Home
Search
Collections
Journals
About
Contact us
My IOPscience
Improvement of energy expenditure prediction from heart rate during running
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 Physiol. Meas. 35 253 (http://iopscience.iop.org/0967-3334/35/2/253) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 155.69.4.4 This content was downloaded on 21/05/2015 at 03:24
Please note that terms and conditions apply.
Institute of Physics and Engineering in Medicine Physiol. Meas. 35 (2014) 253–266
Physiological Measurement
doi:10.1088/0967-3334/35/2/253
Improvement of energy expenditure prediction from heart rate during running ´ emy ´ Keyne Charlot 1,4 , Jer Cornolo 2 , Rachel Borne 1 , 3 Julien Vincent Brugniaux , Jean-Paul Richalet 1 , ´ Pichon 1 Didier Chapelot 1 and Aurelien 1
Laboratoire ‘R´eponses cellulaires et fonctionnelles a` l’hypoxie’, Universit´e Paris 13, Sorbonne Paris Cit´e UFR SMBH EA2363, 74 rue Marcel Cachin, F-93017 Bobigny Cedex, France 2 Oxylane Research, 4, boulevard de Mons, Villeneuve d’Ascq, France 3 Neurovascular Research Laboratory, Faculty of Health, Science and Sport, University of Glamorgan, Glyntaff Campus, Pontypridd CF37 4AT, UK E-mail: [email protected] Received 14 May 2013, revised 10 December 2013 Accepted for publication 17 December 2013 Published 16 January 2014 Abstract
We aimed to develop new equations that predict exercise-induced energy expenditure (EE) more accurately than previous ones during running by including new parameters as fitness level, body composition and/or running intensity in addition to heart rate (HR). Original equations predicting EE were created from data obtained during three running intensities (25%, 50% and 70% of HR reserve) performed by 50 subjects. Five equations were conserved according to their accuracy assessed from error rates, interchangeability and correlations analyses: one containing only basic parameters, two containing VO2max or speed at VO2max and two including running speed with or without HR. Equations accuracy was further tested in an independent sample during a 40 min validation test at 50% of HR reserve. It appeared that: (1) the new basic equation was more accurate than pre-existing equations (R2 0.809 versus. 0,737 respectively); (2) the prediction of EE was more accurate with the addition of VO2max (R2 = 0.879); and (3) the equations containing running speed were the most accurate and were considered to have good agreement with indirect calorimetry. In conclusion, EE estimation during running might be significantly improved by including running speed in the predictive models, a parameter readily available with treadmill or GPS. Keywords: VO2max, indirect calorimetry, heart rate monitor, metabolic cost
4
Author to whom any correspondence should be addressed.
0967-3334/14/020253+01$33.00
© 2014 Institute of Physics and Engineering in Medicine Printed in the UK
253
K Charlot et al
Physiol. Meas. 35 (2014) 253
1. Introduction Laboratory devices (indirect calorimetry) are inconvenient to measure energy expended during an aerobic exercise and are therefore not applicable in general on the field. However, energy expenditure (EE) might be calculated from real-time heart rate (HR), which is easy to measure with affordable HR monitors. This estimation has been proposed because a linear relationship exists between oxygen uptake (VO2) and HR (Berggren and Hohwu Christensen 1950). However, HR alone is not sufficient to correctly assess EE (Lee et al 2011). To improve the EE estimation, several models including individual parameters such as age, weight, sex, resting HR (Hilloskorpi et al 1999, 2003) or even fitness level assessed by maximal oxygen uptake (VO2max) (Dugas et al 2005, Keytel et al 2005) have been proposed for running or cycling activities. Among them, the equations of Keytel et al (2005) designed for a large population (in age, weight and fitness level) are scientifically validated and frequently cited (Smolander et al 2011, Brage et al 2007). These equations were therefore considered as good references for new models. The only sustainable solution proposed actually to improve predictive equations accuracies seems to be complex models combining HR monitor and accelerometer (Brugniaux et al 2010) or only accelerometer (Brandes et al 2012, Koehler et al 2011). However, accelerometry combined or not with HR measurements gave contrasted results (Brandes et al 2012, Spierer et al 2011, Nichols et al 2010) and its high accuracy still remains to demonstrate. Moreover, accelerometers are more expensive than HR monitors alone and do not represent a popular solution. Thus, it might be wise to attempt to improve HR-based equations in introducing original parameters. Indeed, resting HR which has been suggested to reflect fitness level (Bjørnstad et al 1993, Sundberg and Elovainio 1982) is easy to measure and could be a good alternative to the expensive determination of VO2max. Similarly, the smallest speed at which VO2max is reached (S-VO2max) and real maximal HR (HRmax) are easy to measure in the field (Thebault et al 2011, Ahmaidi et al 1992). Body composition is also of great interest to specify EE (Choquette et al 2009) since it could predict the energetic cost of locomotion more precisely than absolute weight (Ruiz et al 2011). Finally, HR is an internal marker of exercise intensity but real-time running speed represents the external intensity and could be a major parameter of EE during exercise. Therefore, the objective of the present study was to improve the accuracy of EE estimation from HR based-equations by including new parameters describing fitness level, body composition and/or running intensity during flat running exercise in subjects.
2. Material and methods 2.1. Design
This study was composed of two arms. During part 1, 50 participants performed a 3-stage test from which several predictive equations of EE were created by using HR and additional parameters. During part 2, a separate sample of 15 participants was used to test the accuracy of the new equations. The subjects were informed about the nature and risks of the experimental procedure prior to giving their written informed consent. The protocol was approved by the representatives of the local arm of the National Ethics Committee (Comit´e de Protection des Personnes Ile-de- France n◦ X). 254
K Charlot et al
Physiol. Meas. 35 (2014) 253
2.2. Part 1: creation of the new predictive equations
Fifty healthy participants were recruited from the University of Paris 13. In order to comprehensively test the usefulness of the new parameters implemented in the equations, subjects with varied age, body composition and fitness level were involved. Demographic characteristics of the subjects are displayed in table 1. Height was measured by an electronic height gauge (Soehnle Nassau an der Lahn, Germany) and weight and fat mass (%) were assessed by an 8-electrode bioelectrical impedance analyzer (Tanita BC 418MA, Tanita Co). Body mass index (BMI) was then calculated. As required by the manufacturer, subjects had to urinate before the measurement to avoid any bias. The age-predicted HRmax was calculated using the Bruce formula (Bruce et al 1974). Resting HR was measured before the first test during the last 2 min of a 5 min resting period in a supine position. Real HRmax, VO2max and S-VO2max were measured at the end of the first test. A treadmill familiarization session was offered to the participants when required.
2.2.1. Subjects.
Subjects came to the laboratory either at 10:00 or at 16:00 at least 2 h after a meal to avoid meal-induced cardiorespiratory and metabolic fluctuations (Charlot et al 2011). Upon arrival, participants were equipped with a HR monitor (Suunto T6) and a facemask (Hans Rudolph, 8940 Series, Kansas City, KA, USA), and gas exchange was measured using a validated open-circuit spirometry (Vmax Encore,Viasys Healthcare, Palm Springs, CA, USA) (Cooper et al 2009). Subjects walked/ran on a treadmill (h/p/cosmos mercury med 4.0 Nussdorf-Traunstein, Deutschland) during 10 min at 25%, 10 min at 50% then 5 min at 75% of HR reserve (age-predicted HRmax – resting HR), respectively. Before each steady-state period, intensity was progressively increased over 5 min to reach the desired HR. Overall EE was deduced from the collected data of VO2 and VCO2 steady-state periods using the Jeukendrup and Wallis equations (Jeukendrup and Wallis 2005). These equations have been validated for moderate and high-intensity exercises since these equations take into account the part of the glucose oxidized derived from glycogen.
2.2.2. Three-stage test.
At the end of the 3-stage test, speed was increased by 0.5 km h−1 every 30 s (Noakes et al 1990) until two of the three following criteria were met: (1) plateau in VO2 despite increase in workload; (2) a respiratory exchange ratio > 1.15; and (3) the attainment of age-predicted HRmax calculated using the Bruce equation (Bruce et al 1974). Speed at VO2max was considered as the smallest speed at which VO2max was reached. Real HRmax was considered as the highest mean HR value measured during 15 s. It has been already shown that previous severe exercise (1 h at 82% VO2max on a treadmill) had no effect on average work, VO2max or real HRmax (Aitken and Thompson 1988).
2.2.3. Maximal oxygen uptake determination.
Parameters that may have significantly contributed to the relationship between HR and/or real-time running speed and EE during the 3-stage tests were determined by multiple regression analyses. We chose to measure or calculate parameters that were never included in models predicting EE and that might significantly influence EE. We therefore chose to add resting HR which might reflect fitness level (Bjørnstad et al 1993, Sundberg and Elovainio 1982). This parameter could be easily measured before exercise in a sitting position during at least 1 min and could be a surrogate to VO2max to define indirectly fitness level. Age-predicted HRmax is an indirect marker of age which is generally included in equations predicting EE (Keytel et al 2005, Hiilloskorpi et al 2003) since it influences resting metabolic rate (Krems et al 2005). We also included VO2maxrelated parameters like S-VO2max and HRmax. VO2max influences EE during exercise (Dugas
2.2.4. Creation of the predictive equations.
255
Physiol. Meas. 35 (2014) 253
Table 1. Subjects’ characteristics.
