Saturday, February 5, 2022

AlphaHRV - the first native Garmin DFA a1 data field

Updated 2/7/22, 2/14/22, 2/22/22, 2/23/22, 4/11/22

New readers who would like to learn more about DFA a1 during dynamic exercise please read the FAQ for background.

 New - Update Review in Frontiers in Physiology

A little over a year ago an established HRV app (HRV Logger) was modified to display DFA a1 as part of it's feature set.  Since that time, two websites (Runalyze, AIEndurance) and one Android app (Fatmaxxer) have advanced the a1 recording field further by essentially reverse engineering Kubios premium's feature of calculating a1 every 5-10s using the same preprocessing methods.  In these three modalities, I've shown that a1 agreement with Kubios is quite close, reassuring us that the results will be relevant to conclusions drawn from published studies.  The intent of this post is to do the same type of analysis and agreement comparisons with the new a1 data field for Garmin products, AlphaHRV.  


 

Because of the severe processor, memory and programming limitations of a Garmin device, the method of data preprocessing is different than Kubios (and Fatmaxxer, Runalyze and AIEndurance) which means results may not match as well as hoped for.  I've discussed before why preprocessing is so important and we saw how different results can be between the initial and final version of both Fatmaxxer and Runalyze.  Here is an comment from a study by Voss et al., that is instructive:

In addition - the following was included in my post on Fatmaxxer regarding "detrending" -

Since we are basing training thresholds on the value of DFA a1, precision is of utmost importance.  Other HRV indexes (SD1, SDNN) rely on looking for a nadir during an exercise ramp for AeT determination.  With a1, we use a value - that's both a blessing and a curse.  The blessing is that we don't need a ramp to failure and could probably estimate from a mixed exercise session.  The curse of course is that we are relying on accuracy of the HRM in getting the RR timing correct and the computation methods of the software.  The process of obtaining the a1 is complex.  The HRV software needs to first "detrend" the data.  This has nothing to do with "detrended fluctuation analysis", AKA "DFA".  The detrending of RR data is done to remove "stationaries".  These are slow changes in beat pattern from other causes.  If they are not removed properly, the DFA a1 will appear more "ordered" than it should be and a bias upward will be seen at very low values (for example a true a1 of .4 may appear as a .6 or higher).  This has been the bugaboo of some of the other software approaches including the initial python related packages.  For instance, Runalyze was somewhat inaccurate before they enhanced their approach to using the "smoothness priors" approach of Kubios.  HRV logger does not use the "smoothness priors" of Kubios fame method either.  The issue is worse for recordings with more stationaries of course - so YMMV.  We saw very good accuracy with HRV logger in the Frontiers study, but in my personal use, it is not there.  After a fair amount of work, Fatmaxxer has been significantly upgraded to use the Kubios method of "smoothness priors".  Despite the calculation load, the speed is not affected even with a re-computation of every 5 seconds

The detrending method employed in the Alpha HRV app can be varied between 3 options - Kalman filter (normal or strong) and "MA".  Although Kubios uses smoothness priors, the same development team did review the Kalman technique and their opinion was very favorable.

The current detrending method used by this app is based on a "Low pass FIR filter (Butterworth order 30) applied twice, forward and backwards".

In addition to the above, we also have the importance of artifact identification and correction.  Even if everything is perfect in the preprocessing department, inadequate artifact removal can bias the a1.  Although many readers just want to get to the bottom line on whether this app works - it's critical we understand that the data we rely upon for training decisions (the a1) may not be as valid as we think.  Therefore, we have the following "hurdles" to jump to yield a1 values in line with Kubios:

  • Preprocessing - detrending (moderate to high difficultly)
  • Artifact recognition and correction (variable difficulty)
  • The actual DFA calculation (not that difficult)
  • Presentation, graphing and recalculation interval (dependent on the developer)

With this in mind, let's take a look at the app.

Protocol - indoor cycling, no formal ramp but gradually raised power by simply changing gears at steady cadence to past the AeT.  This was the day after a hard session for me, so I knew the a1 would drop easily.  I wore the Polar H10 (bluetooth to Fatmaxxer and Garmin) and the Movesense ECG for rhythm analysis.  Bluetooth is recommended for avoiding artifacts at high HR but I decided to use Ant+ for the Garmin alpha 1 app.

Install - download, install, then go to the "activity" of interest (cycling, running etc), add a data screen and populate the data screen with a single connect IQ field, namely alpha HRV.  Use the Garmin Express PC app to change the settings.

Settings:

For this test I used bluetooth. 

