The evidence discussed in my recent posts indicates that for moderately fit individuals there is only a narrow gap between the training load required to produce useful improvement in performance and that which results in the over-reaching that leads to reduced benefits of training. Hence to obtain optimum benefit from training, it is necessary to have a reliable way of estimating recovery from previous training sessions. Furthermore, the evidence, especially the evidence for the recent studies by Kiviniemi and colleagues [1, 2] indicates that adjusting training according to heart rate variability HRV) measured in the resting state each morning might lead to more efficient training. As discussed in my post on 13th July, I think this evidence is promising but not compelling.
Heart rate during exercise
Many athletes use heart rate during exercise itself as a guide to progress with their training. Assessment of fitness such as that proposed by Hadd, which employs the relationship between pace and heart rate, assessed over a range of different heart rates, are based on the assumption that a lower heart rate at a given pace is an indication of increased fitness. On the other hand, when fatigue begins to build up during moderately heavy training, heart rate at a given pace tend to rise indicating the first phase of over-reaching. In contrast, during very heavy training, a decreased heart rate at a given pace can actually indicate parasympathetic over-reaching. Provided the measurements are interpreted in context, gradual changes in heart rate at a given pace over a period of many weeks can be a useful guide to changing level of fitness, while short term changes can provide an indication of over-training.
HRV during exercise
These observations raise the possibility that HRV during exercise might be a useful guide to the degree of exhaustion during a training session. However, there is surprisingly little published information regarding the interpretation of HRV during exercise. Why should this be so? The main problem is that whereas HRV at rest is very strongly influenced by the autonomic nervous system, it is not clear what are the main factors influencing HRV during vigorous exercise.
Figure 1 shows a trace which I recorded last year using my Polar RS800cx in R-R mode, during a graded exercise session on the elliptical cross trainer in July. (Note the figure shows beat-by-beat values of heart rate, which is the reciprocal of R-R interval). After two minutes of standing still, I exercised at a cadence of 80 cycles per minute and adjusted the resistance such that the power output increased from 30 watts in 7 approximately equal steps, each lasting 4 minutes, up to 230 watts. After completing 4 minutes at 230 watts, I maintained an output of 30 watts for a further 4 minutes and then stood still for another 4 minutes.
There are a number of idiosyncratic features of the trace which I will ignore for the present discussion. These include sharp spikes which might be premature atrial beats (or artifact due to disrupted electrode contact). The dip in heart rate to a value below 80 bpm lasting for around 20 seconds at 5 minutes is a typical feature which I observe in many of my recordings while in the lower part of the aerobic zone, and at present remains a mystery to me. I have seen this feature in the recordings of other athletes but have not found any descriptions of these dips in published research reports. However, for present purposes, let us ignore both the spikes and the dips.
The main features of note are:
1) a large amount of high frequency fluctuation (indicated by the thickness of the fuzzy line) when standing still, in the first two minutes;
2) the amount of high frequency fluctuation reduces subsantially (the line becomes less fuzzy) as HR increases into the range from 80 to 120 bpm, if we ignore the aberrant spikes and dips. By HR 120 bpm, at around 12 minutes, there is relativley little high frequency variability, though of course HR is continuig to rise slowy under the influence of the sympathetic nervous system.
3) the high frequency variability then builds up again as HR approaches 140 (my ventilatory threshold, where breathing rate rapidly increases from 40 breaths per minute to 80 breaths per minute), as indicated by the increased fuzziness at around 24 minutes;
4) and finally the variability diminishes slightly as HR approaches its maximum, at around 28 minutes. (My true maximum HR is probably around160 bpm though I have not actually pushed myself to the maximum in recent years) .