Part 1 Males (n = 25)
256
Age (y) Height (cm) Weight (kg) BMI (kg m−2) Fat mass (%) Resting HR (bpm) T-HRmax (bpm) R-HRmax (bpm) VO2max (mL min−1 kg−1) S-VO2max (km h−1)
32.2 178 75.3 23.7 13.5 66 189 189 53.3 16.7
± ± ± ± ± ± ± ± ± ±
8.4 (21–51) 8 (160–197) 9.2 (59–100) 2.9 (20.9–34.1) 5.3 (5.4–26.4) 11 (48–88) 6 (176–196) 6 (170–198) 8.6 (35–67) 2.1 (13–20)
Part 2
Females (n = 25) 31.9 165 62.6 22.8 26.5 78 189 191 42.1 12.2
± ± ± ± ± ± ± ± ± ±
7.8 (21–50) 6 (156–182) 11.7 (45–92) 3.8 (16.7–33.1) 7.8 (13.6–46.9) 13 (60–105) 5 (177–196) 6 (180–202) 6.3 (28–52) 1.9 (8–15)
Males (n = 8) 28.1 179 70.4 21.9 9.9 62 192 190 55.4 17.3
± ± ± ± ± ± ± ± ± ±
6.3 (20–37) 7 (173–187) 7.2 (65–83) 0.8 (20.8–23.8) 3.2 (7.3–15.5) 6 (51–72) 4.0 (186–196) 8 (181–198) 7.5 (44–65) 2.0 (14–19)
Females (n = 7) 24.2 168 59.8 21.2 21.7 79 194 193 44.5 13.6
± ± ± ± ± ± ± ± ± ±
3.9 (21–36) 9 (157–180) 11.7 (48–76) 2.8 (17.4–24.8) 4.0 (15.6–24.1) 16 (60–92) 3 (186–196) 4 (188–199) 4.3 (40–51) 1.4 (11–15)
Mean ± SD (lower value-higher value). BMI: body mass index. T-HRmax: theoretical maximal heart rate, R-HRmax: real maximal heart rate, VO2max: maximal oxygen uptake and S-VO2max: speed at VO2max.
K Charlot et al
K Charlot et al
Physiol. Meas. 35 (2014) 253
et al 2005, Keytel et al 2005). However, its determination requires laboratory equipment. Real HRmax and S-VO2max might be easily assessed with field tests without expensive devices (Thebault et al 2011, Ahmaidi et al 1992). Then, fat mass might predict EE more precisely than body weight (Ruiz et al 2011, Choquette et al 2009) and is assessable with impedance analyzers in health-related structures. Finally, we included real-time running speed since it represents the external intensity of exercise (Saibene and Minetti 2003, Heglund et al 1982) and might therefore largely influence EE. Real-time running speed might be measured with a treadmill, a GPS device or calculated with a chronometer during running on a known circuit. To create predictive equations, we determined a priori three groups of equations. Group A: equations containing HR and basic parameters; Group B: the equations containing HR, basic and additional parameters; and Group C: those containing all the parameters plus realtime running speed (in km h−1). Basic parameters are those that all users knew (age, height, weight), or that could be easily calculated (BMI, age-predicted HRmax) or measured (resting HR). Additional parameters might be determined in a laboratory or medical environment (VO2max and fat mass) or during a specific physical test (S-VO2max and real HRmax) and are therefore not known by every user. Analyses were then performed in each group. Redundant parameters (for example, age and age-predicted HRmax) could not be included in the same equation and all equations configurations were tested. Parameters that significantly explained variation of EE were conserved in the models. Based on these analyses, we therefore selected predictive equations of EE (kcal h−1) and assessed determination coefficient (R2) of the relation between estimated EE and measured EE from multiple regression analyses in each group. For multiple regression analysis, we used six predictors at maximum to assess EE. According to Cohen (1988), with six predictors, α = 0.05, anticipated effect size = 0.35 (large) and a level of desired statistical power = 0.8, a number of subject larger than 46 is needed. Therefore our sample of 50 subjects gives a high statistical power to our results. A p value less than 0.05 was considered to be statistically significant. Finally, among the multiple generated equations, five were finally preserved on all the equations generated: one basic, two containing VO2max or VO2max and two containing real-time running speed (figure 1). Details of the coefficients used in each equation are displayed in table 2. Other equations were excluded because determination coefficients of the models were not improved. 2.3. Part 2: validation of the new predictive equations on an independent sample
Fifteen participants were recruited to perform the second test. Indeed, with α = 0.05, a level of desired statistical power = 0.7, a SD of 110 kcal and an anticipated difference of 75 kcal (15% of difference for a 500 kcal-exercise), a number of subjects equal to 15 is needed. Anthropometric characteristics were similar to the participants of the part 1 (table 1). This test was performed with the same equipment as during part 1. Speed was progressively increased over 5 min to reach 50% of HR reserve, then was maintained during 40 min. Then, speed was increased by 0.5 km h−1 every 30 s to assess VO2max, S-VO2max and real HRmax. Comparisons of EE between the reference technique (indirect calorimetry) and the predictive equations were performed using ANOVA. The absolute mean error rate ([predicted EE - reference EE]/reference EE) was also calculated. Linear Pearson’s product-moment correlation analyses were performed to correlate quantitative variables. The Bland and Altman method (1986) was used to assess the agreement between the estimated and the reference EE. Confidence intervals (CIs) were determined for the mean bias and for the upper (UL) and lower limits (LL) of agreement. As requested by Bland and Altman (1986), we defined a priori 257
Physiol. Meas. 35 (2014) 253
Table 2. Details of the coefficients used in the new equations for each significant predictor obtained after multiple regression analyses.
Equation
HR Height Weight Sex Resting T-HRmax R-HRmax S-VO2max VO2max Running speed Intercept (bpm) (cm) (kg) (1 = M, 2 = F) HR (bpm) (bpm) (bpm) (km h−1) (mL min−1.kg−1) (km h−1)
258
Basic equation 171.62 6.87 BP + S-VO2max + R-HRmax 738.90 6.89 BP + VO2max + R-HRmax 113.20 6.88 BP + speed (without HR) −624.53 BP + speed −534.26 1.38
3.99
2.30 5.48 7.45 5.93 5.43
−139.89
−37.13
−4.26 −2.81 −3.15 1.09
−4.87
−9.50 −6.88
32.31 10.14 88.70 73.78
BP: basic parameter; HR: heart rate, T-HRmax: theoretical maximal HR, R-HRmax: real maximal heart rate, VO2max: maximal oxygen uptake, S-VO2max: speed at VO2max. By example, the basic equation is: EE (kcal h−1) = 171.62 + 6.87 × HR (bpm) + 3.99 × Height (cm) + 2.30 × weight (kg) −139.89 × Sex (1 or 2) −4.26 × resting HR (bpm) − 4.87 × T-HRmax (bpm).
K Charlot et al
K Charlot et al
Physiol. Meas. 35 (2014) 253
1,00 0,90
0.874 0.809
0.879
0.913
0.919
0.821
0,80 0.737
0,70
R²
0,60 0,50 0,40 0,30 0,20 0,10 0,00 Keytel (without VO2max)
BP
Keytel (with VO2max)
BP + S.VO2max BP + VO2max + + R-HRmax R-HRmax
BP + speed (without HR)
BP + speed
Figure 1. Determination coefficients (R2) of new predictive equations. BP: basic
parameter.
acceptable limits of agreement for interchangeability between estimated EE and reference EE. We considered that a variability of 15% could be considered as a good interchangeability with indirect assessments of EE from HR and additional parameters. This threshold corresponded to a 75 kcal-error for a 500-kcal exercise. This point will be discussed later. The same statistical analyses were performed with the EE estimated with the existing equations (Keytel et al 2005) containing basic parameters with or without VO2max to compare our new equations with already validated ones. 3. Results 3.1. Part 1: predictive equations
The coefficient of determination (R2 expressed in percent) was higher in the basic equation than in the Keytel equation without VO2max and slightly lower than the Keytel equation with VO2max (R2 values are displayed in figure 1). The R2 also increased as compared with basic equation with the inclusion of real maximal HR and S-VO2max or VO2max. Real-time running speed largely contributed to the prediction of EE and increased the coefficient of determination compared to the other equations. The model containing real-time speed but not HR was only slightly less accurate than the equation containing both. 3.2. Part 2: validation of predictive equations
Details of statistical tests are displayed in table 3. ANOVA revealed that estimated EE calculated by the Keytel equations underestimated reference EE measured during the validation 259
Physiol. Meas. 35 (2014) 253
Table 3. Summary of the part 2 of the study: the validation of the five most pertinent predictive equations.
