Validity of predictive equations for resting energy expenditure in US and Dutch overweight and obese class I and II adults aged y 1 3

Transcription

1 Validity of predictive equations for resting energy expenditure in US and Dutch overweight and obese class I and II adults aged y 1 3 Peter JM Weijs ABSTRACT Background: Individual energy requirements of overweight and obese adults can often not be measured by indirect calorimetry. Objective: The objective was to analyze which resting energy expenditure (REE) predictive equation was the best alternative to indirect calorimetry in US and Dutch adults aged y with a body mass index (in kg/m 2 )of25to4. Design: Predictive equations based on weight, height, sex, age, fat-free mass, and fat mass were tested. REE in Dutch adults was measured with indirect calorimetry, and published data from the Institute of Medicine were used for US adults. The accuracy of the equations was evaluated on the basis of the percentage of subjects predicted within 1% of the REE measured, the root mean squared prediction error (RMSE), and the mean percentage difference (bias) between predicted and measured REE. Results: Twenty-seven predictive equations (9 of which were based on FFM) were included. Validation was based on 18 women and 158 men from the United States and on 154 women and 54 men from the Netherlands aged 65 y with a body mass index (in kg/m 2 )of25 to 4. Most accurate and precise for the US adults was the Mifflin equation (prediction accuracy: 79%; bias: 1.%; RMSE: 136 kcal/ d), for overweight Dutch adults was the FAO/WHO/UNU weight equation (prediction accuracy: 68%; bias: 2.5%; RMSE: 178), and for obese Dutch adults was the Lazzer equation (prediction accuracy: 69%; bias: 3.%; RMSE: 215 kcal/d). Conclusions: For US adults aged y with a body mass index of 25 to 4, the REE can best be estimated with the Mifflin equation. For overweight and obese Dutch adults, there appears to be no accurate equation. Am J Clin Nutr 28;88: INTRODUCTION The prevalence of overweight and obesity is high and increasing (1, 2). Any weight-reduction program will try to establish a reachable goal for weight loss and a reachable goal for dietary intake. This requires knowledge of individual energy requirements and relies on accurate methods of assessment. Because the gold standard, indirect calorimetry, is hardly feasible in most dietetic settings, it remains important to use the most accurate predictive equation to determine resting energy expenditure (REE) in overweight and obese persons (3). Predictive equations have generally been developed in healthy subjects on the basis of regression analysis of body weight, height, sex, and age as independent variables and measured REE by indirect calorimetry as a dependent variable. On the basis of a comparison of published evidence from Harris and Benedict (4), FAO/WHO/UNU weight or weight and height equations (5), and the equations of Mifflin (6) and Owen (7, 8), Frankenfield et al (9) have advised the use of the Mifflin equation for overweight and obese subjects. However, this expert panel also acknowledges that there are limited data to support the use of the Mifflin equation in overweight and obese subjects. The level of overweight might be an important factor in the accuracy of the predictive equation, but the level of overweight varies among studies. For most equations, overweight and obese subjects were included, but their relative contribution to the final equation often remains unclear. Therefore, validation of predictive equations should be performed in specific overweight and obese groups of subjects. Recent evaluations of the validity of REE predictive equations have been published for overweight and obese subjects (1 13) and for extremely obese subjects with a body mass index (BMI; in kg/m 2 ) 4 (14 18). Only a few studies have validated equations for a clearly defined overweight group (BMI 25 3) (12) or obese group (BMI 3 4) (1). The range of published REE predictive equations has not been validated, including equations based on body composition [fatfree mass (FFM) and fat mass (FM)]. As part of evidence-based practice, the literature was systematically searched for REE predictive equations, and subsequently included REE equations were validated with indirect calorimetry data from adults aged y with a BMI of 25 to 4 to find the most accurate and precise REE predictive equation. To prevent an overgeneralization of our conclusions, all equations were applied to both US (published) and Dutch (new) data. 1 From the Department of Nutrition and Dietetics, Hogeschool van Amsterdam, University of Applied Science, and the Department of Nutrition and Dietetics, VU University Medical Center, Amsterdam, Netherlands. 2 Supported by the Hogeschool van Amsterdam, Amsterdam, Netherlands. 3 Address correspondence to PJM Weijs, Department of Nutrition and Dietetics, Hogeschool van Amsterdam, University of Applied Science, Dr Meurerlaan 8, 167 SM Amsterdam, Netherlands. p.j.m.weijs@hva.nl. Received November 28, 27. Accepted for publication June 25, 28. Am J Clin Nutr 28;88: Printed in USA. 28 American Society for Nutrition 959

