measure cardiac sympatho-vagal balance. Frontiers in physiology, 4, 26.
Plews, DJ, Laursen, PB, Kilding, AE, & Buchheit, M. (2013). Evaluating training adaptation with heart-rate measures: a methodological comparison. International journal of sports physiology and performance, 8 (6), 688-691.
Goldberger JJ (1999). Sympathovagal balance: how should we measure it?. The American journal of physiology, 276 (4), H1273 – H1280. https://doi.org/10.1152/ajpheart.1999.276.4.H1273
Buchheit, M. (2014). Monitoring training status with HR measures: do all roads lead to Rome ?. Frontiers in physiology, 5, 73.
Sassi, R., Cerutti, S., Lombardi, F., Malik, M., Huikuri, HV, Peng, CK, … & Macfadyen, R. (2015). Advances in heart rate variability signal analysis: joint position statement by the e-Cardiology ESC Working Group and the European Heart Rhythm Association co-endorsed by the Asia Pacific Heart Rhythm Society. Ep Europace, 17 (9), 1341-1353.
Dong, JG (2016). The role of heart rate variability in sports physiology. Experimental and therapeutic medicine, 11 (5), 1531-1536.
Gronwald, T., Rogers, B., & Hoos, O. (2020). Fractal correlation properties of heart rate variability: A new biomarker for intensity distribution in endurance exercise and training prescription ?. Frontiers in Physiology, 11, 1152.
Gąsior, JS, Hoffmann, B., Silva, LEV, Małek, Ł., Flatt, AA, Baranowski, R., & Werner, B. (2020). Changes in Short-Term and Ultra-Short Term Heart Rate, Respiratory Rate, and Time-Domain Heart Rate Variability Parameters during Sympathetic Nervous System Activity Stimulation in Elite Modern Pentathlonists — A Pilot Study. Diagnostics, 10 (12), 1104.
Earnest CP, Jurca R, Church TS, et al. Relation between physical exertion and heart rate variability characteristics in professional cyclists during the Tour of Spain British Journal of Sports Medicine 2004; 38: 568-575.
Schmitt, L., Regnard, J., & Millet, GP (2015). Monitoring Fatigue Status with HRV Measures in Elite Athletes: An Avenue Beyond RMSSD ?. Frontiers in physiology, 6, 343. https://doi.org/10.3389/fphys.2015.00343
Schmitt, L., Regnard, J., Auguin, D., & Millet, GP (2016). Monitoring training and fatigue with heart rate variability: case study in a swimming Olympic champion. Journal of Fitness Research, 5 (3), 38-45.
Bourdillon, N., Schmitt, L., Yazdani, S., Vesin, JM, & Millet, GP (2017). Minimal window duration for accurate HRV recording in athletes. Frontiers in neuroscience, 11, 456.
Gilgen-Ammann, R., Schweizer, T. & Wyss, T. RR interval signal quality of a heart rate monitor and an ECG Holter at rest and during exercise. Eur J Appl Physiol 119, 1525-1532 (2019). https://doi.org/10.1007/s00421-019-04142-5
Georgiou, K., Larentzakis, AV, Khamis, NN, Alsuhaibani, GI, Alaska, YA, & Giallafos, EJ (2018). Can wearable devices accurately measure heart rate variability? A systematic review. Medica foil, 60 (1), 7-20.
Dobbs, WC, Fedewa, MV, MacDonald, HV, Holmes, CJ, Cicone, ZS, Plews, DJ, & Esco, MR (2019). The Accuracy of Acquiring Heart Rate Variability from Portable Devices: A Systematic Review and Meta-Analysis. Sports medicine (Auckland, NZ), 49 (3), 417-435. https://doi.org/10.1007/s40279-019-01061-5
Kiviniemi, AM, Hautala, AJ, Seppänen, T., Mäkikallio, TH, Huikuri, HV, & Tulppo, MP (2004). Saturation of high-frequency oscillations of RR intervals in healthy subjects and patients after acute myocardial infarction during ambulatory conditions. American journal of physiology. Heart and circulatory physiology, 287 (5), H1921 – H1927. https://doi.org/10.1152/ajpheart.00433.2004
Goldberger, JJ, Challapalli, S., Tung, R., Parker, MA, & Kadish, AH (2001). Relationship of heart rate variability to parasympathetic effect. Circulation, 103 (15), 1977–1983. https://doi.org/10.1161/01.cir.103.15.1977
Schneider, C., Wiewelhove, T., Raeder, C., Flatt, AA, Hoos, O., Hottenrott, L., Schumbera, O., Kellmann, M., Meyer, T., Pfeiffer, M., & Ferrauti, A. (2019). Heart Rate Variability Monitoring During Strength and High-Intensity Interval Training Overload Microcycles. Frontiers in physiology, 10, 582. https://doi.org/10.3389/fphys.2019.00582
Le Meur, Y., Pichon, A., Schaal, K., Schmitt, L., Louis, J., Gueneron, J., Vidal, PP, & Hausswirth, C. (2013). Evidence of parasympathetic hyperactivity in functionally overreached athletes. Medicine and science in sports and exercise, 45 (11), 2061–2071. https://doi.org/10.1249/MSS.0b013e3182980125
Barrero, A., Le Cunuder, A., Carrault, G., Carré, F., Schnell, F., & Le Douairon Lahaye, S. (2020). Modeling stress-recovery status through heart rate changes along a cycling grand tour. Frontiers in Neuroscience, 14, 1192.