However: If you were to use bluetooth, you must first make sure that the H10 is not paired with the Garmin device (if it is, delete the pairing).  The app apparently has it's own RR acquisition mode and if the H10 is already paired to the watch, it will not work in bluetooth (but will in ant+).  If you want to record RR to both your Garmin device for later analysis in Runalyze or AIendurance, use Ant+ to the app, use bluetooth to the Garmin device so as to be able to confirm results (after uploading to Runalyze or AIEndurance.)  Note - the app requires it's own bluetooth channel and can't use the same HRM module as paired to the Garmin device.  Therefore, to have bluetooth RR recordings to both the app and your Garmin unit, you must use 2 different H10s.

Edit 2/23/22

Potential game changer - it appears this app takes advantage of a higher Ant+ sample rate than the Garmin native device.  The result being - no more rate related missed beat artifact.  I have confirmed this in my own recording and another from a friend. At this point we can have the best of both worlds - Ant+ to the alphaHRV app, bluetooth to the watch and Fatmaxxer.


 

Next:





  • RR series duration - The conventional method for the a1 "measurement window" is to use a time of 2 minutes worth of data.  It turns out we need a certain minimum beat count, and a 2 min window gets us there.  This app does not use a 2 min window, instead using a fixed beat count.  This has been proven before to be valid and 200 beats is fine to use.  
  • It is important to note that the corresponding Kubios (or equivalent) value is not going to be identical simply based on a different span of beats.  However, this method is fine IMO and was validated by Hautala et al:



Period of consecutive DFA analysis - I did cut down the re computation to every 10 seconds and will probably go even further down (every 20s) to save battery.


  • Yes, very cool!
  • There is a listing for artifacts as well.


  • What we see is that the black (Garmin alpha HRV) and the red (Kubios) track remarkably well.  The a1 dynamic range on this session was from 1.75 to about .25 - wide enough to get a good idea if the app fell short in under or over estimating those critical values near .75 or .5.
  • Recall that Kubios used 2 minute window measurements (so 140bpm x 2 min = 280 beats) vs alpha HRV used 200 beat windows - the data points are not going to be identical.  But are they close enough for meaningful conclusions?

What about artifact correction?

  • Artifacts now show in Garmin connect!

Kubios comparison:

  • The high artifacts after the ramp were all APCs - alpha HRV had no difficulty in identification or correction. 13% vs 15% based on beat count vs 2 min window - near identical.
  • Successful implementation of artifact correction - check!


Zoomed view of the "ramp" 



  • The ramp data agreement looks quite solid!
  • This is an amazing achievement considering the limitations on a Garmin device!
  • Remember that Kubios is using 2 min windows and the alpha HRV app is using 200 beat windows - so they should not synch exactly.
  • That said, this is certainly "clinically" useful.  I would have confidence that I was staying in zone 1 of 3 (or 2 of 5) with a1 values above .75.  
  • For most precise threshold determination, I would still recommend Runalyze or AIEndurance if available. These options are closest to published a1 verification studies that used Kubios HRV software.

Added 2/23/22

A friends ramp plus cycling session using Ant+ for the alpha app, bluetooth to the Garmin device for analysis into Kubios 


  • There were no appreciable artifacts seen with Ant+, agreeing with my experience above - the alphaHRV app has a higher Ant+ sample rate than the Garmin device


Update 4/11/22

It's been awhile since the last head to head test with Kubios and I thought another look could be interesting.

This was a 2 hour session combining a brief ramp, 20 minutes just below the AeT (20w less than LT1) and 2 Wingate 30s near max intervals later on (all indoors).  Artifacts below 5%, Polar H10 recording, Kubios software vs alphaHRV.


Zoomed view of the 20 minute stable zone 2 of 5 (high zone 1 of 3):


  • Really impressive matching with Kubios despite the difference in window length (2 minutes vs 200 beats) and different detrending methodology.
  • Certainly would be of use for proper zone enforcement for low-moderate intensity days.


 

Conclusion:

  • Hats off to the developers for making the first DFA a1 native data field for Garmin devices (you guys should be proud).  This brings DFA a1 monitoring during dynamic exercise for intensity assessment, training management, zone identification, readiness and durability to a large portion of Garmin users.
  • Initial test results are very promising and further exploration is certainly needed.  The Detrending method is different than Kubios, but appears to capture the data value direction and range.  
  • The Ant+ transmission difficulty appears to be Garmin software specific - alternate Ant+ acquisition using max sample rate allowed gives excellent results. My recommendation is to use Ant+ to the alphaHRV app and bluetooth to Garmin.
  • Clinically meaningful decisions about thresholds, readiness and durability/fatigue appear to be possible with this app.


