reliability of heart rate measures used to assess post-exercise parasympathetic reactivation

Reliability of heart rate measures used to assesspost-exercise parasympathetic reactivationOlivier Dupuy1,2, Saı̈d Mekary1,3, Nicolas Berryman1,2,3, Louis Bherer3,4, Michel Audiffren2 and LaurentBosquet1,2,3

1Department of Kinesiology, University of Montreal, Montreal, Qc, Canada, 2Faculty of Sport Sciences, University of Poitiers, Poitiers, France, 3Research Center,

Montreal Institute of Geriatrics, Montreal, Qc, Canada, and 4Department of Psychology, University of Quebec at Montreal, Montreal, Qc, Canada

Summary

CorrespondenceLaurent Bosquet, Faculty of sport sciences, 8 Jean

Monnet road, 86000 Poitiers, France

E-mail: [email protected]

Accepted for publicationReceived 06 November 2011;

accepted 25 January 2012

Key words

autonomic nervous system; exercise intensity; heart

rate recovery; heart rate variability

Purpose: Postexercise HRR (heart rate recovery) and HRV (heart rate variability) arecommonly used to asses non-invasive cardiac autonomic regulation and more par-ticularly reactivation parasympathetic function. Unfortunately, the reliability ofpostexercise HRR and HRV remains poorly quantified and is still lacking. The aimof this study was to examine absolute and relative reliability of HRR and HRVindices used to assess postexercise cardiac parasympathetic reactivation.Methods: We studied 30 healthy men, who underwent 10-minute heart raterecording after cessation of maximal and submaximal intensity exercises. Eachcondition of testing was repeated twice within 5 ± 2 days after the first one.Standard indexes of HRR and HRV were computed from heart rate and RR inter-vals.Results: We found no significant bias between repeated measures. Relative reliabil-ity was assessed with the intraclass coefficient correlation (ICC) and absolute reli-ability with the standard error measurement (SEM) and coefficient of variation(CV). A large range for ICC was observed for both indexes of HRR and HRV(0·12 < ICC < 0·87 and 0·14 < ICC < 0·97, respectively). The same heterogene-ity was observed for absolute reliability (5% < CV < 72% for HRR parametersand 24% < CV < 141% for HRV parameters).Conclusion: According to our results, Δ60 (the absolute difference between heartrate immediately at the end of exercise and after 60 s) and HFnu (High Fre-quency expressed in normalized unit; that is, in a percentage of LF+HF) representthe most reliable parameters. In conclusion, we found that the measures used toasses cardiac parasympathetic reactivation were characterized by large randomvariations and their reliability remains moderate.

Introduction

Postexercise heart rate recovery (HRR) and heart rate vari-

ability (HRV) are commonly used to assess non-invasive

cardiac autonomic regulation and more particularly cardiac

parasympathetic reactivation. Heart rate decreases mono-

exponentially towards resting values when exercise ceases

(Perini et al., 1989). The exponential nature of this kinetics

is an intrinsic property of the cardiovascular system and is

modulated by the autonomic nervous system (ANS) (Savin

et al., 1982). The rapid decrease in HR that occurs during

the first 2–3 min after exercise cessation is thought to be

mainly determined by the restoration of parasympathetic

activity at the sinus node level (Savin et al., 1982; Kannank-

eril & Goldberger, 2002). Heart rate indices calculated from

this part of the signal are therefore considered as indirect

measures of cardiac parasympathetic reactivation (Imai et al.,

1994; Kannankeril et al., 2004; Buchheit et al., 2007b). The

slow decrease in HR that takes place from the third minute

after exercise cessation until the return to resting values is a

complex interplay between the decrease in cardiac sympa-

thetic activity and the respiratory modulation of cardiac

parasympathetic activity (Perini et al., 1989; Buchheit et al.,

2007b). This interaction between both branches of ANS can

be assessed through the analysis of HRV, either in the

time or frequency domains (Task Force, 1996). Postexercise

HRR and HRV measures can therefore be considered

as non-invasive measures of cardiac autonomic regulation

Clin Physiol Funct Imaging (2012) 32, pp296–304 doi: 10.1111/j.1475-097X.2012.01125.x

296© 2012 The Authors

Clinical Physiology and Functional Imaging © 2012 Scandinavian Society of Clinical Physiology and Nuclear Medicine 32, 4, 296–304

and more especially cardiac parasympathetic reactivation

(Buchheit et al., 2007b) and have been consistently used in

this purpose.

In a clinical standpoint, a delayed HRR and/or a reduced

HRV has been associated with several health outcomes such as

heart failure, hypertension and diabetes (Lishner et al., 1987;

Tsuji et al., 1996; Cole et al., 1999) and are considered as

being reflective from an increased risk of all-cause mortality

(Bigger et al., 1992; Tsuji et al., 1996; La Rovere et al., 1998;

Cole et al., 1999). These measures are also commonly used in

the follow-up of athletes (Lamberts et al., 2009). In fact, a

large bulk of studies have shown that positive adaptation to

exercise training was associated to a faster HRR and an

increased HRV (Yataco et al., 1997; Triposkiadis et al., 2002).

Although it remains to be confirmed by more experimental

data, it has been suggested that these measures could be used

to monitor acute fatigue and to prevent chronic negative

adaptations to exercise training, such as overreaching or over-

training (Hedelin et al., 2000; Bosquet et al., 2008b; Lamberts

et al., 2010). The potential interest of postexercise HRR and

HRV measurement is therefore undeniable, either in clinical

or physiological settings.