260
Equation
EE estimated (kcal h−1)a
Basic equation Keytel (without VO2max) BP + S-VO2max + R-HRmax BP + VO2max + R-HRmax Keytel (with VO2max) BP + speed (without HR) BP + speed
512 405 513 505 380 494 500
± ± ± ± ± ± ±
119 50c 124 127 63c 93 92
Error rate (%) 14.3 19.4 10.6 10.4 24.1 6.9 7.1
± ± ± ± ± ± ±
13.1 11.6 10.5 10.3 9.6 4.8 4.4
Correlation coefficient (R)
Bias (CI)
Lower limits of agreement (CI)
Upper limits of agreement (CI)
% of limits of agreementb
0.657 0.632 0.809 0.840 0.812 0.933 0.924
−4 (−45 to 37) 104 (66 to 141) −5 (36 to 26) 3 (−27 to 32) 128 (98 to 158) 14 (−3 to 32) 9 (−10 to 28)
−191 (−262 to −120) −69 (−135 to -4) −149 (−204 to −95) −132 (−184 to −81) −9 (−61 to 43) −67 (−97 to -36) −77 (−109 to -45)
183 (113 to 254) 276 (211 to 342) 139 (86 to 194) 138 (87 to 189) 265 (213 to 317) 95 (65 to 126) 94 (62 to 126)
36.8 34.0 28.4 26.6 27.0 15.9 16.8
Bias: mean difference between EE measured and EE predicted. Lower and upper limits of agreement: bias ± 1.96 × SD. BP: basic parameters. ALA: acceptable limits of agreement (i.e. 10%). CI: confidence interval.% of limits of agreement: differences between limits of agreement and the mean energy expenditure (EE) measured. BP: basic parameter.a The reference value of energy expenditure (measured with indirect calorimetry) was 508 ± 111 kcal. b To conclude to interchangeability, the limits of agreement had to remain under 15%. c p < 0.001 different from reference EE and predicted EE with new equations.
K Charlot et al
K Charlot et al
Physiol. Meas. 35 (2014) 253
Keytel
+ R-HRmax
(with VO2max)
100
100
100
100
0
0
0
0
-100
-100
-100
-100
-200
-200
-200
-200
-300
-300
-300
-300
750
500
250
R² = 0,5623
R² = 0,5054
750
200
750
200
500
200
250
200
750
300
500
300
250
300
500
(without VO2max)
300
Mean Reference EE – Estimated EE (kcal.h-1) BP + VO2max
BP + speed BP + speed
(without HR)
100
100
0
0
0
-100
-100
-100
-200
-200
-200
-300
-300
-300
Limits of agreement Bias
750
100
Acceptable limits of agreement
500
200
750
200
500
200
250
300
750
300
500
300
250
+ R-HRmax
250
Difference Reference EE – Estimated EE (kcal.h-1)
BP + S-VO2max
250
Keytel
Basic equation
Mean Reference EE – Estimated EE (kcal.h-1) Figure 2. Interchangeability between real energy expenditure (EE) and estimated EE
according to the Bland and Altman analysis (6). Upper and lower limits of agreement had to contain acceptable limits (25% of mean real energy expenditure) to conclude a good interchangeability. BP: basic parameter.
test without (−18.2 ± 13.6%, p < 0.001) and with VO2max (−24.1 ± 9.6%; p < 0.001). Estimated EE from the new equations generated were not different from the reference EE. The error rate was lower in the basic equation than both Keytel equations and decreased with the addition of real HRmax, VO2max, S-VO2max and/or real-time running speed. All estimated EE were significantly correlated with the reference EE (p < 0.001 for all). Correlation coefficients of the Keytel equations were close to those of the basic equation and models containing VO2max and S-VO2max, but lower than the equations containing real-time running speed. Finally, interchangeability analyses (figure 2) revealed that the limits of agreement were close to the acceptable ones only with the equations containing speed (15.9 and 16.5 versus 15.0% respectively). Moreover, the Keytel equations showed a systematic increase in the bias with the increase of the measured EE (p < 0.001). 261
K Charlot et al
Physiol. Meas. 35 (2014) 253
4. Discussion In this study, we created HR-based predictive equations of EE during running with the inclusion of original parameters representing body composition, fitness level and/or real-time running speed. The results showed that our equation including basic parameters was at least as accurate as the formerly validated equation (Keytel et al 2005). Even more interestingly, inclusion of maximal oxygen uptake, speed at maximal oxygen uptake and real maximal HR, but not fat mass, significantly improved accuracy. Finally, real-time running speed was the parameter that contributed the most to explaining the relationship between HR and EE and this predictor might even replace HR without loss of accuracy during running without slope. 4.1. Basic equation
One team has more particularly contributed to the creation of models predicting exercise EE (Dugas et al 2005, Keytel et al 2005, Hiilloskorpi et al 2003, Hilloskorpi et al 1999). Their models that did not contain VO2max accounted between 73.4% and 78.4% of the variance in EE and those with VO2max accounted between 83.3% and 84.1% of this variance. Given that unusual parameters were introduced in the present paper in most of the new equations, it is difficult to directly compare our equations with the existing ones (Keytel et al 2005). However, we still could compare our basic equation with those of Keytel with or without VO2max. It appeared that the determination coefficients (R2) of our equations were higher than Keytel equations (figure 1) and our basic equation without VO2max was even close to the most advanced Keytel equation (i.e. with VO2max). These results were confirmed during the validation test in which error rates were higher with both Keytel’s equations than the new created equations. One might explain these differences with the fact that age was not included in our equations but replaced by age-predicted HRmax. Moreover, we chose to include resting HR in basic parameters because it is easy to determine before exercise with a HR monitor and correctly corresponds to fitness level (Bjørnstad et al 1993, Sundberg and Elovainio 1982). The fact that our basic equation was almost as accurate as the Keytel equation with VO2max suggests that resting HR is an acceptable index of fitness level. These two differences might have improved the model and should be considered to estimate EE from HR measurements. Protocol choices might also explain the differences. Based on the assumption that the exercise modalities do not influence the prediction of EE (Hilloskorpi et al 1999), Keytel et al (2005), with a similar population than ours, used a combination of tests on treadmill and on ergocycle to generate their equations. They further extrapolated the usefulness of their equations to rowing activities. However, it is well established that differences in energetic cost of locomotion exist between these two modes of exercise. For instance, fat oxidation is higher during running than cycling (Capostagno and Bosch 2010) and HR might differ despite the same relative intensity (Millet et al 2009). It is, therefore, likely that the generation of equations combining running and cycling data might have induced some additional degree of variability. 4.2. Advanced equations
The models containing only basic parameters are very important for the general population since they provide a good estimation of EE with parameters easy to implement into the HR monitor. Yet, the addition of more advanced parameters could be of interest for specific users if a major gain in precision can be obtained. For instance, competitive athletes are 262
K Charlot et al
Physiol. Meas. 35 (2014) 253
more likely to have access to parameters such as VO2max, S-VO2max or real HRmax. Equations included such parameters were deemed more accurate than the basic one, explaining ∼87% of the variance and generating only 10.5% of errors during the validation test. However, the interchangeability with the real EE was only modest since the limits of agreement (26.6 to 28.4%) exceeded a priori acceptable limits fixed at 15%. Moreover, addition of fat mass in the models was not pertinent suggesting that body composition has little influence on EE during exercise when HR and basic parameters are considered. This was unexpected since fat mass is known to influence running economy and therefore EE during exercise (Bunc 2000). It should be noticed that VO2max contributed more to the estimation of EE than the other parameters but the difference with S-VO2max was sparse and equations containing VO2max might be confidently replaced by those containing S-VO2max, which is easier to assess since it does not require a laboratory set up and could be assessed on the field. Since VO2max and consequently S-VO2max were determined following the steady-state test, one might argue that S-VO2max could have been underestimated since it has been proposed that the duration of the steps as well as the speed increment during the VO2max test are to be considered to determine S-VO2max (Roffey et al 2007). This limitation should be taken into account for people using the equation including the S-VO2max.
