Which REE prediction equation should we use in normal-weight, overweight and obese women?

Clinical Nutrition (2003) 22(2): 193–204 r 2003 Elsevier Science Ltd. All rights reserved. doi:10.1054/clnu.2002.0625

ORIGINAL ARTICLE

Which REE prediction equation should we use in normal-weight, overweight and obese women? M. SIERVO, V. BOSCHI, C. FALCONI Department of Neuroscience and Behaviour, Dietetic and Nutrition Section, School of Medicine, University ‘‘Federico II’’, Naples, Italy. (Correspondence to: VB, 91 PianoTorre Biologica (Edi¢cio 19),Via Pansini 5, 80131Napoli, Italy)

AbstractFBackground & Aims: In our modern society, there is a growing and increasing prevalence of overweight, obesity and eating disorders and young female subjects frequently ask for nutritional counselling. Resting energy expenditure (REE) is essential to provide a sound diet to subjects seeking nutritional support.We perform a critical selection of accurate and reliable prediction equations employed on normal-weight, overweight and obese young women. Methods: REE of 157 young women of Caucasian race (18^35 years)was measured with indirect calorimetry and was compared with the principal prediction equations (Harris and Benedict, Owen, Mi¥in,WHO, Bernstein and Robertson and Reid).The statistical analysis used to compare measured and the predicted REE was paired t-test, 795% con¢dence interval and Bland and Altman method.The in£uence of weight loss on the prediction error was estimated in 31subjects. An additional REE measurement was performed on patients who had lost Z5% of the initial body weight due to a sound low-calorie diet. Results:The equations more reliable in our study are Owen’s equation in normal-weight subjects, Bernstein’s equation in overweight subjects and Robertson and Reid’s equation in obese subjects.Weight was a signi¢cant variable according to the stepwise regression analysis resulting in the following equation: 542.2 + 11.5 kg; R2: 0.59.Weight loss decisively increased the overestimation of the equations and only Owen’s equation maintained the error of prediction within acceptable limits. Conclusions:The equation of Owen in normal weight, Bernstein in overweight and of Robertson and Reid in obese subjects should be chosen when we have to predict REE in young women. Due to metabolic adaptation occurring during therapeutic or spontaneous energy restriction, we suggest to use Owen’ s equation. r 2003 Elsevier Science Ltd. All rights reserved.

on the basis that age, weight, height and body surface area (BSA) can be easily and regularly monitored in clinical practice. Nowadays, a soaring number of young women are under dietary treatments mainly for aesthetic purposes rather than for nutritional impairments. Moreover, with the exception of some extreme forms of malnutrition as anorexia and obesity, the number of girls with normal or slightly higher body mass index (BMI) values undergoing dietary treatments is gradually increasing (6, 7). Therefore, it is important when considering the rising demands for nutritional counselling, to plan and prepare diets that are balanced specifically for each individual. Thus, the subject’s REE should be accurately determined, but the process for achieving such data is quite complex. Consequently, one has to turn to predictive equations in order to obtain REE estimates. Harris and Benedict’s equation is one of the most commonly used in clinical practice, even though many studies have demonstrated its inaccuracy. The first purpose of this study was to examine those equations that would permit to formulate more accurate REE estimates within a young female population. The second objective was to revise the chosen equation in order to adjust individual

Key words: resting energy expenditure; indirect calorimetry; prediction equations; obesity

Introduction Resting energy expenditure (REE) constitutes the major part of the daily energy expenditure for people leading a sedentary life style. REE is a key factor for prescription of food intakes specifically tailored to the basal metabolism of people undergoing a therapeutic diet (1, 2). REE may be measured with different techniques (direct and indirect Calorimeters, double-labelled water, etc.); however, their complex nature, the lack of skilled staff, and the high costs for their application limit their regular use at clinical levels, both for diagnostic and prognostic purposes. Since the beginning of the century, some authors (Harris and Benedict (3), Boothby et al. (4), Robertson and Reid (5)) have tried to address the problem by developing equations for predicting REE on the basis of anthropometric and body composition parameters. The aim of the present work was to examine the level of accuracy of the anthropometric predictive equations 193

194

REE PREDICTION IN OBESITY

BMI to the subject’s nutritional state (normal weight, overweight and obese). The third objective was to examine how weight loss might influence the REE prediction of anthropometrical equations after a period of dietary treatment. Since REE is closely bound to energy restriction and this affects remarkably the energy expenditure of each subject, it appears to be interesting to assess the error of prediction. It is an open question how to choose one statistical analysis instead of another to determine the accuracy of a predictive equation compared to the REE measured with a reference method. Bland and Altman have utilized a statistical analysis, which allowed them to obtain clear conclusions regarding the precision of new methodologies (8, 9). Thus, we chose Bland and Altman’s technique, along with other statistical analyses (paired t-test, 795% confidence interval (CI)) to achieve reasonable conclusions and improve the prediction analysis. Finally, in our study we tried to give useful advice, from a clinical point of view, regarding the criteria to use when applying equations of Harris and Benedict, Owen, Mifflin, Bernstein, WHO and Robertson and Reid in a population of normal-weight, overweight and obese young women.