The reliability of HRV (Sinnreich et al., 1998; Schroeder

et al., 2004; Sandercock et al., 2005; Pinna et al., 2007) and

HRR (Christenfeld et al., 2000; Bosquet et al., 2008a; Buchheit

et al., 2008; Al Haddad et al., 2011; Arduini et al., 2011) after

exercise cessation has been assessed in different conditions

and populations. However, several limitations have been

raised including (i) an inadequate protocol, (ii) an inadequate

signal selection and analysis, (iii) an inadequate or incomplete

statistical analysis and (iv) the lack of practical implications

for clinicians and researchers (Sandercock et al., 2005; Pinna

et al., 2007; Sandercock 2007). An additional limit is that

there is no study that assessed the relative and absolute reli-

ability of simultaneously measured HRR and postexercise

HRV, while they bring complementary information about car-

diac parasympathetic reactivation (Buchheit et al., 2007b).

Finally, previous reports have shown that HRR indices derived

from exponential curve fitting were subject to large random

variations (Bosquet et al., 2008a; Buchheit et al., 2008; Arduini

et al., 2011). It is well established that lowering the signal-to-

noise ratio is an effective way to decrease the variability of

these indices. A strategy that is commonly used in this pur-

pose is to average several time-aligned signals of the same

exercise protocol (Koga et al., 2005). However, the effective-

ness of this method to improve the reliability of HRR indices

has not been assessed.

The aim of this study was therefore to examine absolute

and relative reliability of HRR and HRV indices that are com-

monly used to assess postexercise cardiac parasympathetic

reactivation in healthy individuals with a protocol and data

analysis that account for the limitations previously raised. A

secondary aim of this study was to determine whether averag-

ing several signals of the same exercise protocol improved the

reliability of HRR indices.

Methods

Subjects

Thirty healthy men with no smoking history and no known

cardiovascular disease participated in the study. Their

mean ± SD age, stature and body mass were 27 ± 8 years,

177 ± 7 cm and 74 ± 10 kg, respectively. The protocol has

been reviewed and approved by the Research Ethics Board in

Health Sciences of the University of Montreal, Canada.

Experimental design

Following a thorough briefing and medical screening, all par-

ticipants signed a written statement of informed consent. Once

included, they completed successively two maximal continu-

ous graded exercise tests and two submaximal square-wave

constant duration tests on a motorized treadmill (Quinton,

Bothell, WA, USA), which was calibrated at 8 and 16 km h�1

before each session with an in-house system using an optical

sensor connected to an acquisition card. All tests were sepa-

rated by at least 72 h and were performed in a laboratory

room of constant temperature (21°C) and humidity (45%)

within a 2-week period. Participants were asked to refrain

from strenuous training and to abstain from caffeine contain-

ing foods and beverages 24 h before the test, to avoid any

influence on the autonomic control of the myocardium. They

were also asked to arrive fully hydrated to the laboratory, at

least three hours after their last meal.

Exercise testing

Maximal continuous graded exercise test

Initial velocity was set at 12 km h�1 and increased by 1 km

h�1 every 2 min until exhaustion. Less than 5 s after exercise

cessation, participants sat on a chair for a 10-min passive

recovery period. The grade of the treadmill was set at zero

throughout the test. The velocity of the last completed stage

was considered as the peak treadmill speed (PTS). Oxygen

uptake ( _VO2) and related gas exchange measures were deter-

mined continuously using an automated cardiopulmonary

exercise system (Moxus, AEI Technologies, Naperville, IL,

USA). Gas analysers (S3A and CD3A, AEI Technologies) were

calibrated before each test, using a gas mixture of known con-

centration (15% O2 and 5% CO2) and ambient air. Accuracy

of analysers was ± 0·003% for oxygen and ± 0·02% for car-

bon dioxide (data provided by the manufacturer). The turbine

was calibrated before each test using a motorized syringe

(Vacu-Med, Ventura, CA, USA) with an accuracy of ± 1%

(Huszczuk et al., 1990). The tidal volume was set at 3 l and

the stroke rate at 40 cycles per minute. Heart rate was moni-

tored continuously beat by beat during exercise and recovery

using a Polar S810 heart rate monitor (Polar Electro Oy,

Kempele, Finland), with an accuracy of 0·3% (Kingsley et al.,

© 2012 The AuthorsClinical Physiology and Functional Imaging © 2012 Scandinavian Society of Clinical Physiology and Nuclear Medicine 32, 4, 296–304

Postexercise parasympathetic reactivation, O. Dupuy et al. 297

2005) during exercise and 0·4% at rest (Gamelin et al., 2006).

Participants were asked to match their breathing frequency to

an auditory metronome set at 0·20 Hz (12 breaths min�1)

from the 5th to the 10th minute of the passive recovery per-

iod. The second maximal continuous graded exercise test was

performed within 5 ± 2 days after the first one.

Submaximal square-wave constant duration test

Velocity was set at 75% of PTS for a period of 6 min. Less

than 5 s after exercise cessation, participants sat on a chair for

a 10-minute passive recovery period. This protocol (i.e.

6 min of exercise and 10 min of passive recovery) was per-

formed twice, without rest in between. Heart rate was moni-

tored continuously beat by beat during exercise and recovery

using a Polar S810 heart rate monitor (Polar Electro Oy). Par-

ticipants were also asked to match their breathing frequency

to an auditory metronome according to the same modality

than following the maximal continuous graded exercise test.

