Individual versus group HR-V̇O2 equations.
The first question addressed was whether a single HR-V̇O2 prediction equation for all children is appropriate. As can be seen in Table 2, there is considerable individual variation in HR during the different laboratory assessments as well as in potential cut points, such as the Flex-HR. These results underscore the need for individual HR-V̇O2 prediction equations rather than a single HR-V̇O2 equation for all children.
HR-V̇O2 correlations were calculated for each child. These zero-order correlations ranged from 0.856 to 0.988, demonstrating strong HR-V̇O2 relationships across the range of HR values and individual variations in this relationship. Using RCM, we tested whether the variance of individual slopes using HR to predict V̇O2 is significantly different from zero in the population. A significant variance in the slopes (0.0029, with χ2(36) = 442.11, P < 0.001) indicated that there was sufficient individual variation in the HR-V̇O2 prediction equation to require individualized regression equations. We repeated these analyses of the subjects, controlling for age, gender, and BMI, to determine whether a conditional prediction equation might still be sufficient across individuals. Although gender (P < 0.05) and BMI (P < 0.05) were significant predictors of slopes for individual HR-V̇O2 prediction equations, there was still significant individual variation (χ2 (33) = 382.08, P < 0.001).
It is possible that individual differences in the HR-V̇O2 relationship at the low end of the HR-V̇O2 equation (i.e., below groupwide HR thresholds for moderate PA) are primarily responsible for the significant variance in slopes across individuals. If so, a single groupwide HR-V̇O2 equation could be used for HR values that correspond to moderate-to-vigorous PA (i.e., above group HR thresholds for PA). We repeated RCM, limiting the analyses to cases where HR values were above two cut points and controlling for individual age, gender, and BMI. The two groupwide cut points selected were HR values greater than 100 and 120. In all three analyses, individual slopes resulting from the HR-V̇O2 equation still had variances significantly different from zero in the population (for cut points of 100 and 120: χ2 (33) = 224.35, P < 0.001; and χ2 (4) = 134.78, P < 0.001, respectively). Thus we concluded that individualized HR-V̇O2 prediction equations significantly improve EE estimates from HR monitoring.
Can HR be used to estimate EE during low-intensity activities?
A number of investigators have suggested that the HR-V̇O2 relationship is not maintained across the full range of HR values (i.e., at lower levels of EE). If this were the case, it would not be necessary to examine the HR-V̇O2 relationship in activities other than the treadmill; HR levels below those assessed during the treadmill activity would be ignored, and EE could be estimated with a substitute value, such as the resting metabolic rate (RMR) (2,4,5,10,12,19).
Controlling for individual differences as well as age, gender and BMI, RCM was used to examine whether there was a significant relationship (slope) between HR and V̇O2 during the resting stage (lying and sitting) or during tasks involving low-intensity activity (standing, standing and writing, and standing and throwing). As expected, the mean slope of the HR-V̇O2 prediction equation was not significantly different from zero for the resting stage; t(301) = 0.05. However, the relationship between HR and V̇O2 was significant during tasks involving a range of upper body activities; t(394) = 6.33, P < 0.0001. It would appear that HR monitoring during low-intensity activities, including activities primarily limited to upper body activity, does provide information about EE.
The question remains, however, whether the HR values generated from the treadmill activity effectively covered the full range for which there is a significant HR-V̇O2 relationship. To answer this question, the HR-V̇O2 equation was examined for HR below the minimum value measured during the treadmill assessment for each individual. Even when restricted to values not commonly used to calibrate the HR-V̇O2 prediction equation, there was a significant relationship, t(600) = 15.21, P < 0.0001.
Is there a threshold below which a HR-V̇O2 prediction equation is inappropriate?
For those investigators interested in using a groupwide threshold for estimating EE with HR monitoring, we attempted to identify the HR level below which the HR-V̇O2 relationship is no longer significant. RCM controlling for age, gender, and BMI was used to examine whether there was a significant slope when the HR was below 105 bpm and repeated in 5-bpm intervals. This rate was selected as a starting point because it approximates the mean of a commonly used Flex-HR (Table 2). At 95 bpm, the HR-V̇O2 prediction equation still had a significant slope, t(351) = 4.78, P < 0.0001. When the analysis was repeated for HR values below 90 and controlled for age, gender and BMI, the HR-V̇O2 prediction equation no longer had a significant slope, t(256) = 1.79, P > 0.05.
What is the optimum equation for predicting EE?