2 96 WEIJS SUBJECTS AND METHODS Subjects For the US group, indirect calorimetry data were obtained from the National Academy of Sciences report on Dietary Reference Intakes (19). Only adults aged y with a BMI of 25 to 4 were included. The Dutch subjects were included from different weight-loss studies at the Nutrition Lab of the Department of Nutrition and Dietetics, Hogeschool van Amsterdam, University of Applied Science, Amsterdam, Netherlands between 3 April 26 and 27 March 28. Inclusion criteria for the original weight-loss studies were a BMI 25 and an age of y. Data were only included when the subject s BMI at the time of indirect calorimetry was between 25 and 4. All participants gave informed consent. All procedures were in accordance with the ethical standards of the institution. Indirect calorimetry and anthropometric measures The indirect calorimetry measurements were performed with a ventilated hood system (Vmax Encore n29; Viasys Healthcare, Houten, Netherlands), which was calibrated for volume and with 2 standard gases every day before use. Additionally, the ventilated hood system was automatically recalibrated every 5 min. Measurements were standardized by internal guidelines. The subjects were in a supine position and awake and had fasted overnight or for 4 h before the measurement was made if the measurements could not be performed before noon. Subjects had not been physically active. Oxygen consumption and carbon dioxide production were measured, and energy expenditure was calculated by using the Weir formula (2). The measurements took 3 min, and only steady state periods of measurement were selected according to the procedures for the ventilated hood system. The first 5 min of the measurements were discarded. An acceptable CV was 1%. Body weight, FFM, and FM were assessed by using the BodPod system (Life Measurement Inc, Concord, CA). BodPod was calibrated immediately before each measurement. Height was measured by using a stadiometer (Seca 222; Seca, Hamburg, Germany). REE predictive equations PubMed was used for a systematic search for publications on Mesh-derived keys Energy metabolism, Basal metabolism, and Indirect calorimetry and additional terms ( predict*, estimat*, equation*, and formula* ) in every possible combination. Applied limitations were english language and humans and age of 18 y. More references were obtained by screening publications cited. Only equations developed in adults were retrieved. Inclusion criteria were as follows: equations based on body weight, height, age, sex and/or FFM and FM. Exclusion criteria were as follows: age range (only young adults or only elderly), (critically ill) patients, mean BMI 25 (indication of small proportion of overweight and obese; not applicable to large databases of Harris and Benedict, Schofield, and Oxford), insufficient information, specific ethnic group, small sample size (n 5), impractical or suspect body composition as variable (including percentage ideal body weight), plasma values of glucose or insulin or diabetes as variable, suspect indirect calorimetry, total energy expenditure, athletes, duplicate publications. From each included study the best performing equations, based on the highest value for explained variance (R 2 ), were included. However, extra equations were included when based on weight and height (versus weight only) or FFM and also when equations were BMI group specific (different equations for BMI 25 3 and BMI 3 4). After this selection, the studies were judged for the methodologic quality of the calorimetry TABLE 1 Subject characteristics 1 US Dutch Total group Women: BMI 25-3 Women: BMI 3-4 Men: BMI 25-3 Men: BMI 3-4 Total group Women: BMI 25-3 Women: BMI 3-4 Men: BMI 25-3 Men: BMI 3-4 No. of subjects Age (y) Height (cm) Body weight (kg) BMI (kg/m 2 ) REE (kcal/d) REE (kcal/kg body wt) Fat mass (%) Fat mass (kg) FFM (kg) REE (kcal/kg FFM) RQ FFM, fat-free mass; REE, resting energy expenditure; RQ, respiratory quotient. 2 x SD (all such values).

3 TABLE 2 Predictive equations for resting energy expenditure 1 PREDICTING ENERGY EXPENDITURE FOR OBESE ADULTS 961 Reference No. of subjects, sex, age range or mean, BMI range or mean; body-composition method when applicable; remarks on large databases Statistics and cross-validation REE predictive equations United States HB1919 (4) n 239 (136 M, 13 F) M: r.86, CL M: WT HTCM 5.33 AGE F: r.77, CL 212 F: WT HTCM AGE HB1984 (27) n 337 (168 M, 169 F) M: r.86, CL 213 M: WT HTCM AGE F: r.83, CL 21 F: WT 3.98 HTCM 4.33 AGE Bernstein et al (28) n 22 (48 M, 154 F); mean age 4 y; BMI 37; 6 24 kg M: R M: 11.2 WT 1.23 HTCM 5.8 AGE 132 F: R F: 7.48 WT.42 HTCM 3 AGE 844 M: R FFM 3.72 FM 1.55 F: R AGE Bernstein et al, FFM (28) n 22 (48 M, 154 F), mean age 4 y, BMI 37; 6 24 kg, 3 H 2 O dilution Owen et al (7, 8) n 14 (6 M, 44 F), y M: r.71 M: WT BMI 18 5 F: r.74 F: WT Owen et al, FFM (7, 8) n 14 (6 M, 44 F), y M: r.74 M: 22.3 FFM 29 BMI 18 5, underwater weighing F: r.71 F: 19.7 FFM 334 Mifflin et al (6) Mifflin et al, FFM (6) Livingston and Kohlstadt (29) World Schofield, weight (3), MJ/d Schofield, weight and height (3), MJ/d n 498 (251 M, 248 F), n 264 normal weight (129 M, 135 F), n 234 obese (122 M, 112 F), y, BMI 17 42, skinfolds n 498 (251 M, 248 F), n 264 normal weight (129 M, 135 F), n 234 obese (122 M, 112 F), y, BMI 17 42, n 655 (299 M, 356 F), age y, kg (data from reference 4: n 239; references 7 and 8: n 14, measured n 312) n 7173, n y, mean BMI of these 6 groups: 21 24; n 3388 Italians (47%), n 615 tropical residents, n 322 Indian; 114 published studies, n 7173 subjects (11 values, includes group mean values); most European and North American subjects (Italian, closed circuit calorimetry) n 7173, n y, mean BMI of these 6 groups: 21 24; n 3388 Italians (47%), n 615 tropical residents, n 322 Indian; 114 published studies, n 7173 subjects (11 values, includes group mean values); most European and North American subjects (Italian, closed circuit calorimetry) R WT 6.25 HTCM 4.92 AGE 166 SEX 161 R FFM 413 Development (M: R 2.77, F: R 2.71); cross-validation (data from reference 19: n 767, 334 M, 433 F; age 2 96 y; kg; M: R 2.66; F: R 2.64) M: 293 WT AGE F: 248 WT AGE r.65, SE.64; n 2879 M: AGE 18 3 y:.63 WT r.6, SE.7; n 646 M: AGE 3 6 y:.48 WT r.74, SE.66; n 5 M: AGE 6 y:.49 WT r.73, SE.49; n 829 F: AGE 18 3 y:.62 WT 2.36 r.68, SE.47; n 372 F: AGE 3 6 y:.34 WT r.73, SE.43; n 38 F: AGE 6 y:.38 WT r.65, SE.64; n 2879 M: AGE 18 3 y:.63 WT.42 HTM r.6, SE.7; n 646 M: AGE 3 6 y:.48 WT.11 HTM 3.67 r.74, SE.66; n 5 M: AGE 6 y:.38 WT 4.68 HTM r.73, SE.49; n 829 F: AGE 18 3 y:.57 WT HTM.411 r.68, SE.47; n 372 F: AGE 3 6 y:.34 WT.6 HTM 3.53 r.73, SE.43; n 38 F: AGE 6 y:.33 WT HTM.74 (Continued; additional data columns shown on next page)

4 962 WEIJS TABLE 2 (Continued) Reference No. of subjects, sex, age range or mean, BMI range or mean; body-composition method when applicable; remarks on large databases Statistics and cross-validation REE predictive equations FAO, weight (5) FAO, weight and height (5) Henry, weight (31), MJ/d Henry, weight and height (31), MJ/d Germany Müller et al (12), MJ/d Müller et al, BMI (12), MJ/d Müller et al, FFM (12), MJ/d Müller et al, BMI and FFM (12), MJ/d This report mentions that the equations are based on Schofield et al (1985); however, the database was extended to 11 subjects This report mentions that the equations are based on Schofield et al (1985); however, the database was extended to 11 subjects n 1552 (5794 M, 472 F) Oxford database (166 separate investigations, only individual data points; all Italian, closed circuit data excluded) n 1552 (5794 M, 472 F) Oxford database (166 separate investigations, only individual data points; all Italian, closed circuit data excluded) n 2528 (127 M, 151 F); development: n 146 (388 M, 658 F), age 5 8 y; mean BMI 27 n 2528 (127 M, 151 F); development: n 146 (388 M, 658 F), age 5 8 y; mean BMI 27 n 2528 (127 M, 151 F); development: n 146 (388 M, 658 F), age 5 8 y; mean BMI 27; bioimpedance analysis (different equations, multicenter study) n 2528 (127 M, 151 F); development: n 146 (388 M, 658 F), age 5 8 y; mean BMI 27; bioimpedance analysis (different equations, multicenter study) r.65, SD 151 M: AGE 18 3 y: 15.3 WT 679 r.6, SD 164 M: AGE 3 6 y: 11.6 WT 879 r.79, SD 148 M: AGE 6 y: 13.5 WT 487 r.72, SD 121 F: AGE 18 3 y: 14.7 WT 496 r.7, SD 18 F: AGE 3 6 y: 8.7 WT 829 r.74, SD 18 F: AGE 6 y: 1.5 WT 596 r.65, RSD 151 M: AGE 18 3 y: 15.4 WT 27 HTM 717 r.6, RSD 164 M: AGE 3 6 y: 11.3 WT 16 HTM 91 r.84, RSD 132 M: AGE 6 y: 8.8 WT 1128 HTM 171 r.73, RSD 12 F: AGE 18 3 y: 13.3 WT 334 HTM 35 r.7, RSD 18 F: AGE 3 6: 8.7 WT 25 HTM 865 r.82, RSD 94 F: AGE 6 y: 9.2 WT 637 HTM 32 r.76, SE.652; n M: AGE 18 3 y:.669 WT r.742, SE.693; n M: AGE 3 6 y:.592 WT r.776, SE.685; n 534 M: AGE 6 y:.563 WT 2.15 r.7, SE.564; n F: AGE 18 3 y:.546 WT r.69, SE.581; n F: AGE 3 6 y:.47 WT r.786, SE.485; n 334 F: AGE 6 y:.424 WT 2.38 r.764, SE.645; n M: AGE 18 3 y:.6 WT HTM.473 r.756, SE.678; n M: AGE 3 6 y:.476 WT HTM r.789, SE.668; n 533 M: AGE 6 y:.478 WT 2.26 HTM 1.7 r.724, SE.542; n F: AGE 18 3 y:.433 WT HTM 1.18 r.713, SE.564; n F: AGE 3 6 y:.342 WT HTM.486 r.85, SE.472; n 324 F: AGE 6 y:.356 WT 1.76 HTM.448 Development: R 2.73, SEE.83; cross-validation: n 159 (41 M, 649 F), r.83 Development: R 2.62, SEE.77 (n 266); R 2.75, SEE.91 (n 278); crossvalidation: r.79, r.84 Development: R 2.71, SEE.77; cross-validation: r.83 Development: R 2.65, SEE.62 (n 267); R 2.7, SEE.87 (n 261); crossvalidation: r.79, r WT.1452 AGE 1.9 SEX 3.21 BMI 25 3:.457 WT.1553 AGE 1.6 SEX 3.47 BMI 3:.5 WT.1586 AGE 1.13 SEX FFM.436 FM.869 SEX.1181 AGE BMI 25 3:.3776 FFM.313 FM.93 SEX.1196 AGE BMI 3:.5685 FFM.422 FM.88 SEX.142 AGE (Continued; additional data columns shown on next page)

