For an excellent review of HRV during exercise and recovery see this paper.
The statistical analysis of beat to beat variability can be done in many different ways.
Time-domain indices of HRV quantify the amount of variability in measurements of the interbeat interval (IBI), which is the time period between successive heartbeats (see Table Table1). These values may be expressed in original units or as the natural logarithm (Ln) of original units to achieve a more normal distribution.
Frequency-domain measurements estimate the distribution of absolute or relative power into four frequency bands. The Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology (1996) divided heart rate (HR) oscillations into ultra-low-frequency (ULF), very-low-frequency (VLF), low-frequency (LF), and high-frequency (HF) bands.
Power is the signal energy found within a frequency band. Frequency-domain measurements can be expressed in absolute or relative power. Absolute power is calculated as ms squared divided by cycles per second (ms2/Hz). Relative power is estimated as the percentage of total HRV power or in normal units (nu), which divides the absolute power for a specific frequency band by the summed absolute power of the LF and HF bands. This allows us to directly compare the frequency-domain measurements of two clients despite wide variation in specific band power and total power among healthy, age-matched individuals
The LF band (0.04–0.15 Hz) is comprised of rhythms with periods between 7 and 25 s and is affected by breathing from ~3 to 9 bpm. Within a 5 min sample, there are 12–45 complete periods of oscillation (9). The HF or respiratory band (0.15–0.40 Hz) is influenced by breathing from 9 to 24 bpm (11). The ratio of LF to HF power (LF/HF ratio) may estimate the ratio between sympathetic nervous system (SNS) and parasympathetic nervous system (PNS) activity under controlled conditions. Total power is the sum of the energy in the ULF, VLF, LF, and HF bands for 24 h and the VLF, LF, and HF bands for short-term recordings
The HF band is usually associated with the parasympathetic (PS) and the LH band with the sympathetic nervous system (SNS) . At rest the PS predominates and as exercise intensity rises it is withdrawn and the SNS is activated. Therefore we would expect the LF band to be useful in more intense exercise domains. However, it turns out that the respiratory rate effects on HRV occur in the HF band which would be expected to play a larger role as exercise intensity rises. In recent years it has also been noted that the parasympathetic system affects both the LF and HF bands making interpretation more complex.
Non-linear measurements (Table3) allow us to quantify the unpredictability of a time series:
Nonlinear data significance (SD1):
Non-linearity means that a relationship between variables cannot be plotted as a straight line. Non-linear measurements index the unpredictability of a time series, which results from the complexity of the mechanisms that regulate HRV. Non-linear indices correlate with specific frequency- and time-domain measurements when they are generated by the same processes.
A Poincaré plot (return map) is graphed by plotting every R–R interval against the prior interval, creating a scatter plot. Poincaré plot analysis allows researchers to visually search for patterns buried within a time series (a sequence of values from successive measurements). Unlike frequency-domain measurements, Poincaré plot analysis is insensitive to changes in trends in the R–R intervals.
We can analyze a Poincaré plot by fitting an ellipse (curve which resembles a squashed circle) to the plotted points. After fitting the ellipse, we can derive three non-linear measurements, S, SD1, and SD2. The area of the ellipse which represents total HRV (S) correlates with baroreflex sensitivity (BRS), LF and HF power, and RMSSD.
The standard deviation (hence SD) of the distance of each point from the y = x axis (SD1), specifies the ellipse’s width. SD1 measures short-term HRV in ms and correlates with baroreflex sensitivity (BRS), which is the change in IBI duration per unit change in BP, and HF power. The RMSSD is identical to the non-linear metric SD1, which reflects short-term HRV (37). SD1 predicts diastolic BP, HR Max − HR Min, RMSSD, pNN50, SDNN, and power in the LF and HF bands, and total power during 5 min recordings.
Background on VT1 and VT2:
HRV during dynamic exercise has been explored as a means of determining the first and second ventilatory thresholds. Let's explore the basic science behind each one. Some of the following figures were taken from a sports physiology textbook by Plowman.
As exercise load increases, aerobic glucose metabolism eventually is unable to meet all the energy requirements to continue. There comes a point when lactate begins to be accumulated from incomplete glucose breakdown.
This causes an increase in CO2 through the following:
The VT1 and VT2:
The author makes it very clear (with multiple reasons why) that the VT1 and the LT1 as well as the VT2 and the LT2 are not caused by the same process:
With the conclusions:
The reason I have gone to some length in getting proper definitions spelled out is that although one can use lactate associated breakpoints LT1, LT2 during a cycling ramp to get a rough idea about the VT1, VT2, they are not necessarily the same. To actually get a true VT1 and VT2, you will need to do a gas exchange study in a lab. Here is a figure from the study looking at equivalency of VT vs LT:
So although there is some correlation, it is not precise. For instance the highest value for LT1 in the figure is near 275w with the true VT1 is about 230w.
In the rest of the post, for ballpark purposes, I will use the MLSS power as a VT2 surrogate, but it is a only a similarity.
Back in 2007, Cottin et al looked at a derived index (HF power x HF frequency peak) in relation to the VT1 and VT2. Typical ramp testing was done in moderately trained soccer players with gas exchange and HRV parameters. They were particularly interested in the HF band since respiratory HRV occurs there. Apparently, the RSA (respiratory sinus arrhythmia) has 2 main components:
Furthermore, spectral analysis of beat-to-beat RR series allows the computation of the two main components of the -1.The first spectral component of the RSA is its magnitude. It is given by the spectral power (HF) contained within the high frequency band. HF is computed as the sum of the power spectral density ranging from 0.15 Hz to fmax during exercise. The fmax is the maximal frequency induced by the sampling scale of the RR signal.2.The second spectral component of the RSA is the frequency of the HF peak (fHF). It has been shown that fHF closely corresponded to the breathing frequency (BF).
HF thresholds assessment: The present study performed two VTs assessment methods from RSA:1.From fHF: successive values of fHF were averaged to provide a data point for each 20-s period synchronous with the ventilatory data. HF thresholds were detected from the curve of fHF plotted vs. the work rate by an independent investigator. The first HF threshold (TRSA1) corresponded to the last point before a first increase in fHF. The second HF threshold (TRSA2) corre-sponded to a second nonlinear increase in fHF.2.From HF·fHF: successive values of HF·fHF were averaged to provide a data point for each 20-s period synchronous with the ventilatory data. Therefore, HF thresholds were detected from the curve of HF·fHF plotted vs. the work rate by an independent investigator. The first HF threshold (HFT1) corresponded to the first nonlinear increase in HF·fHF after it had reached a minimum. The second HF threshold (HFT2) corresponded to a second nonlinear increase in HF·fHF
They showed a reasonable correlation of their index with both thresholds:
They also found a very steady linear relation of the HF peak frequency with running speed:
The study conclusion:
The main result of the present study is that it is possible to assess the ventilatory thresholds from HRV components during an incremental exhaustive running test. However, the VTs assessment was possible from the HF·fHF index, whereas it could not be determined from fHF only.In addition they did comment that it may be possible to estimate breakpoints using the HF frequency only in cycle ramp tests due to the difference in breathing rate entrainment in running vs cycling:
BF(breathing freq) data occurring at VTs presented in the present study (Table3) seemed to be higher than the corresponding examples of BF data obtained at VTs during cycle ergometer tests [3,5,42]. This observation confirms the different relationships between BF vs. running speed during running (linear) and cycling (nonlinear). While it is possible to assess VTs from fHF during a cycle ergometer test, it is not possible during a running test.How does my ramp data look?
This is a portion of a ramp test I did some months ago using the Hexoskin shirt (with near perfect signal quality), RR intervals extracted, analysis with the Kubios HRV software. This was on an indoor turbo trainer back in January. I am using the premium package ($$) and therefore am able to plot "time varying" trends. The basic version will not allow this, but you can manually create interval blocks (with the free version), look at the HRV metrics (all of them) and then make your own graph.
