Evaluation of body fat in fatter and leaner 10-y-old African American and white children: the Baton Rouge Children s Study 1 3

Original Research Communications Evaluation of body fat in fatter and leaner 10-y-old African American and white children: the Baton Rouge Children s Study 1 3 George A Bray, James P DeLany, David W Harsha, Julia Volaufova, and Catherine C Champagne ABSTRACT Background: Only a few published studies in children used several methods to compare body fat in large groups of fatter and leaner multiethnic children. We hypothesized that the preferred methods of determining body fat may differ in children with larger compared with smaller amounts of body fat, in boys compared with girls, and in African Americans compared with whites. Objective: Our objective was to evaluate several methods of predicting body fat in 10 12-y-old white and African American boys and girls. Design: The body fat of 129 African American and white boys and girls aged 10 12 y, distributed equally by sex and race, was measured with use of dual-energy X-ray absorptiometry (DXA), underwater weighing (densitometry), isotope dilution (H 2 18 O), bioelectrical impedance, skinfold thicknesses, corporal diameters, and circumferences. Results: With use of DXA as the criterion variable, body fat was bimodally distributed in the boys and skewed to higher values in the girls. Biceps skinfold thickness had the highest predictive value of any single skinfold thickness compared with DXA fat. All formulas for estimating body fat from skinfold thicknesses, body density, or impedance performed better in the children in the upper one-half of the fat distribution (the fatter children) than in those in the lower one-half (the leaner children). Body mass index was highly correlated with body fat (R 2 = 0.77); there was a good correlation for the fatter children (R 2 = 0.66) and no correlation for the leaner children (R 2 = 0.09). The hydration of the fat-free mass was significantly higher in the fatter children than in the leaner ones (79.2% compared with 76.7%). Conclusions: These data are consistent with the hypothesis that all methods of estimating body fat work better in children with larger amounts of body fat. The best formulas use skinfold thicknesses, bioelectrical impedance, and a 4-compartment model. Am J Clin Nutr 2001;73:687 702. KEY WORDS DXA, dual-energy X-ray absorptiometry, bioelectrical impedance analysis, skinfold thicknesses, densitometry, isotope dilution, children, body fat, ethnicity, race INTRODUCTION The preadolescent and adolescent years are a period of rapid growth in the body fat and nonfat compartments. Gonadal hormones modify the rapidity of growth and the pattern of fat deposition during adolescence. The prevalence of obesity in children is increasing and childhood is a time of increasing risk of developing obesity and its attendant complications (1, 2). Several factors are suspected in this trend, including increased food intake, reduced amounts of physical activity in childhood, and a pattern of food intake in which fast foods with a high energy density play an important role (3). Not all children are at risk of developing obesity (2). Most studies suggest that 30% of the total burden of obesity begins in childhood and the rest in adult life (4). The prevalence of childhood obesity also shows ethnic and generational differences (4). In a cross-sectional analysis of the distribution of body mass index (BMI) in 2 samples of 6- and 14-y-old children over an interval of 10 y, Troiano and Flegal (2) showed that the children in the upper part of the BMI distribution had the greatest risk of obesity. When the difference in mean BMI at each interval of BMI was calculated for 2 samples of 6-y-olds or 2 samples of 14-y-olds separated by a decade and this mean BMI difference was plotted against BMI, the lower part of the BMI curve was parallel to the x axis, meaning that the children in this part of the weight distribution were increasing similarly in BMI. For the upper part of the BMI distribution, however, there was an upward slope, meaning that during this 10-y interval BMI had increased more in individuals with higher BMIs. The current study of body composition in preadolescent children was designed to oversample children from the upper and lower parts of the body fat distribution to test the hypothesis that predictive equations for body fat are better at predicting fat in fatter children. We also tested the hypothesis that prediction of body fat differs by race and sex. Underwater weighing has been the gold standard for measurement of body composition in humans (5). The 2 widely used formulas by Brozek et al (5) and by Siri (6) were developed in young white men. The applicability of these formulas to growing children of different races has been questioned because the density of lean body mass changes in growing children. For this reason, skinfold-thickness formulas specific for children were developed (7 11). With the advent of dual-energy 1 From the Pennington Biomedical Research Center, Baton Rouge, LA. 2 Supported in part by NIH HD 28020. 3 Address reprint requests to GA Bray, Pennington Biomedical Research Center, 6400 Perkins Road, Baton Rouge, LA 70808. E-mail: brayga@pbrc.edu. Received September 8, 1999. Accepted for publication August 1, 2000. Am J Clin Nutr 2001;73:687 702. Printed in USA. 2001 American Society for Clinical Nutrition 687

