# Predicted and measured resting metabolic rate in young, non-obese women.

INTRODUCTION

Determining the resting metabolic rate (RMR) is a key step in nutritional assessment; because the RMR accounts for over 65% of total daily energy requirements (1), obtaining accurate measurements is essential. Traditional indirect calorimeters (ICs), such as the metabolic cart, have been established as the reference standard; however, they are expensive and cumbersome, and require careful calibration. Use of these devices is impractical, costly, and difficult in many settings (2); therefore, clinicians rely on prediction equations (PEs) to calculate energy expenditure. Commonly used PEs include the Harris-Benedict (3), Mifflin (4), Owen (5), [Schofield.sub.weight], and [Schofield.sub.weight and height] (6) equations; however, accuracy of these equations has been questioned (7-11).

A handheld IC, such as the MedGem, is portable, affordable, and easy to use, and therefore feasible for use by health care and research professionals. Measurements of RMR are obtained by using a mouthpiece and nose clip, and are less invasive than metabolic cart measurements and also faster (10 versus 40 minutes). In several studies, the MedGem IC has been validated in comparisons with the metabolic cart for use in normal-weight, healthy adults (12-15), and in comparison with PEs in overweight women (16) and in a group heterogeneous for age, sex, body mass index (BMI), and ethnicity (17).

PURPOSE

Resting metabolic rate measured by the MedGem handheld IC was compared with RMR predicted by five commonly used PEs: the Harris-Benedict, Mifflin, Owen, [Schofield.sub.weight], and [Schofield.sub.weight and height] equations. To our knowledge, ours is the first study in which PEs have been compared with a handheld IC device in young, non-obese Canadian women.

METHODS

Participants

Female students enrolled in a senior undergraduate class in nutritional assessment were recruited from the University of Guelph, Ontario, Canada. Women were eligible to participate if they were enrolled in the course, had no known health conditions (determined by questionnaire), were younger than 30, had a BMI below 30 kg/[m.sup.2], and were not pregnant. Participation was voluntary and did not influence grades. The study was approved by the University of Guelph Research Ethics Board. Informed written consent was obtained before participants enrolled in the study.

Procedures

All tests were performed by the same trained technician in the University of Guelph Body Composition and Metabolism Laboratory (http://www.uoguelph.ca/bodycomp/). Tests were done between 0930 and 1430, from October to December 2010. Before testing, participants confirmed with a questionnaire that they had fasted from food and alcohol for 12 hours, and had had no exercise in the previous 24 hours. Weight (in light clothing) was measured to the nearest 0.1 kg with an electronic scale (Life-Source MD, A&D Engineering, San Jose, CA), when participants had voided within the preceding 30 minutes. Height without shoes was measured to the nearest 0.1 cm with a wall-mounted stadiometer (Seca Corp., Ontario, CA).

Resting metabolic rate: The RMR (kcal/day) was measured in a private, brightly lit room with a temperature of 22.5 [+ or -] 0.7 oC and 28 [+ or -] 6% humidity. The MedGem handheld IC (MedGem, Microlife Medical Home Solutions, Inc., Golden, CO) was used while participants were awake, motionless, and supine. Participants watched a DVD to stay awake and were monitored throughout the test, which took 10 to 12 minutes.

Autocalibration was performed on a flat surface before each use of the device, according to the manufacturer's instructions. Before measurement, participants rested for at least 10 minutes in the supine position. A new nose clip and disposable mouthpiece were used for each test. Participants breathed through the mouthpiece with a firm lip seal, while holding the MedGem IC in place until it beeped and the indicator light changed to amber, which signalled the end of the measurement. The MedGem IC uses a proprietary algorithm to detect achievement of steady state; this determines termination and calculation of the RMR, which is displayed on the MedGem IC LCD screen (18). The RMR was calculated from measured VO2, using the manufacturer's suggested modification of the Weir equation (19), as follows: kcal /day = (3.941 x V[O.sub.2]) + 1.106 (0.85 x V[O.sub.2]) - 0.365y, where VO2 (mL/min) was assumed at a constant respiratory quotient of 0.85 and 0.365y is the correction for protein metabolism.

In addition, the RMR was predicted with five published equations: the Harris-Benedict (3), Mifflin (4), Owen (5), [Schofield.sub.weight], and [Schofield.sub.weight and height] equations (6).

Data analyses

For statistical analysis, we used IBM SPSS Statistics (version 20.0, IBM Corp., Somers, NY, 2011), with significance set at p [less than or equal to] 0.05. Several tests were conducted to determine the accuracy of the RMR predicted by the five equations in comparison with that measured by the MedGem device.

The RMR data were normally distributed, as confirmed by Jarque-Bera tests. Paired /-tests were conducted to compare group mean differences between measured RMR and predicted RMR for each equation. The proportion of participants for whom measured RMR and predicted RMR differed by 10% or more was also determined. These two tests permitted a general comparison of results at the group level.

For each equation, a Bland-Altman plot was created to provide a more accurate comparison of methods than correlation alone (20). Bland-Altman plots can detect proportional and fixed biases, allowing the detection of random error of both predicted and measured methods (21). Each Bland-Altman plot was created by plotting the residual (predicted RMR minus measured RMR) against the mean ([predicted + measured RMR]/2), and generating limits of agreement ([+ or -] 2 standard deviations, 95%), to permit assessment of systematic bias between measured RMR and predicted RMR.

Additionally, the magnitude of association between measured RMR and predicted RMR was determined using Pearson's correlation coefficient, r; further, [R.sup.2] was adjusted to account for the number of variables in the regression equations. The standard error of estimate (SEE) for each PE was determined using linear regression analysis.

Last, multiple analysis of variance (MANOVA) was used to assess whether age, height, weight, or BMI was associated with differences between predicted RMR and measured RMR. Regression analysis followed to detect whether significant associations existed for any particular variable.

RESULTS

Fifty-five eligible students participated. Three were excluded from analyses: one had consumed alcohol within 12 hours and two had exercised within 24 hours of testing. Mean BMI of the 52 participants (aged 22 [+ or -] 2 years) was 21.8 [+ or -] 2.7 kg/[m.sup.2]; six (12%) were classified as underweight (BMI below 18.5 kg/[m.sup.2]), 39 (75%) were normal weight (BMI 18.5 to 24.9 kg/[m.sup.2]), and seven (13%) were overweight (BMI 24.9 to 29.9 kg/[m.sup.2]).