5 PREDICTING ENERGY EXPENDITURE FOR OBESE ADULTS 963 TABLE 2 (Continued) Reference No. of subjects, sex, age range or mean, BMI range or mean; body-composition method when applicable; remarks on large databases Statistics and cross-validation REE predictive equations Korth et al (32), kj/d n 14 (5 M, 54 F), y, BMI 26 (18 41) Korth et al, FFM (32), kj/d n 14 (5 M, 54 F), 37 (21 68) y, BMI 26(18 41), air-displacement plethysmography 4 Italy De Lorenzo et al (33), kj/d n 32 (127 M, 193 F), y, BMI 27 (17 4) Lazzer et al (16, 34), MJ/d n 346 (164 M, 182 F); men: age 2 65 y, mean BMI 45 (5% FM) 5 ; women: age 19 6 y, mean BMI 45 (57% FM) 5 Lazzer et al, FFM (16, 34), MJ/d Australia Huang et al (15) Huang et al, FFM (15) United Kingdom Johnstone et al, FFM (35), kj/d n 346 (164 M, 182 F); men: age 2 65 y, mean BMI 45 (5% FM) 5 ; women: age 19 6 y, mean BMI 45 (57% FM) 5 n 188 (279 M, 759 F), n 142 diabetic (61 M, 81 F), mean age 52 y; n 896 nondiabetic (218 M, 678 F), mean age 44 y; BMI 35 (mean BMI 46) n 188 (279 M, 759 F), n 142 diabetic (61 M, 81 F), mean age 52 y; n 896 nondiabetic (218 M, 678 F), mean age 44 y; BMI 35 (mean BMI 46); bioimpedance analysis n 15 (43 M, 17 F), age y, BMI (54% overweight or obese), airdisplacement plethysmography r.84, R 2.71, SEE WT 35. HTCM SEX 19.1 AGE r.86, R 2.74, SEE FFM 1231 F: R 2.597, SEE 65 M: WT HTCM AGE 487 M: R 2.597, SEE 581 F: WT HTCM AGE 944 Development: M (R 2.68, SE 1.14; n 82), F (R 2.66, SE.56; n 91); cross-validation: M (R 2.54; n 82, 66% 6 ),F(R 2.7; n 91, 6% 6 ); external cross validation: n 5 F (BMI 4 6), 38% 6 Development: M (R 2.65, SE 1.15; n 82), F (R 2.63, SE.581; n 91); cross-validation: M (R 2.53; n 82, 62% 6 ),F(R 2.69; n 91, 6% 6 ); external cross-validation: n 5 F (BMI 4 6), 46% 6 Development: R 2.737, n 71; cross-validation: n 328 (statistics unclear) M:.48 WT HTM.2 AGE 3.65; F:.42 WT HTM 2.678C M:.81 FFM.49 FM.19 AGE F:.67 FFM.46 FM WT HTCM 1.44 AGE SEX R FFM FM AGE SEX R (cross-validation: n 39 M, not applied to this equation) 9.2 FFM 31.6 FM 12.2 AGE WT, weight in km; HTCM, height in cm; HTM, height in meters; FFM, fat-free mass; FM, fat mass; AGE, age in y; SEX (M 1, F ); REE, resting energy expenditure; FAO, Food and Agriculture Organization; CL, confidence limits; RSD, residual SD. 2 From reference 27 (not the original publication). 3 Only sex-specific R 2 values were reported. 4 Air-displacement plethysmography was selected over other available body-composition methods because it was used for assessment in the present study. 5 BMI is not specified for the development and cross-validation groups. 6 Percentage accurate predictions, defined as within 5% from measured REE. procedure as recently described by Frankenfield et al (21) and Compher et al (22). For each patient the REE was predicted for all equations in kilocalories per day and compared with measured REE. The actual body weight or FFM at the time of the indirect calorimetry measurement was used for this calculation. Statistics Subject characteristics were analyzed with an independentsamples t test. A prediction between 9% and 11% of the REE measured was considered an accurate prediction, a prediction 9% of the REE measured was classified as an underprediction, and a prediction 11% of the REE measured was classified as an overprediction. The percentage of patients that had an REE predicted within 1% of the REE measured was considered a measure of accuracy on an individual level (21). The mean percentage difference between the REE predicted and that measured (bias) was considered a measure of accuracy on a group level. The root mean squared prediction error (RMSE) was used to indicate how well the

6 964 WEIJS model predicted in our data set (23 25). The concordance correlation coefficient (CCC) was used to show the precision and bias of the predictive equations (26). The CCC is calculated by multiplying precision (Pearson correlation coefficient) by accuracy (deviation from line of identity). Data were analyzed by using SPSS 14. (SPSS Inc, Chicago, IL), CCC, and Bland-Altman with MedCalc software (version 8..2.; Mariakerke Belgium). RESULTS The subject characteristics for the 239 US and 28 Dutch adults with a BMI of 25 4 are shown in Table 1. The Dutch adults were slightly younger than the US adults (women: P.6; men: P.9). Weight and height were significantly higher in the Dutch than in the US adults, even within female and male overweight (BMI 25 3) and obese (BMI 3 4) subgroups. REE (kcal/d) was 2 kcal higher in the Dutch than in the US adults. Unfortunately, no body composition data for the US adults were available for further evaluation of REE (in kcal /kg FFM). A total of 63 scientific papers or reports were retrieved for adult REE predictive equations. Forty-eight papers were excluded (see Table under Supplemental data in the online issue): age range, 1; patients, 1; insufficient information, 5; ethnic group, 5; small sample size (n 5); impractical or suspect body composition as variable, 4; glucose, insulin, or diabetes as a variable, 3; suspect indirect calorimetry, 3; total energy expenditure, 1; athletes, 1; mean BMI 25, 1; and duplicate publications with same data and same equation, 1. Fifteen papers or reports were included with 27 equations, 18 weight-based equations and 9 FFM-based equations (Table 2). The studies included had 1 subjects. The quality of these studies according to the procedure of Frankenfield et al (21) resulted in no further exclusion. Studies TABLE 3 Indirect calorimetry conditions for the included studies Reference IC instrument Calibration Subject training Pretest fasting