688 BRAY ET AL X-ray absorptiometry (DXA) (12, 13), which measures bone mineral content, lean mass, and fat mass, the assumptions made in the development of the formulas based on density can be examined critically in children (10). SUBJECTS AND METHODS Subjects The subjects for this study were children from the Baton Rouge, LA, public school system. With the approval of the Superintendent of Schools, the East Baton Rouge School Board, the Louisiana State University Institutional Review Board, and the principals of the 8 schools included, a letter describing the screening study was sent to all children in the 5th grade. We later had to add the 4th grade because the number of black girls below Tanner stage 3 was too small among 5th-grade girls. Those who signed the first consent form (n = 330) underwent an initial screening. During this screening evaluation, triceps and subscapular skinfold thicknesses were measured and Tanner was stage determined by an examiner of the same sex as the participant. Only children at a Tanner stage of 1 2 were included. An equal number of African Americans and whites and an equal number of boys and girls were selected from the screening cohort by sampling from each end of the sum of the triceps plus subscapular skinfold-thickness distribution, which provided a bimodally distributed sample. The final selected sample consisted of 33 white boys, 32 white girls, 32 African American boys, and 32 African American girls who volunteered for the study and whose parents signed a second consent form, which was also approved by the Institutional Review Board and which described the procedures detailed below. Procedures for measuring body composition Anthropometry While the children wore hospital gowns, body weight was measured to the nearest 0.1 kg with an electronic scale (Detecto, Webb City, MO) that was calibrated quarterly and checked with a standard 25-kg weight daily. Height was measured to the nearest 0.5 cm with a wall-mounted stadiometer and BMI was calculated as wt (in kg)/ht 2 (in m). Skinfold thicknesses at the triceps, biceps, subscapula, suprailium, lateral midcalf, and midthigh were measured in duplicate to the nearest 1 mm with a Lange Caliper (Cambridge Scientific Instruments, Cambridge, MD). If the measurements differed by >2 mm, a third measurement was taken and the measurements were averaged. Sites on the right side of the body were used and identified from the atlas of Lohman et al (14). All measurements were made by the same individual (DWH). The CV for the sum of 4 skinfold thicknesses was 8.3%. Circumferences were measured in duplicate to the nearest 0.1 cm with a steel tape at the wrist, midthigh, and midcalf by using the positions in the atlas of Lohman et al (14). Diameters at the shoulder, hip, knee, ankle, and elbow were measured with a stadiometer to the nearest 0.1 cm (Gneupel, Basel, Switzerland) and the average of 2 readings was used. Underwater weighing Underwater weight was measured while the subjects wore bathing suits and sat on a submerged chair that rested on 4 force cubes, similar to the method of Akers and Buskirk (15) except that our system used a permanent tank. Residual lung volume was measured with use of a helium dilution technique (Sensormedics PFT, Fullerton, CA) while the subjects were submerged. The precision of this procedure was determined in 10 college-aged students. In this group, the CV for underwater weighing was 2.6% and the CV for residual lung volume was 5.2%, giving a CV for the overall measurement of percentage of body fat of 5.1%. Isotope ( 18 O) dilution Total body water (TBW) was measured by H 18 2 O dilution with a dose of 0.3 g H 18 2 O/kg TBW. The 18 O isotope abundances were measured on a Finnigan MAT 252 gas-inlet isotope ratio mass spectrometer (Bremen, Germany) with a carbon dioxide water equilibration device (16). Salivary enrichment was measured by comparing a baseline sample with the average of the 2-h and 3-h postdose samples (SD of TBW with use of the 2-h and 3-h samples = 0.16 kg) corrected for revised dilution space constants (17 19). It was shown previously that 18 O is fully equilibrated within 2 3 h (19). Bioelectrical impedance analysis Whole-body reactance was measured by using a variable frequency impedance machine (Xitron, San Diego). For the wrist, one electrode was placed to bisect the ulnar head and the other electrode was placed just behind the middle finger. One of the ankle electrodes was placed to bisect the medial malleolus and the other was placed just behind the middle toe. The CV for the measurement of body fat by bioelectrical impedance analysis (BIA) was 4.2%. Dual-energy X-ray absorptiometry Whole-body scans were made with the Hologic QDR 2000 in the fan beam mode (Hologic Inc, Waltham, MA). The scans were analyzed by using enhanced WHOLE BODY software (version 5.60; Hologic Inc). For ethical reasons, replicate measurements were not made in children, but repeat scans were made in 5 young adults. In this group, the CV for weight was 0.09%; the CVs for bone mineral content and bone mineral density were 0.8% and 1.3%, respectively; and the CVs for fat-free mass (FFM), fat mass, and percentage of body fat were 0.8%, 1.6%, and 1.7%, respectively. Multicompartment models With the data obtained from these studies, several 2-compartment models are possible, as are a 3-compartment and a 4-compartment model. Because several other models in the literature were used to assess the body composition of children in the age range and pubertal status of the children reported here, we compared our data with some of these published models (Table 1). Data analysis All body-composition variables were analyzed by simple analysis of variance with respect to race and sex, with a medial split for body fat. Percentage of body fat as measured by DXA was chosen as the criterion variable for the analyses in this study because it provides a 3-compartment model with use of a single instrument and was validated in several previous studies (30 34). Using DXA, we generated 2 populations with respect to percentage of body fat. The median value was chosen to divide the children into 2 fat groups: those in the upper one-half of the body fat distribution (the fatter children) and those in the lower

