Thursday, December 28, 2023

DFA a1 2023 year in review

As the last days of 2023 approach, I would like to briefly point out several key findings published over the past 12 months.  This is not meant to be totally definitive and I may have missed some good studies.  However, from my viewpoint, the following subjects stand out:

  • The value of combining surrogate thresholds.
  • The need for better agreement of the HRVT1 with VT1/LT1.
  • The use of DFA a1 for assessment and monitoring of fatigue.
  • The day to day reliability of DFA a1 thresholds and behavior.
  • Effects of ramp slope on a1 threshold behavior

Ramp Slope - does it matter?

One of the key observations concerning a1 threshold determination was that ramp slope (between 15 and 45w/min) does not affect HRVT1/2.  Before this study was done, I would have bet that ramp slope would affect threshold determination and certainly 45w/min should be a dealbreaker.  After all, 45w/min is a massive 90 watt rise over the 2 minute DFA a1 measurement window!  Instead, we found that the a1:HR profile was nearly identical for the different ramp slope tests:

This is the first (and only) published evaluation of (any) ramp slope on HRV thresholds.  Many thanks to my co authors including Juan Murias and Pablo Fleitas- Paniagua for making this possible.

Day to day reliability of a1 testing

In other words, does a1 behave similarly on a day to day repetitive basis, either during an exercise ramp or just (short term) constant load testing?

This article done by Bas VanHooren and Thomas Gronwald looked at a1 during both easy and post ramp exercise:

  • It was found that both HR and a1 were very similar on repeated testing (2 different days), both on the warm-up and after the exhaustive ramp.

Although not specifically designed as a day to day reliability study, the ramp slope paper essentially indicated the same result - that a1 thresholds will be similar for a given individual on repeat testing (in this case the slope was different on the 3 test days):

The need for better agreement between the HRVT1 and VT1/LT1/GET

This leads us to the next issue, why do some studies show tight agreement of the HRVT1 to the LT1/VT1/GET and some don't?  As part of the triple ramp study, there was a large discrepancy between the VT1 (or GET) with the HRVT1 (yellow).  However, the HRVT2 relationship/agreement with the RCP (or maximal metabolic steady state) continued to be excellent (red):

On the other hand, this study by Bas VanHooren showed good VT1:HRVT1 agreement:


  • One of my priority projects for the next year is to work on this issue and explore the nuances of the HRVT1 to VT1/GET relationship, hopefully improving the HRV threshold agreement.  Which leads us to the next section...

Combining surrogate threshold markers for better agreement to gold standards:

I was involved with 2 papers that approached this hypothesis, which I explained in depth previously.  In short - It was proposed that combining 2 disparate surrogate markers (HRVT plus something else) would produce better agreement with both the VT1/GET and the RCP.  The "something else" was either muscle O2 desaturation breakpoints for RCP localization and/or EDRT (ECG derived respiratory rate) for both VT1/GET and RCP.  Both hypotheses were shown to be valid and I was also happy to see that the Kubios team has picked up on that idea as seen in this abstract (from ECSS 2023):

My belief is that multiple surrogates will be better than single measures for threshold determination, and hope to see more about this over the next year.

The use of DFA a1 for assessment and monitoring of fatigue:

Last but not least of the DFA a1 findings for 2023, was the paper with Bas VanHooren using two ramps back to back separated by a very short active rest, looking at threshold behavior.

Method summary - a ramp to exhaustion, active rest then another ramp - Whew!

As anticipated, a1 was markedly suppressed on the second ramp despite similar gas kinetics:

The practical implications of this study are twofold:

  • Don't do a threshold ramp when fatigued/ill or post heavy/severe exercise.  In fact, given the previously noted finding after an ultramarathon, even low intensity exercise, if performed long enough, may suppress the a1.
  • Second - a1 shows great promise as an indicator of "autonomic" fatigue/durability.  One may see variable degrees of a1 "drift" depending on individual fitness characteristics, length and intensity of exercise.  Therefore, monitoring of either a1 during longer training sessions is of potential use as far as zone distribution goes.  For instance, does the lower a1 post HIT "count" for a higher training zone than what the wattage shows?  