Methods The study was carried out on a population of 157 Caucasian female subjects (age range: 18–35 years). They attended the functional area of Nutritional Science at the Department of Neurobehavioral Sciences from September 2000 to November 2001. The individuals under study were not on any particular diet, maintained a quite steady weight for at least 2 months (varying 73 kg from the initial weight), had a low level of physical activity and underwent no pharmacological treatment. Moreover, they showed no signs of physical or mental disorders. Each subject underwent the laboratory tests for blood glucose, plasma protein, red and white blood cells, platelets, liver, thyroid and kidney function. Unhealthy subjects were excluded from the study. For the anthropometric measurements, the patients were asked to wear only underclothes. The women’s weight and height were measured by a precision balance and a stadiometer, using close values of 0.1 kg and 0.5 cm, respectively. Their waist and hips were also measured: the waist between the inferior margin of the last rib and the iliac crest, the hips at the tronchater level (10). BMI, BSA (11) and waist and hip ratio (WHR) were then calculated. The subjects under study were then separated into three groups according to their BMI: BMI r25 (normal weight), 25 rBMI o30 (overweight) and BMI Z30 (obese). Three subjects had a BMI o20; since the level

of BMI was above 17.5, they were equally included in the study. REE was measured by indirect calorimetry using a Canopy System gas collector (V MAX 29n, SensoMedics, Yorba Linda, CA, USA). The device was calibrated before each measurement. The calibration of the analyzers was performed with a gas mixture of known standardized concentration (Cal 1: 26% O2, 0% CO2, Cal 2: 16% O2, 4% CO2). The calibration of the flow was performed with a standard 3 l syringe. The device was calibrated every 2 months by means of the ethanol burning test. O2 was examined with paramagnetic analyzers while CO2 with infrared analyzers in our laboratory. The coefficient of variation for the intersubject measurements was 73%. The protocol for measurements required that women would have drunk only water; been fasting for at least 12 h; slept for at least 8 h; kept off smoking, sweet beverages, coffee, tea and drugs for at least 12 h. Furthermore, the subjects had to refrain from doing any physical activity in the 24 h preceding the morning of the measurements (12). The session of measurements began at 8 o’clock in the morning. The subjects were invited to get accustomed to the environmental conditions and to the measurement procedures. The subjects were asked, then, to lie down supine for at least 20 min. The measurements were taken in a peaceful and relaxing environment kept at a constant temperature (25.070.51C) and level of humidity (65) due to air conditioning and dehumidification systems. The measurements were made only when the subjects had reached a steady-state condition, pointed out automatically by the calorimeter when all the following variables (respiratory quotient, oxygen consumption, minute ventilation) were stable for at least 5 min. Otherwise, the test was discarded and repeated again on a different day. Coughing and yawning were noted but were not taken into account for the subjects’ final metabolic values. The metabolic test lasted for at least 25 min and was extended to a maximum of 45 min. During the study, nine subjects repeated the test. REE was then calculated from the oxygen consumption and carbon dioxide production, according to Weir’s equation (13). The measured REE was then compared with the principal prediction equations previously reported (Harris and Benedict, Owen (14), Mifflin (15), WHO (16), Bernstein (17) and Robertson and Reid) and the one derived from our study. In Table 2, we report the equations used in our study to highlight the characteristics of the population, the variables included in the equation and statistical analysis. A mean error of 74% represented the threshold level for assessing the accuracy of each equation. The influence of weight loss on the prediction error was estimated in 31 subjects. An additional REE measurement was performed on patients that had lost Z5% of the initial body weight (IW) due to a sound low-calorie diet.

CLINICAL NUTRITION

The data are reported as mean 7SD. The ANOVA test was performed to examine the statistical significant differences among the three BMI groups. The statistical analysis for comparison of the REE measured and the prediction equations was performed by taking into account Bland’s and Altman’s indications. The paired t-test was used to determine the statistically significant differences between the measured and predicted REE derived from the applied equations. Due to the limit of the mean, the 795% CI and Bland and Altman’s analysis were used for comparison between measured and predicted REE. Such technique was utilized with the entire cohort and, for every equation, the degree of correlation (Pearson coefficient correlation r) between measured and predicted REE was calculated. The reported graphs were made by comparing and plotting the numerical bias with the mean REE [(measured and predicted)/2] and by identifying the limits of agreement according to Bland’ and Altman’s analysis. In order to develop a REE prediction equation, a stepwise multiple regression was performed by using the anthropometrical measurements as independent variables and REE as the dependent one. For the statistical analysis, Statistics 5 for Windows was used (StatSoft, Inc., 1995. Tulsa, USA). Statistically significance was set at Po0.05.