TABLE 1 Summary of published formulas used to evaluate body fat in the present study 1 BODY COMPOSITION IN CHILDREN 689 Author Study group Age and number Formula Siri (6) Theoretical paper Fat (%) = (4.950/D 4.500) 100 Lohman et al (7) Prepubescent children n = 12 WF, 3 BF, Fat (%) = (5.30/D 4.89) 100 (used for prepubescent children 31 WM, 8 BM rather than Siri) Weststrate and Boys and girls 0 23.9 mo Fat (%) = {585 4.2[age (mo)] 0.5 }/D {550 5.1[age (mo)] 0.5 } Deurenberg (8) Boys 2 18 y Fat (%) = {562 4.2[age (y) 2}/D {525 4.7[age (y) 2]} Girls 2 10 y Fat (%) = {562 1.1[age (y) 2]/D {525 1.4[age (y) 2} Girls 10 18 y Fat (%) = {553 7.3[age (y) 10}/D {514 8.0[age (y) 10} Schutte et al (9) African Americans 20.3 ± 0.44 2 y, n = 15 M Fat (%) = (4.374/D 3.928) 100 Whites 22.5 ± 0.80 y, n = 19 M Fat (%) = (4.570/D 4.142) 100 Slaughter et al (11) Prepubescent males 9.8 ± 1.3 y, n = 50 For triceps and calf: Prepubescent females 10.0 ± 1.0 y, n = 16 Fat (%) (M) = 0.735(Tri + Calf) + 1.0 Pubescent males 12.2 ± 1.4 y, n = 30 Fat (%) (F) = 0.610(Tri + Calf) + 5.1 Pubescent females 11.4 ± 1.9 y, n = 29 For triceps and subscapular: Postpubescent males 15.8 ± 1.6 y, n = 58 Fat (%) (WM) = 1.21(Tri + Sub) 0.008(Tri + Sub) 2 1.7 Postpubescent females 15.3 ± 1.6 y, n = 59 Fat (%) (BM) = 1.21(Tri + Sub) 0.008(Tri + Sub) 2 3.2 Adult males 23.1 ± 2.6 y, n = 36 Fat (%) (WM) = 1.21(Tri + Sub) 0.008(Tri + Sub) 2 3.4 Adult females 22.5 ± 2.9 y, n = 32 Fat (%) (BM) = 1.21(Tri + Sub) 0.008(Tri + Sub) 2 5. Fat (%) (WM) = 1.21(Tri + Sub) 0.008(Tri + Sub) 2 5.5 Fat (%) (BM) = 1.21(Tri + Sub) 0.008(Tri + Sub) 2 6.8 All females Fat (%) = 1.33(Tri + Sub) 0.013(Tri + Sub) 2 2.5 All males Fat (%) = 0.783(Tri + Sub) + 1.6 All females Fat (%) = 0.783(Tri + Sub) + 9.7 Kushner et al (20) Neonates 0.02 y, n = 32 TBW (L) = 0.59(Ht 2 /R) + 0.065[Wt (kg)] + 0.04 Preschool children 1.1 y, n = 30 M, 14 F Prepubertal children 7.6 y, n = 23 M, 14 F Adults 41 y, n = 16 M, 36 F Ellis (21) White girls 3 18 y, n = 14 Fat (kg) = 0.642(Wt) 0.120(Ht) 0.606[age (y)] + 8.98 African American girls 3 18 y, n = 104 Fat (kg) = 0.653(Wt) 0.163(Ht) 0.298[age (y)] + 10.7 Hispanic girls 3 18 y, n = 68 Fat (kg) = 0.677(Wt) 0.217(Ht) + 15.5 Ellis (22) White boys 3 18 y, n = 145 Fat (kg) = 0.534(Wt) 1.59[age (y)] + 3.03 African American boys 3 18 y, n = 78 Fat (kg) = 0.594(Wt) 0.381(Ht) + 36.0 Hispanic boys 3 18 y, n = 74 Fat (kg) = 0.591(Wt) 1.82[age (y)] + 3.26 Goran et al (10) Prepubescent girls 6.5 ± 1.4 y, n = 49 FFM (kg) = 0.15(Sub) + 0.36(Wt) + 0.12(Tri) 0.2(Ht 2 /R) 2.3 Prepubescent boys 6.6 ± 1.4 y, n =49 Deurenberg et al (23) Group 1: boys 8.6 ± 0.6 y, n = 18 FFM (kg for age 7 9 y M & F) = 0.640(Ht 2 /R) + 4.83 Group 1: girls 8.3 ± 0.9 y, n = 21 FFM (kg for age 10 12 y F & 10 15 y M) = 0.483(Ht 2 /R) + 0.221(Wt) + 0.1277(Ht) 14.7 Group 2: boys 12.8 ± 1.5 y, n = 71 FFM (kg for F > 13 y; M > 16 y) = 0.258 (Ht 2 /R) + 0.375(Wt) + 6.3(sex) + 10.50(Ht) 0.164[age (y)] 6.5 Group 2: girls 10.7 ± 1.0 y, n =20 Group 3: men 21.6 ± 2.8 y, n = 41 FFM (all ages) = 0.483(Ht 2 /R) + 0.308(Wt) + 1.6(sex) + 7.04(Ht) 8.5 Group 3: women 17.6 ± 3.6 y, n =75 Schaefer et al (24) Group A 12.0 ± 3.7 y, n = 30 M, 27 F FFM (kg) = 0.22 + 0.62(Ht 2 /R) + 0.76[age (y)] Group B 11.7 ± 3.7 y, n = 29 M, 34 F FFM (kg) = 0.45 + 0.68(Ht 2 /R) + 0.58[age (y)] Combined equation FFM (kg) = 0.15 + 0.65(Ht 2 /R) + 0.68[age (y)] Supraspongin et al (25) Prepubertal children 10.6 ± 0.3 y, n = 8 M, 10 F FFM (kg) = 0.632(Ht 2 /R) + 289(Wt) + 1.445 Pubertal children 13.7 ± 0.3 y, n = 8 M, 8 F Adults 22 ± 0.7 y, n = 5 M, 3 F Houtkooper et al (26) Girls 12.3 ± 1.1 y, n = 41 F (%) = {(2.057/D) [0.786(Wt)] 1.286} 100 Boys 12.3 ± 1.4 y, n = 53 FFM (kg) = 0.713(Ht 2 /R) + 0.150(Chest circ) + 0.493 (hip skeletal width) + 0.121(React) 21.41 Hewit et al (27) Prepubescent males 8.5 ± 1.0 y, n = 17 Fat (%) = 0.51[fat(%) from TBW)] + 11.12 Prepubescent females 8.4 ± 1.7 y, n =11 Young adult males 31.1 ± 6.3 y, n =12 Young adult females 32.6 ± 6.0 y, n =19 Older adult males 69.8 ± 3.7 y, n =30 Older adult females 70.0 ± 4.7 y, n =32 (Continued)

690 BRAY ET AL TABLE 1 (Continued) Friedl et al (28) Soldiers 21 ± 2 y, n = 9 M, 1 F 3-Compartment model Fat (%) = [2.118/D 0.780(TBW/Wt) 1.354] 100 4-Compartment model Fat (%) = [2.559/D 0.734(TBW/Wt) + 0.983(TBBM/Wt) 1.841] 100 Wells et al (29) Boys 9.7 ± 1.3 y, n = 16 2-Compartment model Girls 10.1 ± 1.4 y, n = 14 Fat (kg) = [(D FFM /D) 1]/(D FFM / 0.9007) 1] 3-Compartment model Fat (%) = (2.220BV) (0.764TBW) + [1.465BW] 4-Compartment model Fat (kg) = (2.747BV) (0.710TBW) + [1.460BMC (kg)] (2.050BW) Present study Boys and girls 10.79 ± 0.05 y, n = 129 4-Compartment model Fat (%) = [ 0.423 + 1.353/D + 0.83TBW/Wt 4.56BMC (kg)/wt] 100 1 B, black; BMC, bone mineral content by dual-energy X-ray absorptiometry (DXA); BV, body volume; BW, body weight; Calf, calf skinfold thickness in mm; Circ, circumference; D, density; F, females; FFM, fat-free mass; Ht, height; M, males; R, resistance; React, reactance; Sub, subscapular skinfold thickness in mm; TBBM, total-body bone mass; TBW, total body water; Tri, triceps skinfold thickness in mm; W, women; Wt, weight in kg. 2 x ± SEM. one-half (the leaner children). All body-composition variables were analyzed with respect to the 2 fat groups by using unpaired t tests. Linear regression analysis was used to compare several different methods of measuring body fat: DXA and the others proposed by different authors. Multiple linear regression with the R 2 variable selection method was used to create a prediction model for percentage of body fat with use of skinfold thicknesses (35), breadths (35), and circumferences. Because the graphic inspection of percentage of body fat versus the sum of skinfold thicknesses indicated a nonlinear relation, we logarithmically transformed the sum of skinfold thicknesses. Eventually, considering skinfold-thickness measurements, body density measures, BIA measures, and DXA measures of bone mineral content, all variables were used in an R 2 variable selection