Runalyze has a new experimental feature: for activities with HRV data (or RR intervals) the aerobic threshold can be estimated (or at least it’s tried to do so).
The idea for this comes from two recent publications by Gronwald  and Rogers  and a python workbook by Marco Altini. For this, the activity is divided into sections of a certain length (currently: 3 minutes) and the so-called DFA-alpha1 value is determined section by section. In the mentioned publications it was determined that an alpha1 of about 0.75 corresponds to the aerobic threshold. At higher exertion, the value drops to about 0.5.
This makes it possible – in theory – to determine section by section whether one has trained below or above the aerobic threshold. Furthermore, if enough reliable data are available, the exact aerobic threshold (in terms of power/speed/heart rate) can be estimated by a regression line. In the first publication , this was successfully done for data from incremental tests as an example. Currently, it is hoped that this estimation is also possible for arbitrary activities.
In theory, this sounds excellent – in practice, however, this has not been confirmed in our tests so far. Nevertheless, we would like to make this experimental feature available to you in order to collect more experience with it together. The currently chosen scalar parameters as well as the procedure for the regression and possible prerequisites that the data or the structure of the training must fulfill are not set in stone.
Example: Progression run
Let’s just look at an example. This is about an hour-long activity on the treadmill, most of which involves a progression run from 4:45/km to 3:45/km (or 68-87%HRmax).
Now when we open the Aerobic Threshold Estimation for this activity, by default only the results section is open. However, we also take a look at the other sections here.
These settings can be changed by experts or for testing purposes. Especially Window Length and SDNN Threshold can have a big influence on the results.
- Window Length: Length of the sections in seconds. By default, the activity is divided into sections of 3 minutes each.
- SDNN Threshold: Threshold value for SDNN. Sections with larger values are not used for regression.
- Artifact Correction Threshold: Relative threshold to remove artifacts from RR intervals. (default: 10%)
- DFA-alpha1 for Aerobic Threshold: At which alpha1 is the aerobic threshold assumed and estimated?
The most important basis for the whole procedure is of course the raw data, i.e. the RR intervals (or beat-to-beat intervals, the intervals between the individual heartbeats). These can be examined graphically here, whereby in the diagram of the RR intervals shown here the values recognized as artifacts are marked in red. The other two graphs show only the acceptable data points. In addition, the data (both RR intervals and time series for velocity, etc.) can be exported in JSON format for further processing with custom tools.
The results are then presented. For each section, DFA-alpha1, SDNN, RMSSD and the average values for heart rate, power and speed are determined. Sections whose SDNN value is above the permissible limit are directly grayed out. This table can also be exported as a CSV or JSON file for possible further processing.
In the graphical representation, we initially only see the alpha values over time. It is already noticeable here: The values are almost all close to the threshold value of 0.75 or below. We would have hoped that the values would decrease over time with increasing intensity. Possibly the intensity at the beginning of the increase with 4:45/km was already too high, the aerobic threshold thus exceeded (from the athlete’s point of view, however, this is still a comfortable pace).
Using the radio buttons, you can switch the view to the regression for power, for example. However, the regression line should decrease and not increase as in this case. Therefore also the warning: These results are not feasible.
We have seen from this example that we do not yet obtain reliable estimates for all activities with the approach we have used so far. Possibly the activities we have tested so far are insufficient, but perhaps the parameters are still unfavorably chosen, the artifact cleaning is not yet good enough, or the regression is not yet optimally adapted to the conditions.
Nevertheless, we would be happy if you already try the tool and possibly have your own suggestions for improvement.
I don’t see a HRV recording at my Garmin activities
First: Only chest straps supports HRV recordings and almost all garmin watches do not record them by default. See our help article to activate the hrv recording.
Where do I find this tool?
If your activity has a HRV recording you will find it at the activity view
- Gronwald, T., Rogers, B., Hoos, O.: Fractal Correlation Properties of Heart Rate Variability: A New Biomarker for Intensity Distribution in Endurance Exercise and Training Prescription?, Frontiers in Physiology, 11, p. 1152, 2020 doi:10.3389/fphys.2020.550572
- Rogers, B., Giles, D., Draper, N., Hoos, O., Gronwald, T.: A new detection method defining the aerobic threshold for endurance exercise and training prescription based on fractal correlation properties of heart rate variability, Frontiers in Physiology, 2020 doi:10.3389/fphys.2020.596567
19 thoughts on “Aerobic threshold estimation based on HRV”
Hi, there are device, artefact and artefact correction issues. We are evaluating these issues currently and will come up with recommendations.