Pretest restriction 1 Pretest rest Measurement length (min) Temperature Steady state Steady state defined Bernstein et al (28) Owen et al (7, 8) Mifflin et al (6) Livingston et al (29) De Lorenzo et al (33) Lazzer et al (16, 34) Müller et al (12) Beckman 12 h 3 min C Beckman mouthpiece or face mask Sensormedics hood; standard protocol h 3 min 1 min (first 5 discharged, second 5 recorded; min interval) C 12 h No exercise, 2 min; music Yes 3 min no smoking Sensormedics Yes Overnight, early morning Sensormedics face mask Vmax 29 Sensormedics hood Multicenter (n 8): Deltatrac, Beckman, metabolic chamber Korth et al (32) Vmax 29n, Sensormedics Huang et al (15) Johnstone et al (35) Deltatrac hood Deltatrac II hood (Elia and Livesay 1992 equation) 1 No alcohol, nicotine, or caffeine. Yes 25 3 min No exercise 2 min 3 min; quiet room Yes 2 min, 3 min (first 5 1 supine discharged) 8 h 15 min Yes Overnight, Yes Overnight No exercise, no smoking, no coffee Overnight, min (first 5 discharged), datapoint per 2 s 22 C Yes 2 min mean value 2 min 4 min, standard conditions Constant CV 1% 3 min 3 4 min Neutral 15 min

7 PREDICTING ENERGY EXPENDITURE FOR OBESE ADULTS 965 based on the Harris and Benedict, Schofield, and Oxford databases were used as published, but, for smaller studies, information on standardization, calibration, and steady state is provided (Table 3). When judged by evidence-based guidelines for REE measurement with indirect calorimetry (22), reported criteria were always met but unfortunately were not always reported. Mifflin et al (6) reported the shortest measuring time (2 min) and steady state time used (3 min). The absence of bias in subject selection and subject training were rarely found. None of the included equations were based on Dutch adults. REE data are provided as kcal/d, the percentage bias, the maximum values found for negative error (underprediction) and positive error (overprediction), the RMSE (in kcal/d), the percentage of accurate predictions, the percentage of underpredictions, and the percentage of overpredictions (Table 4 and Table 5). The percentage of accurate predictions, percentage bias, and RMSE for overweight women, obese women, overweight men, and obese men (further referred to as sex and BMI subgroups) for US and Dutch adults are shown in Figure 1. For US adults, the equation of Mifflin et (6) provided 79% accurate predictions, 11% underpredictions, and 9% overpredictions and performed well across sex and BMI subgroups; the RMSE was 136 kcal/d, and the percentage bias was 1.%. For overweight Dutch adults, the FAO/WHO/UNU weight equation (5) provided 64% and 8% accurate predictions for women and men, respectively. For obese Dutch adults, the Lazzer equation (16, 34) provided 68% and 72% accurate predictions for women and men, respectively. Only the De Lorenzo equation (33) provided 6% accurate predictions for all sex and BMI subgroups (overall, 65% accurate predictions). The percentage of accurate predictions varied from 79% to 23% for US adults and from 64% to 13% for Dutch adults. The bias for equations varied from 15% to 9% for the US adults and from 2% to 3% for the Dutch adults, and RMSE varied from 136 to 298 kcal/d for US adults and from 193 to 471 kcal/d for Dutch adults (Figure 1). FFM provided no benefit to REE prediction (Figure 2 and Table 5). Body-composition methods were very different among the studies (Table 2). However, even the use of air-displacement plethysmography (32), consistent with the present Dutch study group, did not improve REE prediction. The inclusion of height or BMI-specific equations did not improve the percentage of accurate predictions (Figure 2). Bland-Altman plots for 6 selected equations are shown in Figure 3. TABLE 4 Evaluation of resting energy expenditure (REE) predictive equations in 239 US adults based on bias, root mean squared prediction error (RMSE), and percentage accurate prediction REE predictive equation REE 1 SD Bias 2 negative error 3 Maximum Maximum Accurate positive error 4 RMSE CCC 5 predictions 6 Under predictions 7 Over predictions 8 kcal/d % % % kcal/d % % % REE measured HB1919 (4) HB1984 (27) Bernstein et al (28) Owen et al (7, 8) Mifflin et al (6) Livingston and Kohlstadt (29) Schofield, weight (3) Schofield, weight and height (3) FAO, weight (5) FAO, weight and height (5) Henry, weight (31) Henry, weight and height (31) Müller et al (12) Müller et al, BMI (12) Korth et al (32) De Lorenzo et al (33) Lazzer et al (16, 34) Huang et al (15) As measured. 2 Mean percentage error between predictive equation and measured value. 3 The largest underprediction that was found with this predictive equation as a percentage of the measured value. 4 The largest overprediction that was found with this predictive equation as a percentage of the measured value. 5 CCC, concordance correlation coefficient. 6 The percentage of subjects predicted by this predictive equation within 1% of the measured value. 7 The percentage of subjects predicted by this predictive equation 1% of the measured value. 8 The percentage of subjects predicted by this predictive equation 1% of the measured value.