process. The best prediction model was identified for all children regardless of sex-race subgroups because none of the tests for different slopes and intercepts among these groups were statistically significant. Within both fat groups, predicted values from this model were compared with observed models in terms of squared correlations. We used SAS (version 6.12; SAS Institute Inc, Cary, NC) to perform all analyses and to create the tables and graphs. RESULTS Study population Summarized in Table 2 are the characteristics of all subjects in the study population and of the subgroups according to sex, race, and the median split of body fat as measured by DXA. The age range of the children was narrow because almost all were recruited from the 5th grade. All children were at Tanner stage 1 (n = 101) or Tanner stage 2 (n = 28). A comparison of variables according to Tanner stage showed no significant differences, so the 2 groups were collapsed for all remaining analyses. Heights were variable and body weights even more so, as anticipated by the study design. The boys were significantly taller and heavier and had significantly more bone mineral content than the girls, but body fat was not significantly different between the sexes. Race was 49.6% African American and 50.4% white. There were no significant differences in height or weight between the races but bone mineral content and bone mineral density were significantly higher in the African Americans than in the whites. The mean weight of the fatter children was 14 kg higher than that of the leaner children. Percentage of body fat was nearly double in the fatter children. Children were selected from the lower and upper halves of the distribution of the sum of triceps plus subscapular skinfold thickness. The selection strategy was intended to provide a bimodal distribution of body fat; as is shown in Figure 1, this was accomplished for both sexes and both races. In contrast with the bimodality of body fat, lean body mass as measured by DXA was not bimodal but was normally distributed in the girls, in the boys, and in each race (data not shown). In the remainder of the discussion, the population will be considered as a whole, and each half of the bimodal distribution will be treated separately because prediction formulas that use density, skinfold thicknesses, BIA, or isotope dilution may be different or better in the lower or upper part of the fat distribution. Anthropometry The data on skinfold thicknesses, circumferences, and breadths are presented in Table 3. The boys had larger suprailiac skinfold thicknesses than did the girls, but the other skinfold thicknesses did not differ significantly. Calf skinfold thickness and pelvic breadth were larger in the whites than in the African Americans, but all other anthropometric measures were similar. As planned, the skinfold thicknesses, breadths, and circumferences were greater in the fatter children than in the leaner ones. We compared the triceps skinfold-thicknesses distribution for the children in our sample with normative data from the National Center for Health Statistics (36). As shown in Table 4, the top 2 categories of the distribution were overrepresented in the selected sample (n = 129); there were fewer individuals in the middle categories and in the lowest category. The larger screening sample (n = 309) of children in our study matched the normative data more closely but still overrepresented children in the highest and lowest levels of body fat. We also compared the BMI data of our sample with normative data obtained by Rossner (37), who pooled 3 large samples. In that data set, BMI categories were provided for each sex and for

BODY COMPOSITION IN CHILDREN 691 TABLE 2 Characteristics of all of the children and of the subgroups 1 Sex Race Body fat All children Girls Boys African Americans Whites Leaner Fatter (n = 129) (n = 64) (n = 65) (n = 64) (n = 65) (n = 65) (n = 64) Age (y) 10.79 ± 0.05 10.70 ± 0.08 10.87 ± 0.08 10.81 ± 0.09 10.76 ± 0.06 10.78 ± 0.07 10.79 ± 0.08 Weight (kg) 2,3 40.93 ± 0.95 37.78 ± 1.20 44.02 ± 1.37 41.26 ± 1.38 40.59 ± 1.32 33.89 ± 0.52 48.07 ± 1.34 Height (cm) 2,3 145.09 ± 0.62 143.84 ± 0.91 146.33 ± 0.81 145.48 ± 0.87 144.71 ± 0.87 143.47 ± 0.79 146.75 ± 0.91 BMI (kg/m 2 ) 2,3 19.25 ± 0.35 18.12 ± 0.46 20.36 ± 0.48 19.35 ± 0.52 19.16 ± 0.46 16.42 ± 0.16 22.15 ± 0.45 DXA fat (%) 3 26.63 ± 0.90 26.13 ± 1.26 27.12 ± 1.29 25.63 ± 1.31 27.61 ± 1.23 17.89 ± 0.44 35.50 ± 0.79 Lean (kg) 2,3 27.39 ± 0.39 25.48 ± 0.49 29.28 ± 0.50 27.98 ± 0.59 26.81 ± 0.50 26.08 ± 0.39 28.72 ± 0.64 Fat (kg) 3 11.59 ± 0.64 10.44 ± 0.86 12.72 ± 0.94 11.28 ± 0.92 11.90 ± 0.91 5.99 ± 0.20 17.28 ± 0.81 BMC (kg) 2 4 1.28 ± 0.02 1.20 ± 0.03 1.36 ± 0.03 1.34 ± 0.03 1.23 ± 0.03 1.20 ± 0.02 1.37 ± 0.03 BMD (g/m 2 ) 2,4 0.91 ± 0.01 0.89 ± 0.01 0.92 ± 0.01 0.93 ± 0.01 0.88 ± 0.01 0.90 ± 0.01 0.91 ± 0.01 Resistance at 50 khz ( ) 2,3 663.49 ± 6.97 696.16 ± 10.70 631.33 ± 7.02 672.04 ±11.32 655.10 ± 8.15 681.91 ± 8.76 644.79 ± 10.43 Reactance at 50 khz ( ) 2,3 72.96 ± 0.82 74.01 ± 1.33 71.94 ± 0.97 75.21 ± 1.27 70.75 ± 0.99 75.16 ± 1.13 70.73 ± 1.14 Total body water (L) 2,3 22.43 ± 0.35 20.90 ± 0.41 23.94 ± 0.49 23.02 ± 0.53 21.84 ± 0.43 20.98 ± 0.32 23.90 ± 0.56 Density (g/l) 3 1043 ± 2 1045 ± 3 1042 ± 3 1044 ± 3 1043 ± 3 1060 ± 2 1027 ± 2 1 x ± SE. 2 Significant effect of sex, P < 0.05 (three-way ANOVA). 3 Significant effect of body fat group, P < 0.05 (three-way ANOVA). 