Data on statistical tests used to determine agreement between measured RMR and predicted RMR are shown in Table 1. Paired /"-tests indicated that four of the five PEs led to significant overestimation of measured RMR. The smallest difference was 16 [+ or -] 133 kcal/day (Owen equation), and the largest was 225 [+ or -] 133 kcal/day (Harris-Benedict equation). Predicted RMR values exceeded 10% of measured RMR values for 35% (Owen equation) to 76% (Harris-Benedict equation) of participants.

Bland-Altman analyses revealed good agreement for all five equations, as indicated by the fact that 94% to 96% of the data points fell within two standard deviations (Figures 1a to 1e). However, three of the five PEs (the Owen, Mifflin, and Harris-Benedict equations) showed a systematic bias (p<0.05). That is, the agreement between the residual and the mean of measured and predicted RMR depended on the measured RMR value. According to the agreement shown in Bland-Altman analysis with the mean of the two methods as the independent variable, as RMR increased, agreement decreased. No bias was detected in the two Schofield equations.

For further investigation of biases shown by Bland-Altman analysis for the Owen, Mifflin, and Harris-Benedict equations, measured RMR and predicted RMR were separately plotted against percentage difference (error). A linear trend line was inserted to show the average pattern. Predicted RMRs were not strongly correlated with percentage difference (data not shown), but measured RMRs were. For every 100 kcal increase in measured RMR, the average percentage error decreased by 6% to 8% (Figures 2a to 2c). The RMR range in which percentage error was the lowest (i.e., in which predicted RMR agreed with measured RMR within [+ or -] 10%) was different for each equation. The Owen equation range was 1105 to 1400 kcal/ day with 0% error at approximately 1250 kcal/day; the Mifflin range started at 1280 kcal/day and extended to 1595 kcal/ day with a 0% error at approximately 1275 kcal/day; the range for the Harris-Benedict equation began at 1345 kcal/day and extended to 1630 kcal/day, with 0% error at approximately 1480 kcal/day.

Relative to the RMR measured by the MedGem IC, Pearson's correlations for RMR, predicted by all five equations, were significant. Predicted RMR accounted for 27% (Harris-Benedict) to 34% (Owen) of the variance of measured RMR for all equations. The SEE, determined by linear regression analysis, was similar for all equations.

MANOVA showed no significant association of age, height, weight, or BMI with the residuals for any equation. MANOVA [R.sup.2] indicated that no variable in any equation accounted for the residual at statistically significant levels.

DISCUSSION

Accuracy of equations

In this study, the performance of PEs was variable in comparison with measured RMR. Variations were especially large at the individual level over the range of measured RMRs (SD [+ or -] 133 to 139 kcal/day). At the group level, the Owen equation (5) performed best in this population of young, non-obese women, as the mean RMR predicted for the group differed least from the measured RMR (16 kcal/day). In addition, this equation captured the greatest proportion of individuals (65%) for whom the predicted RMR was within 10% of the measured RMR. Differences between measured RMR and RMR predicted by the other four equations--the Harris-Benedict (3), Mifflin (4), [Schofield.sub.weight] (6), and [Schofield.sub.weight and height] (6) equations--ranged from 159 kcal/day with the Mifflin equation to 225 kcal/day with the Harris-Benedict equation. The RMR predicted by these four equations differed by more than 10% from measured RMR for more than two-thirds of the young women included in this study. A recent United States study, similar to ours, included 186 women aged 18 to 60 and indicated higher overall MedGem-measured RMR (141 [+ or -] 214 kcal/day) and higher BMI (24.8 [+ or -] 4.9 kg/[m.sup.2]) than we found for our participants (17).

No equation accurately predicted individual RMR consistently across the full range of measured RMRs, a finding similar to those of studies in overweight women (16) and in a heterogeneous group of women aged 18 to 60 (17). Rather, in particular for the Owen, Mifflin, and Harris-Benedict equations, a negative linear relationship existed between measured RMR and the residual (predicted RMR minus measured RMR). That is, as measured RMR decreased, the likelihood of overestimation increased. As measured RMR increased, overestimation decreased; however, once RMR increased beyond the point where the residual was nil or near nil (approximately 1250, 1440, or 1480 kcal/day, depending on the equation), the equations began to underestimate measured RMR. In other words, if an individual's measured RMR fell below 1250 to 1480 kcal/day, the equations tended to overestimate RMR; if measured RMR was above 1250 to 1480 kcal/day, the equations tended to underestimate RMR. These results suggest that only a narrow range of measured RMR exists for which PEs can be used with any degree of confidence. Outside this narrow range, predictions become inaccurate (either overestimating or underestimating measured RMR) and may not be suitable for clinical or research use, especially at the individual level.

Over- and underestimation

Systematic bias in the Owen, Mifflin, and Harris-Benedict equations was consistent with findings from Spears et al.'s study, in which handheld IC measurements were compared with PEs (16). Similarly, Spears et al. reported the same relationship of overestimation trending to underestimation as measured RMR increased (16). In our study, the slope of this relationship was relatively similar across all three equations, which suggests that, given a sufficiently wide RMR range, all three equations would perform equally poorly at both the individual and the group levels.

The reason that the Owen equation appeared to perform better at the group level in our study can be attributed to the fact that our participants' mean measured RMR was closest to the RMR at the point of the nil residual in the Owen equation. As such, this equation would have been able to predict the largest proportion of our participants' RMR within 10% of their measured RMR. Additionally, the proportion of individuals with a measured RMR above the nil residue point of the Owen equation was relatively equal to the proportion below. The degree of underestimation countered the degree of overestimation, and therefore the difference between measured RMR and Owen equation-predicted RMR appeared reduced at the group level.

The RMR of the Mifflin and Harris-Benedict equations at the point of the nil residual was higher than our group's mean measured RMR. Therefore, application of those two equations led to greater overestimations (more than 10% difference). Underestimation was not as common for the Mifflin and Harris-Benedict equations, as our sample did not include many women whose measured RMR exceeded the RMR predicted by the Mifflin or Harris-Benedict RMR of the nil residual. As a result, the Mifflin and Harris-Benedict equations, more than the Owen equation, appeared to differ from measured RMR at the group level, even though the slope of percentage error change is approximately equal for all three equations.