The second submaximal square-wave constant duration test

was performed within 4 ± 2 days after the first one.

Data analysis

Determination of maximal oxygen uptake ( _VO2 max)

Mean values of _VO2 (in ml min�1 kg�1) were displayed every

30 s during the test. The primary criteria for the attainment

of _VO2 max were a plateau in _VO2 despite an increase in run-

ning speed. A plateau was considered present when _VO2

increased by <50% of the slope of the _VO2 – speed relation-

ship established during the test (Midgley et al., 2007). In the

absence of a plateau, secondary criterion included a respira-

tory exchange ratio of 1·10 or greater and an apparent

exhaustion of the participants.

Determination of postexercise heart rate recovery

R–R series were edited and visually inspected so that ectopic

beats could be replaced by interpolated data from adjacent

normal-to-normal (NN) intervals. Considering that the confi-

dence of HRR measures, particularly kinetic parameter esti-

mates, is greatly influenced by the beat-by-beat variability of

the data, the signals of the two consecutive 10-min passive

recovery periods were analysed independently one to each

other or time-aligned and averaged to produce a single

response with a higher signal-to-noise ratio. Several indexes

were used to characterize HRR during the 10-min passive

recovery period: D60 (in b min�1), T30 (in s) and the

parameters of a mono-exponential function. D60 was the

absolute difference between heart rate immediately at the end

of exercise (mean of 5 s) and after 60 s of passive recovery

(mean of 5 s) (Cole et al., 1999; Huang et al., 2005). T30 was

the negative reciprocal of the slope of the regression line

between the natural logarithm of heart rate and elapsed time

from the 10th to the 40th second of exercise (Imai et al.,

1994; Buchheit et al., 2007a). The overall kinetics of heart rate

during the 10-min transition from exercise to rest was

described by the following mono-exponential function:

HRðtÞ ¼ a0þ a1 � eð�t=TauÞ ð1Þwhere a0 is the asymptotic value of heart rate (in bpm), a1 is

the decrement below the heart rate value at the end of exercise

for t = ∞ (in bpm) and Tau is the time constant (i.e. the time

needed to reach 63% of the gain, in s) (Perini et al., 1989).

Determination of postexercise heart rate variability

R–R intervals from the 5th to the 10th minute of the passive

recovery period were edited and visually inspected so that

ectopic beats could be replaced by interpolated data from

adjacent normal-to-normal (NN) intervals. HRV was assessed

in the time and frequency domains. The mean HR, the stan-

dard deviation of NN intervals (SDNN) and the root-mean-

square difference of successive normal NN intervals (RMSSD)

were calculated from a segment of 256 s taken in the 5-min

period retained for HRV analysis. In accordance with the

report by Buchheit et al. (2008), the same segment of 256 s

was resampled at 2 Hz and detrended for subsequent analyses

in the frequency domain. As shown in Fig. 1, which presents

all the measures used in this study to assess cardiac parasym-

pathetic reactivation, the condition of signal stationarity for

spectral analysis is fulfilled. As recommended by the Task

Force (1996), spectral analysis was performed with a Fast

Fourier Transform (FFT) to quantify the power spectral den-

sity of the low-frequency (LF; 0·04–0·15 Hz) and the high-

frequency (HF; 0·15–0·40 Hz) bands. Additional calculations

included LF+HF, LF and HF expressed in normalized unit (i.e.

in a percentage of LF+HF) and the LF/HF ratio.

Statistical analysis

Standard statistical methods were used for the calculation of

means and standard deviations. Normal Gaussian distribution

of the data was verified by the Shapiro–Wilk test and homo-

scedascticity by a modified Levene Test.

Systematic bias, which refers to a general trend for mea-

surements to be different in a particular direction between

repeated tests (Atkinson & Nevill, 1998), was assessed with a

paired Student’s t-test. Relative reliability was assessed with

the intraclass correlation coefficient (ICC) (model 2, 1) and

absolute reliability with the standard error of measurement

(SEM) and the coefficient of variation (CV). Both the ICC and

the SEM were computed from the breakdown of a two-way

ANOVA (trials 9 subjects) with repeated measures, using the

following equations (Weir, 2005):

ICC ¼ MSS �MSE

MSS þ ðk� 1ÞMSE þ kðMST�MSEÞn

ð2Þ


Postexercise parasympathetic reactivation, O. Dupuy et al.298

where MSS is the mean-squared subjects, MSE = mean-squared

error, MST = mean-squared trials, k = number of trials and

n = number of subjects. We considered an ICC over 0·90 as

very high, between 0·70 and 0·89 as high and between 0·50and 0·69 as moderate (Munro, 1997). Currier (1990) has

suggested that an ICC value higher than 0·80 is acceptable for

clinical work.

Coefficient of variation is a measure of the discrepancy and

expresses error as a percentage of the mean. It was calculated

as follows:

CV ¼ ðSDD/MÞ � 100 ð3Þ

where SDD is the standard deviation of the differences

between test 1 and test 2 and M the mean for all observations.

SEM ¼ ffiffiffiffiffiffiffi

MSEp ð4Þ

where MSE is the mean-squared error.

Standard error measurement can also be used to determine

the minimum difference to be considered ‘real’ (MD). MD

represents the limit under which the observed difference is

within what we might expect to see in repeated testing just

attributed to the noise in the measurement and can be calcu-

lated as follows (Weir, 2005):

MD ¼ SEM� 1 � 96� ffiffiffi

2p ð5Þ

where SEM is the standard error of measurements computed

from Eq. 3.