4.3. Equations with speed
There are several means to determine real-time running speed: with an accelerometer, with a GPS included in smartphones, when using a treadmill or even by calculating the mean speed from time and distance on a known circuit. In the present study, real-time running speed, which might be an index of the external mechanical work of locomotion (Saibene and Minetti 2003, Heglund et al 1982), was a strong predictor of EE while inclusion of other parameters (VO2max, S-VO2max or real HRmax) in the models did not enhance EE prediction in this case. The equation containing HR, weight, sex and real-time running speed improved the variance by 4.0% compared to the most accurate model of the advanced equations (basic parameters + VO2max + real HRmax) and by 11.0% compared to the basic equation (91.9% versus 87.9% versus 80.9%, respectively). Results from the validation test confirmed this tendency with error rate being only 7.1%. The limits of agreement were still slightly higher than the acceptable ones (16.8% versus 15.0%, respectively), therefore, meaning that the agreement between EE measured by indirect calorimetry and EE estimated by the model containing real-time running speed is at best correct. Moreover, it appeared that equations containing real-time running speed but not HR show a similar level of accuracy. This means that EE might be estimated accurately using only resting HR (needed in this model), i.e. without a measurement of internal load during exercise but only of external load. The addition of supplementary parameters marginally increased R2 suggesting that the equations containing basic parameters and real-time running speed with or without HR are sufficient. We therefore recommend using the equations containing real-time running speed when this measure is available. It should be noted that accuracies of all models presented in this study were larger than those from triaxial accelerometers during treadmill exercises. Indeed, R2 with accelerometry was only ∼ 0.77 (Brandes et al 2012, Howe et al 2009) and limits of agreement were approximately 70% (Brandes et al 2012) while we obtained R2 values from 0.81 to 0.92 and limits of agreement between 16.8% and 36.8%. This clearly indicated that HR-based EE estimations should be preferred over accelerometer-based models in walking and running activities. 263
K Charlot et al
Physiol. Meas. 35 (2014) 253
4.4. Limitations
These results and the application of these equations need to be tempered. Firstly, the application of these equations should be restricted to running without slope. Naturally, implementing realtime running speed into the equations prevent use beyond running without slope. Secondly, these equations have been tested for intensities between 25% and 75% of HR reserve or between 40% and 75% of VO2max. The relationship between HR and VO2 during exercise is linear from about 20% of HR reserve (Ceesay et al 1989). Thus, the accuracy of these equations might diminish at lower intensities. Moreover, given that indirect calorimetry starts to deviate from 75% of VO2max, the relationship between HR and VO2 seems uncertain from this level (Jeukendrup and Wallis 2005). Thirdly, these equations are suitable for a large population (healthy active and sedentary males or females between 18 and 50 years old) but need to be confirmed for specific individuals such as elderly, obese people or athletes. 5. Conclusion These new equations represent an improvement over existing ones in the estimation of energy expenditure (EE) from heart rate (HR) during running without slope. Indeed, inclusions of original parameters very easy to measure or calculate (heart rate at rest or age-related maximal HR), measurable with field tests (and speed at VO2max and real maximal HR) and/or available on treadmill or GPS devices (real-time running speed) improve the agreement between estimated EE from the new created equations and real EE assessed with indirect calorimetry. Based on the significant increase in accuracy, we recommend using equations containing real-time running speed as much as possible. Moreover, we recommend using these equations for intensities between 25% and 75% of HR reserve. Acknowledgments We hereby disclaim any competing interests. This study has been supported by Oxylane research (Villeneuve d’Ascq, France) but KC, RB and AP from the University of Paris 13 performed all the experiments and analyzed the data independently. References Ahmaidi S, Collomp K, Caillaud C and Prefaut C 1992 Maximal and functional aerobic capacity as assessed by two graduated field methods in comparison to laboratory exercise testing in moderately trained subjects Int. J. Sports Med. 13 243–8 Aitken J C and Thompson J 1988 The effects of previous severe exercise upon the respiratory VCO2/VO2 exchange ratio as a predictor of maximum oxygen uptake Eur. J. Appl. Physiol. Occup. Physiol. 57 720–5 Berggren G and Hohwu Christensen E 1950 Heart rate and body temperature as indices of metabolic rate during work Arbeitsphysiologie 14 255–60 Bjørnstad H, Storstein L, Dyre Meen H and Hals O 1993 Electrocardiographic findings according to level of fitness and sport activity Cardiology 83 268–79 Bland J M and Altman D G 1986 Statistical methods for assessing agreement between two methods of clinical measurement Lancet 1 307–10 Brage S, Ekelund U, Brage N, Hennings M A, Froberg K, Franks P W and Wareham N J 2007 Hierarchy of individual calibration levels for heart rate and accelerometry to measure physical activity J. Appl. Physiol. 103 682–92 Brandes M, Van Hees V T, Hann¨over V and Brage S 2012 Estimating energy expenditure from raw accelerometry in three types of locomotion Med. Sci. Sports Exerc. 44 2235–42 264
K Charlot et al
Physiol. Meas. 35 (2014) 253
Bruce R A, Fisher L D, Cooper M N and Gey G 1974 Separation of effects of cardiovascular disease and age on ventricular function with maximal exercise Am. J. Cardiol. 34 757–63 Brugniaux J V, Niva A, Pulkkinen I, Laukkanen R M, Richalet J P and Pichon A P 2010 Polar activity watch 200: a new device to accurately assess energy expenditure Br. J. Sports Med. 44 245–9 Bunc V 2000 Energy cost of treadmill running in non-trained females differing in body fat J. Sports Med. Phys. Fitness 40 290–6 Capostagno B and Bosch A 2010 Higher fat oxidation in running than cycling at the same exercise intensities Int. J. Sport Nutr. Exerc. Metab. 20 44–55 Ceesay S M, Prentice A M, Day K C, Murgatroyd P R, Goldberg G R, Scott W and Spurr G B 1989 The use of heart rate monitoring in the estimation of energy expenditure: a validation study using indirect whole-body calorimetry Br. J. Nutr. 61 175–86 Charlot K, Pichon A and Chapelot D 2011 Exercise prior to a freely requested meal modifies pre and postprandial glucose profile, substrate oxidation and sympathovagal balance Nutr. Metab. Lond. 8 66 Choquette S, Chuin A, Lalancette D A, Brochu M and Dionne I J 2009 Predicting energy expenditure in elders with the metabolic cost of activities Med. Sci. Sports Exerc. 41 1915–20 Cohen J 1988 Statistical Power Analysis for the Behavioral Sciences 2nd edn (Hillsdale, NJ: Lawrence Erlbaum Associates) p 567 Cooper J A, Watras A C, O’Brien M J, Luke A, Dobratz J R, Earthman C P and Schoeller D A 2009 Assessing validity and reliability of resting metabolic rate in six gas analysis systems J. Am. Diet. Assoc. 109 128–32 Dugas L R, van der Merwe L, Odendaal H, Noakes T D and Lambert E V 2005 A novel energy expenditure prediction equation for intermittent physical activity Med. Sci. Sports Exerc. 37 2154–61 Heglund N C, Fedak M A, Taylor C R and Cavagna G A 1982 Energetics and mechanics of terrestrial locomotion. IV. Total mechanical energy changes as a function of speed and body size in birds and mammals J. Exp. Biol. 97 57–66 Hilloskorpi H, Fogelholm M, Laukkanen R, Pasanen M, Oja P, Manttari A and Natri A 1999 Factors affecting the relation between heart rate and energy expenditure during exercise Int. J. Sports Med. 20 438–43 Hiilloskorpi H K, Pasanen M E, Fogelholm M G, Laukkanen R M and Manttari A T 2003 Use of heart rate to predict energy expenditure from low to high activity levels Int. J. Sports Med. 24 332–6 Howe C A, Staudenmayer J W and Freedson P S 2009 Accelerometer prediction of energy expenditure: vector magnitude versus vertical axis Med. Sci. Sports Exerc. 41 2199–206 Jeukendrup A E and Wallis G A 2005 Measurement of substrate oxidation during exercise by means of gas exchange measurements Int. J. Sports Med. 26 (Suppl. 1) S28–37 Keytel L R, Goedecke J H, Noakes T D, Hiiloskorpi H, Laukkanen R, van der Merwe L and Lambert E V 2005 Prediction of energy expenditure from heart rate monitoring during submaximal exercise J. Sports Sci. 23 289–97 Koehler K, Braun H, de Mar´ees M, Fusch G, Fusch C and Schaenzer W 2011 Assessing energy expenditure in male endurance athletes: validity of the SenseWear Armband Med. Sci. Sports Exerc. 43 1328–33 Krems C, L¨uhrmann P M, Strassburg A, Hartmann B and Neuh¨auser-Berthold M 2005 Lower resting metabolic rate in the elderly may not be entirely due to changes in body composition Eur. J. Clin. Nutr. 59 255–62 Lee C M, Gorelick M and Mendoza A 2011 Accuracy of an infrared LED device to measure heart rate and energy expenditure during rest and exercise J. Sports Sci. 29 1645–53 Millet G P, Vleck V E and Bentley D J 2009 Physiological differences between cycling and running: lessons from triathletes Sports Med. 39 179–206 Nichols J F, Aralis H, Merino S G, Barrack M T, Stalker-Fader L and Rauh M J 2010 Utility of the actiheart accelerometer for estimating exercise energy expenditure in female adolescent runners Int. J. Sport Nutr. Exerc. Metab. 20 487–95 Noakes T D, Myburgh K H and Schall R 1990 Peak treadmill running velocity during the VO2 max test predicts running performance J. Sports Sci. 8 35–45 Roffey D M, Byrne N M and Hills A P 2007 Effect of stage duration on physiological variables commonly used to determine maximum aerobic performance during cycle ergometry J. Sports Sci. 25 1325–35 Ruiz J R, Ortega F B, Rodriguez G, Alkorta P and Labayen I 2011 Validity of resting energy expenditure predictive equations before and after an energy-restricted diet intervention in obese women PLoS One 6 e23759 265
K Charlot et al
Physiol. Meas. 35 (2014) 253
Saibene F and Minetti A E 2003 Biomechanical and physiological aspects of legged locomotion in humans Eur. J. Appl. Physiol. 88 297–316 Smolander J, Ajoviita M, Juuti T, Nummela A and Rusko H 2011 Estimating oxygen consumption from heart rate and heart rate variability without individual calibration Clin. Physiol. Funct. Imaging 31 266–71 Spierer D K, Hagins M, Rundle A and Pappas E 2011 A comparison of energy expenditure estimates from the Actiheart and Actical physical activity monitors during low intensity activities, walking, and jogging Eur. J. Appl. Physiol. 111 659–67 Sundberg S and Elovainio R 1982 Resting ECG in athletic and non-athletic adolescent boys: correlations with heart volume and cardiorespiratory fitness Clin. Physiol. 2 419–26 Thebault N, Leger L A and Passelergue P 2011 Repeated-sprint ability and aerobic fitness J. Strength Cond. Res. 25 2857–65
266
Search
Collections
Journals
About
Contact us
My IOPscience
Improvement of energy expenditure prediction from heart rate during running
This content has been downloaded from IOPscience. Please scroll down to see the full text. 2014 Physiol. Meas. 35 253 (http://iopscience.iop.org/0967-3334/35/2/253) View the table of contents for this issue, or go to the journal homepage for more
Download details: IP Address: 155.69.4.4 This content was downloaded on 21/05/2015 at 03:24
Please note that terms and conditions apply.