4 Significant effect of race, P < 0.05 (three-way ANOVA). both races. As shown in Table 4, the upper BMI categories were oversampled in our selected sample and the lower ones were less well represented. The larger (n = 330) screening sample more closely matched the normative data but still overrepresented the fatter group. When BMI was compared with body fat by DXA for the entire sample, the correlation was high (R 2 = 0.77), similar to that observed with underwater weighing or skinfold-thickness measures. BMI was also a good predictor of body fat in the fatter children (R 2 = 0.66), whereas there was no significant relation between body fat and BMI in the leaner children (R 2 = 0.09). Shown in Table 5 are the regression models for percentage of body fat as measured by DXA with use of the R 2 variable selection method against all of the anthropometric variables listed in Table 3. Only the best models with 1, 2, or 3 variables are presented, although models were run for all 14 variables. Biceps skinfold thickness entered the regression model at step 1 and by step 3 was present in all but 1 of the 3-variable models. Calf skinfold thickness and triceps skinfold thickness also entered early and rapidly became dominant. The same regression model method was used to evaluate anthropometric models for predicting body fat in the subgroups (data not shown). In this model, biceps skinfold thickness was the dominant variable for the white girls and the African American boys but not for the other 2 subgroups. For the African American girls, triceps skinfold thickness was the dominant skinfold thickness; for the white boys, it was a combination of hip circumference, pelvic breadth, and thigh skinfold thickness. The only other group in which breadths or circumferences were important was the African American girls, in whom shoulder breadth entered early. The R 2 variable selection procedure was also used with the skinfold thicknesses alone, excluding the breadths and circumferences. Biceps skinfold thickness was again the best single variable correlated with percentage of body fat measured by DXA. It also entered into the model with any combination of 2 skinfold thicknesses for all subjects. In the subgroups, the picture was somewhat different. Biceps skinfold thickness was the best predictor of fat in the white girls, second behind calf skinfold thickness in the African American boys, third behind FIGURE 1. Distribution of percentage of body fat as measured by dual-energy X-ray absorptiometry (DXA) in the whites, African Americans, boys, and girls.

692 BRAY ET AL TABLE 3 Anthropometric characteristics of all of the children and of the subgroups 1 Sex Race Body fat All children Girls Boys African Americans Whites Leaner Fatter Skinfold thickness (mm) Triceps 2 15.74 ± 0.64 14.67 ± 0.93 16.80 ± 0.88 14.87 ± 0.90 16.60 ± 0.91 10.27 ± 0.36 21.30 ± 0.76 Biceps 2 10.26 ± 0.53 9.69 ± 0.76 10.83 ± 0.73 9.54 ± 0.71 10.98 ± 0.77 5.61 ± 0.23 14.99 ± 0.61 Subscapular 2 11.40 ± 0.67 10.34 ± 0.83 12.45 ± 1.05 11.38 ± 0.92 11.43 ± 0.99 5.96 ± 0.16 16.94 ± 0.93 Suprailiac 2,3 15.07 ± 0.84 13.33 ± 1.07 16.78 ± 1.27 14.59 ± 1.13 15.54 ± 1.25 7.64 ± 0.31 22.61 ± 1.00 Thigh 2 24.50 ± 1.00 23.79 ± 1.34 25.20 ± 1.49 24.08 ± 1.48 24.92 ± 1.35 15.77 ± 0.63 33.37 ± 1.10 Calf 2,4 15.28 ± 0.73 14.75 ± 1.12 15.80 ± 0.93 13.77 ± 0.85 16.76 ± 1.15 9.56 ± 0.38 21.09 ± 0.97 Circumferences (cm) Waist 2,3 64.66 ± 0.98 59.93 ± 1.07 69.31 ± 1.42 64.04 ± 1.36 65.26 ± 1.42 57.19 ± 0.53 72.24 ± 1.36 Hip 2,3 79.58 ± 0.91 77.16 ± 1.24 81.87 ± 1.27 80.01 ± 1.32 79.17 ± 1.26 72.13 ± 0.62 87.16 ± 1.09 Calf 2,3 30.05 ± 0.30 29.03 ± 0.38 31.05 ± 0.43 30.05 ± 0.45 30.04 ± 0.40 27.91 ± 0.25 32.22 ± 0.40 Breadth (cm) Elbow 2,3 6.52 ± 0.05 6.30 ± 0.07 6.75 ± 0.07 6.58 ± 0.07 6.47 ± 0.08 6.24 ± 0.05 6.81 ± 0.08 Shoulder 2,3 32.02 ± 0.28 31.11 ± 0.47 32.91 ± 0.26 32.01 ± 0.51 32.03 ± 0.24 31.24 ± 0.25 32.81 ± 0.48 Hip 2 24.83 ± 0.25 24.47 ± 0.41 25.18 ± 0.28 24.42 ± 0.42 25.23 ± 0.27 23.39 ± 0.33 26.29 ± 0.26 Pelvic 2,4 22.19 ± 0.18 21.89 ± 0.27 22.49 ± 0.23 21.54 ± 0.23 22.84 ± 0.25 21.43 ± 0.21 22.96 ± 0.26 Ankle 6.51 ± 0.20 6.49 ± 0.40 6.52 ± 0.04 6.73 ± 0.40 6.29 ± 0.04 6.64 ± 0.39 6.37 ± 0.06 1 x ± SE. 2 Significant effect of body fat group, P < 0.05 (three-way ANOVA). 3 Significant effect of sex, P < 0.05 (three-way ANOVA). 4 Significant effect of race, P < 0.05 (three-way ANOVA). TABLE 4 Comparison of triceps skinfold-thickness and BMI categories in the present study with normative data 1 triceps and thigh skinfold thicknesses in the white boys, and third behind triceps and subscapular skinfold thicknesses in the African American girls. When the breadths and circumferences were examined by the R 2 variable selection method, hip circumference and hip breadth were most strongly correlated with body fat (R 2 = 0.70 for hip circumference and R 2 = 0.35 for hip breadth). Waist circumference was the second circumference to enter the model but added essentially nothing (R 2 = 0.86 compared with R 2 = 0.85). Formulas for estimating body fat with use of 1 skinfold thickness or a combination of 2, 3, or 4 skinfold thicknesses are shown in Table 6 for the entire population. For the entire group, the R 2 ranged from 0.77 (subscapular and calf) to 0.85 (biceps), with a minimal mean square error of 15.36 24.23. As shown in Screening sample Selected sample Triceps or BMI category (n = 309) (n = 129) Normative n (%) n (%) % Triceps category (mm) Boys Girls 0 6 0 8 50 (16.2) 12 (9.3) 10 6 8 8 10 61 (19.7) 16 (12.4) 15 8 10 10 12 50 (16.2) 22 (17.1) 25 10 14 12 17 54 (17.5) 22 (17.1) 25 14 18 17 23 42 (13.6) 19 (14.7) 15 18 21 23 27 14 (4.5) 12 (9.3) 5 >21 >27 38 (12.3) 26 (20.2) 5 BMI category (kg/m 2 ) Boys Girls Black White Black White 0 14.6 0 14.6 0 14.2 0 13.9 12 (3.9) 2 (1.6) 5 14.6 15.4 14.6 15.4 14.2 15.2 13.9 14.9 11 (3.6) 8 (6.2) 10 15.4 17.1 15.4 17.1 15.2 17.4 14.9 16.9 100 (32.4) 42 (32.6) 35 17.1 19.0 17.1 19.2 17.4 20.2 16.9 19.3 73 (23.6) 28 (21.7) 25 19.0 20.6 19.2 20.9 20.2 22.3 19.3 21.0 39 (12.6) 9 (7.0) 10 20.6 24.4 20.9 24.5 22.3 26.1 21.0 24.1 43 (13.9) 22 (17.1) 10 >24.4 >24.5 >26.1 >24.1 31 (10.0) 18 (14.0) 5 1 Normative data for triceps skinfold thickness are from the National Center for Health Statistics (36). Normative data for BMI are pooled data (37). Chi-square test for normative percents, P < 0.001 for all observed data.