Modifications needed

Our results from plotting percentage error against measured RMR suggest that the multipliers of weight, height, and/or age may need to be modified for the Owen, Mifflin, and Harris-Benedict equations, in order to increase accuracy across a wide range of RMRs. However, we did not attempt to modify equations in the current study as MANOVA did not show a strong association between any of the variables and the equations.

Owen equation

The Owen equation was originally derived from a sample of healthy women aged 38 [+ or -] 16 (5), which was older than our sample (22 [+ or -] 2); otherwise, the two sample populations were similar, which perhaps explains the better performance of the Owen equation in comparison with the others. Consistent with our findings are those of Siervo et al., who also found the Owen equation more accurate than others in their sample of normal-weight women whose age (23.8 [+ or -] 3.8) was similar to our study population's (22). In that study, the Owen equation predicted RMR, measured by metabolic cart, to within 10 [+ or -] 145 kcal/day at the group level. Interestingly, Siervo et al.'s results also showed large variability, as indicated by the large standard deviation, which suggests inaccuracies would still arise with application of the equation at the individual level. Owen et al. also concluded that their PE for RMR in healthy women was better than those previously available for women; however, its usefulness remained questionable because data derived from their population showed large confidence limits (5).

Harris-Benedict equation

The Harris-Benedict equation produced the largest mean difference from measured RMR and the greatest number of participants whose predicted RMR exceeded 10% of measured RMR. The Harris-Benedict equation was originally validated in a female sample that was older than ours (31 [+ or -] 14 versus 22 [+ or -] 2) (3), and in a population that may be dissimilar to contemporary populations, as the study was published in 1919 (4). Previous studies have shown its inaccuracy in various populations, including those who are obese and clinically ill (23-25), as well as the non-obese (5,8).

Mifflin and Schofield equations

Poor performance of the Mifflin and the two Schofield equations may be attributed to the fact that the populations in which these equations were developed differed considerably from ours. For example, the Mifflin equation was originally developed in a population that included overweight, obese, and severely obese individuals (4); in addition, the mean measured RMR and standard deviation were higher than ours, by 139 and 54 kcal, respectively. This equation is perhaps therefore best applied to a group of people with higher body weights. Indeed, the Mifflin equation may be useful in assessing caloric needs for the US population at large, as BMIs continue to rise (4). The Schofield equations originated from a meta-analysis of approximately 100 studies with data from 7173 participants with a wide age range (three to more than 60 years) (6). Further, the mean measured RMR of the population was 277 kcal higher than our population's. Taken together, these findings suggest that deviation of a sample population from the characteristics of the sample in which an equation is developed may result in inaccurate predictions of RMR.

Study strengths and weaknesses

A unique strength of our study was the investigation of the dependency of the residual on the mean of predicted RMR and measured RMR for each individual. We found that the multiplier of weight, height, or age in the Owen, Mifflin, and Harris-Benedict equations may need to be modified in order to increase accuracy. Otherwise, performance of these equations may be useful in predicting RMRs in only a narrow range.

Because of homogeneity among the participants, we were unable to pinpoint any specific variables for which the multiplier should be adjusted, and we also were unable to verify the extension of underestimation for the Harris-Benedict and Mifflin equations in comparison with measured RMR. Nevertheless, our findings have indicated factors for investigation in future research. Our study would have been strengthened by the use of a metabolic cart.

Finally, having a group of participants with relatively similar characteristics increases our confidence in concluding that, when measuring RMR is not feasible even with a practical handheld device such as the MedGem IC, the Owen equation is most appropriate for young, non-obese women.

RELEVANCE TO PRACTICE

Equations to predict RMR do not perform well against measured RMR in young, reportedly healthy women. Accurate assessment of RMR is important for weight management, even for such a population. For example, young Canadian female university students have been reported to gain weight (2.4 kg), at least within the first six months of university attendance (26). Assessment of RMR with an easy-to-use handheld device, such as the MedGem IC, would therefore be beneficial in dietetic practice in such settings. Weight loss has been demonstrated to be more successful with energy goals developed using MedGem IC-measured RMR than with goals developed from RMR estimated from PEs, a finding that suggests clinical feasibility (27).

Prediction equations should be modified by factors that reduce corresponding percentage error; otherwise, RMR should be measured. If measurement is not possible, we recommend that for the Owen, Mifflin, and Harris-Benedict equations, the multiplier of weight, height, and/or age be modified on the basis of data generated from large studies of participants varying in weight, height, and age, where adjustments in predicted RMR are made according to the percentage error that corresponds with their measured RMR. In the interim, we recommend that the Owen equation be used for young, non-obese women because measured RMR is mostly within 1105 to 1400 kcal/day, the optimal range for the Owen equation. For a female population whose RMR may be higher, the Mifflin equation (optimal range, 1280 to 1595 kcal/day) or the Harris-Benedict equation (optimal range, 1345 to 1630 kcal/day) may be used.

Acknowledgements

We thank the study participants, Nicole Holland for data collection, and Peter Mantha at ManthaMed for use of the MedGem IC. Funding was provided by the University of Guelph.

References

(1.) FAO/WHO/UNU Expert Consultation. Energy and protein requirements. Geneva: World Health Organization; 1985.

(2.) Kacmarek RM, Hess D, Stoller JK. Monitoring in respiratory care. St. Louis, MO: Mosby; 1993.

(3.) Harris JA, Benedict FG. A biometric study of basal metabolism in man. Washington: Carnegie Institution of Washington; 1919.

(4.) Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Clin Nutr. 1990;51:241-7.

(5.) Owen OE, Kavle E, Owen RS, Polansky M, Caprio S, Mozzoli MA, et al. A reappraisal of caloric requirements in healthy women. Am J Clin Nutr. 1986;44:1-19.

(6.) Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr. 1985;39(Suppl 1):5-41.

(7.) Frankenfield D, Roth-Yousey L, Compher C. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review. J Am Diet Assoc. 2005;105:775-89.