Esco, MR, & Flatt, AA (2014). 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), 535–541.
Flatt, AA and Esco, MR (2016a), Heart rate variability stabilization in athletes: towards more convenient data acquisition. Clin Physiol Funct Imaging, 36: 331-336. https://doi.org/10.1111/cpf.12233
Alcantara, J., Plaza-Florido, A., Amaro-Gahete, FJ, Acosta, FM, Migueles, JH, Molina-Garcia, P., Sacha, J., Sanchez-Delgado, G., & Martinez-Tellez, B. (2020). Impact of Using Different Levels of Threshold-Based Artefact Correction on the Quantification of Heart Rate Variability in Three Independent Human Cohorts. Journal of clinical medicine, 9 (2), 325. https://doi.org/10.3390/jcm9020325
Flatt, AA, & Esco, MR (2013). Validity of the ithlete ™ Smart Phone Application for Determining Ultra-Short-Term Heart Rate Variability. Journal of human kinetics, 39, 85–92. https://doi.org/10.2478/hukin-2013-0071
Esco, MR, Flatt, AA, & Nakamura, FY (2017). Agreement Between a Smartphone Pulse Sensor Application and Electrocardiography for Determining lnRMSSD. Journal of strength and conditioning research, 31 (2), 380–385. https://doi.org/10.1519/JSC.0000000000001519
Perrotta, AS, Jeklin, AT, Hives, BA, Meanwell, LE, & Warburton, D. (2017). Validity of the Elite HRV Smartphone Application for Examining Heart Rate Variability in a Field-Based Setting. Journal of strength and conditioning research, 31 (8), 2296–2302. https://doi.org/10.1519/JSC.0000000000001841
Gambassi, BB, Neves, VR, Brito, EZA, da Silva Fernandes, DS, Sá, CA, da Rocha Nogueira, RM,… & Schwingel, PA (2020). A validation study of a smartphone application for heart rate variability assessment in asymptomatic adults. American Journal of Cardiovascular Disease, 10 (3), 219.
Shaffer, F., McCraty, R., & Zerr, CL (2014). A healthy heart is not a metronome: an integrative review of the heart’s anatomy and heart rate variability. Frontiers in psychology, 5, 1040. https://doi.org/10.3389/fpsyg.2014.01040
Jiménez Morgan, S., Molina Mora, JA Effect of Heart Rate Variability Biofeedback on Sport Performance, a Systematic Review. Appl Psychophysiol Biofeedback 42, 235-245 (2017). https://doi.org/10.1007/s10484-017-9364-2
Deschodt-Arsac, V., Lalanne, R., Spiluttini, B., Bertin, C., & Arsac, LM (2018). Effects of heart rate variability biofeedback training in athletes exposed to stress of university examinations. PloS one, 13 (7), e0201388. https://doi.org/10.1371/journal.pone.0201388
Plews, DJ, Scott, B., Altini, M., Wood, M., Kilding, AE, & Laursen, PB (2017). Comparison of Heart-Rate-Variability Recording With Smartphone Photoplethysmography, Polar H7 Chest Strap, and Electrocardiography. International journal of sports physiology and performance, 12 (10), 1324–1328. https://doi.org/10.1123/ijspp.2016-0668
Altini M. (2015). HRV measurements: duration. https://www.hrv4training.com/blog/hrv-measurements-duration [access: 19/04/2021]
Stone, JD, Ulman, HK, Tran, K., Thompson, AG, Halter, MD, Ramadan, JH, Stephenson, M., Finomore, VS, Jr, Galster, SM, Rezai, AR, & Hagen, JA (2021 ). Assessing the Accuracy of Popular Commercial Technologies That Measure Resting Heart Rate and Heart Rate Variability. Frontiers in sports and active living, 3, 585870. https://doi.org/10.3389/fspor.2021.585870
Martinmäki, K., Rusko, H., Kooistra, L., Kettunen, J., & Saalasti, S. (2006). Intraindividual validation of heart rate variability indexes to measure vagal effects on hearts. American journal of physiology. Heart and circulatory physiology, 290 (2), H640 – H647. https://doi.org/10.1152/ajpheart.00054.2005
Mourot L (2018) CODESNA_HRV, a new tool to assess the activity of the autonomic nervous system from heart rate variability. Phys Med Rehabil Res 3: DOI: 10.15761 / PMRR.1000165
Flatt, AA, & Esco, MR (2016b). Evaluating Individual Training Adaptation With Smartphone-Derived Heart Rate Variability in a Collegiate Female Soccer Team. Journal of strength and conditioning research, 30 (2), 378–385. https://doi.org/10.1519/JSC.0000000000001095
Fatisson, J., Oswald, V., & Lalonde, F. (2016). Influence Diagram of Physiological and Environmental Factors Affecting Heart Rate Variability: An Extended Literature Overview. Heart International, 11 (1). https://doi.org/10.5301/heartint.5000232
Plews, D. (2014). The practical application of heart rate variability-monitoring training adaptation in world class athletes (Doctoral dissertation, Auckland University of Technology).