Although it is clear that there is a significant relationship between HR and V̇O2 at HR levels below that measured during the treadmill activity and below cut points frequently used in the literature, it is still possible that the HR-V̇O2 prediction equation generated from the treadmill data is sufficient to predict the V̇O2 for the full range of potential HR values, including those for low-intensity activities. Five potential methods for generating HR-V̇O2 prediction equations were identified: 1) a single, generic equation based on the treadmill HR data from the entire sample and applied to all children; 2) individualized HR-V̇O2 prediction equations generated with treadmill data; 3) the Flex-HR method (2,4,5,10,12,19), using RMR for all HR values below the Flex-HR and the treadmill equation for all HR values above; 4) individualized HR-V̇O2 prediction equations calculated on the full range of activities (resting, low-intensity, and treadmill); or 5) two individualized equations, one based on treadmill data to predict V̇O2 values above the HR range of the nontreadmill activities, the other based on all nontreadmill activities to predict V̇O2 values in this lower range. In all five methods, the HR-V̇O2 prediction equations included both linear and quadratic terms of HR. To determine whether there was a significant difference between the five different HR-V̇O2 prediction methods, the five methods were compared based on SEE and the percentages of variance explained by each method. A smaller SEE value and more variance indicate better prediction. The results of these analyses are presented in Table 3.
The HR-V̇O2 prediction equation derived from the treadmill data had a SEE of 5 and accounted for 85% of the variance in the observed V̇O2 readings for all activities. Thus, reasonably good prediction of EE can be obtained from a general HR-V̇O2 prediction equation applied to all children. However, each subsequent iteration of individualized HR-V̇O2 prediction equations accounted for significantly more of the variance in the observed V̇O2 readings (Table 3). In addition, including data from the low-level activities in the equation improved prediction: 96% of the variance was accounted for by the combination of two individualized equations, one for low-level activity and one for moderate to vigorous activity.
Consistent with previous research, there were significant individual differences in the HR-V̇O2 prediction equations, suggesting that individual equations for each child improve prediction. BMI and gender also contributed to the prediction equation; however, these individual characteristics had minimal effects in reducing individual differences in the HR-V̇O2 prediction equation. Individual variations in the HR-V̇O2 prediction equation were not attributable to variation in low-intensity activities. Significant individual variations in slope persisted when HR values were restricted to those above commonly used cut point values. These results suggest that individual equations significantly improve prediction.
One limitation of these analyses is that the calibration equations were used to predict V̇O2 during a prescribed set of activities in a laboratory setting. Furthermore, the equations that were used to predict V̇O2 for the same set of activities were used to derive the prediction equation. The question remains as to whether there is any practical consequence when these different predictive equations are applied to the distribution of activity levels and corresponding HR in a free-living population. That is, the laboratory analyses do not address whether HR-V̇O2 prediction equations developed in the laboratory also produce significantly different estimates of free-living EE. In order to answer this question, a second study was performed to apply these equations to HR representative of a typical day.
Thirty-seven of the children from study 1 (27 girls, 10 boys) participated in study 2.
Free-living HR monitoring.
Toward the end of the study 1 laboratory visit, children and parents were instructed on how to use a Polar Vantage XL heart rate monitor. Appointments were then made for three separate days of recording, including two weekdays and one weekend day; for each day of recording, parents were instructed to attach the HR monitor immediately after the child awoke in the morning. The goal was to obtain 12 h of minute-by-minute HR data for each day scheduled. HR data was uploaded to a laptop computer during a home visit on the evening of the same day.
The focus of the analyses was to compare different methods for estimating daily EE (specifically, V̇O2) by applying the HR-V̇O2 equations generated in study 1 to the free-living HR data gathered in study 2. The mean estimated daily EE for each child was then compared using paired t-tests.
The characteristics of the study 2 sample are displayed in Table 4. Girls who did not complete study 2 differed from girls who completed the study on one of 13 characteristics; girls who did not complete study 2 had significantly lower biceps skinfold readings; t(24) = −3.17, P < 0.01.
Estimating EE from free-living HR data.
There were two criteria for acceptable HR data. First, at least 90% of the data had to be free of artifacts, i.e., HR < 40 or >220 (16). Second, each child had to have at least 12 h of data without artifacts. Less than 1% of the data included artifacts. After artifacts were removed, the mean minutes of observations per child were 1813 (SD = 408).