Honorable mentions (in no order):

Continued improvement of the alphaHRV app for Garmin devices.  Despite major hardware and software limitations, the developers have achieved good a1 agreement to Kubios standards.

Web app enhancements to both AIEndurance and Runalyze.

Finally, my thanks to all co-authors over the past years for making many of these studies possible.

Happy New Year!

Tuesday, August 8, 2023

Ramp slope and HRV thresholds, Physiological Reports 2023

Have you ever wondered what is the best type of cycling incremental ramp (RI) to estimate DFA a1 based HRVT's?  Is it better to use a ramp with a steep slope (rapid rise in cycling power) or one with a shallow slope (slow rise in power)?  Or something in between? Although I have done some limited personal testing myself (N=1), that's a poor substitute for a formal investigation.  A search of the literature reveals very little about the effects of ramp slope on HRV thresholds (actually, it reveals nothing).  

Is this an important question to resolve, or just an esoteric research data point?  In my opinion, it is of paramount importance since without this knowledge we may get inaccurate results and can't reliably compare studies looking at HRVT agreement to gold standards.  For example - research group A looks at HRVT agreement in teen women using a ramp slope of 45 watts/min and finds a major discrepancy with historical values previously seen.  However, there has never been any HRVT assessment at that ramp slope, so we don't know if the issue is the ramp increment being so high, or is it truly something related to females (estrogen status for example).  Another example would be a meta analysis done on all HRVT trials (with different RI slopes) - if slope does matter, we can't hope to get an accurate analysis!  Currently, the range of published works describing DFA a1 during RI have used slopes ranging from about 7 to 25 w/min. 

Since the following study evaluating ramp slope was an offshoot of a topic near and dear to my interests (muscle O2 desats and HRV thresholds), ideally it would have been nice to have the "parent" article published before this came out.  As I mentioned in the last post, some journals take longer to get through their queue of acceptances to finally be published.  In our case, the "sequel" came out before the first book of the series.  Having said that, I can give a small hint of what the original study goal was by way of this abstract from ECSS:

  • The full paper is accepted and should appear shortly in a popular sports physiology journal
  • The accompanying post to that has been written for quite some time. 😁
  • And yes, the combo of HRVT2 and HHb BP to assess the critical intensity is article #1 (from here)
  • Would I like to say more - definitely - but it's not fair to the journal, so this is it.
  • I would again like to thank Juan Murias, the co authors and Pablo Fleitas-Paniagua for making this possible.  The complexity of managing such a large pool of participants performing three different ramps was not trivial.  Therefore, although future studies confirming these results would certainly be welcome, from a practical standpoint, they may not be quickly forthcoming.

Fortuitously, the data used for the prior HRVT/HHb BP Combo project was part of a larger set of material encompassing the same group of individuals performing 3 different RIs with slopes of 15, 30 and 45 w/min.  Therefore, we have the ideal framework for evaluating the effects of RI slope on HRVT behavior.  This has now been formally published in Physiological Reports.


Some highlights of the study are shown below:

Method highlights - the Polar H10 (with ECG guidance for belt placement) was used to record RRs with Fatmaxxer.  We used the standard 2 minute measurement windows with every 5 sec recalculation to plot the a1 HRVT1 and 2.

This is an example of the a1 behavior vs HR during the 3 ramps in one individual:

  • The left plot is the 30 w/min ramp, and on the right are all three ramps in the same individual.  As noted, they are fairly close.

Correlation and Bland Altman agreement between ramps were all excellent:

  • There seems to be no pattern of the degree of correlation or agreement with rising slope difference.  In other words, the 15 vs 45 watt ramps were no worse off on a statistical basis than the 15 to 30 or 30 to 45 watt ramps.