Results The subjects under study were divided into three groups according to the individual BMI: BMI o25 (n=41: normal weight), 25 rBMI o30 (n=58: overweight) and BMI Z30 (n=58: obese). The anthropometric characteristics of each subject are reported in Table 1. Weight (Po0.001), BMI (Po0.001), waist (Po0.001), hip (P o0.001), WHR (Po0.001) and BSA (Po0.001) (Table 1) were statistically different among the three groups. It is worth noting the difference of the measured REE in the three BMI groups (Po0.001). The description of the error of prediction is rather complex because of the simultaneous use of different methods of statistical analysis. Examining each single group separately and gradually analyzing the development of the errors of prediction, the paired t test indicated that Owen’s prediction equation and ours (OUR) were the most accurate in the normal weight group. The mean numerical bias and the mean percentage bias estimates were 9.7 (underestimation = k 0.75%) and 8.81 (k0.76%) for Owen’s and OUR, respectively. In the overweight group, Bernstein’s equation and OUR proved to be the most accurate reporting a mean numerical bias and a mean percentage bias of 11.15 (overestimation = m 0.93%) and 4.75 (m 0.4%), respectively. Finally, in the third group of the obese subjects, the most reliable equations were Robertson and Reid’s and

195

Table 1 Anthropometric data of subjects (n=157)

Age (years) Height (cm) Weight (kg)*** BMI (kg/m2)*** Waist circumference (cm)*** Hip circumference (cm)*** WHR*** BSA (m2)***

Normal weight (n=41)

Overweight (n=58)

Obesity (n=58)

23.7873.79 160.5875.7 58.7376.04 22.7571.71 71.4876.78

25.3975.42 161.2776.86 71.3777.77 27.3771.43 82.4677.21

23.8275.49 161.4576.75 90.88710.55 34.8773.64 97.8978.9

97.1475.78

105.1675.75

118.5377.71

0.7870.06 1.7570.13

0.8270.07 1.9470.13

0.7370.04 1.6170.103

body mass Index (BMI): weight (kg)/height (m)2; waist hip ratio (WHR): waist circumference (cm)/hip circumference (cm); body surface area (BSA): 0.007184*weight (kg) 0.425 * height 0.725. Statistical difference among normal weight, overweight and obese subjects on ANOVA test (P level: *Po0.05; **Po0.01; ***Po0.001).

OUR with errors of -10.47 (m 0.66 %) and 1.28 (m0.1%), respectively (Table 3). To avoid misleading conclusion, an additional evaluation of the range of variability for each BMI group was made by using 795% CI. Taking into account the measured REE 7 95% CI, those equations whose 795% CI of prediction fell within the one measured were considered the most accurate. In the normal-weight group, the 795% CI of the measured REE was 1177.35/1275.61. Owen’s equation and OUR stayed within such range, reporting REE prediction values of 1203.04/1230.41 and of 1195.74/ 1239.59, respectively. The same two equations showed a 795% CI of the bias% within 3.01%/4.53% and 2.98%/4.52%. In the overweight group, the 795% CI for the measured REE was 1313.69/1402.92. Bernstein’s for the first and OUR for the second had REE prediction in this interval: 795% CI of 1352.81/1386 and 1339.54/ 1387.57, respectively. Moreover, they showed a 795% CI of the bias % of 3.64%/1.77% and 3.03%/2.21%. In the obese group, the 795% CI for measured REE was 1531.69/1640; the most accurate equations were Robertson and Reid’s (1566.42/1626.42) and OUR (1555.42/1619.25) with 795% CI of 3.51%/2.19 % and 2.88 %/2.67 %, respectively. (Table 4). In Tables 1–6, we report the most accurate equations omitting those with wider range of errors. Bland and Altman provided useful methods for comparison of the results of different statistical analyses. Their analysis was applied to the total population under study to assess the error of prediction. Such approach stressed the visual assessment of the error: the error of prediction was determined by accurately evaluating the overestimation, the underestimation and the gradual development of the error as the REE increased. According to Bland and Altman, the error was evident when the limits of agreements were included. Analyzing each equation separately, we observed that the simple correlation between the measured and predicted REE

196

Prediction equation to estimate RMR: sample characteristics, formula and statistical aspects Subjects/gender

Harris and Benedict (1919) Owen (1987) Mifflin (1990) Bernstein (1983) WHO (1985) Robertson and Reid (1952) OUR

Weight status

Age

Formula for female subjects

239/m–f 44/f

NW NW, OW, OB

29714 (X7SD) Range (18–82)

655+9.5 wt (kg)+1.9 ht (cm) -4.7 age (year) 795+7.18 wt (kg)

498/m–f 202/m–154 11,000/m–f

NW, OW, OB OW, OB NW, OW OB

Range (19–78 anni) 40712 (X7SD) Variable

2310/m–f 157/f

NS NW, OW, OB

Range (3–80 anni) 23.7873.79 (X7SD)

9.99 wt (kg)+6.25 ht (cm) 4.92 age (year) -161 7.48 wt (kg) 0.42 ht (cm) 3.0 age (year)+844 18–30 years: 55.6 wt (kg)+1397.4 ht (m)+146 30–60 years: 36.4 wt (kg) 104.6 ht (m)+3619 BSA (m2)  24  age-specific value 11.5 wt (kg) + 542.2

Statistics and other 2

RMR; R : 0.53; 95% CI7210 kcal/day RMR; predicted RMR over or underestimate measured RMR by 21–33% RMR; R2: 0.71 RMR; R2: 0.66 RMR; prediction coefficient of variation: 12.5% RMR; curvilinear function included in BSA formula RMR; R2: 0.59

m=male; f=female; NS=not specified; NW=normal weight; OW: overweight; OB=obesity; X=mean; SD=standard deviation; BSA=body surface area; RMR=resting metabolic rate; wt=weight; ht=height.