Statistical significance was set at P<0·05 level for all analysis.

All calculations were made with Statistica 6·0 (Statsofts, Tulsa,

OK, USA).

Results

Information regarding the maximal graded exercise test is

presented in Table 1. We found no difference between test 1

and test 2 for _VO2 max, PTS and HRmax. These measures were

highly to very highly reliable (0·86 < ICC < 0·99; 0·4 < CV

< 5·9%). We found no systematic bias for HRR indices. Rela-

tive reliability was high for Δ60, a0 and Tau (0·74 < ICC

< 0·87) and moderate for a1 (ICC = 0·60). The CV was

<11·5% for all these indices. T30 was not reliable

(ICC = 0·12, SEM = 116 s and CV = 72·6%). We neither

found systematic bias for HRV indices. Reliability was very

high for SDNN (ICC = 0·88 and 0·97, respectively) and mod-

erate for HF, LF, LF/HF and HFnu (0·50 < ICC < 0·68). TheCV of these measures ranged from 26·6 to 68·1%. RMSSD

and LF+HF were not reliable (ICC = 0·14 and 0·27,SEM = 7·6 ms and 13 ms2, CV = 141% and 26·6%, respec-

tively).

Information regarding the submaximal square-wave con-

stant duration test is presented in Tables 2 and 3. Table 2

summarizes data obtained during the first 6-min exercise

bout. Relative reliability was high for a0 and a1 (ICC = 0·81and 0·80, respectively) and moderate for Δ60 (ICC = 0·69).The CV of these measures ranged between 8·2 and 12·5% of

average response. Other HRR indices such as T30 and Tau

were not reliable (ICC = 0·23 and 0·47, respectively;

SEM = 52 and 11 s, CV = 50% and 29·8%, respectively). Rel-ative reliability of most HRV indices was high (0·70 < ICC

< 0·88 for RMSSD, HF, LF/HF and HFnu) or moderate

(ICC = 0·57 and 0·58 for SDNN and LF, respectively). The CV

of these measures was higher than 24·1% of average response.

Figure 1 Graphical representation of heart rate recovery and heart rate variability measures used to assess cardiac parasympathetic reactivation inthis study.



LF+HF was not reliable (ICC = 0·36, SEM = 13 ms2 and

CV = 24·1%).Table 3 presents the reliability analysis of HRR indices cal-

culated from the combined signals (Fig. 2). Results are similar

to those calculated from a single exercise bout (Table 2).

Overall reliability is not improved by this procedure.

Discussion

The aim of this study was to examine absolute and relative

reliability of HRR and HRV indices that are commonly used to

assess postexercise cardiac parasympathetic reactivation in

healthy individuals with a protocol and data analysis that

Table 1 Reliability of autonomic indices during the recovery from maximal exercise.

Parameter Test 1 (mean ± SD) Test 2 (mean ± SD) ICC (value ± CL) SEM (value ± CL) MD CV (%%) (value ± CL)

ExercisePTS (km.h�1) 16·2 ± 1·3 16·3 ± 1·3 0·99 ± 0·01 0·1 ± 0·2 0·4 1·2 ± 2·4_VO2 (ml.min�1.kg�1) 53 ± 6 54 ± 6 0·86 ± 0·10 2·1 ± 4·1 5·9 5·7 ± 11·2HR max (b.min�1) 189 ± 10 188 ± 10 0·96 ± 0·03 2·1 ± 4·1 5·7 1·5 ± 2·9

Heart rate recoveryT30 (s) 204 ± 60 250 ± 165 0·12 ± 0·36 116 ± 227 321 72·6 ± 142D 60 (b.min�1) 50 ± 7 49 ± 8 0·77 ± 0·16 4·0 ± 7·8 11 10·8 ± 21·2a0 (b.min�1) 101 ± 12 99 ± 10 0·87 ± 0·10 4·0 ± 7·8 11 5·4 ± 10·6a1 (b.min�1) 95 ± 8 96 ± 9 0·60 ± 0·24 5·7 ± 11·2 16 8·2 ± 16·2Tau (s) 69 ± 11 70 ± 11 0·74 ± 0·18 6·3 ± 12·3 17 11·5 ± 25·6

Heart rate variability in the time domainHR (b.min�1) 101 ± 12 99 ± 11 0·88 ± 0·09 4·0 ± 7·8 11 5·2 ± 10·2SDNN (ms) 72 ± 337 61 ± 265 0·97 ± 0·02 51 ± 100 142 109 ± 213RMSSD (ms) 6·8 ± 4·4 8·3 ± 10·7 0·14 ± 0·35 7·6 ± 14·9 21 141 ± 277

Heart rate variability in the frequency domainHF (ms2) 30 ± 16 29 ± 18 0·62 ± 0·23 11 ± 22 29 50·5 ± 99·1LF (ms2) 41 ± 20 41 ± 16 0·50 ± 0·28 13 ± 26 35 43·8 ± 85·9LF+HF (ms2) 72 ± 17 70 ± 14 0·27 ± 0·34 13 ± 26 37 26·6 ± 52·2LF/HF 2·0 ± 1·8 2·3 ± 1·9 0·68 ± 0·21 1·0 ± 2·0 2·8 68·1 ± 133HF nu (%) 43 ± 20 40 ± 21 0·66 ± 0·22 12 ± 24 32 39·5 ± 77·4

SD, standard deviation; ICC, intraclass correlation coefficient; SEM, standard error of measurement; MD, minimum difference to be considered real;CL, confidence limits; CV, coefficient of variation; PTS, peak treadmill speed. HRmax, heart rate maximal. HR, heart rate; for the others abbrevia-tions see the ‘Methods’ section.