Here is the HF power (black dots) along with the heart rate(blue dots) in 30 second blocks:
- The HF power is relatively stable throughout the early ramp and increases once it is over. To simplify analysis, I will just show the HF power and HF peak on separate graphs
The HF frequency peak is interesting:
If we ignore the possible early artifact, there is a notable breakpoint at a heart rate of about 120, and at 143 watts (artifact removed below):
A closer look from the Kubios software package using 30 second windows before and after the breakpoint:
- The pre HF frequency peak is .156 vs .656 for the post (in yellow). The FFT tracing does not have a visually dominant band easily discernible.
- Note should be taken of the HF band upper limit being 2 Hz (green).
For comparison here is a 30 sec window later on the ramp with a heart rate of 160:
- The red arrow shows the dominant frequency on the FFT graph. The magnitude and frequency are much different than the lower heart rate values.
- I would be have been more comfortable with the data analysis if the lower heart rate frequency band was as visually dominant.
Hexoskin ventilation comparison:
A comparison of the power, heart rate and ventilation is here:
- I circled the 135 watt/120 heart rate area as the point where ventilation starts to pick up, similar to the 143 watt noted above.
- Whether or not these are actual breakpoints is arguable. The tracing looks more curvilinear to me than segmented. This is a problem with threshold detection in general, many studies use multiple "examiners" to decide where the breakpoints are.
Threshold assessment by HRV:
An issue with many of the studies done is that they generally all use different parameters to look for breakpoints and thresholds. I am not even sure if the above study has been reproduced by other groups. There are several other papers on HRV and VT/MLSS but they all use different heart rate derived metrics.
In fact a comment from the review mentioned above caught my eye:
During exercise, HRV measures demonstrate a curvilinear decay
as a function of exercise intensity, that is closely related
to exercising HR. HRV measures associated with cardiac
parasympathetic activity (e.g., RMSSD and HF) usually reach
a near-zero minimum at moderate intensity (possibly being
associated with the first ventilation/lactate threshold). These
measures are sometimes observed to increase slightly as exercise
intensity increases toward maximum, although this is likely
mediated by non-neural mechanisms such as direct mechanical
effects of respiration on the SA node. The data also leads
to further questioning of the use of frequency domain ratio
and normalized measures as indicators of sympathetic activity
or “sympatho-vagal balance.” In addition to demonstrating
inconsistent responses to exercise, the response of these measures
is rarely consistent with our current understanding of autonomic
control during exercise, namely progressive parasympathetic
withdrawal and sympathetic activation.
A study in basketball players was done looking at the relationship of HRV to both ventilatory threshold 1 (VT1) and the VT2 (near the MLSS). They chose a parameter called the SD1 (see above) to use as the indicator for VT1 and the HF power band to correlate with the MLSS.
The basketball players completed an incremental test to exhaustion on a treadmill (Run MedTechnogym, Cessena, Italy) in standard environmental conditions,with the grade set at 1%.
HF power (HFp) trend, as a function of time and frequency over the entire exercise period, was calculated from R-R interval series using aResults:
time-varying short-term Fourier transform with 64 s moving
window. VT2 was determined from HFp at the final abrupt
increase in the HF band (HRVT2). HFp range was extended
from resting recordings (>0.15–0.5 Hz to >0.15–2 Hz).
Although the results did have statistical significance, the correlation coefficient was not very high:
The correlation was much better here:
We concluded that the ability of the HRV timeA very bold statement indeed!
varying spectral analysis to estimate VT2 during incremental
running test, in professional basketball players, was demonstrated
to provide sufficient reliability and validity and thus may
be implemented into a training session without the use of a gas
analyzer to determine HR, speed, or VO2.
(the HF power band is useful to estimate the VT2/MLSS)
My data - I plotted the HF power in normalized units and came up with this.