Institute of Physics and Engineering in Medicine Physiol. Meas. 35 (2014) 253–266
Physiological Measurement
doi:10.1088/0967-3334/35/2/253
Improvement of energy expenditure prediction from heart rate during running ´ emy ´ Keyne Charlot 1,4 , Jer Cornolo 2 , Rachel Borne 1 , 3 Julien Vincent Brugniaux , Jean-Paul Richalet 1 , ´ Pichon 1 Didier Chapelot 1 and Aurelien 1
Laboratoire ‘R´eponses cellulaires et fonctionnelles a` l’hypoxie’, Universit´e Paris 13, Sorbonne Paris Cit´e UFR SMBH EA2363, 74 rue Marcel Cachin, F-93017 Bobigny Cedex, France 2 Oxylane Research, 4, boulevard de Mons, Villeneuve d’Ascq, France 3 Neurovascular Research Laboratory, Faculty of Health, Science and Sport, University of Glamorgan, Glyntaff Campus, Pontypridd CF37 4AT, UK E-mail: [email protected] Received 14 May 2013, revised 10 December 2013 Accepted for publication 17 December 2013 Published 16 January 2014 Abstract
We aimed to develop new equations that predict exercise-induced energy expenditure (EE) more accurately than previous ones during running by including new parameters as fitness level, body composition and/or running intensity in addition to heart rate (HR). Original equations predicting EE were created from data obtained during three running intensities (25%, 50% and 70% of HR reserve) performed by 50 subjects. Five equations were conserved according to their accuracy assessed from error rates, interchangeability and correlations analyses: one containing only basic parameters, two containing VO2max or speed at VO2max and two including running speed with or without HR. Equations accuracy was further tested in an independent sample during a 40 min validation test at 50% of HR reserve. It appeared that: (1) the new basic equation was more accurate than pre-existing equations (R2 0.809 versus. 0,737 respectively); (2) the prediction of EE was more accurate with the addition of VO2max (R2 = 0.879); and (3) the equations containing running speed were the most accurate and were considered to have good agreement with indirect calorimetry. In conclusion, EE estimation during running might be significantly improved by including running speed in the predictive models, a parameter readily available with treadmill or GPS. Keywords: VO2max, indirect calorimetry, heart rate monitor, metabolic cost
4
Author to whom any correspondence should be addressed.
0967-3334/14/020253+01$33.00
© 2014 Institute of Physics and Engineering in Medicine Printed in the UK
253
K Charlot et al
Physiol. Meas. 35 (2014) 253
1. Introduction Laboratory devices (indirect calorimetry) are inconvenient to measure energy expended during an aerobic exercise and are therefore not applicable in general on the field. However, energy expenditure (EE) might be calculated from real-time heart rate (HR), which is easy to measure with affordable HR monitors. This estimation has been proposed because a linear relationship exists between oxygen uptake (VO2) and HR (Berggren and Hohwu Christensen 1950). However, HR alone is not sufficient to correctly assess EE (Lee et al 2011). To improve the EE estimation, several models including individual parameters such as age, weight, sex, resting HR (Hilloskorpi et al 1999, 2003) or even fitness level assessed by maximal oxygen uptake (VO2max) (Dugas et al 2005, Keytel et al 2005) have been proposed for running or cycling activities. Among them, the equations of Keytel et al (2005) designed for a large population (in age, weight and fitness level) are scientifically validated and frequently cited (Smolander et al 2011, Brage et al 2007). These equations were therefore considered as good references for new models. The only sustainable solution proposed actually to improve predictive equations accuracies seems to be complex models combining HR monitor and accelerometer (Brugniaux et al 2010) or only accelerometer (Brandes et al 2012, Koehler et al 2011). However, accelerometry combined or not with HR measurements gave contrasted results (Brandes et al 2012, Spierer et al 2011, Nichols et al 2010) and its high accuracy still remains to demonstrate. Moreover, accelerometers are more expensive than HR monitors alone and do not represent a popular solution. Thus, it might be wise to attempt to improve HR-based equations in introducing original parameters. Indeed, resting HR which has been suggested to reflect fitness level (Bjørnstad et al 1993, Sundberg and Elovainio 1982) is easy to measure and could be a good alternative to the expensive determination of VO2max. Similarly, the smallest speed at which VO2max is reached (S-VO2max) and real maximal HR (HRmax) are easy to measure in the field (Thebault et al 2011, Ahmaidi et al 1992). Body composition is also of great interest to specify EE (Choquette et al 2009) since it could predict the energetic cost of locomotion more precisely than absolute weight (Ruiz et al 2011). Finally, HR is an internal marker of exercise intensity but real-time running speed represents the external intensity and could be a major parameter of EE during exercise. Therefore, the objective of the present study was to improve the accuracy of EE estimation from HR based-equations by including new parameters describing fitness level, body composition and/or running intensity during flat running exercise in subjects.
2. Material and methods 2.1. Design
This study was composed of two arms. During part 1, 50 participants performed a 3-stage test from which several predictive equations of EE were created by using HR and additional parameters. During part 2, a separate sample of 15 participants was used to test the accuracy of the new equations. The subjects were informed about the nature and risks of the experimental procedure prior to giving their written informed consent. The protocol was approved by the representatives of the local arm of the National Ethics Committee (Comit´e de Protection des Personnes Ile-de- France n◦ X). 254
K Charlot et al
Physiol. Meas. 35 (2014) 253
2.2. Part 1: creation of the new predictive equations
Fifty healthy participants were recruited from the University of Paris 13. In order to comprehensively test the usefulness of the new parameters implemented in the equations, subjects with varied age, body composition and fitness level were involved. Demographic characteristics of the subjects are displayed in table 1. Height was measured by an electronic height gauge (Soehnle Nassau an der Lahn, Germany) and weight and fat mass (%) were assessed by an 8-electrode bioelectrical impedance analyzer (Tanita BC 418MA, Tanita Co). Body mass index (BMI) was then calculated. As required by the manufacturer, subjects had to urinate before the measurement to avoid any bias. The age-predicted HRmax was calculated using the Bruce formula (Bruce et al 1974). Resting HR was measured before the first test during the last 2 min of a 5 min resting period in a supine position. Real HRmax, VO2max and S-VO2max were measured at the end of the first test. A treadmill familiarization session was offered to the participants when required.
2.2.1. Subjects.
Subjects came to the laboratory either at 10:00 or at 16:00 at least 2 h after a meal to avoid meal-induced cardiorespiratory and metabolic fluctuations (Charlot et al 2011). Upon arrival, participants were equipped with a HR monitor (Suunto T6) and a facemask (Hans Rudolph, 8940 Series, Kansas City, KA, USA), and gas exchange was measured using a validated open-circuit spirometry (Vmax Encore,Viasys Healthcare, Palm Springs, CA, USA) (Cooper et al 2009). Subjects walked/ran on a treadmill (h/p/cosmos mercury med 4.0 Nussdorf-Traunstein, Deutschland) during 10 min at 25%, 10 min at 50% then 5 min at 75% of HR reserve (age-predicted HRmax – resting HR), respectively. Before each steady-state period, intensity was progressively increased over 5 min to reach the desired HR. Overall EE was deduced from the collected data of VO2 and VCO2 steady-state periods using the Jeukendrup and Wallis equations (Jeukendrup and Wallis 2005). These equations have been validated for moderate and high-intensity exercises since these equations take into account the part of the glucose oxidized derived from glycogen.
2.2.2. Three-stage test.
At the end of the 3-stage test, speed was increased by 0.5 km h−1 every 30 s (Noakes et al 1990) until two of the three following criteria were met: (1) plateau in VO2 despite increase in workload; (2) a respiratory exchange ratio > 1.15; and (3) the attainment of age-predicted HRmax calculated using the Bruce equation (Bruce et al 1974). Speed at VO2max was considered as the smallest speed at which VO2max was reached. Real HRmax was considered as the highest mean HR value measured during 15 s. It has been already shown that previous severe exercise (1 h at 82% VO2max on a treadmill) had no effect on average work, VO2max or real HRmax (Aitken and Thompson 1988).
2.2.3. Maximal oxygen uptake determination.
Parameters that may have significantly contributed to the relationship between HR and/or real-time running speed and EE during the 3-stage tests were determined by multiple regression analyses. We chose to measure or calculate parameters that were never included in models predicting EE and that might significantly influence EE. We therefore chose to add resting HR which might reflect fitness level (Bjørnstad et al 1993, Sundberg and Elovainio 1982). This parameter could be easily measured before exercise in a sitting position during at least 1 min and could be a surrogate to VO2max to define indirectly fitness level. Age-predicted HRmax is an indirect marker of age which is generally included in equations predicting EE (Keytel et al 2005, Hiilloskorpi et al 2003) since it influences resting metabolic rate (Krems et al 2005). We also included VO2maxrelated parameters like S-VO2max and HRmax. VO2max influences EE during exercise (Dugas
2.2.4. Creation of the predictive equations.
255
Physiol. Meas. 35 (2014) 253
Table 1. Subjects’ characteristics.