Table 2 Inter-session reliability of autonomic indices during the recovery from submaximal intensity exercise (first trial of session 1 comparedwith the first trial of session 2).

Parameter Test 1 (mean ± SD) Test 2 (mean ± SD) ICC (value ± CL) SEM (value ± CL) MD CV(%%) (value ± CL)

ExerciseHR (b.min�1) 166 ± 12 164 ± 14 0·82 ± 0·13 5·4 ± 10·6 15 4·7 ± 9·3

Heart rate recoveryT30 (s) 148 ± 56 145 ± 61 0·23 ± 0·34 52 ± 102 143 50 ± 98D60 (b.min�1) 56 ± 9 56 ± 8 0·69 ± 0·20 4·9 ± 9·6 14 12·5 ± 24·6a0 (b.min�1) 85 ± 11 83 ± 12 0·81 ± 0·14 4·9 ± 9·6 13 8·2 ± 16·2a1 (b.min�1) 81 ± 11 81 ± 11 0·80 ± 0·14 5·0 ± 9·8 14 8·8 ± 17·2Tau (s) 50 ± 13 56 ± 18 0·47 ± 0·29 11 ± 22 31 29·8 ± 58·4

Heart rate variability in the time domainHR (b.min�1) 84 ± 12 83 ± 11 0·86 ± 0·10 4·3 ± 8·4 12 7·4 ± 14·5SDNN (ms) 36 ± 26 36 ± 19 0·57 ± 0·25 15 ± 30 42 59·5 ± 117RMSSD (ms) 21 ± 14 24 ± 18 0·70 ± 0·20 8·7 ± 17·1 24 53·7 ± 105·2

Heart rate variability in the frequency domainHF (ms2) 29 ± 17 27 ± 16 0·70 ± 0·20 8·8 ± 17·2 24 44·4 ± 87·0LF (ms2) 47 ± 14 47 ± 15 0·58 ± 0·25 9·4 ± 18·4 26 28·1 ± 55·1LF+HF (ms2) 75 ± 16 73 ± 15 0·36 ± 0·32 13 ± 26 35 24·1 ± 47·1LF/HF 2·7 ± 2·6 2·7 ± 2·7 0·88 ± 0·09 0·9 ± 1·8 2·6 48·7 ± 95·5HF nu (%) 37 ± 17 36 ± 19 0·80 ± 0·14 8·3 ± 16·3 23 32·0 ± 62·8

SD, standard deviation; ICC, intraclass correlation coefficient; SEM, standard error of measurement; MD, minimum difference to be considered real;CL, confidence limits; CV: coefficient of variation HR, heart rate. For the others abbreviations see the ‘Methods’ section.



account for the limitations previously raised. We found a high

heterogeneity among HRR and HRV parameters. Although

there does not exist a specific scale for the interpretation of

SEM and CV, absolute reliability could be considered as mod-

erate, while relative reliability ranged from low to very high.

According to our results, Δ60 and HFnu represent the most

reliable HRR and HRV parameters to assess parasympathetic

reactivation. It should be kept in mind, however, that random

variations and the influence of measurement error remained

relatively high, making these parameters only moderately

reliable.

Relative reliability

Relative reliability represents the degree to which individuals

maintain their position in a sample with repeated measure-

ments (Weir, 2005). A classical measure of relative reliability

is the ICC, which permits estimation of the percentage of the

observed variance that is attributable to the true score vari-

ance. The higher the ICC the higher the relative reliability,

and the lower the influence of measurement error. We used

the scale of Munro (1997) to interpret the magnitude of the

ICC. Relative reliability was considered as very high when ICC

was higher than 0·90, high when it ranged from 0·70 to 0·89and moderate when it ranged from 0·50 to 0·69. Data were

considered as being not reliable when ICC was lower than

0·50.The ICC of parameters calculated in this study ranged from

0·12 to 0·87 for HRR and 0·14 to 0·97 for HRV. By the

exception of Tau, which was found to be more reliable after

the maximal graded exercise test, exercise intensity did not

appear to affect significantly random error. This observation is

in agreement with previous reports (Bosquet et al., 2008a;

Arduini et al., 2011). Currier (1990) suggested that an ICC

value higher than 0·80 was acceptable for clinical work. None

of the HRR parameters usually used to assess cardiac parasym-

pathetic reactivation reached this critical level. D60 appeared

to be the measure of choice, while T30 was clearly not reli-

able. Tau could also be considered after an exercise of maxi-

mal intensity, as its ICC was close to that calculated for Δ60(ICC = 0·74 and 0·77, respectively). These results are in

agreement with the report by Bosquet et al. (2008a), who

assessed the reliability of Δ60 and Tau after a maximal and a

submaximal intensity exercise in thirty physically active partic-

ipants (0·43 < ICC < 0·71), and the report by Arduini et al.