(8.) Frankenfield D, Rowe W, Smith S, Cooney R. Validation of several established equations for resting metabolic rate in obese and nonobese people. J Am Diet Assoc. 2003;103:1152-9.

(9.) Garrel DR, Jobin N, DeJonge LHM. Should we still use the Harris-Benedict equations? Nutr Clin Pract. 1996;11:99-103.

(10.) Daly JM, Heymsfield SB, Head CA, Harvey LP, Nixon DW, Katzeff H, et al. Human energy requirements: overestimation by widely used prediction equation. Am J Clin Nutr. 1985; 42:1170-4.

(11.) Feurer ID, Crosby LO, Mullen JF. Measured and predicted resting energy expenditure in clinically stable patients. Clin Nutr. 1984;3:27-34.

(12.) Rubenbauer JR, Johannsen DL, Baier SM, Litchfield R, Flakoll J. The use of a handheld calorimetry unit to estimate energy expenditure during different physiological conditions. J Parenter Enteral Nutr. 2006;30:246-50.

(13.) Stewart C, Goody C, Branson R. Comparison of two systems of measuring energy expenditure. J Parenter Enteral Nutr. 2005;29:212-7.

(14.) Alam DS, Hulshof PJ, Roordink D, Meltzer M, Yunus M, Salam MA, et al. Validity and reproducibility of resting metabolic rate measurements in rural Bangladeshi women: comparison of measurements obtained by MedGem and by Deltatrac device. Eur J Clin Nutr. 2005;59:651-7.

(15.) McDoniel SO. A systematic review on use of a handheld indirect calorimeter to assess energy needs in adults and children. Int J Sport Nutr Exerc Metab. 2007;17:491-500.

(16.) Spears K, Kim H, Behall K, Conway J. Hand-held indirect calorimeter offers advantages compared with prediction equations, in a group of overweight women, to determine resting energy expenditures and estimated total energy expenditures during research screening. J Am Diet Assoc. 2009;109:836-45.

(17.) Hasson RE, Howe CA, Jones BL, Freedson PS. Accuracy of four resting metabolic rate prediction equations: effects of sex, body mass index, age, and race/ ethnicity. J Sci Med Sport. 2011;14:344-51.

(18.) Reeves M, Capra S, Bauer J, Davies P, Battistutta D. Clinical accuracy of the MedGem indirect calorimeter for measuring resting energy expenditure in cancer patients. Eur J Clin Nutr. 2005;59:603-10.

(19.) Weir JB. New methods for calculating metabolic rate with special reference to protein metabolism. J Physiol. 1949 Aug;109(1-2):1-9.

(20.) Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;327:307-10.

(21.) Ludbrook J. Statistical techniques for comparing measurers and methods of measurement: a critical review. Clin Exp Pharmacol Physiol. 2002;29:527-36.

(22.) Siervo M, Boschi V, Falconi C. Which REE prediction equation should we use in normal-weight, overweight and obese women? Clin Nutr. 2003;22: 193-204.

(23.) Frankenfield D, Hise M, Malone A, Russell M, Gradwell E, Compher C. Prediction of resting metabolic rate in critically ill adult patients: results of a systematic review of evidence. J Am Diet Assoc. 2007;107:1552-61.

(24.) Weijs PJ. Validity of predictive equations for resting energy expenditure in US and Dutch overweight and obese class I and II adults aged 18-65 years. Am J Clin Nutr. 2008;88:959-70.

(25.) Pavlou KN, Hoefer MA, Blackburn GL. Resting energy expenditure in moderate obesity. Predicting velocity of weight loss. Ann Surg. 1986;203:136-41.

(26.) Edmonds MJ, Ferreira KJ, Nikiforuk EA, Finnie AK, Leavey SH, Duncan AM, et al. Body weight and percent body fat increase during the transition from high school to university in females. J Am Diet Assoc. 2008;108:1033-7.

(27.) McDoniel S, Nelson H, Thomson C. Employing RMR technology in a 90-day weight control program. Obes Facts. 2008;1:298-304.

ELAINE YAO, BASc, ANDREA C. BUCHHOLZ, PhD, RD, Department of Family Relations and Applied Nutrition, University of Guelph, Guelph, ON; A. MICHELLE EDWARDS, PhD, MLIS, Data Resource Centre, university of Guelph, Guelph, ON; JANIS A. RANDALL SIMPSON, PhD, RD, Department of Family Relations and Applied Nutrition, university of Guelph, Guelph, ON

Determining the resting metabolic rate (RMR) is a key step in nutritional assessment; because the RMR accounts for over 65% of total daily energy requirements (1), obtaining accurate measurements is essential. Traditional indirect calorimeters (ICs), such as the metabolic cart, have been established as the reference standard; however, they are expensive and cumbersome, and require careful calibration. Use of these devices is impractical, costly, and difficult in many settings (2); therefore, clinicians rely on prediction equations (PEs) to calculate energy expenditure. Commonly used PEs include the Harris-Benedict (3), Mifflin (4), Owen (5), [Schofield.sub.weight], and [Schofield.sub.weight and height] (6) equations; however, accuracy of these equations has been questioned (7-11).

A handheld IC, such as the MedGem, is portable, affordable, and easy to use, and therefore feasible for use by health care and research professionals. Measurements of RMR are obtained by using a mouthpiece and nose clip, and are less invasive than metabolic cart measurements and also faster (10 versus 40 minutes). In several studies, the MedGem IC has been validated in comparisons with the metabolic cart for use in normal-weight, healthy adults (12-15), and in comparison with PEs in overweight women (16) and in a group heterogeneous for age, sex, body mass index (BMI), and ethnicity (17).

PURPOSE

Resting metabolic rate measured by the MedGem handheld IC was compared with RMR predicted by five commonly used PEs: the Harris-Benedict, Mifflin, Owen, [Schofield.sub.weight], and [Schofield.sub.weight and height] equations. To our knowledge, ours is the first study in which PEs have been compared with a handheld IC device in young, non-obese Canadian women.

METHODS

Participants

Female students enrolled in a senior undergraduate class in nutritional assessment were recruited from the University of Guelph, Ontario, Canada. Women were eligible to participate if they were enrolled in the course, had no known health conditions (determined by questionnaire), were younger than 30, had a BMI below 30 kg/[m.sup.2], and were not pregnant. Participation was voluntary and did not influence grades. The study was approved by the University of Guelph Research Ethics Board. Informed written consent was obtained before participants enrolled in the study.

