Saturday, January 2, 2021

DFA a1 and exercise intensity FAQ

With the recent interest in using the DFA a1 HRV index to determine aerobic thresholds, follow exercise intensity and the use of real time DFA a1 output from HRV Logger, I've decided to put together a "frequently asked questions" list.  This is by no means totally inclusive and will be updated on a regular basis as new questions (and answers) come up.  Here we go...

I have an hour to listen to something but can't read the blog right now.

What is DFA a1?

  • Simply put, it's an index of heart rate beat to beat, fractal related self similarity.  Although your heart rate may be 60 bpm, the beats are not occurring exactly every 1.000 seconds.  The pattern of self similarity changes as exercise intensity rises, from values well above 1, moving down to .75 near the aerobic threshold and dropping even further above this exercise intensity.  See the articles below for details.
Can it be used to determine the aerobic threshold?
  • Yes.  In fact we have 2 articles describing this process.
  • https://www.frontiersin.org/articles/10.3389/fphys.2020.550572/full
  • https://www.frontiersin.org/articles/10.3389/fphys.2020.596567/full
  • Below is a YouTube video I did for a conference going over the advantages of DFA a1 over other indexes
 
 
 

 

How accurate is it?

  • Before answering, we need to think about how accurate the comparison "gold standards" methods are.  As discussed in the articles above, there are real issues in both lactate and gas exchange tests, making them subject to various errors and inconsistencies.  Some gas exchange results are so confusing that they are not interpretable.  Machine based gas exchange results are not always accurate.  From the limited study data so far, it seems the DFA a1 is a reasonable surrogate for the AT.  Below is the Bland Altman analysis and regression plot from our validation study:


 
  • As you can see, some folks had more or less agreement with the gas exchange AT, but for the most part the differences were small (several bpm)

What sports can this be applied to?

  • This is a very valid question.  So far, only running and cycling have been well explored.  Other activities such as those using upper and lower extremities (xc skiing, kayaking, rowing) may not follow the same relationship with the AT. 

What can affect the numbers I get?

  • A very wide range of factors.  Stress, caffeine, caffeine withdrawal, food, fasting and over-training are some of the factors before we even process the data.  Preprocessing algorithms, software settings are also critical.  Kubios may give different results from a python based method. We will need to do formal comparison testing between Kubios and python methods eventually.

Do I need clean, artifact free data?

  • A very important item that will affect the DFA a1 is artifact in the RR series.  Missed beat artifact is the most common, and if above 3% could, but if above 6% will affect the values you get.  A single APC may also dramatically drop the DFA a1 for that window of measurement.  Correction methods help with this but are now perfect.  One of the strengths of our Frontiers study was that we used ECG data with almost no artifact.  YMMV using a chest belt with artifact.
  • We recently had an article accepted at the journal "Sensors".  Below is the abstract from that study:
 
Recent study points to the value of a non-linear heart rate variability (HRV) biomarker using detrended fluctuation analysis (DFA a1) for aerobic threshold determination (HRVT). Significance of recording artefact, correction methods and device bias on DFA a1 during exercise and HRVT is unclear. Gas exchange and HRV data were obtained from 17 participants during an incremental treadmill run using both ECG and Polar H7 as recording devices. First, artefacts were randomly placed in the ECG time series to equal 1, 3 and 6% missed beats with correction by Kubios software’s automatic and medium threshold method. Based on linear regression, Bland Altman analysis and Wilcoxon paired testing, there was bias present with increasing artefact quantity. Regardless of artefact correction method, 1 to 3% missed beat artefact introduced small but discernible bias in raw DFA a1 measurements. At 6% artefact using medium correction, proportional bias was found (maximum 19%). Despite this bias, the mean HRVT determination was within 1 bpm across all artefact levels and correction modalities. Second, the HRVT ascertained from synchronous ECG vs. Polar H7 recordings did show an average bias of minus 4 bpm. Polar H7 results suggest that device related bias is possible but in the reverse direction as artefact related bias.