Table 3

Measured and predicted REE of subjects Normal weight

REE measured (kcal/day)*** Harris and Benedict (kcal/day)

1226.487155.64 1406.15767.15111

Owen (kcal/day)

1216.72743.36

Mifflin (kcal/day)

1312.43790.12111

Bernstein (kcal/day)

1279.45747.711

WHO (kcal/day)

1349.84793.55111

Robertson and Reid (kcal/day)

1319.83784.88111

OUR (kcal/day)

1217769.46

Bias Bias% 179.667147.05 m12.76710.48% 9.757144.77 k 0.75711.94% 85.947148.54 m 6.41711.22% 52.967146.54 m 4.16 7 11.54% 123.367146.55 m 9.13710.68% 93.347143.98 m 7.07710.77% 8.817144.24 k 0.76711.89%

Overweight 1358.317169.69 1521.17789.66111 1307.5755.8411 1435.87120.09111 1369.46763.1 1501.697120.38111 1436.027109.04111 1363789.44

Bias Bias% 162.867136.78 m 10.7879.07% 50.87143.02 k 3.74710.85% 76.797136.56 m 5.2879.53% 11.157141.42 m 0.93710.32% 143.387143.73 m 9.5479.48% 77.717135.05 m 5.479.43% 4.757135.8 m 0.479.97%

m: Overestimationk: underestimation; bias: measured-predicted; Bias%: [(REE measured REE predicted).*100]/REE predicted. Statistical difference among normal-weight, overweight and obese subjects on ANOVA Test (P level: *Po0.05; **Po0.01; ***Po0.001). Statistical difference between measured and predicted REE on a two-tailed paired t-test (P level: 1Po0.05; 11Po0.01; 111Po0.001).

Obesity 1586.057206.73 1716.09711.76111 1447.53775.77111 1638.787135.971 1520.12782.4411 1750.917160.3111 1596.527114.16 1587.347121.37

Bias Bias% 130.047169 m 7.63710.04% 138.527171.46 k 9.42711.84% 52.737171.17 m 3.13710.68% 65.927172.09 k 4.21711.40% 164.867186.39 m 9.41710.71% 10.477170.72 m 0.66710.84% 1.287164.04 m 0.1710.57%

REE PREDICTION IN OBESITY

Table 2

10.27/ 4.98 6.3/12.53 5.94/ 0.32 1.22/7.22 12.02/ 6.38 3.51/2.19 2.88/2.67 174.4/ 85.52 93.43/183.6 97.65/ 7.63 20.73/111.22 213.87/ 115.85 55.28/34.49 44.42/41.84 1531.69/1640.41 1686.62/1745.41 1427.60/1467.45 1602.93/1674.45 1498.39/1541.75 1708.76/1793.06 1566.42/1626.46 1555.42/1619.25 13.16/ 8.39 0.89/6.59 7.78/ 2.7 3.64/1.77 12.04/ 7.06 7.87/ 2.91 3.03/2.21 198.74/ 126.81 13.2/88.41 112.61/ 40.8 48.28/26.08 182.45/ 106.86 113.08/ 42.06 40.46/30.95 1313.69/1402.92 1497.5/1544.67 1292.82/1322.18 1403.42/1466.6 1352.81/1386 1471.41/1534.51 1407.2/1464.56 1339.54/1386.57 16.06/ 9.44 3.01/4.52 9.95/ 2.86 7.8/ 0.51 12.4/ 5.65 10.39/ 3.59 2.98/4.52 225.96/ 133.13 35.93/55.45 132.71/ 38.94 99.14/ 6.63 169.61/ 77.1 138.69/ 47.8 36.71/54.34

Normal weight

REE 7 95% CI Bias 7 95% CI Bias% 7 95% CI REE 7 95% CI Bias 7 95% CI Bias% 7 95% CI REE 7 95% CI Bias 7 95% CI Bias% 7 95% CI

1177.35/1275.61 1384.84/1427.23 1203.04/1230.41 1283.87/1340.76 1264.32/1294.43 1320.31/1379.37 1292.95/1346.51 1195.74/1239.59 Measured Harris and Benedict (kcal/day) Owen (kcal/day) Mifflin (kcal/day) Bernstein (kcal/day) WHO (kcal/day) Robertson and Reid (kcal/day) OUR (kcal/day)