One question is whether, during a typical day, a sufficient time is spent in low-level activity to have an appreciable effect on daily EE. It was established in study 1 that there is a significant relationship between HR and V̇O2 for HR below Flex-HR. Approximately 55% of free-living HR were below the Flex-HR. The cut point used in the two-equation solution in study 1 was the highest HR during low-level activities; approximately 73% of free-living HR fell below this cut point. That is, the traditional method of estimating daily EE (i.e., applying the treadmill HR-V̇O2 equation to all HR above the Flex-HR) ignores the potential HR-V̇O2 relationship for 55% of the HR values. Furthermore, the low-level activity cut point suggests that it may be inappropriate to apply the treadmill HR-V̇O2 equation for an additional 18% of the free-living HR values.
The primary question is whether the two individualized equations that were most successful in estimating EE using lab data produce significantly different estimates for EE using free-living HR data. The five methods for predicting EE in study 1 were applied to the full range of HR data to estimate daily EE (Table 5). Application of these five methods to free-living HR data results in significantly different estimates of daily EE. The two best solutions from study 1, individual single equations derived from all activities, and the individual two-equation method with a low-level activity cut point, differed significantly from each other and from the traditional Flex-HR solution.
The results of study 1 suggest that individual calibrations of the HR-V̇O2 equation significantly improve prediction. Introduction of low-level activities into the calibration process further improves prediction; there was a significant HR-V̇O2 relationship within the range of HR values during low-level activity. This significant relationship persisted below the commonly used Flex-HR cut point. Furthermore, results from the two-equation method suggest that the HR-V̇O2 equation derived from treadmill data does not adequately describe the HR-V̇O2 relationship during low-level activities. In study 2, fully 73% of the HR values in the free-living sample were within the range of the low-level activity equation; therefore, using the treadmill equation alone may result in an overestimation of daily EE. These findings have practical implications for assessing daily EE and for making dietary and physical activity recommendations for youth at risk for obesity and health-related problems.
HR monitoring of children is frequently used to estimate free-living EE, to identify patterns of PA, and to validate other methods for assessing EE and PA such as accelerometers and self-report. There is still some disagreement as to how best to calculate the HR-V̇O2 prediction equation. Excluding the use of labor-intensive and burdensome 24 h assessment of EE as a means of calibrating the HR monitor (20), previous efforts have tended to be limited to adult samples, relatively small, homogeneous samples of children (usually males), or a set of activities that do not include a range of low-intensity activities (1-6,10-12,17,19,21). Study 1 addresses some of the problems of previous studies with a larger sample of children that includes an oversampling of girls and a procedure for calculating the HR-V̇O2 prediction equation that adds a range of low-intensity activities to the treadmill test. Although study 1 was not designed to include a representative sample at each age group, all of the major findings remained significant when analyses controlled for age, gender, and BMI.
Flex-HR cut points are frequently used to identify the HR above which there is a significant HR-EE relationship (1,5,6,11). Flex-HR values are usually calculated as between the highest RHR and the minimum HR during exercise, such as a treadmill test or equivalent. Consistent with previous work, there was no significant relationship between HR and V̇O2 during resting periods. However, HR was a significant predictor of EE during low-intensity activities. Furthermore, HR predicted EE below the threshold for HR during treadmill activities and well below a commonly used Flex-HR cut point. This is strong evidence for a HR-V̇O2 relationship during low-intensity activities. One limitation to these conclusions is that HR and V̇O2 were assessed using primarily upper body activities. However, these HR were fairly low, below the level of most cutoffs used in research estimating EE and below HR generated from a fairly slow rate of walking. These findings are consistent with those of Bitar et al. (3), who included at least one low-intensity physical activity in their procedures. They found that the best equation for predicting EE included values below the Flex-HR, rather than an equation limited to HR above the Flex-HR. Further research is needed to identify whether these findings are limited by the activities selected for evaluation in this protocol. However, it should be noted that the sample used in this study is larger, and includes a larger proportion of females than many of the previous studies in this area.
An important question for the researcher is: How much is gained by using individualized HR-V̇O2 prediction equations? In addition, is it necessary to calibrate the HR-V̇O2 prediction equation with a range of activities, or is the treadmill test or its equivalent sufficient? The good news for the investigator wishing to use a single cut point or single HR-EE prediction equation is that in this sample the single, general equation derived from the group treadmill data accounted for 85% of the variance in V̇O2 across all tasks. These results are similar to those of Sarton-Miller et al. (17), who used change in HR and V̇O2 as the predictive and outcome elements in their equation, and found that a single regression equation based on group data accounted for approximately 84% of the variance. We are not suggesting that the equation derived from this sample can be used for other samples. Group differences in Flex-HR and the HR-V̇O2 relationship due to age, gender, body composition, etc. suggest that investigators would still need to calculate a HR-V̇O2 prediction equation for their sample by testing a representative subsample (11,19).