(2011), who assessed the reliability of these indices after two

exercises of different submaximal intensity in twenty healthy

participants (0·78 < ICC < 0·81). However, they differ some-

what from the results of Buchheit et al. (2008), who exam-

ined the reliability of T30, Δ60 and Tau after a submaximal

intensity exercise in fifteen adolescent handball players. The

main and most significant difference concerned T30, as it was

found to be highly reliable (ICC = 0·73) in the study by

Buchheit et al. (2008) and moderately reliable (0·55 < ICC

< 0·56) in the study by Arduini et al. (2011), while it was

simply not reliable in our participants (ICC = 0·23). Differ-

ences in the physical fitness and the age of the participants or

in the model used to calculate ICC probably account for this

difference (Bunc et al., 1988; Carnethon et al., 2005; Weir,

2005).

Among HRV parameters, SDNN was the single one to reach

an ICC > 0·80 after the maximal intensity exercise, while

Table 3 Inter-session reliability of autonomic indices during the recovery from submaximal intensity exercise (combined signal of session 1compared with the combined signal of session 2)

Parameter Test 1 (mean ± SD) Test 2 (mean ± SD) ICC (value ± CL) SEM (value ± CL) MD CV(%%) (value ± CL)

ExerciseHR (b.min�1) 165 ± 12 165 ± 13 0·85 ± 0·11 5·0 ± 9·8 14 4·2 ± 8·2

Heart rate indicesT30 (s) 151 ± 78 141 ± 51 0·51 ± 0·28 46 ± 90 129 45 ± 88D60 (b.min�1) 55 ± 8 55 ± 9 0·65 ± 0·22 9·3 ± 18·2 26 13·1 ± 25·6

Exponential modellinga0 (b.min�1) 84 ± 10 83 ± 12 0·80 ± 0·14 4·9 ± 9·6 14 7·2 ± 14·2a1 (b.min�1) 80 ± 11 81 ± 14 0·78 ± 0·15 5·8 ± 11·4 16 9·0 ± 17·7Tau (s) 58 ± 20 58 ± 20 0·49 ± 0·28 15 ± 29 40 35·5 ± 69·5

SD, standard deviation; ICC, intraclass correlation coefficient; SEM, standard error of measurement; MD, minimum difference to be considered real;CL, confidence limits; CV, coefficient of variation HR, heart rate; for the others abbreviations see the ‘Methods’ section.

Figure 2 Graphical representation of a combined signal.



HFnu and LF/HF reached this critical level after the submaxi-

mal intensity exercise. The magnitude of ICCs calculated in

our study was very close to that reported by Buchheit et al.

(2008) and confirms observations made at rest (Pitzalis et al.,

1996; Gerritsen et al., 2003).

Absolute reliability

Absolute reliability represents the degree to which repeated

measurements vary for individuals (Weir, 2005). Two classical

measures of absolute reliability are the (SEM), which provides

an index of the expected trial-to-trial noise in the data

(Yamamoto et al., 2001), and the CV that is a measure of the

discrepancy and expresses error as a percentage of the mean.

These measures are very useful in interventional studies to

determine whether an observed change on reassessment is

within the boundaries of measurement error or whether a true

change has occurred. The lower the SEM the lower the ran-

dom variations and the higher the precision of individual

scores. A high absolute reliability is an important characteristic

of a measure when assessing the treatment effect in individual

participants, as the change on reassessment needed to exceed

measurement error will be lower.

By the exception of T30, the random error was quite

acceptable in HRR parameters after maximal intensity exercise.

However, we observed an increase in the random variation of

Tau when exercise intensity was submaximal, while it

remained the same for other parameters. The presence of a

moderate to large random error is in accordance with previ-

ous reports (Bosquet et al., 2008a; Buchheit et al., 2008; Al

Haddad et al., 2011).

As consistently observed in the literature, either at rest

(Pinna et al., 2007) or after exercise cessation (Buchheit

et al., 2008), HRV parameters were associated with large

random variations, whatever the domain (i.e. time or fre-

quency). This poor absolute reliability may be explained in

part by the physiological profile of the participants, but also

by mood state, fatigue, diet, protocol and other factors

known to affect cardiac autonomic regulation (Pinna et al.,

2007). It should be noted, however, that testing conditions

in our study were highly standardized, as we controlled the

time of day (±1 h), physical activity of the previous day,

breathing frequency, the composition of the meal, environ-

mental conditions and the consumption of sympathomimetic

beverages.

Signal-to-noise ratio

Considering that the signal-to-noise ratio was an important

determinant of the confidence of HRR indices, and more par-

ticularly of the parameter estimates from exponential curve fit-

ting (i.e. Tau, a0 and a1), we wanted to determine whether

reliability would be improved by analysing the average

response of two exercise-to-rest transitions instead of a single

one. Contrary to our hypothesis, neither absolute nor relative

reliability was increased by this procedure, as test–retest ICC,

SEM and CV were similar to data obtained from a single test.

Practical implications

Although it deserves nuance, the moderate reliability of most

HRR and HRV parameters measures after exercise cessation

underscore the need for standardization. Digestion, tempera-

ture, noise, infections and pharmacological or non-pharmaco-

logical substances known to affect cardiac autonomic

regulation should be controlled to improve reliability (Bosquet

et al., 2008a). Other factors such as age, sex or physical fitness

are also known to affect postexercise HRR and HRV (Bunc

et al., 1988; Carnethon et al., 2005) and should be considered

in inclusion criteria to decrease inter-individual variability and

therefore decrease the sample size required to reach a given

statistical power in group comparison studies.