Part 1 Males (n = 25)
256
Age (y) Height (cm) Weight (kg) BMI (kg m−2) Fat mass (%) Resting HR (bpm) T-HRmax (bpm) R-HRmax (bpm) VO2max (mL min−1 kg−1) S-VO2max (km h−1)
32.2 178 75.3 23.7 13.5 66 189 189 53.3 16.7
± ± ± ± ± ± ± ± ± ±
8.4 (21–51) 8 (160–197) 9.2 (59–100) 2.9 (20.9–34.1) 5.3 (5.4–26.4) 11 (48–88) 6 (176–196) 6 (170–198) 8.6 (35–67) 2.1 (13–20)
Part 2
Females (n = 25) 31.9 165 62.6 22.8 26.5 78 189 191 42.1 12.2
± ± ± ± ± ± ± ± ± ±
7.8 (21–50) 6 (156–182) 11.7 (45–92) 3.8 (16.7–33.1) 7.8 (13.6–46.9) 13 (60–105) 5 (177–196) 6 (180–202) 6.3 (28–52) 1.9 (8–15)
Males (n = 8) 28.1 179 70.4 21.9 9.9 62 192 190 55.4 17.3
± ± ± ± ± ± ± ± ± ±
6.3 (20–37) 7 (173–187) 7.2 (65–83) 0.8 (20.8–23.8) 3.2 (7.3–15.5) 6 (51–72) 4.0 (186–196) 8 (181–198) 7.5 (44–65) 2.0 (14–19)
Females (n = 7) 24.2 168 59.8 21.2 21.7 79 194 193 44.5 13.6
± ± ± ± ± ± ± ± ± ±
3.9 (21–36) 9 (157–180) 11.7 (48–76) 2.8 (17.4–24.8) 4.0 (15.6–24.1) 16 (60–92) 3 (186–196) 4 (188–199) 4.3 (40–51) 1.4 (11–15)
Mean ± SD (lower value-higher value). BMI: body mass index. T-HRmax: theoretical maximal heart rate, R-HRmax: real maximal heart rate, VO2max: maximal oxygen uptake and S-VO2max: speed at VO2max.
K Charlot et al
K Charlot et al
Physiol. Meas. 35 (2014) 253
et al 2005, Keytel et al 2005). However, its determination requires laboratory equipment. Real HRmax and S-VO2max might be easily assessed with field tests without expensive devices (Thebault et al 2011, Ahmaidi et al 1992). Then, fat mass might predict EE more precisely than body weight (Ruiz et al 2011, Choquette et al 2009) and is assessable with impedance analyzers in health-related structures. Finally, we included real-time running speed since it represents the external intensity of exercise (Saibene and Minetti 2003, Heglund et al 1982) and might therefore largely influence EE. Real-time running speed might be measured with a treadmill, a GPS device or calculated with a chronometer during running on a known circuit. To create predictive equations, we determined a priori three groups of equations. Group A: equations containing HR and basic parameters; Group B: the equations containing HR, basic and additional parameters; and Group C: those containing all the parameters plus realtime running speed (in km h−1). Basic parameters are those that all users knew (age, height, weight), or that could be easily calculated (BMI, age-predicted HRmax) or measured (resting HR). Additional parameters might be determined in a laboratory or medical environment (VO2max and fat mass) or during a specific physical test (S-VO2max and real HRmax) and are therefore not known by every user. Analyses were then performed in each group. Redundant parameters (for example, age and age-predicted HRmax) could not be included in the same equation and all equations configurations were tested. Parameters that significantly explained variation of EE were conserved in the models. Based on these analyses, we therefore selected predictive equations of EE (kcal h−1) and assessed determination coefficient (R2) of the relation between estimated EE and measured EE from multiple regression analyses in each group. For multiple regression analysis, we used six predictors at maximum to assess EE. According to Cohen (1988), with six predictors, α = 0.05, anticipated effect size = 0.35 (large) and a level of desired statistical power = 0.8, a number of subject larger than 46 is needed. Therefore our sample of 50 subjects gives a high statistical power to our results. A p value less than 0.05 was considered to be statistically significant. Finally, among the multiple generated equations, five were finally preserved on all the equations generated: one basic, two containing VO2max or VO2max and two containing real-time running speed (figure 1). Details of the coefficients used in each equation are displayed in table 2. Other equations were excluded because determination coefficients of the models were not improved. 2.3. Part 2: validation of the new predictive equations on an independent sample
Fifteen participants were recruited to perform the second test. Indeed, with α = 0.05, a level of desired statistical power = 0.7, a SD of 110 kcal and an anticipated difference of 75 kcal (15% of difference for a 500 kcal-exercise), a number of subjects equal to 15 is needed. Anthropometric characteristics were similar to the participants of the part 1 (table 1). This test was performed with the same equipment as during part 1. Speed was progressively increased over 5 min to reach 50% of HR reserve, then was maintained during 40 min. Then, speed was increased by 0.5 km h−1 every 30 s to assess VO2max, S-VO2max and real HRmax. Comparisons of EE between the reference technique (indirect calorimetry) and the predictive equations were performed using ANOVA. The absolute mean error rate ([predicted EE - reference EE]/reference EE) was also calculated. Linear Pearson’s product-moment correlation analyses were performed to correlate quantitative variables. The Bland and Altman method (1986) was used to assess the agreement between the estimated and the reference EE. Confidence intervals (CIs) were determined for the mean bias and for the upper (UL) and lower limits (LL) of agreement. As requested by Bland and Altman (1986), we defined a priori 257
Physiol. Meas. 35 (2014) 253
Table 2. Details of the coefficients used in the new equations for each significant predictor obtained after multiple regression analyses.
Equation
HR Height Weight Sex Resting T-HRmax R-HRmax S-VO2max VO2max Running speed Intercept (bpm) (cm) (kg) (1 = M, 2 = F) HR (bpm) (bpm) (bpm) (km h−1) (mL min−1.kg−1) (km h−1)
258
Basic equation 171.62 6.87 BP + S-VO2max + R-HRmax 738.90 6.89 BP + VO2max + R-HRmax 113.20 6.88 BP + speed (without HR) −624.53 BP + speed −534.26 1.38
3.99
2.30 5.48 7.45 5.93 5.43
−139.89
−37.13
−4.26 −2.81 −3.15 1.09
−4.87
−9.50 −6.88
32.31 10.14 88.70 73.78
BP: basic parameter; HR: heart rate, T-HRmax: theoretical maximal HR, R-HRmax: real maximal heart rate, VO2max: maximal oxygen uptake, S-VO2max: speed at VO2max. By example, the basic equation is: EE (kcal h−1) = 171.62 + 6.87 × HR (bpm) + 3.99 × Height (cm) + 2.30 × weight (kg) −139.89 × Sex (1 or 2) −4.26 × resting HR (bpm) − 4.87 × T-HRmax (bpm).
K Charlot et al
K Charlot et al
Physiol. Meas. 35 (2014) 253
1,00 0,90
0.874 0.809
0.879
0.913
0.919
0.821
0,80 0.737
0,70
R²
0,60 0,50 0,40 0,30 0,20 0,10 0,00 Keytel (without VO2max)
BP
Keytel (with VO2max)
BP + S.VO2max BP + VO2max + + R-HRmax R-HRmax
BP + speed (without HR)
BP + speed
Figure 1. Determination coefficients (R2) of new predictive equations. BP: basic
parameter.
acceptable limits of agreement for interchangeability between estimated EE and reference EE. We considered that a variability of 15% could be considered as a good interchangeability with indirect assessments of EE from HR and additional parameters. This threshold corresponded to a 75 kcal-error for a 500-kcal exercise. This point will be discussed later. The same statistical analyses were performed with the EE estimated with the existing equations (Keytel et al 2005) containing basic parameters with or without VO2max to compare our new equations with already validated ones. 3. Results 3.1. Part 1: predictive equations
The coefficient of determination (R2 expressed in percent) was higher in the basic equation than in the Keytel equation without VO2max and slightly lower than the Keytel equation with VO2max (R2 values are displayed in figure 1). The R2 also increased as compared with basic equation with the inclusion of real maximal HR and S-VO2max or VO2max. Real-time running speed largely contributed to the prediction of EE and increased the coefficient of determination compared to the other equations. The model containing real-time speed but not HR was only slightly less accurate than the equation containing both. 3.2. Part 2: validation of predictive equations
Details of statistical tests are displayed in table 3. ANOVA revealed that estimated EE calculated by the Keytel equations underestimated reference EE measured during the validation 259
Physiol. Meas. 35 (2014) 253
Table 3. Summary of the part 2 of the study: the validation of the five most pertinent predictive equations.