Figure 2a shows the Poincare plot of each interval between consecutive beats against the interval between the preceding pair of consecutive beats, over a 60 second interval around 24 minutes (power output, 200watts). If a series of beats were equally spaced, the point representing each pair of consecutive beats in the Poincare plot would lie on a straight line inclined at 45 degrees. The degree of scatter away from the 45 degree line indicates beat by beat fluctuations in HR. The amount of scatter can be quantified by calculating the quantity, SD1, the standard deviation of the scatter away from the 45 degree line. In this plot, sd1 is 4.4 ms which is typical of values I observe at the upper end of the aerobic zone. Figure 2b shows the natural log of SD1 calculated for 60 second epochs at the end of each of the 4 minute steps, when HR is relatively stable.
It should be noted that estimating HRV during a period when HR is varying due to increasing work load presents problems. Various measures other than SD1, such as a scaling factor designated alpha computed on the basis of the theory of nonlinear systems (i.e. chaos theory), have been proposed, but the interpretation of such measurements is fraught with difficulty. This is illustrated by poor consistency between measurement of alpha in the same person in sessions a week apart . Hence I think it is best to employ SD1 despite the limitations of this measurement. The reason that it is conventional to plot the log of SD1 is that during the early phase of increasing exercise intensity, parasympathetic influence falls away exponentially. An exponential decay would be expected to produce a straight line when plotted on a logarithmic scale.
Figure 2b confirms that high frequency HRV (as quantified by the log of SD1) falls rapidly from standing to low intensity exercise; continues to diminish gradually in the lower aerobic zone and then increases around the anaerobic threshold before decreasing again as HR approaches maximum. This is fairly typical of what is seen in other athletes, though I exhibit a more abrupt initial withdrawal than I observe in others.
What determines HRV during exercise?
It is widely accepted that the initial fall off in high frequency HRV (as HR increases from its resting value) is due to withdrawal of parasympathetic influence. Once in the aerobic zone, parasympathetic input is minimal and further increase in heart rate is driven by increased sympathetic activity . However, high frequency variation at around respiratory frequency is not entirely abolished even in the mid-aerobic zone, suggesting that some other factor is contributing. It is quite likely that intrinsic variability in the function of the sinoatrial node (the collection of specialized muscle cells in the right atrial wall which fire spontaneously and normally act as pace-maker) plays a substantial role. In a transplanted heart, which does not receive any input for the parasympathetic or sympathetic nervous system, heart rate variability during exercise is similar to that in a heart with intact input from the nervous system. Casadei and colleagues from Oxford carried out a study in which they used the pharmaceutical agent, atropine, to block parasympathetic nervous activity, and observed that about a third of the variability of heart rate in the lower aerobic zone, can be accounted for by non-neural mechanisms . Perini and colleagues  have argued that the most plausible alternative mechanism is mechanical. For example, respiration would be expected to produce rhythmic stretching of tissues that might change the excitability of the pace-maker leading to variability at the respiratory frequency. Furthermore, mechanical effects are a plausible explanation for the prominent increase in the magnitude of high frequency fluctuation around the anaerobic threshold, at which point respiratory effort increases markedly.
However, I remain a little skeptical that mechanical effects account fully for the high frequency variation in the upper aerobic and anaerobic zones. In particular, excessive high frequency fluctuation during vigorous exercise appears to be a significant predictor of poor long term health. For example, in a large study of 1335 subjects, mainly males, Dewey and colleagues from Stanford University  demonstrated that increasing magnitude of high frequency fluctuation during vigorous exercise (and also during the recovery phase) was a significant predictor of mortality in general, and especially of cardiac mortality, in the following five years. This suggests that HRV during vigorous exercise might reflect some physiological process that is indicative of impaired cardiac well being, probably something more subtle than the simple mechanical effects of respiration. Might that putative physiological marker of impaired cardiac well-being be sensitive to the state of recovery during training?