Table 4

BMI groups (normal weight, overweight, obesity): REE 7 95% CI, bias and bias%

Over weight

Obesity

CLINICAL NUTRITION

197

and the correlation coefficient r were neither able to quantify nor assess the type of error. Furthermore, the introduction of the line of equality and the different representation of the data (histogram and scatterplot) were unable to give a clear indication of the type of error observed (Fig. 1a). Harris and Benedict’s equation showed an REE overestimation with a slight decrease in the level of error with increasing REE. Conversely, Owen’s equation showed a trend to underestimation which was manifest as REE gradually increased. Prediction was accurate for low REE values and the trend to overestimation was observed with REE values below 1100 kcal/day. Mifflin’s equation also showed an overestimation error and a higher accuracy with increasing REE, but compared to Harris and Benedict’s equation, the error was quantitatively lower. Bernstein’ s equation presented an overestimation error in low metabolic conditions and an increase in underestimation with increasing REE, as those observed with Owen’s. The WHO equation showed an REE overestimation regardless of REE variation. Finally, along with the equation of Owen and Bernstein, OUR revealed, good REE prediction, even though the range of the prediction error was wider, and a trend to underestimation with increasing REE (Fig. 1). Taking into account all study population, the correlation analysis between BMI and bias of each prediction equation was statistically significant for those of Owen (r=0.34; Po0.000), Bernstein (r=0.32; Po0.000) and Robertson and Reid (r=0.26; Po0.001) (Table 5). The effect of weight loss on the error of prediction was assessed in a group of 31 subjects. The anthropometric and metabolic characteristics are reported in Table 6. The patients were re-evaluated every 15 days/1 month as outpatients and were given no less than 4 weeks for losing 5% of their initial weight. We calculated a mean loss of about 1 kg/week. There was a statistical difference in weight (Po0.001), BMI (Po0.001), BSA (Po0.001) and measured REE (Po0.001) obtained at the first observation and after losing 5% of weight. The paired t-test, 795% CI range of variability and visualization of the mean %bias indicated how the error of prediction was altered by BMI variation. It was also evident how weight loss increased the overestimation of the equations. Only Owen’s equation maintained the error of prediction within acceptable limits (Fig. 2). Finally, weight was a significant variable according to the stepwise regression analysis resulting in the following equation: 542.2+11.5 kg; R2: 0.59.

Discussion The present study was focused on a cohort of young women who are likely to develop nutritional disorders

198

REE PREDICTION IN OBESITY

Table 5 Correlation between BMI and bias of each equation in the total sample Bias Harris and Benedict

Bias Owen

Bias Mifflin

Bias Bernstein

0.14 0.077

0.34 0.000

0.10 0.190

0.32 0.000

BMI (r) P Level

Bias WHO

Bias OUR

0.11 0.143

Bias Robertson and Reid

0.03 0.684

0.26 0.001

body mass index (BMI): weight (kg)/height (m)2; bias: measured-predicted; (r): Pearson’s coefficient correlation.

Table 6 Anthropometric variables and analysis of the prediction error after weight loss Z5% First meeting (n=31) X 7 SD Age (years) Weight (kg) Height (cm) BMI (kg/m2) BSA (m2) Metabolic data REE (kcal/day) Harris and Benedict Owen Mifflin Bernstein WHO Robertson and Reid OUR

Weight loss Z5% (n=31)

REE (7 95% CI) Bias (7 95% CI)

24.037 4.43 81.72713.51 160.3278.04 31.8174.97 1.8470.17 1533.127243.10 1625.207141.71** 1381.8797.05 *** 1539.237168.39 1450.557105.67 * 1646.417195.93 ** 1508.117142.71 1482.067155.44

X 7 SD

REE (7 95% CI) Bias (7 95% CI)

73.95711.48 28.7874.15 1.7770.16 1443.95/1622.3 1573.22/1677.19 1346.2/1417.39 1477.46/1601 1411.79/1489.32 1574.54/1718.28 1455.76/1560.46 1425.04/1539.07

157.05/ 27.1 83.58/219.07 70.67/58.47 14.54/150.59 179.55/ 47.01 39.93/89.96 10.47/112.6

1311.87169.55 1550.567122.98 *** 1325.97782.48 1461.557150.63 *** 1392.39790.68 *** 1546.657169.52 *** 1448.447132.49 *** 1392.647132.1 ***

1249.61/1373.99 1505.45/1595.67 1295.71/1356.22 1406.3/1516.8 1359.13/1425.66 1484.47/1608.83 1399.84/1497.04 1344.18/1441.1

281.53/ 195.98 57.67/29.34 195.86/ 103.63 124.44/ 36.73 280.69/ 189 181.29/ 91.97 120.67/ 41

X7SD: mean7standard deviation; BMI: body mass index; BSA: body surface area. Statistical difference between measured and predicted REE on a two-tailed paired t-test (P level: *Po0.05; **Po0.01; ***Po0.001).