260
Equation
EE estimated (kcal h−1)a
Basic equation Keytel (without VO2max) BP + S-VO2max + R-HRmax BP + VO2max + R-HRmax Keytel (with VO2max) BP + speed (without HR) BP + speed
512 405 513 505 380 494 500
± ± ± ± ± ± ±
119 50c 124 127 63c 93 92
Error rate (%) 14.3 19.4 10.6 10.4 24.1 6.9 7.1
± ± ± ± ± ± ±
13.1 11.6 10.5 10.3 9.6 4.8 4.4
Correlation coefficient (R)
Bias (CI)
Lower limits of agreement (CI)
Upper limits of agreement (CI)
% of limits of agreementb
0.657 0.632 0.809 0.840 0.812 0.933 0.924
−4 (−45 to 37) 104 (66 to 141) −5 (36 to 26) 3 (−27 to 32) 128 (98 to 158) 14 (−3 to 32) 9 (−10 to 28)
−191 (−262 to −120) −69 (−135 to -4) −149 (−204 to −95) −132 (−184 to −81) −9 (−61 to 43) −67 (−97 to -36) −77 (−109 to -45)
183 (113 to 254) 276 (211 to 342) 139 (86 to 194) 138 (87 to 189) 265 (213 to 317) 95 (65 to 126) 94 (62 to 126)
36.8 34.0 28.4 26.6 27.0 15.9 16.8
Bias: mean difference between EE measured and EE predicted. Lower and upper limits of agreement: bias ± 1.96 × SD. BP: basic parameters. ALA: acceptable limits of agreement (i.e. 10%). CI: confidence interval.% of limits of agreement: differences between limits of agreement and the mean energy expenditure (EE) measured. BP: basic parameter.a The reference value of energy expenditure (measured with indirect calorimetry) was 508 ± 111 kcal. b To conclude to interchangeability, the limits of agreement had to remain under 15%. c p < 0.001 different from reference EE and predicted EE with new equations.
K Charlot et al
K Charlot et al
Physiol. Meas. 35 (2014) 253
Keytel
+ R-HRmax
(with VO2max)
100
100
100
100
0
0
0
0
-100
-100
-100
-100
-200
-200
-200
-200
-300
-300
-300
-300
750
500
250
R² = 0,5623
R² = 0,5054
750
200
750
200
500
200
250
200
750
300
500
300
250
300
500
(without VO2max)
300
Mean Reference EE – Estimated EE (kcal.h-1) BP + VO2max
BP + speed BP + speed
(without HR)
100
100
0
0
0
-100
-100
-100
-200
-200
-200
-300
-300
-300
Limits of agreement Bias
750
100
Acceptable limits of agreement
500
200
750
200
500
200
250
300
750
300
500
300
250
+ R-HRmax
250
Difference Reference EE – Estimated EE (kcal.h-1)
BP + S-VO2max
250
Keytel
Basic equation
Mean Reference EE – Estimated EE (kcal.h-1) Figure 2. Interchangeability between real energy expenditure (EE) and estimated EE
according to the Bland and Altman analysis (6). Upper and lower limits of agreement had to contain acceptable limits (25% of mean real energy expenditure) to conclude a good interchangeability. BP: basic parameter.
test without (−18.2 ± 13.6%, p < 0.001) and with VO2max (−24.1 ± 9.6%; p < 0.001). Estimated EE from the new equations generated were not different from the reference EE. The error rate was lower in the basic equation than both Keytel equations and decreased with the addition of real HRmax, VO2max, S-VO2max and/or real-time running speed. All estimated EE were significantly correlated with the reference EE (p < 0.001 for all). Correlation coefficients of the Keytel equations were close to those of the basic equation and models containing VO2max and S-VO2max, but lower than the equations containing real-time running speed. Finally, interchangeability analyses (figure 2) revealed that the limits of agreement were close to the acceptable ones only with the equations containing speed (15.9 and 16.5 versus 15.0% respectively). Moreover, the Keytel equations showed a systematic increase in the bias with the increase of the measured EE (p < 0.001). 261
K Charlot et al
Physiol. Meas. 35 (2014) 253
4. Discussion In this study, we created HR-based predictive equations of EE during running with the inclusion of original parameters representing body composition, fitness level and/or real-time running speed. The results showed that our equation including basic parameters was at least as accurate as the formerly validated equation (Keytel et al 2005). Even more interestingly, inclusion of maximal oxygen uptake, speed at maximal oxygen uptake and real maximal HR, but not fat mass, significantly improved accuracy. Finally, real-time running speed was the parameter that contributed the most to explaining the relationship between HR and EE and this predictor might even replace HR without loss of accuracy during running without slope. 4.1. Basic equation
One team has more particularly contributed to the creation of models predicting exercise EE (Dugas et al 2005, Keytel et al 2005, Hiilloskorpi et al 2003, Hilloskorpi et al 1999). Their models that did not contain VO2max accounted between 73.4% and 78.4% of the variance in EE and those with VO2max accounted between 83.3% and 84.1% of this variance. Given that unusual parameters were introduced in the present paper in most of the new equations, it is difficult to directly compare our equations with the existing ones (Keytel et al 2005). However, we still could compare our basic equation with those of Keytel with or without VO2max. It appeared that the determination coefficients (R2) of our equations were higher than Keytel equations (figure 1) and our basic equation without VO2max was even close to the most advanced Keytel equation (i.e. with VO2max). These results were confirmed during the validation test in which error rates were higher with both Keytel’s equations than the new created equations. One might explain these differences with the fact that age was not included in our equations but replaced by age-predicted HRmax. Moreover, we chose to include resting HR in basic parameters because it is easy to determine before exercise with a HR monitor and correctly corresponds to fitness level (Bjørnstad et al 1993, Sundberg and Elovainio 1982). The fact that our basic equation was almost as accurate as the Keytel equation with VO2max suggests that resting HR is an acceptable index of fitness level. These two differences might have improved the model and should be considered to estimate EE from HR measurements. Protocol choices might also explain the differences. Based on the assumption that the exercise modalities do not influence the prediction of EE (Hilloskorpi et al 1999), Keytel et al (2005), with a similar population than ours, used a combination of tests on treadmill and on ergocycle to generate their equations. They further extrapolated the usefulness of their equations to rowing activities. However, it is well established that differences in energetic cost of locomotion exist between these two modes of exercise. For instance, fat oxidation is higher during running than cycling (Capostagno and Bosch 2010) and HR might differ despite the same relative intensity (Millet et al 2009). It is, therefore, likely that the generation of equations combining running and cycling data might have induced some additional degree of variability. 4.2. Advanced equations
The models containing only basic parameters are very important for the general population since they provide a good estimation of EE with parameters easy to implement into the HR monitor. Yet, the addition of more advanced parameters could be of interest for specific users if a major gain in precision can be obtained. For instance, competitive athletes are 262
K Charlot et al
Physiol. Meas. 35 (2014) 253
more likely to have access to parameters such as VO2max, S-VO2max or real HRmax. Equations included such parameters were deemed more accurate than the basic one, explaining ∼87% of the variance and generating only 10.5% of errors during the validation test. However, the interchangeability with the real EE was only modest since the limits of agreement (26.6 to 28.4%) exceeded a priori acceptable limits fixed at 15%. Moreover, addition of fat mass in the models was not pertinent suggesting that body composition has little influence on EE during exercise when HR and basic parameters are considered. This was unexpected since fat mass is known to influence running economy and therefore EE during exercise (Bunc 2000). It should be noticed that VO2max contributed more to the estimation of EE than the other parameters but the difference with S-VO2max was sparse and equations containing VO2max might be confidently replaced by those containing S-VO2max, which is easier to assess since it does not require a laboratory set up and could be assessed on the field. Since VO2max and consequently S-VO2max were determined following the steady-state test, one might argue that S-VO2max could have been underestimated since it has been proposed that the duration of the steps as well as the speed increment during the VO2max test are to be considered to determine S-VO2max (Roffey et al 2007). This limitation should be taken into account for people using the equation including the S-VO2max.