8 966 WEIJS TABLE 5 Evaluation of resting energy expenditure (REE) predictive equations in 28 Dutch adults based on bias, root mean squared prediction error (RMSE), and percentage accurate prediction REE predictive equation REE 1 SD Bias 2 negative error 3 Maximum Maximum Accurate positive error 4 RMSE CCC 5 predictions 6 Under predictions 7 Over predictions 8 kcal/d % % % kcal/d % % % REE measured HB1919 (4) HB1984 (27) Bernstein et al (28) Bernstein et al, FFM (28) Owen et al (7, 8) Owen et al, FFM (7, 8) Mifflin et al (6) Mifflin et al, FFM (6) Livingston and Kohlstadt (29) Schofield, weight (3) Schofield, weight and height (3) FAO, weight (5) FAO, weight and height (5) Henry, weight (31) Henry, weight and height (31) Müller et al (12) Müller et al, BMI (12) Müller et al, FFM (12) Müller et al, FFM and BMI (12) Korth et al (32) Korth et al, FFM (32) De Lorenzo et al (33) Lazzer et al (16, 34) Lazzer et al, FFM (16, ) Huang et al (15) Huang et al, FFM (15) Johnstone et al, FFM (35) As measured. 2 Mean percentage error between predictive equation and measured value. 3 The largest underprediction that was found with this predictive equation as a percentage of the measured value. 4 The largest overprediction that was found with this predictive equation as a percentage of the measured value. 5 CCC, concordance correlation coefficient. 6 The percentage of subjects predicted by this predictive equation within 1% of the measured value. 7 The percentage of subjects predicted by this predictive equation 1% of the measured value. 8 The percentage of subjects predicted by this predictive equation 1% of the measured value. 9 FFM, fat-free mass. DISCUSSION From this study it appears that REE for US overweight and obese class I and II adults can best be predicted with the Mifflin equation (6). For Dutch overweight adults, the FAO/WHO/UNU weight equation (5) can be used with reasonable accuracy up to a BMI of 3. However, for Dutch obese adults with a BMI of 3 4, the Lazzer equation (16, 34) provides improved accuracy. Whereas the Mifflin equation provides almost 8% accurate equations for US adults, this level of accuracy cannot be reached with presently available equations for Dutch adults. Assessment of FFM and FM did not improve the accuracy of REE prediction. Because any choice is less than optimal for the Dutch, the best way to proceed would be to produce a new equation. This extends a previous observation in Dutch patients, which showed that the FAO/WHO/UNU weight and height (FAOwh) equations (5) was most accurate (23), although only 5% of patients had accurate predictions. In this previous validation study, there were not enough overweight and obese patients available to establish the accuracy for a BMI of 25 4.

9 PREDICTING ENERGY EXPENDITURE FOR OBESE ADULTS 967 Percentage accurate predictions Mifflin Livingston HenryWH HenryW Huang MullerBMI Muller HB1919 SchofieldW SchofieldWH HB1984 Lazzer FAOw FAOwh Owen DeLorenzo Korth Bernstein Korth DeLorenzo Lazzer Percentage bias HB1984 HB1919 FAOw FAOwh SchofieldW MullerBMI SchofieldWH HenryW Muller HenryWH Huang Mifflin Livingston Owen Bernstein Bernstein Owen Livingston Mifflin HenryWH Huang HenryW SchofieldWH SchofieldW MullerBMI Muller FAOwh HB1919 Lazzer FAOw HB1984 DeLorenzo Korth Bernstein RMSE Owen Livingston Mifflin Huang HenryWH HenryW MullerBMI SchofieldWH Muller SchofieldW FAOwh HB1919 FAOw HB1984 DeLorenzo Lazzer Korth Mifflin Livingston HenryWH Muller MullerBMI Huang HenryW HB1919 HB1984 SchofieldWH FAOwh DeLorenzo SchofieldW FAOw Lazzer Owen Korth Bernstein FIGURE 1. Percentage of accurate predictions, percentage bias, and root mean squared prediction error (RMSE) for US (left panels) and Dutch (right panels) adults (with 95% CIs), overweight women (E), obese women ( ), overweight men ( ), and obese men ( ) for 18 resting energy expenditure predictive equations. For each panel the data are sorted by values for all adults combined (line). See Table 2 for predictive equations. Lazzer Korth DeLorenzo HB1984 HB1919 FAOw Muller FAOwh MullerBMI SchofieldWH SchofieldW HenryW HenryWH Huang Mifflin Livingston Owen Bernstein A recent review by an expert panel (9) advised that the Mifflin equation be used for overweight and obese subjects. However, this expert panel acknowledges that there are limited data to support the use of the Mifflin equation in overweight and obese subjects. In this review, the evidence for the accuracy of the Mifflin and Owen equations in overweight and obese subjects is based on one study by Frankenfield et al (1). Six studies were used to validate the original Harris and Benedict equation, of which 2 studies were of suboptimal quality, 1 had an unclear BMI range, and the other 3 were from Owen et al (7, 8) and Frankenfield et al (1). No individual accuracy studies for overweight and obese subjects were used to validate the FAOw and FAOwh equations (5). The present study essentially confirms the conclusion by Frankenfield et al (1) for US adults with a BMI of 25 4 and an age 65 y and now provides the strongest evidence for use of the Mifflin equation in the United States. Because African American females were found to have lower REE values than European American females (36, 37), ethnicity should be addressed. The higher REE values for the Dutch than for the US adults are probably due to the higher weight and height values, even within sex and BMI subgroups. If the higher values are due to the Dutch being taller, the low accuracy level might be true for other countries with tall adults. Recent evaluations of the validity of REE predictive equations have been published for overweight and obese subjects (1 13) and for extremely obese subjects (14 18). There is some support for using the Mifflin equation in European American females (36), males (29), and extremely obese females (17). The FAOw equation has been acceptable in females (29) and in extremely obese subjects (14, 16). Also, the HB1919 equation has been found to be acceptable in persons with a broad weight range (12) and in extremely obese persons (14, 16). However, there