For investigators requiring more precision, individual HR-V̇O2 equations improve prediction. Individual equations generated from a treadmill or equivalent activity result in a significant improvement over a single equation based on group data; in this sample, individual HR-V̇O2 prediction equations derived from the treadmill data accounted for 88% of the variance in V̇O2 across all activities. This is higher than the typical equation derived using the Flex-HR method (9). Single individualized HR-V̇O2 prediction equations based on HR values from the full range of activities accounted for 95% of the variance in V̇O2 readings. The greatest precision, accounting for 96% of the variance, resulted from two individualized HR-V̇O2 prediction equations, one for HR in the range of the nontreadmill activities, and one for HR above this range. For investigators hoping to get maximum precision from the HR-V̇O2 prediction equation, there is a clear benefit to individualized HR equations. Calibrating the HR-V̇O2 equation with activities that vary in intensity improves the precision of these estimates.
In study 1, the two best methods—individualized equations using the full range of laboratory HR values, and the two-equation solution—were significantly better than the Flex-HR method in estimating EE within the laboratory protocol. In study 2, these two methods produced significantly different estimates of daily EE than the Flex-HR method. The Flex-HR estimate of daily EE was approximately 7% higher than that of the two-equation solution, and differed by 8% from the individualized equation solution. Since the equations for calculating EE from the upper HR values uniformly rely on the treadmill data across methods, these differences in estimates of daily EE must reflect significant differences in impact of lower-intensity activities. Some investigators have suggested various adjustments to the Flex-HR method to get a better estimate of daily EE, such as increasing the Flex-HR by 10 (19), effectively reducing the estimate of daily EE by limiting the range of HR values to which the treadmill equation is applied.
Correlations between daily EE calculated by the different methods were quite varied (Table 5). Uniformly high correlations would have suggested consistency across individuals in the relative importance of low-level activity, even though the absolute values differed. However, the inconsistent correlations indicate considerable individual variation in the contribution of low-level activity to daily EE.
Exclusive reliance on the HR-V̇O2 treadmill equation for estimating EE appears to result in an overestimation of daily EE. The findings from study 1 and study 2 have implications for research and clinical applications. For example, from a research perspective, misestimation and individual variations in the impact of low-level activity on estimates of daily EE could significantly affect hypothesized relationships among EE, PA, diet, and body composition. From a clinical perspective, the differences could influence recommendations regarding dietary intake. For a child at risk of obesity who is believed to be very active, a 2800-kcal diet might be recommended to maintain weight. However, a 7% error in estimated daily EE would result in a daily recommendation that is almost 200 kcal higher than it should be. This error is sufficient to cause regular weight gain.
Based on a number of factors and constraints (e.g., purpose of assessment, sample size, budget, subject burden, subject retention, subject adherence to the protocol), investigators can decide whether a HR-V̇O2 equation derived from group data and applied to the entire sample is satisfactory compared to individualized prediction equations. Once an investigator decides to use individualized prediction equations, it appears that the optimum method would be to use two HR-V̇O2 prediction equations derived from a range of laboratory activities, with separate equations for HR values above or below the HR cut point identified with nontreadmill activities or their equivalent.
This project was supported by a Research Challenge Award from the Ohio Board of Regents and by the National Institute of Child Health and Human Development. The authors would like to thank Wayne Brooks, Emily Cole, Paco Labrador, Sara Levine, Rachel Mitchell, Caroline Thomas, Adam Van Auken, Kalina Van Every, and Tracy Whittaker for their assistance with the project.
1. Beghin, L., T. Budniok, G. Vaksman, et al. Simplification of the method of assessing daily and nightly energy expenditure in children, using heart rate monitoring calibrated against open circuit indirect calorimetry. Clin. Nutr
. 19:425–435, 2000.
2. Van den Berg-Emons R. J., W. H. M. Saris, K. R. Westerterp, and M. A. Van, Baak. Heart rate monitoring to assess energy expenditure in children with reduced physical activity. Med. Sci. Sports Exerc
. 28:496–501, 1996.