requiring a long-term nutritional counselling. The main objective was to analyze the accuracy of some predictive equations according to the subjects’ BMI. Our results indicate that the most accurate equations were Owen’s for the normal-weight group (k0.75 %); Owen’s (k 3.74%) and Bernstein’s (m 0.93%) for the overweight group and, finally, Robertson and Reid’s for the obese group of young women (m0.66%). Cunningham, previously, evaluated the REE study from the physiological and clinical prospective (18), and we addressed this issue from the same clinical and physiological point of view. In the present study, the equations including anthropometric variables, easy to measure (age, weight, height) and to derive (BSA), and suitable for clinical examination were examined. Moreover, the physiological implications related to the energy expenditure were assessed by examining the REE prediction in relationship to weight loss. Bland and Altman’s analyses, as previously stated by their authors, allow one to visualize the degree of error, the type of error related to the prediction equation (underestimation/overestimation) and its numerical entity, along with its gradual development in relation to the REE weight loss. Furthermore, we evaluated the precision of the predictive equations over the entire population. Our results indicate that in the age group ranging 18–35 years, the most accurate equations were Bernstein’s and OUR, followed by Owen’s and Mifflin’s, which presented slight underestimation and overestimation errors, respectively. Other equations, such as that of Harris and Benedict, were not taken into account in the REE

prediction of this population because of the high error rates, as already reported (18, 20–23). De Lorenzo et al. (24), and Antonini (25) on the other hand, suggest the use of the equations of Harris and Benedict and Schofield for the REE prediction of normal-weight and overweight Italian subjects. Moreover, we explored the degree of error for different nutritional states and we observed that in a sample divided according to the BMI the most accurate equations were those of Owen for normal-weight subjects (previously shown by Scalfi et al. (26) in a group of young females), Bernstein for the overweight and the Robertson and Reid for the obese. OUR equation accurately predicted the REE both for the overall population and for the three BMI groups. In the total population, the bias of prediction was affected by BMI only for those of Owen, Bernstein and Robertson and Reid. Particularly, it is worth noting that BMI and bias are directly correlated and, according to Pearson’s correlation coefficient, BMI increases as well as bias of prediction of Owen, Bernstein and Robertson and Reid. This supports our suggestion to apply the equations of Owen, Bernstein and Robertson and Reid to normal-weight, overweight and obese subjects, respectively. Furthermore, it is important to note that bias of our equation was not affected by BMI. The accuracy of predictions in relationship to the different BMI values might depend on the characteristics of population utilized by the different authors for their original studies (Table 2). For example, Owen in his gender-specific equation, derived from a sample of women characterized by different weight, utilized in his

CLINICAL NUTRITION

199

REE Measured

Harris Benedict

Owen

Mifflin

Bernstein

WHO

Our

Robertson Reid

(a) Pearson’s Correlation Coefficient (r) between REE Measured and REE Predicted Harris Benedict Owen Mifflin Bernstein REE Measured 0.75 0.76 0.75 0.76

WHO 0.73

OUR 0.76

Robertson Reid 0.75

Fig. 1 Bland & Altman Analysis (a) Correlations between REE measured and REE predicted; data presented by means of Pearson’s coefficient correlation, histograms and scatterplot. (b) agreement assessment (Bland and Altman method) between bias against mean value for each equations in all the subjects (n=157). Means of predicted and measured REE [(measured + predicted)/2] are plotted against the differences between the two procedures (measured-predicted); the arrow lines represent two SDs from the mean (agreement lines).

200

REE PREDICTION IN OBESITY

Measured vs Predicted (Harris Benedict) 600

400

200

Bias

0

-200

-400

-600

-800 800

1000

1200

1400

1600

1800

2000

2200

1800

2000

2200

Average (REE+HB)/2 Measured vs Predicted (Owen) 800

600

Bias

400

200

0

-200

-400 800

1000

1200

1400

1600

Average (REE+Owen)/2 Fig. 1 Continued

formula weight only as a variable. This variable in a normal-weight sample successfully reflects the influence that body composition has on REE estimations. Then, our data demonstrate that Bernstein’s equation, based on a large population of overweight and obese patients (109–209% of ideal weight) was suitable for the overweight group. However, its linearity causes inaccuracy for the obese group. For the latter, Robertson and Reid’s equation was the best BSA-dependent prediction equation as previously stated by Heshka et al. (19). The accuracy of the equation is likely related to the curvilinear function

‘weight0.425’ for the calculation of BSA. This is important for correlation between REE and the fat-free physiological gain that increases as the obese subjects gain weight. However, predictive equations derived from one type of population might not retain the same accuracy when applied to other populations. The range of applicability and the accuracy of the predictive equations derived from a wide current and modern population stratified in major variables by high technological devices and by a reasonable correlation level between the REE and the adopted variables. The purpose of this complex process