The moderate absolute reliability we found in this study

clearly impacts the minimum difference needed to be consid-

ered real (MD). MD represents the limit under which the

observed difference is within what would be expected in

repeated testing just attributed to the noise in measurement

(Weir, 2005). Whatever the exercise intensity, none of the

parameters used to assess parasympathetic reactivation permits

the detection of a real change when it is <20% for HRR and

<47% for HRV (see Tables 1 and 2 to obtain MD in the spe-

cific units of measurement). This is clearly a problem, as small

changes that may have important clinical implications, for

example, when a cut-off value exists (Cole et al., 1999), may

not be detectable by these measures.

Conclusion

Heart rate recovery and HRV parameters measured after the

cessation of exercise are commonly used to assess cardiac

parasympathetic reactivation. Although it would be hazardous

to generalize our results to other populations than the partici-

pants tested in this study, it appears that these measures are

poorly to moderately reliable. Care should therefore be taken

to implement highly standardized protocols and to control

some participant’s characteristics known to affect HRR and

HRV parameters in the inclusion criteria. Even though these

precautions should increase statistical power and decrease MD

in interventional studies, the reliability of measures used to

assess cardiac parasympathetic reactivation remains moderate.

Acknowledgments

No funding was received for this work from any organizations

or any institution.

Conflicts of interest

The authors have no conflicts of interest that are directly rele-

vant to the content of this manuscript.



References

Al Haddad H, Laursen PB, Chollet D, Ahmaidi

S, Buchheit M. Reliability of Resting and

Postexercise Heart Rate Measures. Int J SportsMed (2011); 32: 598–605.

Arduini A, Gomez-Cabrera MC, RomagnoliM. Reliability of different models to assess

heart rate recovery after submaximal bicycleexercise. J Sci Med Sport (2011); 14: 352–

357.Atkinson G, Nevill AM. Statistical methods for

assessing measurement error (reliability) invariables relevant to sports medicine. Sports

Med (1998); 26: 217–238.Bigger JT, Fleiss JL, Steinman RC, Rolnitzky

LM, Kleiger RE, Rottman JN. Frequency-domain measures of heart period variability

and mortality after myocardial-infarction.Circulation (1992); 85: 164–171.

Bosquet L, Gamelin FX, Berthoin S. Reliabilityof postexercise heart rate recovery. Int J

Sports Med (2008a); 29: 238–243.Bosquet L, Merkari S, Arvisais D, Aubert AE.

Is heart rate a convenient tool to monitorover-reaching? A systematic review of the

literature. Br J Sports Med (2008b); 42: 709–714.

Buchheit M, Laursen PB, Ahmaidi S. Parasym-pathetic reactivation after repeated sprint

exercise. Am J Physiol-Heart Circ Physiol(2007a); 293(1): H133–H141.

Buchheit M, Papelier Y, Laursen PB, AhmaidiS. Noninvasive assessment of cardiac para-

sympathetic function: postexercise heart raterecovery or heart rate variability? Am J Physiol-

Heart Circ Physiol (2007b); 293: H8–H10.

Buchheit M, Millet GP, Parisy A, Pourchez S,Laursen PB, Ahmaidi S. Supramaximal train-

ing and postexercise parasympathetic reacti-vation in adolescents. Med Sci Sports Exerc

(2008); 40: 362–371.Bunc V, Heller J, Leso J. Kinetics of heart rate

responses to exercise. J Sports Sci (1988); 6:39–48.

Carnethon MR, Jacobs DR, Sidney S, SternfeldB, Gidding SS, Shoushtari C, Liu K. A longi-

tudinal study of physical activity and heartrate recovery: CARDIA, 1987–1993. Med Sci

Sports Exerc (2005); 37: 606–612.Christenfeld N, Glynn LM, Gerin W. On the

reliable assessment of cardiovascular recovery:an application of curve-fitting techniques.

Psychophysiology (2000); 37: 543–550.Cole CR, Blackstone EH, Pashkow FJ, Snader

CE, Lauer MS. Heart-rate recovery immedi-ately after exercise as a predictor of

mortality. N Engl J Med (1999); 341: 1351–1357.

Currier D. Elements of reserch in physical therapy.pp. 1–205 (1990). Williams and Wilkins,

Baltimore.

Gamelin FX, Berthoin S, Bosquet L. Validity

of the polar S810 heart rate monitor tomeasure R-R intervals at rest. Med Sci Sports

Exerc (2006); 38: 887–893.Gerritsen J, TenVoorde BJ, Dekker JM, King-

ma R, Kostense PJ, Bouter LM, HeethaarRM. Measures of cardiovascular autonomic

nervous function: agreement, reproducibility,and reference values in middle age and elderly

subjects. Diabetologia (2003); 46: 330–338.Hedelin R, Kentta G, Wiklund U, Bjerle P,

Henriksson-Larsen K. Short-term overtrain-ing: effects on performance, circulatory

responses, and heart rate variability. Med SciSports Exerc (2000); 32: 1480–1484.

Huang P-H, Leu H-B, Chen J-W, Lin S-J.Heart rate recovery after exercise and endo-

thelial function–two important factors topredict cardiovascular events. Prev Cardiol

(2005); 8: 167–170.Huszczuk A, Whipp BJ, Wasserman K. A

respiratory gas-exchange simulator for rou-tine calibration in metabolic studies. Eur

Respir J (1990); 3: 465–468.Imai K, Sato H, Hori M, Kusuoka H, Ozaki

H, Yokoyama H, Takeda H, Inoue M,Kamada T. Vagally mediated heart rate

recovery after exercise is accelerated in ath-

letes but blunted in patients with chronicheart failure. J Am Coll Cardiol (1994); 24:

1529–1535.Kannankeril PJ, Goldberger JJ. Parasympa-

thetic effects on cardiac electrophysiologyduring exercise and recovery. Am J Physiol-

Heart Circ Physiol (2002); 282: H2091–H2098.