Finally, a "longitudinal" look of HRVT HR and VO2 through the ramps per participant:

  • T testing between all paired groups (15 vs 30, 30 vs 45 and 15 vs 45 watt) showed no differences.

Did we have a pre study hypothesis?

We were requested by the reviewers to add a hypothesis since I did not have one in the original draft.  My reasoning for not having one was based on the definition of hypothesis - "a supposition or proposed explanation made on the basis of limited evidence as a starting point for further investigation".  The problem here is that there was no "limited evidence or starting point" to make the educated guess.  In the end, we added something, but truly, the question of RI slope and HRV was a conundrum from the start.

On one hand, I was not surprised that the 15 vs 30 watt ramps were equivalent since they are relatively close, but was very surprised that the 45 w/min ramp was similar to both.  The 45 w/min ramp represents a 90 watt change over the 2 minute DFA a1 measurement window.  We discuss the significance of that in detail in the discussion:

"Over the past 20 years numerous studies evaluating DFA a1 behavior during dynamic exercise have been performed (17-18, 21-26, 30-32).  However, despite showing potential as a marker defining exercise thresholds through RI testing, there has been no consensus as to what type of ramp protocol is optimal or desirable.  Therefore, the intent of this study was to assess the behavior of DFA a1 related HRVTs during cycling RI with varying slopes.  Literature has shown that fast ramps tend to have the greatest degree of discordance between measurements such as cycling power and corresponding gas exchange derived thresholds (12-15), unless a correction is used to account for the V̇O2 mean response time and slow component (3,14).  In the context of an established ANS marker such as DFA a1, conjecture as to the effect of ramp slope is complex. A slower incremental rise in work rate resulting in a longer ramp may lead to fatigue related effects (45-46) that could result in biased threshold estimation.  On the other hand, a rapid intensity rise may not be able to truly describe an index encompassing a measuring window of 2 minutes.  For example, over the 2-minute DFA a1 measuring window, a full 90 watts of external load increase will have occurred if the RI test was performed at a 45 W·min-1 slope.  Whether or not DFA a1 values done under such non steady state circumstances produce comparable results to those done under a more gradual rise in load is unclear.

Since DFA a1 calculations need about a 2-minute measurement window for validity (31,39-40), fast ramps lasting only several minutes may also present a challenge simply on the basis of limited available data points.  In addition, even though the ANS response is believed to be rapid in relation to the various regulation factors (47-49), there could be a lag between these inputs and their effect on DFA a1 behavior during fast ramps.  Many initial studies measured DFA a1 toward the end of a “step” interval of varying length but always longer than 2-minute steps (30-32).  When DFA a1 was first proposed as a surrogate marker for ventilatory threshold determination (21), a new calculation technique was used, based on the “time varying” method available in Kubios HRV software.  Time varying refers to the index being recalculated continually every 5 seconds throughout the exercise period.  Before this technique, the index was determined either at the end of each interval step or at periodic, non-overlapping points during the exercise test.  Since we are now able to easily measure DFA a1 on a more granular level over the course of increasing load, the question remains whether absolute ramp slope matters for both index behavior and HRVT determination. 