CLINICAL NUTRITION

201

Measured vs Predicted (Mifflin) 600

400

200

Bias

0

-200

-400

-600

-800 800

1000

1200

1400

1600

1800

2000

2200

1800

2000

2200

Average (REE+Mifflin)/2 Measured vs Predicted (Bernstein) 600

400

Bias

200

0

-200

-400 800

1000

1200

1400

1600

Average (REE+Bernstein)/2 Fig. 1 Continued

is to limit as much as possible the error of prediction (Shrinkage) in the other samples compared to the one from which the equation was derived (15, 19). We were interested in the evaluation of prediction error when the BMI changed. From a clinical perspective, the REE prediction after the loss of 5% of the initial weight was inappropriate so that all equations, including OUR, had to be considered inaccurate. In particular, the equations, with the exception of that of Owen which had an acceptable overestimation, showed a remarkable increase of overestimation. On the other hand, from a physiological point of view, the error of

analysis indicated that the phenomena of energy restriction influenced REE with consequent increase of the metabolic efficiency. Many authors, studying REE after a period of spontaneous and/or therapeutic low-energy intake, have suggested that the changes observed were related to the metabolic adjustments not otherwise specified, modifications of the body composition and undefined direct influence of hormonal factors (thyroid, insulin, glucagons, catecholamine, corticosteroids and growth hormones).The metabolic adjustment that occurred during the therapeutic restriction increased the overestimation

202

REE PREDICTION IN OBESITY

Measured vs Predicted (WHO) 600

400

200

Bias

0

-200

-400

-600

-800 800

1000

1200

1400

1600

1800

2000

2200

2000

2200

Average (REE+WHO)/2 Mesaured vs Predicted (Robertson Reid) 600

400

200

Bias

0

-200

-400

-600

-800 800

1000

1200

1400

1600

1800

Average (REE+Robertson Reid)/2 Fig. 1 Continued

of the equations which were derived from the data of subjects with a stable weight and an equal energetic balance (intake = expenditure). During the restriction stage, Owen’s equation revealed an acceptable prediction according to the course of the metabolic adjustment. It is worth noting that within the ambulatory population, two different clinical metabolic status were identified, where REE prediction was strictly bound to an increase of metabolic efficiency after an energy restriction. In the context of a clinical evaluation and therapeutical approach, it is important, thus, to evaluate the intensity and duration of the

nutritional restriction, because the use of prediction equations could yield considerable REE overestimations. Therefore, nutritionists should be aware of the increasing number of subjects affected by the dieting phenomenon which may determine metabolic adjustment influencing the accuracy of REE predictive equations (27–30). Furthermore, dieting might be a triggering factor of eating disorders,. One could observe that our study examined the problem of the error of prediction taking into account data derived only from young women; however, it is

CLINICAL NUTRITION

203

Measured vs Predicted (OUR) 800

600

400

Bias

200

0

-200

-400

-600 800

1000

1200

(b)

1400

1600

1800

2000

2200

Average (REE+OUR)/2

Fig. 1 Continued

BIAS %: First Meeting - Weight Loss >= 5 % of Initial Body Weight (n=31) BMI Variation

BMI

Underprediction

% 10

5.82 5

34 32 30 28 26 24 22 20 18

31.81 28.78

BMI

0.74

BMI Weight Loss >= 5%

0 -0.87

Bias % First Meeting

-2.82 -5

-4.54

-8.65

-10

Overprediction

Bias % WL >= 5 %

-4.68

-9.12

-9.92

-10.83 -12.45 -13.18 -15

-17.81

-18.25 -20 Harris Benedict

Owen

Mifflin

Bernstein

WHO

Robertson Reid

OUR

Fig. 2 Percentage prediction equation (bias%) at the beginning of the treatment and after a weight loss Z5% in a group of 31 subjects. BMI=body mass index; WL=weight loss.

reasonable to conclude that young women most frequently require dietetic treatments. In conclusion, our study suggests that REE measurements are important for assessing REE in subjects

seeking nutritional counselling. When this is not possible, the results of the present study indicate that Owen’s equation for normal weight, Bernstein’s equation for overweight and Robertson and Reid’s equation

204

REE PREDICTION IN OBESITY

for obese subjects should be chosen for prediction of REE in young women. Finally, according to the phenomena of metabolic adaptation during the therapeutic or spontaneous energy restriction, we suggest to use Owen’s equation.

13. 14. 15.

Acknowledgements The authors wish to thank Dr Paola Merolla for her collaboration in the translation of the present work.

16. 17. 18.