 

So what does this mean on a practical basis?  Anything with >6% artifact in the area of interest should not be trusted.  Since both the Kubios threshold correction method and HRV logger use similar techniques, 3% or less artifact containing data will provide reasonable HRVT accuracy.  There is also a chance that 3-5% artifact containing data series will be fine, but you may want to re test yourself.  The effect of missed beat artifact on DFA a1 is to artificially raise the computed value at low DFA a1 ranges (not high ranges).  For example, if the DFA a1 was .5 with no artifact, after adding 6% missed beats with correction (by Kubios), the software will output .65 +-.  

Here is a look at how that works out on a Bland Altman assessment


The solid line is the "average" difference between methods, notice how this process is dependent on what DFA a1 actually is.  There is minimal "bias" between DFA a1 of 1 and .5 which is important for the HRVT.  However, values below .5 are very much altered.

  • Also see below under recording devices.

What artifact correction settings do you recommend?

  • If you are using the Kubios paid premium version, use the "auto" method.  Free version Kubios uses the threshold method (similar to HRV logger).  The medium correction setting is the default and should work well (similar to the 20% setting in Logger).  The exception is with an APC where a sudden drop is seen.  Using the extra strong filter setting (or the "work out mode" in Logger) will filter out the APC but can also filter some physiologic beat to beat variation.  Get a feel if you exhibit frequent APC activity, and if so, use the more aggressive settings.

Does recording device matter?

  • This is something else we are looking at.  The above validation study was done with a research grade ECG.  It is very possible that a chest belt device will detect R peaks differently as well as be affected by preprocessing issues.  Interference with either chest wall or diaphragm related activity can change the ECG waveform.  Disturbance of the pattern of self similarity would then occur after the introduction of this type of distortion.  However, the Polar H10 results appear very close to accurate waveform ECG derived values.
  • In the Sensors study, we found that the Polar H7 "measures" DFA a1 as slightly lower values.  This is in the opposite direction as what missed beat correction induces, which is actually quite convenient!  The end result of a Polar H7 recording with 3-5% missed beat correction may yield values that are very close to those of an ECG.  Below is a figure from our article that shows this very nicely.  The Polar reads lower than the ECG, but the 6% artifact recording reads high - making for a "self correcting" effect.  If you had a Polar RR series with no artifact, yes, you might have some bias.  We are continuing to look into this.

    Time-varying analysis (window width: 120s, grid interval: 5s), DFA a1 for matched time series containing no artefact in one representative participant, ECG (solid triangle), Polar H7 (open circle), ECG 6% MC (open triangle).

     

How do I set up an aerobic threshold test scenario with HRV Logger?

  • I've devoted many posts on doing this in Kubios but lets look at a simple method in HRV Logger. Warm up 15 to 20 minutes then do 6 minute constant load efforts.  Make sure you line up the first effort with an even time number in the Logger.  Since the Logger spits out a value every 2 minutes, it's helpful to have the time under effort match up.  As an example, start the Logger, warm up 20 minutes then at 20 min exactly (on the Logger), start your first interval at a very easy level.  Throw out the first value after the interval start (it's not at steady state yet) but the values at 24 and 26 minutes will be valid.  At 26 minutes, boost your power or speed by a notch (still easy) and measure at 30 and 32 minutes (remember that 28 was non steady state).  Keep this progression up until you pass through .7 to .8 DFA a1.  That would have been your AT related intensity.  Do one more effort at the next stage higher to confirm the DFA a1 is indeed below .7. 

How do I reproduce your published study protocol?