Procedures

All tests were performed by the same trained technician in the University of Guelph Body Composition and Metabolism Laboratory (http://www.uoguelph.ca/bodycomp/). Tests were done between 0930 and 1430, from October to December 2010. Before testing, participants confirmed with a questionnaire that they had fasted from food and alcohol for 12 hours, and had had no exercise in the previous 24 hours. Weight (in light clothing) was measured to the nearest 0.1 kg with an electronic scale (Life-Source MD, A&D Engineering, San Jose, CA), when participants had voided within the preceding 30 minutes. Height without shoes was measured to the nearest 0.1 cm with a wall-mounted stadiometer (Seca Corp., Ontario, CA).

Resting metabolic rate: The RMR (kcal/day) was measured in a private, brightly lit room with a temperature of 22.5 [+ or -] 0.7 oC and 28 [+ or -] 6% humidity. The MedGem handheld IC (MedGem, Microlife Medical Home Solutions, Inc., Golden, CO) was used while participants were awake, motionless, and supine. Participants watched a DVD to stay awake and were monitored throughout the test, which took 10 to 12 minutes.

Autocalibration was performed on a flat surface before each use of the device, according to the manufacturer's instructions. Before measurement, participants rested for at least 10 minutes in the supine position. A new nose clip and disposable mouthpiece were used for each test. Participants breathed through the mouthpiece with a firm lip seal, while holding the MedGem IC in place until it beeped and the indicator light changed to amber, which signalled the end of the measurement. The MedGem IC uses a proprietary algorithm to detect achievement of steady state; this determines termination and calculation of the RMR, which is displayed on the MedGem IC LCD screen (18). The RMR was calculated from measured VO2, using the manufacturer's suggested modification of the Weir equation (19), as follows: kcal /day = (3.941 x V[O.sub.2]) + 1.106 (0.85 x V[O.sub.2]) - 0.365y, where VO2 (mL/min) was assumed at a constant respiratory quotient of 0.85 and 0.365y is the correction for protein metabolism.

In addition, the RMR was predicted with five published equations: the Harris-Benedict (3), Mifflin (4), Owen (5), [Schofield.sub.weight], and [Schofield.sub.weight and height] equations (6).

Data analyses

For statistical analysis, we used IBM SPSS Statistics (version 20.0, IBM Corp., Somers, NY, 2011), with significance set at p [less than or equal to] 0.05. Several tests were conducted to determine the accuracy of the RMR predicted by the five equations in comparison with that measured by the MedGem device.

The RMR data were normally distributed, as confirmed by Jarque-Bera tests. Paired /-tests were conducted to compare group mean differences between measured RMR and predicted RMR for each equation. The proportion of participants for whom measured RMR and predicted RMR differed by 10% or more was also determined. These two tests permitted a general comparison of results at the group level.

For each equation, a Bland-Altman plot was created to provide a more accurate comparison of methods than correlation alone (20). Bland-Altman plots can detect proportional and fixed biases, allowing the detection of random error of both predicted and measured methods (21). Each Bland-Altman plot was created by plotting the residual (predicted RMR minus measured RMR) against the mean ([predicted + measured RMR]/2), and generating limits of agreement ([+ or -] 2 standard deviations, 95%), to permit assessment of systematic bias between measured RMR and predicted RMR.

Additionally, the magnitude of association between measured RMR and predicted RMR was determined using Pearson's correlation coefficient, r; further, [R.sup.2] was adjusted to account for the number of variables in the regression equations. The standard error of estimate (SEE) for each PE was determined using linear regression analysis.

Last, multiple analysis of variance (MANOVA) was used to assess whether age, height, weight, or BMI was associated with differences between predicted RMR and measured RMR. Regression analysis followed to detect whether significant associations existed for any particular variable.

RESULTS

Fifty-five eligible students participated. Three were excluded from analyses: one had consumed alcohol within 12 hours and two had exercised within 24 hours of testing. Mean BMI of the 52 participants (aged 22 [+ or -] 2 years) was 21.8 [+ or -] 2.7 kg/[m.sup.2]; six (12%) were classified as underweight (BMI below 18.5 kg/[m.sup.2]), 39 (75%) were normal weight (BMI 18.5 to 24.9 kg/[m.sup.2]), and seven (13%) were overweight (BMI 24.9 to 29.9 kg/[m.sup.2]).

Data on statistical tests used to determine agreement between measured RMR and predicted RMR are shown in Table 1. Paired /"-tests indicated that four of the five PEs led to significant overestimation of measured RMR. The smallest difference was 16 [+ or -] 133 kcal/day (Owen equation), and the largest was 225 [+ or -] 133 kcal/day (Harris-Benedict equation). Predicted RMR values exceeded 10% of measured RMR values for 35% (Owen equation) to 76% (Harris-Benedict equation) of participants.

Bland-Altman analyses revealed good agreement for all five equations, as indicated by the fact that 94% to 96% of the data points fell within two standard deviations (Figures 1a to 1e). However, three of the five PEs (the Owen, Mifflin, and Harris-Benedict equations) showed a systematic bias (p<0.05). That is, the agreement between the residual and the mean of measured and predicted RMR depended on the measured RMR value. According to the agreement shown in Bland-Altman analysis with the mean of the two methods as the independent variable, as RMR increased, agreement decreased. No bias was detected in the two Schofield equations.

For further investigation of biases shown by Bland-Altman analysis for the Owen, Mifflin, and Harris-Benedict equations, measured RMR and predicted RMR were separately plotted against percentage difference (error). A linear trend line was inserted to show the average pattern. Predicted RMRs were not strongly correlated with percentage difference (data not shown), but measured RMRs were. For every 100 kcal increase in measured RMR, the average percentage error decreased by 6% to 8% (Figures 2a to 2c). The RMR range in which percentage error was the lowest (i.e., in which predicted RMR agreed with measured RMR within [+ or -] 10%) was different for each equation. The Owen equation range was 1105 to 1400 kcal/ day with 0% error at approximately 1250 kcal/day; the Mifflin range started at 1280 kcal/day and extended to 1595 kcal/ day with a 0% error at approximately 1275 kcal/day; the range for the Harris-Benedict equation began at 1345 kcal/day and extended to 1630 kcal/day, with 0% error at approximately 1480 kcal/day.