The zone that would correspond to my approximate MLSS is circled:
And a look at the 30 sec windows pre and post at that section:
- The HF power rises from 2.5 to 4.4. The FFT window shows more HF activity (green) in the post breakpoint section.
- Whether this is real or just imagination is of course the question.
Another study was done to try to confirm the above observation, but in moderately trained subjects.
The incremental test was conducted on a cycle ergometer
(Lode; Corival, Groningen, The Netherlands). The protocol
involved a two-minute warm-up at 50 watts with increments
of 20 watts occurring every minute with subjects instructed to
maintain 80 revolutions per minute (rpm). The test was terminated
in the case of volitional fatigue, a failure to maintain
80 rpm or a plateau in VO2 despite an increase in workload.
Upon completion of the incremental test, ventilatory data were
averaged to every 20 s and exported to a personal computer for
further analysis. The simplified V-slope method was used to
detect the VT (Schneider et al., 1993). Here, VCO2 is plotted on
the y-axis and VO2 on the x-axis (Fig. 1). The first breakpoint
observed was the VT and is considered the point where glycolytic
activity increases so that the lactic acid starts accumulating.
Here, ventilation and VCO2 begin to increase as a result of
hydrogen ion buffering whilst VO2 remains constant (Beaver
et al., 1986).
Nonlinear analysis of the R-R intervals was performedNotes -
adopting the Poincare Plot. This technique allows for a graphical
representation of the systems evolution in phase space
(Lombardi, 2000), where each interval is plotted as a function
of the previous one (R-Rn, R-Rn+1) to create a scatter plot
(Tulppo et al., 1998). Two axes are fitted 45 degree to one
another passing through an ellipse fitted to the centre of the
data points. As a marker of nonlinear short-term variability,
we adopted the SD1 parameter, which is shown to reflect
vagal modulation (Tulppo et al., 1996).
For the spectral analysis, a short-term Fourier transform was
applied with a moving window of 64 s and time shift of 10 s
(Cassirame et al., 2015). The high-frequency band was
extended (>0 15–0 4 Hz to >0 15–2 Hz) to consider spectral
energy content resulting from increases in respiratory frequency
during high-intensity exercise (Cottin et al., 2006a,b).
HFp was captured for further analysis. In addition, to validate
further observed results, frequency peak of HF (fHF) (Hz) in
the power spectral density was accounted for, which allows for
correlation analysis between respiratory frequency and HRV.
There is an extension of the HF band up to 2 Hz to encompass the respiratory frequency effects. When looking at my data later on we will make an adjustment in the software to override the default values. Some investigators also use a value of 1 Hz as the upper limit and as a compromise, I have set the software to this value.
The VT is defined as the VT1, not the MLSS/VT2.
The main findings of the present study were the following:
(i) TSD1 shows no relationship with the VT in moderately
trained healthy males. When expressed as both power output
and heart rate, a significant difference was observed, which
was confirmed by the Bland and Altman plots. (ii) Similarly,
THFp did not exhibit a significant correlation with the VT
when expressed as power output but in contrast exhibited a
strong correlation with the VT when expressed as heart rate.
It appears that the correlation of SD1 with the VT1 was poor but there was a reasonably good relation between HF power and VT1 associated heart rate. Why the watt power did not correlate well with HRV derived indexes was not really discussed.
Conclusion:Lastly, in untrained individuals (VO2 max about 30 ml/kg/min) another study found good correlation using the SD1 and RMSSD metrics with VT1 and VT2 (as opposed to poor agreement to the SD1).
Results of this study revealed that TSD1 derived through the
Poincare plot shows no relationship with the VT in moderately
trained healthy males. However, THFp showed a strong
relationship with the VT when expressed as heart rate. Given
the relationship between HRV and respiration, this may be a
more robust indirect measure of the VT in trained individuals
allowing for both assessment of cardiovascular fitness and prescription
The first criteria, HRVT1, was determined
in the first physical effort intensity in which the SD1 and
RMSSD index were < 3 ms ; that point occurs at the same intensity
of lactate and ventilatory threshold. The second additional
criteria, HRVT2, was determined in the first effort intensity
in which the difference between 2 consecutive intensities was < 1 ms
in the SD1 and RMSSD index, which occurs at same intensity
of ventilatory threshold.