Kannankeril PJ, Le FK, Kadish AH, GoldbergerJJ. Parasympathetic effects on heart rate

recovery after exercise. J Investig Med (2004);52: 394–401.

Kingsley M, Lewis MJ, Marson RE. Compari-son of Polar 810 s and an ambulatory ECG

system for RR interval measurement duringprogressive exercise. Int J Sports Med (2005);

26: 39–44.Koga S, Shiojiri T, Kondo N. Measuring the

VO2 kinetics: the practicalities. In OxygenUptakes Kinetics (Ed Jones, AW, Poole, DC)

(2005); pp. 39–62. Rouledge, Taylor &Francis Inc, New York.

La Rovere MT, Bigger JT, Marcus FI, Mortara A,Schwartz PJ, Investigators A. Baroreflex sensi-

tivity and heart-rate variability in predictionof total cardiac mortality after myocardial

infarction. Lancet (1998); 351: 478–484.Lamberts RP, Swart J, Noakes TD, Lambert MI.

Changes in heart rate recovery after high-intensity training in well-trained cyclists. Eur

J Appl Physiol (2009); 105: 705–713.Lamberts RP, Swart J, Capostagno B, Noakes

TD, Lambert MI. Heart rate recovery as a

guide to monitor fatigue and predict

changes in performance parameters. Scand J

Med Sci Sports (2010); 20: 449–457.Lishner M, Akselrod S, Avi VM, Oz O, Divon

M, Ravid M. Spectral-analysis of heart-ratefluctuations - a noninvasive, sensitive method

for the early diagnosis of autonomic neu-ropathy in diabetes-mellitus. J Auton Nerv Syst

(1987); 19: 119–125.Midgley AW, McNaughton LR, Polman R,

Marchant D. Criteria for determination ofmaximal oxygen uptake - A brief critique

and recommendations for future research.Sports Med (2007); 37: 1019–1028.

Munro B. Statistical methods for health care research,3rd edn. (1997), pp. 1–444. Lippincott,

New York.Perini R, Orizio C, Comande A, Castellano M,

Beschi M, Veicsteinas A. Plasma norepi-nephrine and heart-rate dynamics during

recovery from submaximal exercise in man.Eur J Appl Physiol (1989); 58: 879–883.

Pinna GD, Maestri R, Torunski A, Danilowicz-Szymanowicz L, Szwoch M, La Rovere MT,

Raczak G. Heart rate variability measures: afresh look at reliability. Clin Sci (2007);

113: 131–140.Pitzalis MV, Mastropasqua F, Massari F, Forleo

C, DiMaggio M, Passantino A, Colombo R,

DiBiase M, Rizzon P. Short- and long-termreproducibility of time and frequency

domain heart rate variability measurementsin normal subjects. Cardiovasc Res (1996); 32:

226–233.Sandercock GRH, Bromley PD, Brodie DA.

The reliability of short-term measurementsof heart rate variability. Int J Cardiol (2005);

103: 238–247.Sandercock G. Normative values, reliablity

and sample size estimates in heart ratevariability. Clin Sci (2007); 113: 129–130.

Savin WM, Davidson DM, Haskell WL. Auto-nomic contribution to heart-rate recovery

from exercise in humans. J Appl Physiol(1982); 53: 1572–1575.

Schroeder EB, Whitsel EA, Evans GW, PrineasRJ, Chambless LE, Heiss G. Repeatability of

heart rate variability measures. J Electrocardiol(2004); 37: 163–172.

Sinnreich R, Kark JD, Friedlander Y, Sapozni-kov D, Luria MH. Five minute recordings of

heart rate variability for population studies:repeatability and age-sex characteristics.

Heart (1998); 80: 156–162.Task Force. Task Force of the European Soci-

ety of Cardiology the North American Soci-ety of Pacing Electrophysiology. Heart rate

variability - Standards of measurement,physiological interpretation, and clinical

use. Circulation (1996); 93: 1043–1065.Triposkiadis F, Ghiokas S, Skoularigis I, Kotsa-

kis A, Giannakoulis I, Thanopoulos V.

Cardiac adaptation to intensive training in



prepubertal swimmers. Eur J Clin Invest

(2002); 32: 16–23.Tsuji H, Larson MG, Venditti FJ, Manders ES,

Evans JC, Feldman CL, Levy D. Impact ofreduced heart rate variability on risk for

cardiac events - The Framingham HeartStudy. Circulation (1996); 94: 2850–2855.

Weir JP. Quantifying test-retest reliability

using the intraclass correlation coefficientand the SEM. J Strength Cond Res (2005); 19:

231–240.Yamamoto K, Miyachi M, Saitoh T, Yoshioka

A, Onodera S. Effects of endurance trainingon resting and post-exercise cardiac auto-

nomic control. Med Sci Sports Exerc (2001);

33: 1496–1502.Yataco AR, Fleisher LA, Katzel LI. Heart hate

variability and cardiovascular fitness insenior athletes. Am J Cardiol (1997); 80:

1389–1391.



reliability of heart rate measures used to assess post-exercise parasympathetic reactivation

Documents