10 968 WEIJS Percentage accurate predictions Bernstein Owen Mifflin Lazzer Muller MullerBMI Korth Huang Weight based FFM based Schofield US Schofield NL FAO US FAO NL Henry US Henry NL Muller US Muller NL Weight based or BMI independent Weight & height based or BMI specific FIGURE 2. Comparison of percentage accurate predictions for weightbased compared with fat-free mass (FFM) based resting energy expenditure predictive equations and for weight-based or BMI-independent compared with weight- and height-based or BMI-specific REE predictive equations for US and Dutch adults. seems to be no consensus concerning the use of one preferred REE equation. This might be explained by differences in subject group composition, methods used, or the statistics used, for example. Different statistical methods may have been used to evaluate how well an equation fits the data. Bias, correlation, and regression analysis are not preferred methods for validating equations (25). Suitable methods include the RMSE and an individual measure of accuracy, such as the percentage of subjects predicted within 1% of the measured value. Additionally, the bias obviously has to be small in order for a predictive equation to perform well. However, a statistical comparison (t test or ANOVA with a post hoc test) indicating a nonsignificant difference between group means is not the same as an accurate fit, because high positive errors might counterbalance high negative errors. The methods used for the equation development studies might also have differed (Table 3). Differences in subject selection, subject training, and measurement conditions should be considered. Subject selection refers to the representativeness of the study group relative to the population for which the equation should be used. In the present study, data from Dutch adults that intended to lose weight and agreed to the study procedures were included. This group is not representative of the whole Dutch population and certainly not of the different ethnic groups. However, it seems reasonable to use this group as an approximation for the (white) Dutch adults with a BMI of 25 4 and 65yof age. The US data might be more representative of the US population with a BMI of 25 4 and 65 y of age. Subject training was not reported in most studies, as in the present study, although the procedures had been explained to the subjects beforehand. The effect of stress (activation of the sympathetic nervous system) might be minimized by collecting steady state data, although this might not fully account for the problem. Because Mifflin - REE measured SD 21,8 Mean -174, SD -55,4 Owen - REE measured SD 119, Mean -284, SD -688,4 Lazzer - REE measured SD 338, Mean -35, SD -48, AVERAGE of Mifflin and REE measured AVERAGE of Owen and REE measured AVERAGE of Lazzer and REE measured De Lorenzo - REE measured SD 34,9 Mean -64, SD -434,3 FAO/WHO/UNU weight - REE measured SD 299,9 Mean -96, SD -492,9 HB REEmeasured SD 284,8 Mean -99, SD -484, AVERAGE of De Lorenzo and REE measured AVERAGE of FAO/WHO/UNU weight and REE measured AVERAGE of HB1919 and REEmeasured FIGURE 3. Bland-Altman plots for 6 selected resting energy expenditure (REE) predictive equations (Mifflin, Owen, Lazzer, De Lorenzo, FAO/WHO/ UNU, HB1919) for overweight women (E), obese women (F), overweight men ( ), and obese men (f).

11 PREDICTING ENERGY EXPENDITURE FOR OBESE ADULTS 969 Mifflin had the shortest time of measurement and of steady state duration, it might be preferable not to extend the measurement time beyond 2 min because some subjects get restless. Another factor that might affect REE data is the time of day. However, when measurement conditions as defined by Compher et al (22) are adhered to, this should not be of major concern (38). Inclusion of height into the equation does not systematically improve REE prediction for overweight and obese adults (Table 5), although this has been observed for a group of patients of whom 5% were underweight (BMI 18.5) (23). However, the Mifflin and Lazzer equations both include height as a variable. The FAO/WHO/UNU report (5) showed no statistical advantage of height inclusion; however, there was no evaluation of BMI groups in this 1985 report. Furthermore, there was no improvement in REE prediction by FFM assessment. Müller et al (12) showed that, even for BMI groups, there is no advantage to using FFM-based equations. Korth et al (32) showed that although weight explains less variance in REE than FFM does, inclusion of weight, height, age, and sex together explain a similar amount of variance in REE. This observation is important because weight-based equations are more likely to be used in clinical practice than are FFM-based equations, although there might be other sound reasons for body-composition analysis in weight treatment. It is advisable to validate REE prediction equations for any single specific population, because prediction equations are expected to be valid for the original population only (39). This is especially true for individual accuracy: the percentage of accurate predictions. In conclusion, this study showed that there is a wide variation in the accuracy of REE predictive equations. For overweight and obese class I and II US adults, almost 8% could be accurately predicted with the Mifflin equation. For Dutch adults, however, there is no single accurate REE prediction equation. For now, the FAO/WHO/UNU weight equation can be used for overweight adults, and the Lazzer equation for obese subjects. Bodycomposition assessment is not needed for REE prediction. Whether these equations will also be best for obese patients remains to be assessed. Thanks to Marcel Hesseling (data management), Monique Dekker, Judith Huisman, Fiona van Donselaar, Mieke Kramer, Annemieke Schaap, Ankie van Baar, Renske Strikwerda, Linda Sluijmers, Dennis Bast, Ryanne Zomer, Willemijn Wissekerke, Johanna van Zoonen, Ilona Sonke, and Bianca Ooteman for collecting the data and to Aimee van Dijk, Hinke Kruizenga, and Ageeth Hofsteenge for collecting part of the data for