TABLE 5 Regression (R 2 ) models for estimating percentage body fat from anthropometric variables 1 BODY COMPOSITION IN CHILDREN 693 No. of Mean variables square Coefficient of variables in equations included R 2 error Intercept Ankle Elbow Hip Pelvic Shoulder SI SS TH TRI BI CA Calf Hip Waist 1 0.85 15.36 10.44 1.58 1 0.81 19.71 6.77 1.26 1 0.77 23.91 12.46 0.94 1 0.77 24.21 13.26 1.17 1 0.77 24.23 10.00 1.09 1 0.76 24.77 7.37 0.79 1 0.70 31.58 39.18 0.83 1 0.64 37.64 20.83 0.73 1 0.58 44.11 41.87 2.28 1 0.35 68.32 26.68 2.15 1 0.34 69.67 37.19 9.78 1 0.26 77.85 30.36 2.57 1 0.09 96.15 3.71 0.95 1 0.00 105.15 27.17 0.08 2 0.89 11.93 9.02 1.09 0.42 2 0.88 12.27 8.09 0.29 1.15 2 0.88 12.38 10.07 0.35 1.10 2 0.88 12.50 8.06 0.52 1.01 2 0.88 12.81 10.40 0.42 1.12 2 0.87 13.52 4.88 1.24 0.24 2 0.87 13.87 6.31 0.44 0.62 2 0.86 14.53 3.23 1.36 0.15 2 0.86 14.57 0.48 1.39 0.43 2 0.86 15.10 4.51 0.27 1.50 2 0.86 15.18 8.17 0.53 0.79 2 0.86 15.21 5.29 0.17 1.55 2 0.86 15.31 6.71 0.79 0.49 2 0.86 15.37 13.71 0.89 0.33 3 0.91 9.83 7.26 0.25 0.77 0.36 3 0.90 10.73 9.11 0.25 0.87 0.32 3 0.90 10.79 9.26 0.30 0.85 0.34 3 0.90 10.93 7.80 0.35 0.83 0.32 3 0.90 10.97 8.51 0.31 0.24 0.85 3 0.90 11.01 8.46 0.25 0.21 0.89 3 0.90 11.04 1.89 0.91 0.37 0.17 3 0.89 11.25 8.50 0.25 0.37 0.83 3 0.89 11.33 7.21 0.21 0.35 0.87 3 0.89 11.38 8.60 0.30 0.40 0.81 3 0.89 11.45 3.57 0.46 0.82 0.18 3 0.89 11.56 2.21 0.22 1.04 0.43 3 0.89 11.64 7.33 0.39 0.32 0.45 3 0.89 11.64 1.36 0.24 0.97 0.15 1 All variables were used 1 at a time, then 2 at a time, 3 at a time, etc, with the best-fit model at the top for each variable or group of variables. Ankle, elbow, hip, pelvic, and shoulder are breadths; SI, SS, TH, TRI, BI, and CA are suprailiac, subscapular, thigh, triceps, biceps, and calf skinfold thicknesses; and calf, hip, and waist are circumferences. Alternate lines are bold for easier reading. Figure 2, the relations between percentage of body fat measured by DXA and triceps and biceps skinfold thicknesses were linear. The predictive power of the single skinfold thicknesses was considerably lower in the leaner children (R 2 = 0.17 0.44) than in the fatter ones (R 2 = 0.60 0.71). Use of 2 skinfold thicknesses modestly improved the overall estimate (R 2 = 0.89 for calf and biceps skinfold thicknesses), but again there was a big difference in predictive power between the leaner (R 2 = 0.48) and fatter (R 2 = 0.80) children. Adding a third or fourth skinfold thickness further improved the overall relation somewhat (R 2 = 0.90 0.91), but had only a modest effect in the leaner children (R 2 = 0.48 0.54). As shown in Figure 2, when the skinfold thicknesses were summed (38), the relation with DXA became curvilinear. Hydrodensitometry The density of the children in the selected sample is shown in Table 2. The only significant difference by subgroups was between the leaner and fatter children (1060 ± 2 g/l in the former compared with 1027 ± 2 g/l in the latter). Next, we evaluated percentage fat as measured by DXA against 5 equations for calculating body fat derived primarily from densitometry (Table 1) (5 9). Four of these 5 formulas used body density calculated by hydrodensitometry alone to calculate body fat, whereas 1 also

694 BRAY ET AL TABLE 6 Prediction formulas for percentage body fat from skinfold-thickness measurements in all children and in each body fat group 1 MSE R 2 Percentage fat based on 1 skinfold thickness: All children = 10.44 + 1.58 Biceps 15.36 0.85 = 6.77 + 1.26 Triceps 19.71 0.81 = 12.46 + 0.94 Suprailiac 23.91 0.77 = 13.26 + 1.17 Subscapular 24.21 0.77 = 10.00 + 1.09 Calf 24.23 0.77 Leaner children = 9.63 + 0.80 Triceps 7.27 0.44 = 10.62 + 0.75 Calf 7.63 0.41 = 11.75 + 1.10 Biceps 8.60 0.34 = 10.86 + 1.18 Subscapular 10.68 0.17 Fatter children = 19.18 + 1.09 Biceps 11.92 0.71 = 23.80 + 0.69 Subscapular 14.13 0.65 = 21.90 + 0.64 Calf 15.40 0.62 = 18.30 + 0.81 Triceps 16.19 0.60 Percentage fat based on 2 skinfold thicknesses: All children = 9.02 + 1.09 Biceps + 0.42 Calf 11.93 0.89 = 6.71 + 0.79 Triceps + 0.49 Calf 15.31 0.86 = 8.15 + 0.73 Triceps + 0.38 Subscapular 7.20 0.45 Leaner children = 9.06 + 0.50 Triceps + 0.38 Calf 6.79 0.48 = 9.57 + 0.57 Biceps + 0.54 Calf 7.0 0.47 Fatter children = 17.42 + 0.73 Biceps + 0.34 Calf 8.23 0.80 = 17.24 + 0.46 Triceps + 0.40 Calf 11.19 0.73 = 19.04 + 0.42 Triceps + 0.44 Subscapular 11.06 0.73 Percentage fat based on 3 skinfold thicknesses: All children = 7.26 + 0.76 Biceps + 0.36 Calf + 0.24 Thigh 9.83 0.91 Leaner children = 8.30 + 0.52 Biceps + 0.39 Calf + 0.19 Thigh 6.43 0.52 Fatter children = 17.66 + 0.27 Subscapular + 0.52 Biceps 6.59 0.84 + 0.26 Calf Percentage fat based on 4 skinfold thicknesses: All children = 7.66 + 0.22 Subscapular + 0.21 Thigh 9.29 0.91 + 0.64 Biceps + 0.31 Calf Leaner children = 7.87 + 0.15 Thigh + 0.28 Triceps 6.29 0.54 + 0.39 Biceps + 0.26 Calf Fatter children = 16.48 + 0.24 Subscapular + 0.08 Thigh 6.45 0.85 + 0.46 Biceps +0.26 Calf Percentage fat based on skinfold-thickness measurements of triceps, biceps, subscapula, and suprailium: All children = 8.71 + 0.19 Subscapular + 0.76 Biceps 10.96 0.90 + 0.18 Suprailiac +0.33 Triceps Leaner children = 8.21 + 0.12 Subscapular + 0.41 Biceps 7.05 0.48 + 0.11 Suprailiac + 0.57 Triceps Fatter children = 17.47 + 0.28 Subscapular + 0.57 Biceps 8.26 0.81 + 0.04 Suprailiac + 0.17 Triceps 1 Percentage body fat as measured by dual-energy X-ray absorptiometry was used as the criterion variable; the best models using 1, 2, 3, or 4 skinfold-thicknesses were evaluated for all children and for the fatter and leaner children. MSE, mean square error. included age (8). Some of the formulas were