Relative to the RMR measured by the MedGem IC, Pearson's correlations for RMR, predicted by all five equations, were significant. Predicted RMR accounted for 27% (Harris-Benedict) to 34% (Owen) of the variance of measured RMR for all equations. The SEE, determined by linear regression analysis, was similar for all equations.

MANOVA showed no significant association of age, height, weight, or BMI with the residuals for any equation. MANOVA [R.sup.2] indicated that no variable in any equation accounted for the residual at statistically significant levels.

DISCUSSION

Accuracy of equations

In this study, the performance of PEs was variable in comparison with measured RMR. Variations were especially large at the individual level over the range of measured RMRs (SD [+ or -] 133 to 139 kcal/day). At the group level, the Owen equation (5) performed best in this population of young, non-obese women, as the mean RMR predicted for the group differed least from the measured RMR (16 kcal/day). In addition, this equation captured the greatest proportion of individuals (65%) for whom the predicted RMR was within 10% of the measured RMR. Differences between measured RMR and RMR predicted by the other four equations--the Harris-Benedict (3), Mifflin (4), [Schofield.sub.weight] (6), and [Schofield.sub.weight and height] (6) equations--ranged from 159 kcal/day with the Mifflin equation to 225 kcal/day with the Harris-Benedict equation. The RMR predicted by these four equations differed by more than 10% from measured RMR for more than two-thirds of the young women included in this study. A recent United States study, similar to ours, included 186 women aged 18 to 60 and indicated higher overall MedGem-measured RMR (141 [+ or -] 214 kcal/day) and higher BMI (24.8 [+ or -] 4.9 kg/[m.sup.2]) than we found for our participants (17).

No equation accurately predicted individual RMR consistently across the full range of measured RMRs, a finding similar to those of studies in overweight women (16) and in a heterogeneous group of women aged 18 to 60 (17). Rather, in particular for the Owen, Mifflin, and Harris-Benedict equations, a negative linear relationship existed between measured RMR and the residual (predicted RMR minus measured RMR). That is, as measured RMR decreased, the likelihood of overestimation increased. As measured RMR increased, overestimation decreased; however, once RMR increased beyond the point where the residual was nil or near nil (approximately 1250, 1440, or 1480 kcal/day, depending on the equation), the equations began to underestimate measured RMR. In other words, if an individual's measured RMR fell below 1250 to 1480 kcal/day, the equations tended to overestimate RMR; if measured RMR was above 1250 to 1480 kcal/day, the equations tended to underestimate RMR. These results suggest that only a narrow range of measured RMR exists for which PEs can be used with any degree of confidence. Outside this narrow range, predictions become inaccurate (either overestimating or underestimating measured RMR) and may not be suitable for clinical or research use, especially at the individual level.

Over- and underestimation

Systematic bias in the Owen, Mifflin, and Harris-Benedict equations was consistent with findings from Spears et al.'s study, in which handheld IC measurements were compared with PEs (16). Similarly, Spears et al. reported the same relationship of overestimation trending to underestimation as measured RMR increased (16). In our study, the slope of this relationship was relatively similar across all three equations, which suggests that, given a sufficiently wide RMR range, all three equations would perform equally poorly at both the individual and the group levels.

The reason that the Owen equation appeared to perform better at the group level in our study can be attributed to the fact that our participants' mean measured RMR was closest to the RMR at the point of the nil residual in the Owen equation. As such, this equation would have been able to predict the largest proportion of our participants' RMR within 10% of their measured RMR. Additionally, the proportion of individuals with a measured RMR above the nil residue point of the Owen equation was relatively equal to the proportion below. The degree of underestimation countered the degree of overestimation, and therefore the difference between measured RMR and Owen equation-predicted RMR appeared reduced at the group level.

The RMR of the Mifflin and Harris-Benedict equations at the point of the nil residual was higher than our group's mean measured RMR. Therefore, application of those two equations led to greater overestimations (more than 10% difference). Underestimation was not as common for the Mifflin and Harris-Benedict equations, as our sample did not include many women whose measured RMR exceeded the RMR predicted by the Mifflin or Harris-Benedict RMR of the nil residual. As a result, the Mifflin and Harris-Benedict equations, more than the Owen equation, appeared to differ from measured RMR at the group level, even though the slope of percentage error change is approximately equal for all three equations.

Modifications needed

Our results from plotting percentage error against measured RMR suggest that the multipliers of weight, height, and/or age may need to be modified for the Owen, Mifflin, and Harris-Benedict equations, in order to increase accuracy across a wide range of RMRs. However, we did not attempt to modify equations in the current study as MANOVA did not show a strong association between any of the variables and the equations.

Owen equation

The Owen equation was originally derived from a sample of healthy women aged 38 [+ or -] 16 (5), which was older than our sample (22 [+ or -] 2); otherwise, the two sample populations were similar, which perhaps explains the better performance of the Owen equation in comparison with the others. Consistent with our findings are those of Siervo et al., who also found the Owen equation more accurate than others in their sample of normal-weight women whose age (23.8 [+ or -] 3.8) was similar to our study population's (22). In that study, the Owen equation predicted RMR, measured by metabolic cart, to within 10 [+ or -] 145 kcal/day at the group level. Interestingly, Siervo et al.'s results also showed large variability, as indicated by the large standard deviation, which suggests inaccuracies would still arise with application of the equation at the individual level. Owen et al. also concluded that their PE for RMR in healthy women was better than those previously available for women; however, its usefulness remained questionable because data derived from their population showed large confidence limits (5).

Harris-Benedict equation

The Harris-Benedict equation produced the largest mean difference from measured RMR and the greatest number of participants whose predicted RMR exceeded 10% of measured RMR. The Harris-Benedict equation was originally validated in a female sample that was older than ours (31 [+ or -] 14 versus 22 [+ or -] 2) (3), and in a population that may be dissimilar to contemporary populations, as the study was published in 1919 (4). Previous studies have shown its inaccuracy in various populations, including those who are obese and clinically ill (23-25), as well as the non-obese (5,8).

Mifflin and Schofield equations

Poor performance of the Mifflin and the two Schofield equations may be attributed to the fact that the populations in which these equations were developed differed considerably from ours. For example, the Mifflin equation was originally developed in a population that included overweight, obese, and severely obese individuals (4); in addition, the mean measured RMR and standard deviation were higher than ours, by 139 and 54 kcal, respectively. This equation is perhaps therefore best applied to a group of people with higher body weights. Indeed, the Mifflin equation may be useful in assessing caloric needs for the US population at large, as BMIs continue to rise (4). The Schofield equations originated from a meta-analysis of approximately 100 studies with data from 7173 participants with a wide age range (three to more than 60 years) (6). Further, the mean measured RMR of the population was 277 kcal higher than our population's. Taken together, these findings suggest that deviation of a sample population from the characteristics of the sample in which an equation is developed may result in inaccurate predictions of RMR.

Study strengths and weaknesses

A unique strength of our study was the investigation of the dependency of the residual on the mean of predicted RMR and measured RMR for each individual. We found that the multiplier of weight, height, or age in the Owen, Mifflin, and Harris-Benedict equations may need to be modified in order to increase accuracy. Otherwise, performance of these equations may be useful in predicting RMRs in only a narrow range.

Because of homogeneity among the participants, we were unable to pinpoint any specific variables for which the multiplier should be adjusted, and we also were unable to verify the extension of underestimation for the Harris-Benedict and Mifflin equations in comparison with measured RMR. Nevertheless, our findings have indicated factors for investigation in future research. Our study would have been strengthened by the use of a metabolic cart.

Finally, having a group of participants with relatively similar characteristics increases our confidence in concluding that, when measuring RMR is not feasible even with a practical handheld device such as the MedGem IC, the Owen equation is most appropriate for young, non-obese women.

RELEVANCE TO PRACTICE

Equations to predict RMR do not perform well against measured RMR in young, reportedly healthy women. Accurate assessment of RMR is important for weight management, even for such a population. For example, young Canadian female university students have been reported to gain weight (2.4 kg), at least within the first six months of university attendance (26). Assessment of RMR with an easy-to-use handheld device, such as the MedGem IC, would therefore be beneficial in dietetic practice in such settings. Weight loss has been demonstrated to be more successful with energy goals developed using MedGem IC-measured RMR than with goals developed from RMR estimated from PEs, a finding that suggests clinical feasibility (27).

Prediction equations should be modified by factors that reduce corresponding percentage error; otherwise, RMR should be measured. If measurement is not possible, we recommend that for the Owen, Mifflin, and Harris-Benedict equations, the multiplier of weight, height, and/or age be modified on the basis of data generated from large studies of participants varying in weight, height, and age, where adjustments in predicted RMR are made according to the percentage error that corresponds with their measured RMR. In the interim, we recommend that the Owen equation be used for young, non-obese women because measured RMR is mostly within 1105 to 1400 kcal/day, the optimal range for the Owen equation. For a female population whose RMR may be higher, the Mifflin equation (optimal range, 1280 to 1595 kcal/day) or the Harris-Benedict equation (optimal range, 1345 to 1630 kcal/day) may be used.

Acknowledgements

We thank the study participants, Nicole Holland for data collection, and Peter Mantha at ManthaMed for use of the MedGem IC. Funding was provided by the University of Guelph.

References

(1.) FAO/WHO/UNU Expert Consultation. Energy and protein requirements. Geneva: World Health Organization; 1985.

(2.) Kacmarek RM, Hess D, Stoller JK. Monitoring in respiratory care. St. Louis, MO: Mosby; 1993.

(3.) Harris JA, Benedict FG. A biometric study of basal metabolism in man. Washington: Carnegie Institution of Washington; 1919.

(4.) Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Clin Nutr. 1990;51:241-7.

(5.) Owen OE, Kavle E, Owen RS, Polansky M, Caprio S, Mozzoli MA, et al. A reappraisal of caloric requirements in healthy women. Am J Clin Nutr. 1986;44:1-19.

(6.) Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr. 1985;39(Suppl 1):5-41.

(7.) Frankenfield D, Roth-Yousey L, Compher C. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review. J Am Diet Assoc. 2005;105:775-89.

(8.) Frankenfield D, Rowe W, Smith S, Cooney R. Validation of several established equations for resting metabolic rate in obese and nonobese people. J Am Diet Assoc. 2003;103:1152-9.

(9.) Garrel DR, Jobin N, DeJonge LHM. Should we still use the Harris-Benedict equations? Nutr Clin Pract. 1996;11:99-103.

(10.) Daly JM, Heymsfield SB, Head CA, Harvey LP, Nixon DW, Katzeff H, et al. Human energy requirements: overestimation by widely used prediction equation. Am J Clin Nutr. 1985; 42:1170-4.

(11.) Feurer ID, Crosby LO, Mullen JF. Measured and predicted resting energy expenditure in clinically stable patients. Clin Nutr. 1984;3:27-34.

(12.) Rubenbauer JR, Johannsen DL, Baier SM, Litchfield R, Flakoll J. The use of a handheld calorimetry unit to estimate energy expenditure during different physiological conditions. J Parenter Enteral Nutr. 2006;30:246-50.

(13.) Stewart C, Goody C, Branson R. Comparison of two systems of measuring energy expenditure. J Parenter Enteral Nutr. 2005;29:212-7.

(14.) Alam DS, Hulshof PJ, Roordink D, Meltzer M, Yunus M, Salam MA, et al. Validity and reproducibility of resting metabolic rate measurements in rural Bangladeshi women: comparison of measurements obtained by MedGem and by Deltatrac device. Eur J Clin Nutr. 2005;59:651-7.

(15.) McDoniel SO. A systematic review on use of a handheld indirect calorimeter to assess energy needs in adults and children. Int J Sport Nutr Exerc Metab. 2007;17:491-500.

(16.) Spears K, Kim H, Behall K, Conway J. Hand-held indirect calorimeter offers advantages compared with prediction equations, in a group of overweight women, to determine resting energy expenditures and estimated total energy expenditures during research screening. J Am Diet Assoc. 2009;109:836-45.