the predictive equations. The author s responsibilities were as follows designed the study, performed the literature search, conducted the data analysis, and wrote the manuscript. The author had no conflict of interest. REFERENCES 1. Ogden CL, Carroll MD, Curtin LR, McDowell MA, Tabak CJ, Flegal KM. Prevalence of overweight and obesity in the United States, JAMA 26;295: Schokker DF, Visscher TL, Nooyens AC, van Baak MA, Seidell JC. Prevalence of overweight and obesity in the Netherlands. Obes Rev 27;8: Schoeller DA. Making indirect calorimetry a gold standard for predicting energy requirements for institutionalized patients. J Am Diet Assoc 27;17(3): Harris JA, Benedict FG. A biometric study of basal metabolism in man. Washington, DC: Carnegie Institute of Washington, FAO/WHO/UNU. Energy and protein requirements. Geneva, Switzerland: World Health Organ Tech Rep Ser, Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr 199;51: Owen OE, Kavle E, Owen RS, et al. A reappraisal of caloric requirements in healthy women. Am J Clin Nutr 1986;44: Owen OE, Holup JL, D Alessio DA, et al. A reappraisal of the caloric requirements of men. Am J Clin Nutr 1987;46: 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 25;15: Frankenfield DC, Rowe WA, Smith JS, Cooney RN. Validation of several established equations for resting metabolic rate in obese and nonobese people. J Am Diet Assoc 23;13: Siervo M, Boschi V, Falconi C. Which REE prediction equation should we use in normal-weight, overweight and obese women? Clin Nutr 23;22(2) Müller MJ, Bosy-Westphal A, Klaus S, et al. World Health Organization equations have shortcomings for predicting resting energy expenditure in persons from a modern, affluent population: generation of a new reference standard from a retrospective analysis of a German database of resting energy expenditure. Am J Clin Nutr 24;8: De Luis DA, Aller R, Izaola O, Romero E. Prediction equation of resting energy expenditure in an adult Spanish population of obese adult population. Ann Nutr Metab 26;5: Das SK, Saltzman E, McCrory MA, et al. Energy expenditure is very high in extremely obese women. J Nutr 24;134: Huang KC, Kormas N, Steinbeck K, Loughnan G, Caterson ID. Resting metabolic rate in severely obese diabetic and nondiabetic subjects. Obes Res 24;12: Lazzer S, Agosti F, Silvestri P, Derumeaux-Burel H, Sartorio A. Prediction of resting energy expenditure in severely obese Italian women. J Endocrinol Invest 27;3(1): Dobratz JR, Sibley SD, Beckman TR, et al. Predicting energy expenditure in extremely obese women. JPEN J Parenter Enteral Nutr 27; 31(3): Boullata J, Williams J, Cottrell F, Hudson L, Compher C. Accurate determination of energy needs in hospitalized patients. J Am Diet Assoc 27;17(3): National Academy of Sciences. Dietary Reference Intakes for energy, carbohydrates, fiber, fat, protein and amino acids (macronutrients). Washington, DC: National Academy of Sciences, Weir JB. New methods for calculating metabolic rate with special reference to protein metabolism. J Physiol 1949;19: Frankenfield D, Roth-Yousey L, Compher C, for the Evidence Analysis Working Group. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review. J Am Diet Assoc 25;15(5): Compher C, Frankenfield D, Keim N, Roth-Yousey L, for the Evidence Analysis Working Group. Best practice methods to apply to measurement of resting metabolic rate in adults: a systematic review. J Am Diet Assoc 26;16(6): Weijs PJM, Kruizenga HM, van Dijk AE, et al. Validation of predictive equations for resting energy expenditure in adult outpatients and inpatients. Clin Nutr 28;27(1): Kutner MH, Nachtsheim CJ, Neter J, Li W. Applied linear statistical models. 5th ed. New York, NY: McGraw-Hill/Irwin, Sheiner LB, Beal SL. Some suggestions for measuring predictive performance. J Pharmacokinet Biopharm 1981;9: Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics 1989;45: Roza AM, Shizgal HM. The Harris Benedict equation reevaluated: resting energy requirements and the body cell mass. Am J Clin Nutr 1984; 4: Bernstein RS, Thornton JC, Yang MU, et al. Prediction of the resting metabolic rate in obese patients. Am J Clin Nutr 1983;37: Livingston EH, Kohlstadt I. Simplified resting metabolic rate predicting formulas for normal-sized and obese individuals. Obes Res 25; 13: Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr 1985;39C: Henry CJK. Basal metabolic rate studies in humans: measurement

12 97 WEIJS and development of new equations. Public Health Nutr 25;8: Korth O, Bosy-Westphal A, Zschoche P, Glüer CC, Heller M, Müller MJ. Influence of methods used in body composition analysis on the prediction of resting energy expenditure. Eur J Clin Nutr 27;61(5): De Lorenzo A, Tagliabue A, Andreoli A, Testolin G, Comelli M, Deurenberg P. Measured and predicted resting metabolic rate in Italian males and females, aged y. Eur J Clin Nutr 21;55: Lazzer S, Agosti F, Resnik M, Marazzi N, Mornati D, Sartorio A. Prediction of resting energy expenditure in severely obese Italian males. J Endocrinol Invest 27 Oct;3(9): Johnstone AM, Rance KA, Murison SD, Duncan JS, Speakman JR. Additional anthropometric measures may improve the predictability of basal metabolic rate in adult subjects. Eur J Clin Nutr 26;6: Vander Weg MW, Watson JM, Klesges RC, Eck Clemens LH, Slawson DL, McClanahan BS. Development and cross-validation of a prediction equation for estimating resting energy expenditure in healthy African- American and European-American women. Eur J Clin Nutr 24;58: Douglas CC, Lawrence JC, Bush NC, Oster RA, Gower BA, Darnell BE. Ability of the Harris-Benedict formula to predict energy requirements differs with weight history and ethnicity. Nutr Res 27;27: Weststrate JA, Weys PJM, Poortvliet EJ, Deurenberg P, Hautvast JGAJ. Diurnal variation in postabsorptive resting metabolic rate and dietinduced thermogenesis. Am J Clin Nutr 1989;5: Moreira da Rocha EE, Alves VGF, Silva MHN, Chiesa CA, da Fonseca RBV. Can measured resting energy expenditure be estimated by formulae in daily clinical nutrition practice? Curr Opin Clin Nutr Metab Care 25;8: