Thursday, July 22, 2021

DFA a1 as a biomarker of fatigue - article review

Over the past year we have focused on using an index of short term correlation properties of HRV, also known as DFA a1, as a useful tool in delineating exercise thresholds and controlling training intensity distribution.  But, can we also use the relationship of a1 and exercise intensity to examine other issues?  Since a1 reflects changes in autonomic balance and overall "organismic" demand, there could be a shift in behavior after a fatiguing endurance effort.  In a previous post I discussed a personal observation demonstrating a change in a1 behavior post HIT.   There was a reduction in the HRVT (the a1 based aerobic threshold) after high intensity intervals but not after a session of cycling for 1 hour near the aerobic threshold.  So far, no published work has looked at a1 as a biomarker of endurance exercise fatigue.  However, an article looking at DFA a1 behavior post Ultra-marathon has been released today, which we will review (with thanks to the hard work of the volunteer runners, and coauthors).  Although one should read the original article for details, I wanted to add some additional comments and potential use case scenarios in using a1 to help monitor long distance endurance physiology either retrospectively or in real time. 


The study - In brief, a group of experienced ultra marathon runners either ran for 6 hours or did usual daytime activity. Testing was done pre and immediately post run-(fatigue)/rest-(control).  

Two weeks after collecting baseline demographic and gas exchange data, 11 experienced ultramarathon runners were divided into two groups. Seven runners performed a simulated ultramarathon for 6 h (Fatigue group, FG) and four runners performed daily activity over a similar period (Control group, CG). Before (Pre) and after (Post) the ultramarathon or daily activity, DFA a1, heart rate (HR), running economy (RE) and countermovement jumps (CMJ) were measured while running on a treadmill at 3 m/s  

Running economy refers to the VO2 (oxygen consumption) per unit speed.  Since the speed was fixed Pre/Post, it was essentially the VO2.

In Pre versus Post comparisons, data showed a decline with large effect size in DFA a1 post intervention only for FG (Pre: 0.71, Post: 0.32; d = 1.34), with minor differences and small effect sizes in HR (d = 0.02) and RE (d = 0.21). CG showed only minor differences with small effect sizes in DFA a1 (d = 0.19), HR (d = 0.15), and RE (d = 0.31). CMJ vertical peak force showed fatigue‐induced decreases with large effect size in FG (d = 0.82) compared to CG (d = 0.02). At the completion of an ultramarathon, DFA a1 decreased with large effect size while running at low intensity compared to pre‐race values. DFA a1 may offer an opportunity for real‐time tracking of physiologic status in terms of monitoring for fatigue and possibly as an early warning signal of systemic perturbation.
  • Take home point - DFA a1 declined in almost all runners after the 6 hour run (A), but not in controls (B) while running at the same speed as in the before Ultra run test.  In the one case where it did not decline post Ultra marathon, it was already suppressed beforehand.

  • Note that the a1 values seen after the 6 hour run are quite low, below the .5, uncorrelated value, usually seen near the FTP/MLSS despite the running intensity being very mild (zone 1).

Singe subject example of Ultramarathon Pre Post changes

Although it is difficult to include single subject time varying data as a statistical entity, I would like to show an example of one athlete.  Below is a time varying plot of a1/HR while running on a treadmill at an easy pace - comparing results before and after the Ultra run.

  • The HR values are about the same pre/post 
  • DFA a1 is markedly suppressed (in red) after the Ultra run (while running at the same easy pace) compared to that before the 6 hour run (black).
  • The DFA a1 seen during the post Ultra run are below .5, corresponding to values seen near or above the FTP/MLSS/LT2

What about "conventional" HRV metrics (SD1 as the example):

  • SD1 seen running an an easy pace pre or post Ultra run are about the same.
  • For those of us who track SD1 or it's equivilant RMSSD, this is not a surprise.  Minimal exertion generally suppresses this index and one needs to observe a nadir on an incremental ramp to attempt aerobic threshold measurement.  We can see that it has little or no discrimination as a "fatigue" marker during exercise.


Practical implications of fatigue induced DFA a1 suppression.  Now that we have some solid data showing a1 behavior is affected after a fatiguing event, how can we leverage this information for training and fitness assessments?  Here are some theoretical use case scenarios employing DFA a1 for monitoring fatigue and ways to make use of this property (with statements from the article in quotation):

  • "Real‐time DFA a1 monitoring during endurance exercise could be used to inform an individual about current physiologic (fatigue) status and potential metabolic destabilization".  Although a1 behavior would probably be ignored as a safety measure during a race, knowledge of inappropriate decline during a long training session could prevent unwanted stress and signal when it's time to back off or stop entirely.  For example, in preparing for an ultra marathon or lengthy cycling endeavor, one will be performing very long endurance training sessions leading up to the event.  It certainly would be helpful to have some forewarning of impending metabolic destabilization (organ injury included).  Real time observation of a1 might be particularly valuable in this case.  If one were seeing marked a1 suppression in the anti-correlated range (<.5) at an usually easy pace, that could herald potential unwanted training stress.
  • "It is also possible that altered DFA a1 kinetics such as a delay of its decline over a given pace/distance following a training intervention could signify an improving performance status".  In this case we would be looking at improvements in DFA a1 after HIT or lengthy endurance sessions.  For instance, you note that the a1 values are about ..5 to .7 after a 3 hour cycling bout at an average power near the AeT.  After 6 weeks of "training" (your choice), you find that after the same average power, the DFA a1 is now tracking .8 to 1.0.  Although not a formally tested comparison metric, it would make sense that this improvement in a1 behavior is a reflection of improving ability to tolerate that load.  Moreover, that improvement may translate to less neuromuscular fatigue after the 3 hours, allowing for better sprints and pace.  As far as current news regarding the realm of fatigue recovery, I found it interesting that one of the professional cycling teams is now offering coaching services to all (for a fee of course).  One of their selling points is assessing effects of cycling fatigue/recovery with pre/post metrics:

With a Polar H10, power meter, smartphone and Fatmaxxer/Runalyze/HRV logger we can do something similar.  How?  By monitoring power and DFA a1 after a lengthy exercise session with an eye on achieving a more robust a1 correlation (higher a1 values at the same power/interval workout) as you train and become more fit. 

In addition, with the arrhythmia detection capability of Fatmaxxer and a Polar H10, potentially dangerous cardiac ectopy can also be spotted.

Others have recently written about this concept as a "durability" of performance over time and intensity:

Profiling physiological attributes is an important role for applied exercise physiologists working with endurance athletes. These attributes are typically assessed in well-rested athletes. However, as has been demonstrated in the literature and supported by field data presented here, the attributes measured during routine physiological-profiling assessments are not static, but change over time during prolonged exercise. If not accounted for, shifts in these physiological attributes during prolonged exercise have implications for the accuracy of their use in intensity regulation during prolonged training sessions or competitions, quantifying training adaptations, training-load programming and monitoring, and the prediction of exercise performance. In this review, we argue that current models used in the routine physiological profiling of endurance athletes do not account for these shifts. Therefore, applied exercise physiologists working with endurance athletes would benefit from development of physiological-profiling models that account for shifts in physiological-profiling variables during prolonged exercise and quantify the ‘durability’ of individual athletes, here defined as the time of onset and magnitude of deterioration in physiological-profiling characteristics over time during prolonged exercise.  

This situation has also been spoken of and clear to athletes in the past:

This quote dates well before his last days - when still in his youth.  Although I couldn't find a video of the interview, I remember watching Lance speak during his last TDF.  He described how, at this more advanced age, recovery was not what it was as in his younger days. Although his peak values were still on par, he needed to carefully manage his efforts to preserve himself over the course of 3 weeks of racing - hence recovery potential.  With DFA a1, we have an objective measure of recovery comparisons.

In session "readiness to train"

  • Seeing "suppressed" DFA a1 at what are usually zone 1 power levels.  This may indicate residual fatigue or over reaching.  Backing down power/pace in response to this information may help with proper recovery.  For instance, at a power/pace well into the zone 1 range, seeing a1 already at or below the .75 range is a warning sign that you are not fully recovered from previous training.  Although this remains to be proven, it would be a logical assumption of the above study.  For those of us who don't do resting HRV on a regular basis, this may serve a similar role - in other words, don't do an intensity session if the a1 is already inappropriately low at the beginning of exercise.



  • For the first time, published data indicates the potential usage of DFA a1 monitoring as a biomarker for endurance exercise fatigue.
  • For optimal usage, accurate DFA a1 calculation is essential.  Avoidance of excess artifact, using a precise HRM (Polar H10 or ECG) and appropriate software are all important factors to consider when comparing your results to those seen in the literature.
  • Potential use case scenarios include:
    • Forewarning of physiologic/metabolic derangement during long distance training in real time before disruption occurs while still exercising.  Although one can stop and do a series of CMJ or lab tests, it's hard to imagine any other modality indicating fatigue while the activity is in progress.  Although HR can "drift" over a long exercise session, it is not a reliable measure of endurance fatigue.
    • Assessment of fitness improvement/recovery kinetics after a new or novel training intervention.
    • As an indicator of already existing fatigue/over-reaching early in the course of an exercise session.  In other words - are you ready for an intense session or need a low intensity recovery?

Part 2 -
To train hard or not, that's the question 

Heart rate variability during dynamic exercise

Saturday, July 10, 2021

ECG Arrythmia/Artifact visualization - some tips

Now that we have an easy to implement way to capture HRV artifacts as ECG snippets with Fatmaxxer, I wanted to share a few tips and tricks for going through the visualization sequence quickly and with minimal cost.  If you have Kubios premium, just opening the .ecg file and viewing it is all we need:

But, for those who don't have Kubios premium, we need to rely on Excel.  The issue here is it can be time consuming to go through each segment one at a time to see what's there.  However, with the help of an Excel plugin which allows zoom and plot axis shifting, you can quickly scan through the ECG file and locate the areas in question for further analysis. 

Basic steps:

  • Download and install the plugin.
  • Follow the directions in the prior post on Fatmaxxer ECG strips to obtain the .ecg file
  • Open the file in Excel.
  • Add a new column as "A" - then create an auto number sequence starting at 1 
  • Select the A column as X and the yV column as Y - make a XY scatter plot

  • Change the points to lines but leave the axis alone
  • Make sure the ecg chart is actively selected
  • Open the EnginExcel tab - select Zoom chart
  • Change the X axis scale zoom - use the "Only X" choice for zooming the time, the Y should not need much zoom.

  • Start at the beginning of the file - by first zooming in, then moving to the Left (beginning) and then progressively move to the Right frame by frame.
  • When you find a wide QRS complex, stop and zoom further
  • Locate the segment for analysis
  • Look at the X axis and find the segment that corresponds to the number in column A
  • Match up the time stamp to the heart rate/part of the exercise session for informational purpose.  In this example, I would see what I was doing at 8:29.  Stress, high effort, dehydration, high temps etc should be noted.

  • To get a better look you can isolate the segment
  • Plot out just the segment (using the sampleNr and yV columns) and make sure the axis scale is correct
  • Figure out the beat type - generally a ventricular premature complex will be wide, have a different T wave morphology and not reset the atrial pacemaker.  Therefore the beat sequence will be preserved.  
  • The trace below is from the entire file - The first 2 beats (blue circle) represent the joining of segments, so the timing will be different - do not mistake that for an APC.

  • Below is an early beat, an atrial premature complex (APC) - it has normal morphology, a slightly abnormal appearing P wave and a short PR interval (red circle).  It generally should reset the pacemaker and the pause after would not be compensatory - but it occasionally is.  Although this example is atypical, the narrow complex is fairly diagnostic for an APC.

  • Continue scanning until the end:

Comments on ECG reading from a single lead:

Recently I noted some wide complex beats in my own ECG strips and could not decide whether they were atrial or ventricular in origin.  Although atrial beats are usually of normal narrow appearance, occasionally they will show as wide complexes due to conduction system blockade (usually right bundle branch block).  The reason behind this is that the conduction fibers themselves have a non uniform refractory period, so the ventricle is not excited in the usual pattern.  There are a set of clues in making the decision on abnormal beat origin, of which some rely on the presence of beat timing (VPC does not reset the pacer but APC does).  But, some APCs do not reset the pacer, so we usually look at other leads to help decide.  Without those other leads, it can be problematic.

I asked for several opinions on this from both cardiology and ER docs (they did not agree).

The cardiologist felt this was an APC with conduction delay, followed by a VPC.  However, the first APC did not reset the pacer.

And this was felt to be a VPC.

The reason is based purely on the timing issue:

Bottom line - interpretation of wide complex beats using a single lead depends on timing, P wave visualization and can be difficult.  If you see many of these - get it checked out by an expert.


  • An easy to use Excel plugin makes scanning the .ecg file quick and easy.
  • For most people, some APC activity is to be expected.
  • Having many VPC/wide complexes during rest or exertion may indicate a problem.
  • Reading a single lead tracing can be a challenge and occasionally it is impossible to separate a VPC from an APC with a conduction block.
  • If you are concerned, bringing this data to your physicians attention is a reasonable idea.

Heart rate variability during dynamic exercise