developed specifically for use in children (7 9). Overall, the formulas provided comparably good estimates for boys and girls (R 2 = 0.81 0.92). When the fatter and leaner children were examined, however, the R 2 worsened markedly (Figure 3; R 2 = 0.51 0.66). As shown in Table 7, 3 of the densitometry formulas provided reasonably good estimates of overall body fat compared with DXA (5, 6, 9) and 2 others did not (7, 8). However, these overall estimates may cover up some important differences in the subgroups. The estimate of body fat for boys and girls as a group was almost identical with all 5 formulas, although 3 were better overall. The formulas also provided similar estimates of the body fat of the white and African American children (Figure 4). The slope for the African American children differed slightly from that for the white children but was similar for all 5 formulas. For any given density, the white children had slightly more fat than did the African American children, except at the lower end of the curves. The problem with skinfold-thickness formulas based on density can be seen from the substantially different estimates of fat for the fatter and leaner children (Figure 3). In the area of 15 30% body fat, the formulas derived from density consistently provided higher estimates of body fat in the fatter children than in the leaner ones (with use of percentage of body fat by DXA as the standard). If these formulas provided consistently reliable estimates of body fat, the slope of the regression line relative to DXA would be 1 and the intercept would be 0. We used regression analysis of percentage of body fat measured by DXA to test this hypothesis. The simultaneous F test rejected the hypothesis that the slope was 1 and the intercept was 0 for all 5 formulas (5 9). The slopes of the Brozek et al (5) and Schutte et al (9) formulas, however, were not significantly different from 1 in the total population and in each of the subgroups. Only the Schutte et al formula had an intercept that was not significantly different from 0. However, the Lohman et al (7) formula and the Weststrate and Deurenberg (8) formula resulted in rejection of the null hypothesis. Among the subgroups, the Brozek et al and Schutte et al formulas were also similar in that neither the intercept nor the slope was significantly different from 0 and 1, respectively. The intercept and slope for the Siri (6) formula were also not significant different from 0 and 1 for the white girls and the African American boys, but for the other 2 subgroups the values were either borderline or significantly different. The tests of whether the slope of the relation between percentage of body fat measured by DXA and the estimate from density for the racial groups was different were as follows: Weststrate and Deurenberg, P = 0.049; Lohman et al, P = 0.078; Brozek et al, P = 0.078; Siri, P = 0.078; and Schutte et al, P = 0.28. Thus, only the Weststrate and Deurenberg formula produced significantly different slopes for the 2 racial groups. Total body water by isotope dilution Isotope dilution with H 18 2 O significantly underestimated body fat when 73.2% hydration (the adult value) was used to estimate lean body mass (Table 2 and Table 8). When the average hydration value of 78% for these children was used to calculate body fat, the estimates were good but tended to overestimate fat in the leaner group. The hydration status of the FFM in the fatter children was significantly higher than that in the leaner children (79.2% and 76.7%, respectively). This relation held for both the African Americans (79.8% and 77.1%) and the whites (78.7% and 76.2%) and for both the boys (79.8% and 77.1%) and the girls (78.7% and 76.3%). Percentage of body fat

BODY COMPOSITION IN CHILDREN 695 FIGURE 2. Percentage of body fat as measured by dual-energy X-ray absorptiometry (DXA fat) regressed on 5 different combinations of skinfold thicknesses. The sum of 4 skinfold thicknesses is from the Durnin and Womersley (38) equation. measured by DXA and percentage of body fat calculated by isotope dilution were essentially identical in the sex and race subgroups (Figure 5). Because protein and minerals are the most dense components in the fat-free body, and water is less dense, the density of the FFM would be expected to decrease with increasing body fatness (41). Bioelectrical impedance analysis BIA provides an estimate of body water from which fat can be calculated. A formula developed in children by Deurenberg et al (23) gave estimates of body fat similar to those measured in our subjects with use of DXA (Table 8). A formula based on modeling the results from many frequencies that was developed in adults underestimated body fat in all groups. A BIA formula for determining lean body mass that was validated against measurement of body water and density was published by Houtkooper et al (26, 39). Their formula [TBW = 0.61 (height 2 /R) + 0.25 (body weight) + 1.31, where R is resistance] provided R 2 estimates between 0.81 and 0.95 for all subgroups of our population (sex, race, and fat groups) when compared with FFM by DXA. However, the percentage of body fat obtained with this formula was lower than that measured with use of DXA in all groups (Table 8). A widely used formula for calculating fat from BIA was published by Kushner et al (20, 40). This formula, developed in adults, overestimated body fat in all groups of our children. With use of our data, the best fit formula for TBW with use of BIA is TBW = 0.40 (height 2 /R) + 0.148 (weight) + 3.32 (1) for which R 2 = 0.86. We compared the estimates of body water derived with use of this equation with our