  • Here it is:
  • The following procedure was used to indicate at what level of running intensity (as VO2 or HR) the DFA a1 would cross a value of .75: DFA a1 was calculated from the incremental exercise test RR series using 2 minute time windows with a recalculation every 5 seconds throughout the test. Two minute time windowing was chosen based on the reasoning of Chen et al. (2002). The rolling time window measurement was used to better delineate rapid changes in the DFA a1 index over the course of the test. Each DFA a1 value is based on the RR series 1 minute pre and 1 minute post the designated time stamp. For example, at a time of 10 minutes into the testing, the DFA a1 is calculated from the 2 minute window starting from minute 9 and ending at minute 11 and labeled as the DFA a1 at 10 minutes. Based on a rolling time recalculation every 5 seconds, the next data point would occur at 10:05 minutes (start 9:05 minutes and end 11:05 minutes).
    Plotting of DFA a1 vs time was then performed. Inspection of the DFA a1 relationship with time generally showed a reverse sigmoidal curve with a stable area above 1.0 at low work rates, a rapid, near linear drop reaching below .5 at higher intensity, then flattening without major change. A linear regression was done on the subset of data consisting of the rapid near linear decline from values near 1.0 (correlated) to approximately .5 (uncorrelated). The time of DFA a1 reaching .75 was calculated based on the linear regression equation from that straight section (Figure 1b). The time of DFA a1 reaching .75 was then converted to VO2 using the VO2 vs time relation, resulting in the VO2 at which DFA a1 equaled .75 (HRVT). A similar analysis was done for the HR reached at a DFA a1 of .75. First, ECG data from each 2 minute rolling window was used to plot the average HR and DFA a1. The HR at which DFA a1 equaled .75 was found using the same technique as above, a linear regression through the rapid change section of DFA a1 values of 1.0 to below .5, with a subsequent equation for HR and DFA a1 (Figure 1c). Using a fixed variable of DFA a1 equals .75, the resulting HR was obtained. The HR at DFA a1 .75 (based on ECG data) was then compared to the HR at VT1 GAS obtained from the metabolic cart data (based on the Polar H7).
  • Note should be made of the difference between Kubios "time varying" timestamps and the Logger.  In Kubios, a given time varying window timestamp is centered in the middle of the window (bold print above) whereas in the Logger the timestamp is at the end of the window.

How do I make sure I'm really doing a recovery ride on my rest day?

  • This is an ideal scenario for the Logger in real time.  Just watch the live read out and keep DFA a1 above .8 or even .9 (yes .75 is the cutoff, but there is individual variation and a small buffer is advised).  A single value that falls below .75 then normalizes where it started again was probably due to an APC.

Are my values going to be the same day to day?

  • Probably not.  Although they may be close, it's normal and expected to have some shifting in heart rate or power on a day to day basis.  This would be the case with gas exchange or lactate as well.  As stated above, other factors will change the index result, especially heat, skin temp and humidity.

Can the intensity of exercise where DFA a1 = .75 be used as a way of tracking fitness changes after training?

  • There is nothing published on this as of yet.  It is something we are currently looking at and I will update this when I am able.

I'm on beta blocker therapy, will this change the DFA a1 to intensity relation?

 

Since the DFA a1 drops with intensity, can I use this for my zone 3 cutoff (respiratory compensation point, MLSS, anaerobic threshold)?

  • No, the DFA a1 has great dynamic range but it's centered around the VT1/AT/LT1.  In some individuals, just a bit higher intensity above the AT will suppress the results to a nadir that will not drop any further.  Having a very low DFA a1 generally indicates high intensity but not with precision.  Better non invasive alternatives exist such as the FTP or muscle O2 desaturation.

How do I match up timestamps in Kubios and HRV Logger?

The time stamping is tricky.
The times are all different in each "method".
Logger - time is at the end of a 2 min window - so a timestamp of 2 minutes is from 0 to 2 min elapsed.
Kubios free - the time is from the beginning of a 2 min window - so a timestamp of 2 min is from 2 min to 4 min elapsed
Kubios premium time varying download (enclosed) - the timestamp is in the window center - so a timestamp of 1 min is from zero to 2 min elapsed.
Therefore the logger will be different from Kubios either way.

Pre correction


Make sure you shift Logger 1 minute forward (the Logger at T=2 min equals the time varying Kubios at 1 min)

Post correction


 

Heart rate variability during dynamic exercise


 

2 comments:

  1. What would be a good power between “steps” using HRV logger and the six minute long steps? 10 watts? 20 watts?

    ReplyDelete
  2. I think 20 watts should be fine. Once you get your approximate threshold, you can retest by just doing 3 stages, AT-20, AT and AT+20w.

    ReplyDelete