HRV during exercise when fatigued
Observations of my own HRV during exercise confirm this hypothesis. Last summer, my preparation for the Robin Hood half marathon had been seriously disrupted by quite severe bout of illness in June. I had missed about 4 weeks of training, and attempted to build up training volume in July, in preparation for the race in early September. As described in my blog posting on 31st August 2009, I developed quite marked fatigue during August. In an attempt to determine if the fatigue was affecting my heart, I had repeated the graded increase in exercise on the elliptical cross trainer at the end of August, following an identical schedule with 4 minutes at a series of seven steps spanning a range of power output from 30 to 230 watts, to that employed on 18th July. The R-R trace for 31st August is shown in figure 3, while fig 3a is an expanded view comparing the variability in the period 23-25 minutes on 31st Aug with that on 18th July. Figure 4a depicts the Poincare plot based on R-R intervals at 23-24 minutes on 31st August and Figure 4b shows the natural log of SD1 calculated for 60 second epochs at the end of each of the 4 minute steps.
The most prominent feature is that on 31st August, there was a much more dramatic increase in high frequency HRV as I approached the anaerobic threshold (shown by the marked fuzziness of the line around 24 minutes). Furthermore, despite a subsequent increase in power output from 200 to 230 watts, heart rate scarcely rose in the following 4 minutes. In contrast to the similar test on 18th July, when HR rose to 157 bpm at a power output of 230 watts, on 31st August the peak heart rate was 145 bpm. Subjectively, I experienced tremendous fatigue and found it very difficult to complete the 4 minutes at 230 watts. Thus, on this occasion overwhelming feelings of fatigue limited my power output and were associated with excessive high frequency HRV.
The greater amount of high frequency variability is also clear from the wide scatter of points away from the 45 degree line in the Poincare plot and the relatively large peak in the value of log (SD1) at a heart rate of 145 bpm, visible in the plot of log (SD1) against HR. (Compare fig 4a and 4b with fig 2a and 2b).
Although I have not subsequently experienced such overwhelming fatigue, there have been a number of occasions in which I have observed excessive high frequency HRV in association with moderate fatigue, sometimes when exercising in the lower aerobic zone and sometimes near anaerobic threshold. For example, a few weeks ago I became increasingly tired during a hectic week at work. On the Friday evening (18th June) I set out to do an easy 7.2Km run in the lower aerobic zone. I felt tired and lethargic throughout. The R-R trace (figure 5) for a 15 minute segment in the middle of the run, when my pace was stable at 5:56 min/Km, shows marked high frequency fluctuation in heart rate. Average heart rate was 124 bpm and SD1 was 9.0 ms. For comparison figure 6 shows a similar 15 minute segment in the mid-stage of a 7.2Km low aerobic run on the same path in January. Pace was somewhat faster at 5:39 min/Km, average heart rate was a little lower at 122 bpm, and high frequency variability was much less (SD1= 3.8ms).
The following week I developed signs of an upper respiratory tract infection. The R-R trace during an interval session on the elliptical machine is shown in Figure 7. High amplitude fluctuation in heart rate can be seen in the second, third and fourth effort epochs. During the crest of the second effort epoch, SD1 = 12.8ms
For comparison the R-R trace during a virtually identical elliptical interval session performed a week later when the respiratory tract infection had resolved, is shown in figure 8. The magnitude of HRV is much less, and SD1 at a comparable period in the 2nd effort epoch is 2.5 ms.
Although not all of the data provides quite such clear-cut information as the illustrations I have presented here, on many occasions on which I have felt fatigued or sluggish, there as been excessive high frequency HRV. This feature is observed in various different types of session including steady running in the lower aerobic zone, tempo sessions, interval sessions and progressive increases in exercise intensity from the low aerobic to the anaerobic zone. In contrast to resting HRV, where increased high frequency HRV is generally an indication of good recovery (except during advanced stages of over-reaching) increased high frequency HRV during vigorous exercise generally appears to be an indicator of greater stress.
I do not know what physiological process is responsible for this. Perhaps some central governor generates ‘protective’ parasympathetic activity even at exercise intensities at which parasympathetic influence would normally be minimal. Alternatively, in view of the fact that I suffer from asthma, it is possible that these episodes of fatigue are associated with subliminal constriction of my airways that results in more labored breathing and hence greater mechanical forces on my heart. However, I have observed this phenomenon on many occasions when I am not aware of any breathing difficulty. I would be very interested to hear if anyone else has observed similar increases in high frequency variability during exercise when under stress.