4.3. Equations with speed
There are several means to determine real-time running speed: with an accelerometer, with a GPS included in smartphones, when using a treadmill or even by calculating the mean speed from time and distance on a known circuit. In the present study, real-time running speed, which might be an index of the external mechanical work of locomotion (Saibene and Minetti 2003, Heglund et al 1982), was a strong predictor of EE while inclusion of other parameters (VO2max, S-VO2max or real HRmax) in the models did not enhance EE prediction in this case. The equation containing HR, weight, sex and real-time running speed improved the variance by 4.0% compared to the most accurate model of the advanced equations (basic parameters + VO2max + real HRmax) and by 11.0% compared to the basic equation (91.9% versus 87.9% versus 80.9%, respectively). Results from the validation test confirmed this tendency with error rate being only 7.1%. The limits of agreement were still slightly higher than the acceptable ones (16.8% versus 15.0%, respectively), therefore, meaning that the agreement between EE measured by indirect calorimetry and EE estimated by the model containing real-time running speed is at best correct. Moreover, it appeared that equations containing real-time running speed but not HR show a similar level of accuracy. This means that EE might be estimated accurately using only resting HR (needed in this model), i.e. without a measurement of internal load during exercise but only of external load. The addition of supplementary parameters marginally increased R2 suggesting that the equations containing basic parameters and real-time running speed with or without HR are sufficient. We therefore recommend using the equations containing real-time running speed when this measure is available. It should be noted that accuracies of all models presented in this study were larger than those from triaxial accelerometers during treadmill exercises. Indeed, R2 with accelerometry was only ∼ 0.77 (Brandes et al 2012, Howe et al 2009) and limits of agreement were approximately 70% (Brandes et al 2012) while we obtained R2 values from 0.81 to 0.92 and limits of agreement between 16.8% and 36.8%. This clearly indicated that HR-based EE estimations should be preferred over accelerometer-based models in walking and running activities. 263
K Charlot et al
Physiol. Meas. 35 (2014) 253
4.4. Limitations
These results and the application of these equations need to be tempered. Firstly, the application of these equations should be restricted to running without slope. Naturally, implementing realtime running speed into the equations prevent use beyond running without slope. Secondly, these equations have been tested for intensities between 25% and 75% of HR reserve or between 40% and 75% of VO2max. The relationship between HR and VO2 during exercise is linear from about 20% of HR reserve (Ceesay et al 1989). Thus, the accuracy of these equations might diminish at lower intensities. Moreover, given that indirect calorimetry starts to deviate from 75% of VO2max, the relationship between HR and VO2 seems uncertain from this level (Jeukendrup and Wallis 2005). Thirdly, these equations are suitable for a large population (healthy active and sedentary males or females between 18 and 50 years old) but need to be confirmed for specific individuals such as elderly, obese people or athletes. 5. Conclusion These new equations represent an improvement over existing ones in the estimation of energy expenditure (EE) from heart rate (HR) during running without slope. Indeed, inclusions of original parameters very easy to measure or calculate (heart rate at rest or age-related maximal HR), measurable with field tests (and speed at VO2max and real maximal HR) and/or available on treadmill or GPS devices (real-time running speed) improve the agreement between estimated EE from the new created equations and real EE assessed with indirect calorimetry. Based on the significant increase in accuracy, we recommend using equations containing real-time running speed as much as possible. Moreover, we recommend using these equations for intensities between 25% and 75% of HR reserve. Acknowledgments We hereby disclaim any competing interests. This study has been supported by Oxylane research (Villeneuve d’Ascq, France) but KC, RB and AP from the University of Paris 13 performed all the experiments and analyzed the data independently. References Ahmaidi S, Collomp K, Caillaud C and Prefaut C 1992 Maximal and functional aerobic capacity as assessed by two graduated field methods in comparison to laboratory exercise testing in moderately trained subjects Int. J. Sports Med. 13 243–8 Aitken J C and Thompson J 1988 The effects of previous severe exercise upon the respiratory VCO2/VO2 exchange ratio as a predictor of maximum oxygen uptake Eur. J. Appl. Physiol. Occup. Physiol. 57 720–5 Berggren G and Hohwu Christensen E 1950 Heart rate and body temperature as indices of metabolic rate during work Arbeitsphysiologie 14 255–60 Bjørnstad H, Storstein L, Dyre Meen H and Hals O 1993 Electrocardiographic findings according to level of fitness and sport activity Cardiology 83 268–79 Bland J M and Altman D G 1986 Statistical methods for assessing agreement between two methods of clinical measurement Lancet 1 307–10 Brage S, Ekelund U, Brage N, Hennings M A, Froberg K, Franks P W and Wareham N J 2007 Hierarchy of individual calibration levels for heart rate and accelerometry to measure physical activity J. Appl. Physiol. 103 682–92 Brandes M, Van Hees V T, Hann¨over V and Brage S 2012 Estimating energy expenditure from raw accelerometry in three types of locomotion Med. Sci. Sports Exerc. 44 2235–42 264
K Charlot et al
Physiol. Meas. 35 (2014) 253
Bruce R A, Fisher L D, Cooper M N and Gey G 1974 Separation of effects of cardiovascular disease and age on ventricular function with maximal exercise Am. J. Cardiol. 34 757–63 Brugniaux J V, Niva A, Pulkkinen I, Laukkanen R M, Richalet J P and Pichon A P 2010 Polar activity watch 200: a new device to accurately assess energy expenditure Br. J. Sports Med. 44 245–9 Bunc V 2000 Energy cost of treadmill running in non-trained females differing in body fat J. Sports Med. Phys. Fitness 40 290–6 Capostagno B and Bosch A 2010 Higher fat oxidation in running than cycling at the same exercise intensities Int. J. Sport Nutr. Exerc. Metab. 20 44–55 Ceesay S M, Prentice A M, Day K C, Murgatroyd P R, Goldberg G R, Scott W and Spurr G B 1989 The use of heart rate monitoring in the estimation of energy expenditure: a validation study using indirect whole-body calorimetry Br. J. Nutr. 61 175–86 Charlot K, Pichon A and Chapelot D 2011 Exercise prior to a freely requested meal modifies pre and postprandial glucose profile, substrate oxidation and sympathovagal balance Nutr. Metab. Lond. 8 66 Choquette S, Chuin A, Lalancette D A, Brochu M and Dionne I J 2009 Predicting energy expenditure in elders with the metabolic cost of activities Med. Sci. Sports Exerc. 41 1915–20 Cohen J 1988 Statistical Power Analysis for the Behavioral Sciences 2nd edn (Hillsdale, NJ: Lawrence Erlbaum Associates) p 567 Cooper J A, Watras A C, O’Brien M J, Luke A, Dobratz J R, Earthman C P and Schoeller D A 2009 Assessing validity and reliability of resting metabolic rate in six gas analysis systems J. Am. Diet. Assoc. 109 128–32 Dugas L R, van der Merwe L, Odendaal H, Noakes T D and Lambert E V 2005 A novel energy expenditure prediction equation for intermittent physical activity Med. Sci. Sports Exerc. 37 2154–61 Heglund N C, Fedak M A, Taylor C R and Cavagna G A 1982 Energetics and mechanics of terrestrial locomotion. IV. Total mechanical energy changes as a function of speed and body size in birds and mammals J. Exp. Biol. 97 57–66 Hilloskorpi H, Fogelholm M, Laukkanen R, Pasanen M, Oja P, Manttari A and Natri A 1999 Factors affecting the relation between heart rate and energy expenditure during exercise Int. J. Sports Med. 20 438–43 Hiilloskorpi H K, Pasanen M E, Fogelholm M G, Laukkanen R M and Manttari A T 2003 Use of heart rate to predict energy expenditure from low to high activity levels Int. J. Sports Med. 24 332–6 Howe C A, Staudenmayer J W and Freedson P S 2009 Accelerometer prediction of energy expenditure: vector magnitude versus vertical axis Med. Sci. Sports Exerc. 41 2199–206 Jeukendrup A E and Wallis G A 2005 Measurement of substrate oxidation during exercise by means of gas exchange measurements Int. J. Sports Med. 26 (Suppl. 1) S28–37 Keytel L R, Goedecke J H, Noakes T D, Hiiloskorpi H, Laukkanen R, van der Merwe L and Lambert E V 2005 Prediction of energy expenditure from heart rate monitoring during submaximal exercise J. Sports Sci. 23 289–97 Koehler K, Braun H, de Mar´ees M, Fusch G, Fusch C and Schaenzer W 2011 Assessing energy expenditure in male endurance athletes: validity of the SenseWear Armband Med. Sci. Sports Exerc. 43 1328–33 Krems C, L¨uhrmann P M, Strassburg A, Hartmann B and Neuh¨auser-Berthold M 2005 Lower resting metabolic rate in the elderly may not be entirely due to changes in body composition Eur. J. Clin. Nutr. 59 255–62 Lee C M, Gorelick M and Mendoza A 2011 Accuracy of an infrared LED device to measure heart rate and energy expenditure during rest and exercise J. Sports Sci. 29 1645–53 Millet G P, Vleck V E and Bentley D J 2009 Physiological differences between cycling and running: lessons from triathletes Sports Med. 39 179–206 Nichols J F, Aralis H, Merino S G, Barrack M T, Stalker-Fader L and Rauh M J 2010 Utility of the actiheart accelerometer for estimating exercise energy expenditure in female adolescent runners Int. J. Sport Nutr. Exerc. Metab. 20 487–95 Noakes T D, Myburgh K H and Schall R 1990 Peak treadmill running velocity during the VO2 max test predicts running performance J. Sports Sci. 8 35–45 Roffey D M, Byrne N M and Hills A P 2007 Effect of stage duration on physiological variables commonly used to determine maximum aerobic performance during cycle ergometry J. Sports Sci. 25 1325–35 Ruiz J R, Ortega F B, Rodriguez G, Alkorta P and Labayen I 2011 Validity of resting energy expenditure predictive equations before and after an energy-restricted diet intervention in obese women PLoS One 6 e23759 265
K Charlot et al
Physiol. Meas. 35 (2014) 253
Saibene F and Minetti A E 2003 Biomechanical and physiological aspects of legged locomotion in humans Eur. J. Appl. Physiol. 88 297–316 Smolander J, Ajoviita M, Juuti T, Nummela A and Rusko H 2011 Estimating oxygen consumption from heart rate and heart rate variability without individual calibration Clin. Physiol. Funct. Imaging 31 266–71 Spierer D K, Hagins M, Rundle A and Pappas E 2011 A comparison of energy expenditure estimates from the Actiheart and Actical physical activity monitors during low intensity activities, walking, and jogging Eur. J. Appl. Physiol. 111 659–67 Sundberg S and Elovainio R 1982 Resting ECG in athletic and non-athletic adolescent boys: correlations with heart volume and cardiorespiratory fitness Clin. Physiol. 2 419–26 Thebault N, Leger L A and Passelergue P 2011 Repeated-sprint ability and aerobic fitness J. Strength Cond. Res. 25 2857–65
266