3. Bitar, A., M. Vermorel, N. Fellmann, M. Bedu, A. Chamoux, and J. Coudert. Heart rate recording method validated by whole body indirect calorimetry in 10-yr-old children. J. Appl. Physiol
. 81:1169–1173, 1996.
4. Ceesay, S. M., A. M. Prentice, K. C. Day, et al. The use of heart rate monitoring in the estimation of energy expenditure: a validation study using indirect whole-body calorimetry. Brit. J. Nutr
. 61:175–186, 1989.
5. Ekelund, U., A. Yngve, K. Westerterp, and M. Sjöström. Energy expenditure assessed by heart rate and doubly labeled water in young athletes. Med. Sci. Sports Exerc
. 34:1360–1366, 2002.
6. Emons, H. J. G., D. C. Groenenboom, K. R. Westerterp, and W. H. M. Saris. Comparison of heart rate monitoring combined with indirect calorimetry and the doubly labelled water (2
O) method for the measurement of energy expenditure in children. Eur. J. Appl. Physiol
. 65:99–103, 1992.
7. Goran, M. I., B. A. Gower, T. R. Nagy, and R K. Johnson. Developmental changes in energy expenditure and physical activity in children: Evidence for a decline in physical activity in girls before puberty. Pediatr
. 101:887–891, 1998.
8. Horswill, C. A., C. L. Kien, and W. B. Zipf. Energy expenditure in adolescents during low intensity, leisure activities. Med. Sci. Sports Exerc
. 27:1311–1314, 1995.
9. Leonard, W. R. Measuring human energy expenditure: what have we learned from the flex-heart rate method. Am. J. Hum. Biol
. 15:479–489, 2003.
10. Livingstone, M. B. E., W. A. Coward, A. M. Prentice, et al. Daily energy expenditure in free-living children: comparison of heart-rate monitoring with doubly labeled water (2
O) method. Am. J. Clin. Nutr
. 56:343–352, 1992.
11. Livingstone, M. B. E., P. J. Robson, and M. Totton. Energy expenditure by heart rate in children: an evaluation of calibration techniques. Med. Sci. Sports Exerc
. 32:1513–1519, 2000.
12. Maffeis, C., L. Pinelli, M. Zaffanello, F. Schena, P. Iacumin, and Y. Schutz. Daily energy expenditure in free-living conditions: comparison of doubly labelled water (2
O) method and heart-rate monitoring. Int. J. Obes
. 19:671–677, 1995.
13. Ott, A. E., R. R. Pate, S. G. Trost, D. S. Ward, and R. Saunders. The use of uniaxial and triaxial accelerometers to measure children’s “free-play” physical activity. Pediatr. Exerc. Sci
. 12:360–370, 2000.
14. Pate, R. R., M. Pratt, S. N. Blair, et al. Physical activity and public health: a recommendation from the Centers for Disease Control and Prevention and the American College of Sports Medicine. JAMA
15. Raudenbush, W. R., and A. S. Bryk. Hierarchical Linear Models: Applications and Data Analysis Methods
. 2nd ed. Thousand Oaks: Sage Publications, 2002, pp. 23–24.
16. Sallis, J. F., P. K. Strikmiller, D. W. Harsha, et al. Validation of interviewer- and self-administered physical activity checklists for fifth grade students. Med. Sci. Sports Exerc
. 28:840–851, 1996.
17. Sarton-Miller, I., D. J. Holman, and H. Spielvogel. Regression-based prediction of net energy expenditure in children performing activities at high altitude. Am. J. Hum. Biol
. 15:554–565, 2003.
18. Sirard, J. R., and R. R. Pate. Physical activity assessment in children and adolescents. Sports Med
. 31:439–454, 2001.
19. Spurr, G. B., J. C. Reina, and D. L. Dufour. Comparative study of flex heart rate in Columbian children and in pregnant, lactating, and on-pregnant, nonlactating women. Am. J. Hum. Biol
. 9:647–657, 1997.
20. Treuth, M. S., A. L. Adolph, and N. F. Butte. Energy expenditure in children predicted from heart rate and activity calibrated against respiration calorimetry. Am. J. Physiol
. 275:E12–E18, 1998.
21. Warner, J. T., W. Bell, D. K. H. Webb, and J. W. Gregory. Daily energy expenditure and physical activity in survivors of childhood malignancy. Pediatr. Res
. 43:607–613, 1998.
Keywords:©2004The American College of Sports Medicine
ASSESSMENT; METHODOLOGY; ENERGY E-PENDITURE; E-ERCISE; ADOLESCENCE