Use 2 min windows and a good chest strap… follow the discussion on the HRV logger from Marco on Twitter (Marcos Account). We discussed a few issues there.
Use also the blog from Bruce: http://www.muscleoxygentraining.com/2021/01/dfa-a1-and-exercise-intensity-faq.html
Thanks for having a look at this! Looking forward for your following recommendations.
I’ve added the link to Bruce’s faq to the blog post.
Hello, how do I test the tool?. Can’t find the “Aerobic Threshold Estimation” option in any activity.
Regards from Chile,
Your activity must have hrv data / rr-intervals recorded. Which device are you using?
The DFA a1 value can vary depending on pre processing settings, especially something called “detrending” method (see Kubios preferences and website for description). In addition, artifact presence/prevalence and/or any arrhythmia can affect the DFA a1 depending on the aggressiveness of RR time tolerance in the artifact settings. Finally, precise RR timing is needed so YMMV depending on the HRM.
Thanks for your remarks! I’ll have a look at Kubios detrending methods in the pre pocessing steps.
The above example is from a Garmin fenix3 with a hrm-run strap. Not sure how well that one compares to Polar’s devices in terms of hrv.
Nutze seit 2 Wochen die HRVLogger App von Marco Altini, um den DFAalpha 1 umständlich zu ermitteln, und nun implementiert ihr das als beta. Super!
Könnt ihr auch einen Graphen erstellen, indem die zeitliche Entwicklung dieser Schwellen ersichtlich ist, das ganze nach Sportart getrennt?
IndoorRudern, Fahrradindoor/outdoor, Laufen?
Viele Grüße, Dirk
Bisher können wir das nicht, weil wir bisher die Werte je Aktivität nicht speichern, sondern erst beim Aufruf jeweils “live” berechnen. Da warten wir quasi noch ab, bis wir hinsichtlich der Parameter “sicher” sind und auch klare Bedingungen haben, ob wir die geschätzte Schwelle weiter verwenden wollen oder nicht.
Bei meinen eigenen Aktivitäten bin ich bisher mit den Ergebnissen nicht “zufrieden”, aber das kann auch gut an der Qualität der Eingangsdaten oder der “falschen Struktur” der Workouts liegen. Daher diese “Testphase”.
Wir haben das aber langfristig auf unserem Wunschzettel. Wenn sich herausstellt, dass wir für zumindest einen Teil der Aktivitäten gute Schwellwerte ermitteln, werden wir diese auch speichern und als Graph (nach Sportart etc. getrennt) darstellen können.
Hi there, thanks for such a great application! I am a xc skier, using a garmin vivoactive HR + polar H10 strap – just curious whether this would be available for xc skiing, and if so how would I find it in the application? Thanks, Ben
I don’t think that the Vivoactive HR+ does store rr-intervals in the fit file. You might try what’s specified at https://the5krunner.com/2014/12/11/garmin-920xt-turn-on-hrv/ or have a look at the device’s settings and check if it does work for your device. Otherwise, there’s no chance to do the calculations if rr-intervals are not stored in the file.
Thanks for the reply and link – it seems from the comments of that artical that it may be possible, so I’ll give it a shot next week.
I get different results in Kubios than in Runalyze. The Kubios ones are matching my guestimated VT1 (by other methods) @ DFA a1 of 0.75 pretty well, while Runalyze ones always show higher DFA a1.
As the detrending method was missing at Runalyze you’ve got different results. We have just added this in the end of the last week. Take a look at the outcome of this change here: http://www.muscleoxygentraining.com/2021/05/runalyze-vs-kubios-dfa-a1-agreement.html (by Bruce Rogers)
Hi, does Fenix 6 have HRV recording for activities enabled as a default or do I need to ulpoad that hrv.input.fit file there too just like in F3 series?
Also, I’m usually training with Polar OH1 which should record HRV too, but is that accurate enough?
You just have to enable the HRV recording at the Garmin Fenix 6 in the menu (https://runalyze.com/help/article/enable-hrv-recording-garmin)
> Also, I’m usually training with Polar OH1 which should record HRV too, but is that accurate enough?
We don’t have a Polar OH1 – It shouldn’t be that bad, but we don’t have any test devices (only our own watches)
comment exporter un enregistrement venant de HRV logger dans Runalyse ???
Hrv logger propose d’exporter en
Runalyse d’accepte pas
I think we had contact yesterday via mail. We have added the possibility to upload the csv from HRV logger