Heart rate variability during dynamic exercise







19 comments:

  1. Thank you for the comparison to Kubios, looks quite promising. I also just started trying this out after Coach Wharton mentioned it yesterday. I'm curious to do my own visual comparison between this and Fatmaxxer; after reading this, I'm expecting it to be very close.

    ReplyDelete
  2. Thanks for your continuous work on DFA a1. I've done only 1 test with the alphaHRV app and think it reacts a bit faster (erratic) than Fatmaxxer. Could it be a processing capacity difference between Garmin and a cellphone ?
    Will do some more rides to validate this handy app vs Fatmaxxer Android.

    Thanks

    ReplyDelete
  3. You are welcome. Recall that Fatmaxxer uses 2 min windows but alphaHRV uses 200 beat windows - the results may be close but should not be identical. Stay with the strong Kalman filter for now. If you are prone to early aberrant beats (APCs), some of those will not get corrected (as shown above) and will lead to inappropriate a1 suppression. The developer is working on improving this behavior and the newer version should fix that. Keep us posted.

    ReplyDelete
  4. Thanks, I have been looking for a way to get on my headunit. With my Garmin Edge 530 it did crash about 2 hours in to the ride. I will try again without a course going at the same time.

    ReplyDelete
  5. This is a very demanding app cpu wise. On older Garmin units you may need to sacrifice some concurrent items. Perhaps try to cut back the recalculation to every 20s

    ReplyDelete
  6. Thank you for the great work. Will it support more Garmin devices such as FR935?

    ReplyDelete
    Replies
    1. I don't know, try sending a note to the developer.

      Delete
    2. Number of supported devices has grown in 0.9.6. but FR935 is not in the list (yet)
      https://apps.garmin.com/en-US/apps/1a69b10a-1d31-4afe-a32f-6a579ae20d9f#0

      Delete
  7. Hi Bruce, I wanted to reach out to you to ask if you would like to join us for an interview on our Fitter Radio podcast https://fitter.co.nz/listen-radio We have a weekly listenership of 25,000 downloads and thought that it would be a good way for us to discuss the concept of HRV-based aerobic threshold estimation, DFA alpha 1 as well as your most recent blog on Garmin DFA a1 data field. If you're interested feel free to reply to bevan@fitter.co.nz Cheers, Bevan

    ReplyDelete
  8. Bruce, you should correct the dates. You're still in 2021...
    Only noted this today.

    ReplyDelete
  9. Bruce,
    Fascinating work. Wondering your thoughts on incorporating morning HRV as a load measuring metric (a la HRV4Training) with DFA alpha 1 for in session intensity and Critical Power Work Balance (per Skiba) for total work availability (now available as Garmin app CPScience)
    - there is triangulation into the area of - how intense? for how long? in this session

    ReplyDelete
  10. I have a question regarding the connecting HR belt with Garmin EDGE head unit.
    I deleted the HR strap from EDGE and opened the data field. The unit found the belt and I got the values on the screen. But on other screens where the HR should also be visible it's not and I can't add the HR belt from sensors menu.
    Why is it so?

    ReplyDelete
  11. Great work. With the confirmation that the ant+ data is good does that mean you can use an older ant+ only Garmin strap/sensor or do you still have to use the Garmin dual?

    ReplyDelete
    Replies
    1. Good question - the heart rate monitor is important - I would trust the Polar H10, but can't say how the Garmin straps would do. The issue with ant+ on the Garmin side is the Garmin "receiver" (ie, the watch or head unit) that has a low sample rate (to save battery). The alphaHRV app uses it's own ant+ software protocol with the max sample rate allowed, bypassing the Garmin firmware.

      Delete
  12. I'm having trouble getting this connected. I've set the appropriate settings in my watch (Fenix 6 Pro), but the data field just keeps saying "Searching...". Is there something i am missing? I'm using Garmin HRM Pro. Is that not supported?

    ReplyDelete
    Replies
    1. Please reach out to the app developer, I'm not sure if that device is compatible.

      Delete
    2. I can't get it to work with my Garmin Edge 530 either using Polar H10.

      Delete
    3. Works if the settings are changed -
      If you want to use ANT+mode:

      - DELETE ANT+ sensor pairing with your device if previously exists (sensors menu in your device).

      - (Optional for dual BLE/Ant+ sensors) Connect the sensor to your device using BLE (sensors menu in your device). It will allow HR data to be included in fit file as native parameter.

      - Set ANT+ mode in alphaHRV settings (Garmin Connect app in your Android/iOS mobile or Garmin Express in PC/Mac). Use ANY (device number = 0), when you pair first time a new sensor and change to LAST or provide the sensor ID to keep same sensor for further connections.

      Delete