(17.) Hasson RE, Howe CA, Jones BL, Freedson PS. Accuracy of four resting metabolic rate prediction equations: effects of sex, body mass index, age, and race/ ethnicity. J Sci Med Sport. 2011;14:344-51.

(18.) Reeves M, Capra S, Bauer J, Davies P, Battistutta D. Clinical accuracy of the MedGem indirect calorimeter for measuring resting energy expenditure in cancer patients. Eur J Clin Nutr. 2005;59:603-10.

(19.) Weir JB. New methods for calculating metabolic rate with special reference to protein metabolism. J Physiol. 1949 Aug;109(1-2):1-9.

(20.) Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;327:307-10.

(21.) Ludbrook J. Statistical techniques for comparing measurers and methods of measurement: a critical review. Clin Exp Pharmacol Physiol. 2002;29:527-36.

(22.) Siervo M, Boschi V, Falconi C. Which REE prediction equation should we use in normal-weight, overweight and obese women? Clin Nutr. 2003;22: 193-204.

(23.) Frankenfield D, Hise M, Malone A, Russell M, Gradwell E, Compher C. Prediction of resting metabolic rate in critically ill adult patients: results of a systematic review of evidence. J Am Diet Assoc. 2007;107:1552-61.

(24.) Weijs PJ. Validity of predictive equations for resting energy expenditure in US and Dutch overweight and obese class I and II adults aged 18-65 years. Am J Clin Nutr. 2008;88:959-70.

(25.) Pavlou KN, Hoefer MA, Blackburn GL. Resting energy expenditure in moderate obesity. Predicting velocity of weight loss. Ann Surg. 1986;203:136-41.

(26.) Edmonds MJ, Ferreira KJ, Nikiforuk EA, Finnie AK, Leavey SH, Duncan AM, et al. Body weight and percent body fat increase during the transition from high school to university in females. J Am Diet Assoc. 2008;108:1033-7.

(27.) McDoniel S, Nelson H, Thomson C. Employing RMR technology in a 90-day weight control program. Obes Facts. 2008;1:298-304.

ELAINE YAO, BASc, ANDREA C. BUCHHOLZ, PhD, RD, Department of Family Relations and Applied Nutrition, University of Guelph, Guelph, ON; A. MICHELLE EDWARDS, PhD, MLIS, Data Resource Centre, university of Guelph, Guelph, ON; JANIS A. RANDALL SIMPSON, PhD, RD, Department of Family Relations and Applied Nutrition, university of Guelph, Guelph, ON

Table 1 Agreement between resting metabolic rate (RMR) measured with MedGem handheld indirect calorimetry and RMR predicted with five published equations, for 52 young, non-obese women Variable MedGem Harris and RMR (a) Benedict (b) RMR (kcal/day) (g) 1210 [+ or -] 160 1435 [+ or -] 93 Paired f-test Difference between -- 225 [+ or -] 133 predicted and measured RMR (kcal/day) 95% confidence limits -- 188-262 (p value) -- (<0.001) Bland-Altman % within [+ or -] 2 SD -- 96% [R.sup.2] biases -- 0.32 (p value) -- (<0.001) Pearson's correlations r -- 0.58 (p value) -- (<0.001) Adjusted [R.sup.2] -- 0.30 Standard error of -- 134 estimate (kcal/day) Number of subjects for -- 40 whom predicted and measured RMR differed by [greater than or equal to] 10% Variable Mifflin (c) Owen (d) RMR (kcal/day) (g) 1369 [+ or -] 120 1226 [+ or -] 64 Paired f-test Difference between 159 [+ or -] 139 16 [+ or -] 133 predicted and measured RMR (kcal/day) 95% confidence limits 120-198 -20-53 (p value) (<0.001) (=0.38) Bland-Altman % within [+ or -] 2 SD 94% 96% [R.sup.2] biases 0.09 0.63 (p value) (<0.001) (<0.001) Pearson's correlations r 0.59 0.59 (p value) (<0.001) (<0.001) Adjusted [R.sup.2] 0.27 0.33 Standard error of 136 131 estimate (kcal/day) Number of subjects for 34 18 whom predicted and measured RMR differed by [greater than or equal to] 10% Variable [Schofield.sub. [Schofield.sub. weight] (e) weight and height] (f) RMR (kcal/day) (g) 1376 [+ or -] 132 1385 [+ or -] 134 Paired f-test Difference between 166 [+ or -]135 174 [+ or -] 137 predicted and measured RMR (kcal/day) 95% confidence limits 128-203 137-213 (p value) (<0.001) (<0.001) Bland-Altman % within [+ or -] 2 SD 96% 94% [R.sup.2] biases 0.05 0.05 (p value) (=0.10) (=0.13) Pearson's correlations r 0.59 0.58 (p value) (<0.001) (<0.001) Adjusted [R.sup.2] 0.33 0.34 Standard error of 131 132 estimate (kcal/day) Number of subjects for 35 35 whom predicted and measured RMR differed by [greater than or equal to] 10% SD = standard deviation (a) Mean [+ or -] SD (b) Harris-Benedict equation (3): 65 5.0955 + 9.5634 weight (kg) + 1.8496 height (cm) - 4.6756 age (years) (c) Mifflin equation (4): 9.99 weight (kg) + 6.25 height (cm) - 4.9 age (years) - 161 (d) Owen equation (5): 795 + 7.18 weight (kg) (e) [Schofield.sub.weight] (females 18 to 30 years) equation (6): 14.808 weight (kg) + 486.29 (f) [Schofield.sub.weight and height] equation (6): 13.614 weight (kg) + 2.828 height (cm) + 98.166 (g) Coefficient of variation for repeated measurements = 7% (14)

Printer friendly Cite/link Email Feedback | |

Title Annotation: | Research/Recherche |
---|---|

Author: | Yao, Elaine; Buchholz, Andrea C.; Edwards, A. Michelle; Simpson, Janis A. Randall |

Publication: | Canadian Journal of Dietetic Practice and Research |

Article Type: | Report |

Geographic Code: | 1CANA |

Date: | Sep 22, 2013 |

Words: | 4543 |

Previous Article: | Relations entre les pratiques alimentaires/maternelles et l'alimentation d'enfants quebecois d'age prescolaire. |

Next Article: | A study of challenges and opportunities for long-term care dietitians in British Columbia. |

Topics: |