The dips: could they be due to transient sympathetic withdrawal?
[Note added 29 Nov 2010: the dips are almost certainly due to a transient surge of parasympathetic actviity associated with relaxation of smooth muslce at the the oesophageal-gastric junction - as discussed with Steve in the comment section below]
Although I have not hitherto focused on the dips in HR that I frequently observe when exercising in the lower aerobic zone, or as my heart rate drops during the recovery from intense exercise, I have marked these by black arrows in figure 8. Similar dips are also visible in figs 1,5,6 & 7 . I have no reason to connect these to the increases in HRV associated with fatigue. In fact the duration of these dips is typically 10-20 seconds. If they are due to influence of the autonomic nervous system, they might possibly reflect sympathetic withdrawal, which would be expected to act on this time scale They do not appear to be a marker for fatigue, but nonetheless I wonder whether they too might reflect a protective effect generated by a putative central governor. I have observed these dips in the traces of other athletes, but would also be very pleased to hear from others who have observed similar dips. Because of the relatively long time scale, such dips would in fact be seen more readily in a record of HR averaged over 5 second intervals.
Is ‘real time’ assessment of HRV during exercise practical?
If indeed the excess HRV during exercise is a reliable marker for undue stress, would this be of any practical use? The R-R trace is not generally available for inspection until after the completion of the session. However, in principle it should be possible to produce a continuous read-out of HRV averaged over a preceding period of about 20 seconds. In fact the Polar RS800cx produces a measure called RLX. Although it is difficult to obtain a precise account of the computation on which RLX is based, it appears to be very closely related to a continuous estimate of SD1. Furthermore, according to the Polar manual, it should be possible to present the continuously updated value of RLX in the display of the wrist unit. I have been unable to do this, and think that it is yet another manifestation of the fact that my particular RS800cx is infested with gremlins. However as far as I can tell from retrospective examination of RLX values, the Polar computation of RLX is based on computation over too short an interval to provide reliable values. I suspect that if I can convince myself that it would be worthwhile having access to a continuous read-out of high frequency HRV while exercising, it would probably be necessary to develop some more sophisticated way of performing the computation.
 Antti Kiviniemi, Arto Hautala, Hannu Kinnunen & Mikko Tulppo (2007) Endurance training guided individually by daily heart rate variability measurements. Eur J Appl Physiol. 101(6):743-751.
 Kiviniemi AM, Hautala AJ, Kinnunen H, Nissilä J, Virtanen P, Karjalainen J, Tulppo MP . (2010) Daily exercise prescription based on Heart Rate Variability among men and women. Med Sci Sports Exerc. 42(7):1355-63
 Tan CO, Cohen MA, Eckberg DL, Taylor JA (2009) Fractal properties of human heart period variability:physiological and methodological implications J Physiol 587.15 pp 3929–3941
 Rowell LB, O’Leary DS (1990) Reflex control of the circulation during exercise: chemoreflexes and mechanoreflexes J. Appl. Physiol. 69(2): 407-418
[5} Casadei B, Moon, J, Johnston J, Caizza S, Sleight P (1996) Is respiratory sinus arrhythmia a good index of cardiac vagal tone in exercise? J.Appl. Physiol. 81(2): 556-564,
 Perini R, Veicsteinas A (2003) Heart rate variability and autonomic activity at rest and during exercise in various physiological conditions. Eur J Appl Physiol 90: 317–325
 Dewey FE Freeman JV, Engel G, Oviedo R, Abrol N, Ahmed N, Myers J, Floelicher VF (2007) Novel predictor of prognosis from exercise stress testing: heart rate variability response to the exercise treadmill test. Am Heart J, 153(2) 281-8.