Ramp study, 30 w/min rise:
Now that we have some idea what to look for how does this shape up in reality? I went back to the old ramp I did using the Hexoskin and graphed out the SD1 with both power and heart rate. The RR intervals were of high quality with artifacts below 1% and the Kubios premium software was used for analysis.
Here is the SD1/HR graph from Kubios:
I took the liberty of drawing in a breakpoint (guess) with the corresponding HR near 140 bpm.
Here is the raw data with power averages per interval (30w/min):
Since I really do not know my true VT1/2 (yet) and ramp tests can lead to different values depending on the protocol, I'm not claiming any conclusion.
Since the study above indicated that HF power seemed to track well with VT1, I was curious as to how this looked.
Here I am plotting the HF power in nu (normalized units).
- There does seem to be a defined step up in HF power at about a HR of 120 bpm. The curve continues to rise to a max near the end of the ramp test.
Kubios data with power in watts:
- The tracing here seems even better defined (same data but different scaling) with the jump in HF power occurring when the ramp power went from 143 to 175 watt average. The HR zone corresponding to this is is about 120 at 143 watts to 127 at 175 watts. If my VO2 max power is about 350w, this represents about 50% of the MAP which is at the low end for estimated VT1.
Longer ramp intervals near VT1
HRV parameters are conventionally done over longer windows than 30 or 60 seconds. To get a better handle on whether HRV is useful in looking at VT1 thresholds, longer ramp zones with finer power increments should be used. To this end, I did a limited ramp from 152 up to 218 watts with 4 minute intervals and about 10 watt increments. Since there are arguably better, more established ways of looking at VT2/MLSS power (lactate levels, SmO2), I decided to confine the ramp to the VT1 estimation.
Power profile + heart rate (polar H10)
The zone power, heart rates, cadence are listed in the table (laps 2-7):
HF power in normalized units:
- HR is in blue, HF power in black.
- The initial part of the ramp shows an undulating but relatively stable HF power.
- There may be a shift in the HF power curve (black) that corresponds to the 172 watt interval at an average heart rate of 125.
HF frequency peak:
- HF freq peak in black.
- There does not appear to be much in the way of a pattern here in the power range that was tested.
- However, a closer look at the FFT spectrum shows little in the way of a dominant band, making interpretation problematic:
- SD1 is in black, HR in blue.
- Similar to the HF power, there is some undulation early in the ramp, but it appears to shift at 172 watts, 125 avg HR to a lower undulating pattern.
- HRV analysis is possible during dynamic exercise but is a challenge to implement.
- Several studies have concluded that both the VT1 and VT2 are able to be obtained from appropriate analysis of the HRV data of exercise ramp protocols.
- The exact HRV parameters best suited to looking for these thresholds are somewhat confusing. Almost each study uses either a different HRV variable or a relatively convoluted derivation of multiple variables.
- Some studies have felt that the SD1 does correlate with VT1 and perhaps my longer, fine increment ramp seems to show a threshold in the area of interest.
- The peak HF frequency may also be useful for VT1 as well as VT2 in cycling but not running.
- The HF power has been shown to be associated with VT1 and VT2 in several studies.
- However - Even in some of the more "successful" studies, threshold correlation may only be moderate. Interpretation of thresholds is made difficult by undulating tracings, natural variation, artifact and the limitations of software analysis. Many published works use the popular Kubios software but overlay the analysis with additional layers of derived values.
- There does seem to be a potential place for using this technique in threshold monitoring as well as demarcation of intensity zones.
- Further, using HRV during dynamic exercise could be an "insurance" policy for a true recovery session. The goal being to keep out of the more stressful zones for optimal cardiac recovery.
DFA alpha1, HRV complexity and polarized training