References 1. Livelli di Assunzione Raccomandata di nutrienti per la popolazione italiana (LARN), Societa` Italiana di Nutrizione Umana. Revisione, 1996 2. McArdle W D, Katch F I, Katch V L. Fisiologia Applicata allo Sport, Aspetti energetico, nutrizionale e performance. Milano: Casa Editrice Ambrosiana, 1998 3. Harris J A, Benedict F G. Biometric Studies of Basal Metabolism in Man. Washington, DC: Carnegie Institute of Washington, 1919 (Publication number 297) 4. Boothby W M, Berkson J, Dunn H L. Studies of the energy metabolism of normal individuals: a standard for basal metabolism with a nomogram for clinical application. Am J Physiol 1936; 116: 468–484 5. Robertson J D, Reid D D. Standards for the basal metabolism in normal people in Britain. Lancet 1952; 1: 940–943 6. American Psychiatric Association Practice Guidelines, Practice Guideline for the Treatment of Patient with Eating Disorders (Revision). Am J Psychiatry 2002; 157 (1)(Supplement): 1–39 7. Position of ADA, Nutrition intervention in the treatment of anorexia nervosa, bulimia nervosa and eating disorders non otherwise specified (EDNOS). J Am Diet Assoc, 2001; 101 (7): 810–819 8. Bland J M, Altman D G. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 307–310 9. Bland J M, Altman D G. Comparing methods of measurements: why plotting difference against standard method is misleading. 1995; 1085–1087 10. Lohman T G, Roche A, Martorell R. Manuale di riferimento per la standardizzazione antropometrica. Milano: EDRA Edizioni, 1994 11. Du Bois B S, Du Bois E F. Clinical calorimetry: a formula to estimate the approximate surface area if height and weight be known. Arch Intern Med 1916; 17: 863–871 12. Schutz Y. Terminology, factors and constants in studies on energy metabolism of humans. Glossary of energy terms and factors used for calculations of energy metabolism in human studies. In: Van Submission date: 10 January 2002 Accepted: 29 October 2002

19.

20.

21. 22. 23. 24.

25. 26. 27.

28.

29. 30.

Es A J H (ed). Human Energy Metabolism: Physical Activity and Energy Expenditure Measurements in Epidemiological Research Based Upon Direct and Indirect Calorimetry. EuroFNut Report 5. Wageningen, the Netherlands, 1984; 153–168 Weir J B D. New method for calculating metabolic rate with special reference to protein metabolism. J Physiol 1949; 1–9 Owen O E, Kavle E, Owen S R et al. A reappraisal of caloric requirements in healthy women. Am J Clin Nutr 1986; 44: 1–19 Mifflin M D, St Jeor T S, Hill L A, Scott B J, Daugherty S A, Koh Y O. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr 1990; 51: 241–247 WHO. Energy and protein requirements. Technical Reports Series 724, Geneva, 1985 Bernstein R S, Thornton J C, Yang M U et al. Prediction of the resting metabolic rate in obese patients. 1983; 37: 595–602 Cunningham J J. Body composition as a determinant of energy expenditure: a synthetic review and a proposed general prediction equation. Am J Clin Nutr 1991; 54: 963–969 Heshka S, Feld K, Yang M U, Allison D, Heymsfield S B. Resting energy expenditure in the obese: a cross validation and comparison of prediction equations. J Am Diet Assoc 1993; 93: 1031–1036 Daly J M, Heymsfield S B, Head C A, Harvey L P, Nixon D W, Katzeff H, Grossman G D. Human energy requirements: overestimation by widely used prediction equation. Am J Clin Nutr 1985; 42(6): 1170–1174 Garrel R D, Jobin N, De Jonghe M H L. Should we still use the Harris and Benedict Equations? Nutr Clin Pract 1996; 11: 99–103 Frankenfiled D C, Muth E C, Rowe W A. The Harris Benedict studies of human basal metabolism: history and limitations. J Am Diet Assoc 1998; 98: 439–445 Roza M A, Shizgal H M. The Harris Benedict equation revaluated: resting energy requirements and the body cell mass. Am J Clin Nutr 1984; 40: 168–182 De Lorenzo A, Tagliabue A, Andreoli A, Testolin G, Comelli M, Deurenberg P. Mesaured and predicted resting metabolic rate in Italian males and females, aged 18–59 y. Eur J Clin Nutr 2001; 55(3): 208–214 Antonini R, Tagliabue A, Comelli M, Lanzola E. Stima del metabolismo basale: revisione critica e applicazione sperimentale delle equazioni predittive. Clin Dietol 1999; 27: 93–100 Scalfi L, Coltorti A, Sapio C et al. Predicted and measured resting energy expenditure in healthy young women. Clin Nutr 1993; 12: 1–7 Weinsier R L, Nagy T R, Hunter G R, Darnell B E, Hensrud D D, Weiss H L. Do adaptive changes in metabolic rate favor weight regain in weight reduced individuals: an examination of the set point theory. Am J Clin Nutr 2000; 72: 1088–1094 Valtuena S, Blanch S, Barenys M, Sola` R, Salas Salvado J. Changes in body composition and resting energy expenditure after rapid weight loss: is there an energy metabolism adaptation in obese patient? Int J Obes 1995; 19: 119–125 Waterlow J C. The nature and significance of nutritional adaptation. Eur J Clin Nutr 1999; 53(suppl 1): S2–S5 Shetty P S. Adaptation to low energy intakes: the responses and limits to low intakes in infants, children and adults. Eur J Clin Nutr 1999; 53(suppl 1): S14–S33