The results of this study show that the V̇O2 or HR reached at both HRVT1 and HRVT2 is relatively independent of the ramp slope during incremental exercise testing (for those slopes used in this report).  There was excellent correlation between all three ramp protocols using ICC3,1 with values between 0.88 and 0.93 and no mean differences across all groups with ANOVA.  Pearson’s r was also highly correlated between paired ramp groups with values between 0.84 to 0.95 (Table 2).  Bland Altman analysis showed small mean differences between ramp slopes (Table 2 and Figure 3). There were no statistical differences seen between any ramp slope series looking at either HR or V̇O2 according to paired t testing.  Importantly, there was no major discrepancy in correlation/agreement or t testing in comparing the 15 to the 45 W·min-1 ramp slopes, despite the three-fold difference in power output rate increment.  The observation that DFA a1 is capable of rapidly shifting during the 45 W·min-1 ramp to match that of the 15 W·min-1 ramp is a novel finding of interest.  Like the HR response to RI testing (11-12), there appears to be a prompt matching of “organismic“ demand as represented by DFA a1, to the external exercise load.  This makes sense as both HR and HRV responses are mediated by related and/or linked ANS, central nervous system (CNS) centers, vagal output and effects on the atrial pacemaker cells (16,27,47-49).  However, it has been unclear whether an HRV measurement window encompassing a relatively large span of differing metabolic input would yield usable results.  This similarity in DFA a1 response across disparate ramp slopes is illustrated in a detailed plot of HR vs DFA a1 of a typical participant during the 3 RI tests (Figure 1B).  The pattern of DFA a1 decline as HR rises is similar across the differing ramp slopes.  Since the 45 W·min-1 group had similar agreement to that of the 15 or 30 W·min-1 groups, it seems that DFA a1 measurement of a linear increasing load leads to comparable HR or V̇O2 correspondence no matter the rate of rise (within tested limits).  This has major practical significance as prior and possibly future studies evaluating DFA a1 behavior may employ RI with different slopes.  Since it appears the RI slope does not affect the resultant HRVTs, these studies can be more easily compared and implemented.

How well did HRVT1/2 agree with gas exchange?

This group consisted of the same participants as the Combo NIRS HHb BP/HRVT2 study that appears in JSCR.  In that assessment, the HRVT2 agreed quite well with the RCP/VT2 (with individual variation of course).  However, the group HRVT1 did not agree well with the GET/VT1, with about a 20 bpm differential.  This is discussed in the limitations section in more detail.  Going forward, we will need to be more aware of potential subpopulation differences, device bias and even belt position differences.  Although the significance of an a1 of 0.5 has been discussed (it represents an uncorrelated, random beat to beat state), the 0.75 is a more empirical guesstimate.  Even though the bias was large, the correlation was still reasonable and similar to reported values:

This is a plot of the 30w/min ramp GET/VT1 vs HRVT1 HR from the same ramp.

Hopefully, as time goes on and we better understand the many factors involved with measuring a1 during exercise, this will become clearer.

A word about cycling power thresholds and RI slope

Although we discuss this briefly in the discussion, there is a wealth of data looking at the disconnect between RI gas exchange derived thresholds and their equivalence to constant power interval data.  For an excellent review and discussion see this (by Dr Murias's team).  However, an older study by Weston also provides some information on how thresholds will be skewed, especially with steep ramps in particular:

  • The cycling power difference between the 10 and 50 w/min ramp was 20 watts for VT1 and 60 watts for VT2.


Expected day to day variability

There is now solid evidence that ramp slope has minimal effect on the HR/VO2 of the HRVTs - but what about day-to-day repeatability/reliability of a1 in general?  Recent study data point to minimal shifts in a1


  • Therefore, one may see some "shifting" of the HRVT1/2 when performing different ramp slopes just based on day to day variability (but not necessarily from the slope change).


Summary and Conclusions

  • DFA a1 related HRVT1 and HRVT2 (as HR or VO2) are not affected by ramp slope (at least through the range of 15 to 45 w/min).  
  • Cycling power VT1/VT2 threshold is affected by ramp slope.  If you are interested in an accurate assessment of power at a particular threshold, try to use shallow ramps.  However, there is no difference in HR thresholds with ramp slope.
  • Since HRVTs behave as HR/VO2 gas exchange metrics during RI protocols, the power associated with HRVTs will be slope dependent - therefore, shallow slopes are preferred for threshold power agreement.
  • Although there appears to be no systematic bias in HRVT HR with slope, some ramp to ramp variation is expected.
  • Whether your own particular HRVT agrees perfectly with gold standard (or does not), one should be reassured that ramp slope is not a factor (for HR/VO2).
  • Cross comparisons with prior and future HRVT RI studies can now be reliably compared without slope as a concern.

Heart rate variability during dynamic exercise