Kiviniemi, AM, Hautala, AJ, Kinnunen, H., & Tulppo, MP (2007). Endurance training guided individually by daily heart rate variability measurements. European journal of applied physiology, 101 (6), 743–751. https://doi.org/10.1007/s00421-007-0552-2
Kiviniemi, AM, Hautala, AJ, Kinnunen, H., Nissilä, J., Virtanen, P., Karjalainen, J., & Tulppo, MP (2010). Daily exercise prescription on the basis of HR variability among men and women. Medicine and science in sports and exercise, 42 (7), 1355–1363. https://doi.org/10.1249/mss.0b013e3181cd5f39
Kiviniemi, AM, Hautala, AJ, Kinnunen, H., Nissilä, J., Virtanen, P., Karjalainen, J., & Tulppo, MP (2010). Daily exercise prescription on the basis of HR variability among men and women. Medicine and science in sports and exercise, 42 (7), 1355–1363. https://doi.org/10.1249/mss.0b013e3181cd5f39
Michael, S., Graham, KS, & Davis, GM, Oam (2017). Cardiac Autonomic Responses during Exercise and Post-exercise Recovery Using Heart Rate Variability and Systolic Time Intervals-A Review. Frontiers in physiology, 8, 301. https://doi.org/10.3389/fphys.2017.00301
Granero-Gallegos, A., González-Quílez, A., Plews, D., & Carrasco-Poyatos, M. (2020). HRV-Based Training for Improving VO2max in Endurance Athletes. A Systematic Review with Meta-Analysis. International Journal of Environmental Research and Public Health, 17 (21), 7999. doi: 10.3390 / ijerph17217999
Medellín Ruiz, JP, Rubio-Arias, J. Á., Clemente-Suarez, VJ, & Ramos-Campo, DJ (2020). Effectiveness of Training Prescription Guided by Heart Rate Variability Versus Predefined Training for Physiological and Aerobic Performance Improvements: A Systematic Review and Meta-Analysis. Applied Sciences, 10 (23), 8532. doi: 10.3390 / app10238532
Seiler, S., Haugen, O., & Kuffel, E. (2007). Autonomic recovery after exercise in trained athletes: intensity and duration effects. Medicine and science in sports and exercise, 39 (8), 1366–1373. https://doi.org/10.1249/mss.0b013e318060f17d;
Vesterinen, V., Nummela, A., Heikura, I., Laine, T., Hynynen, E., Botella, J., & Häkkinen, K. (2016). Individual Endurance Training Prescription with Heart Rate Variability. Medicine and science in sports and exercise, 48 (7), 1347–1354. https://doi.org/10.1249/MSS.0000000000000910
Plews, DJ, Laursen, PB, Kilding, AE, & Buchheit, M. (2012). Heart rate variability in elite triathletes, is variation in variability the key to effective training? A case comparison. European journal of applied physiology, 112 (11), 3729–3741. https://doi.org/10.1007/s00421-012-2354-4
Sandercock, GR, Bromley, PD, & Brodie, DA (2005). Effects of exercise on heart rate variability: inferences from meta-analysis. Medicine and science in sports and exercise, 37 (3), 433–439. https://doi.org/10.1249/01.mss.0000155388.39002.9d
Aubry, A., Hausswirth, C., Louis, J., Coutts, AJ, & LE Meur, Y. (2014). Functional overreaching: the key to peak performance during the taper ?. Medicine and science in sports and exercise, 46 (9), 1769–1777. https://doi.org/10.1249/MSS.0000000000000301
Bouchard, C., An, P., Rice, T., Skinner, JS, Wilmore, JH, Gagnon, J., Pérusse, L., Leon, AS, & Rao, DC (1999). Familial aggregation of VO (2max) response to exercise training: results from the HERITAGE Family Study. Journal of applied physiology (Bethesda, Md.: 1985), 87 (3), 1003–1008. https://doi.org/10.115