estimates of body water with isotope dilution. The relation was good, with an R 2 of 0.83. This equation was also used to estimate body water for the fatter and leaner children (R 2 = 0.83 and 0.73, respectively). TBW from our study was plotted against the data obtained with the Kushner et al (20, 40) formula for the 4 subgroups in our study (data not shown). The slope for each of the subgroups was different, indicating the difficulty of using a single formula to calculate TBW in boys and girls of different races. The overall R 2 for the relation of these 2 variables was 0.82. For the subgroups it ranged from 0.81 for the white boys to 0.90 for the white girls. When the slope and intercept were evaluated against the null hypothesis that the slope would be 1 and the intercept 0, the intercept of the equation of Kushner et al was significantly different from 0 but the slope was not significantly different from 1. Multicompartment models A multicompartment model for estimating body fat can be developed by using underwater weighing (body density or volume), TBW by isotope dilution, bone mineral content from DXA, and body weight. Using these measurements from the children in the present study, we obtained the following formula: Percentage of body fat = [2.115 0.436 exp(body density) 0.835 (TBW/body wt) 4.509 (BMC/body wt)] 100 (2) where TBW is by isotope dilution and BMC is bone mineral content measured by DXA in kg. The overall R 2 for this model was 0.93. An equally good fit was obtained by using the following formula (although the previous formula gave slighter better correlations in the leaner group): Percentage of body fat = [ 0.423 + 1.353/density 0.830 (TBW/body wt) 4.5585 (BMC/body wt)] 100 (3) for which R 2 = 0.93.

696 BRAY ET AL FIGURE 3. Percentage of body fat as measured by dual-energy X-ray absorptiometry (DXA fat) in the upper one-half ( ; fatter) and lower one-half (*; leaner) of the fat distribution regressed on percentage of body fat from densitometry calculated by using the formulas of Siri (6), Brozek et al (5), Schutte et al (9), Lohman et al (7), and Weststrate and Deurenberg (8). Next we compared this 4-compartment model with other 3- and 4-compartment models (28, 29, 42, 43), with the skinfoldthickness model of Slaughter et al (11), with the density model of Siri (6, 44), with the BIA model of Kushner et al (20, 40) (using the value of fat from isotope dilution), and with BMI for the fatter and leaner subgroups (Table 9). For the fatter children, all but 2 of the methods of estimating body fat gave R 2 values between 0.66 and 0.84 (6, 21, 22, 44), but only 5 fell within the 95% CI of percentage fat measured by DXA. The value with use of isotope dilution was closest to the value obtained with use of DXA. For the leaner children, the multicompartment model, the Schaefer et al model (24), the Siri model (44), and the Goran et al models (10) were within the 95% CI of percentage fat measured by DXA. The R 2 values, particularly for BMI, were lower in the leaner children. Our multicompartment model, the model of Friedl et al (28), the model of Slaughter et al (11), and the model of Goran et al (10) had the highest R 2 values (0.53 0.67) in the leaner children. DISCUSSION This study is one of the largest cross-sectional studies to use multiple methods of estimating body fat in boys and girls. Our sample of 10 12-y-old boys and girls in Tanner stages 1 and 2 included equal numbers of African Americans and whites. We chose DXA as the criterion method because it entails the use of a single instrument and gives a 3-compartment model that has been widely validated as a reliable estimate of body fat that is relatively independent of hydration (30 34). The DXA technique has been validated by chemical analysis (in pigs), by hydrodensitometry, by anthropometry, by BIA, and by measurement of total body potassium (30 33). Our study population was drawn from the local public school system to represent the upper and lower halves of the triceps plus subscapular skinfold-thickness distribution. The bimodality of the body fat distribution was confirmed in all of the measurements. Lean body mass, on the other hand, was normally distributed. By using the data of Frisancho (36) for triceps skinfold-thickness percentiles and that of Rossner (37) for BMI percentiles, we confirmed that we oversampled the upper one-half of the BMI and triceps skinfold-thickness distributions of American children. According to the standards published by Must et al (45), a smaller percentage of the children in our study had triceps skinfold thicknesses or BMIs above the 95th percentile than expected, and our population had significantly fewer children with BMI or triceps skinfoldthickness values above the 85th percentile. Thus, the overweight part of our sample was largely from the 50th to 85th percentiles. Our sample fits well within the parameters of other studies of body fat in children. The girls in our sample had amounts of body fat similar to those estimated by Forbes (46) (26.1% in this study compared with 24.9% in the study by Forbes) by use of total body potassium counting, but the boys in the present study were fatter (27% compared with 14.6% in the study by Forbes). Overall, our population was almost identical to the children studied by Goulding et al (47). Our children were 1 y younger and were shorter and lighter but were fatter than the children studied by Schaefer et al (24). The triceps, biceps, subscapular, and suprailiac skinfold thicknesses of our population were almost 50% greater than those of the children studied by Schaefer et al (24). Our population was close in age to the 10 14-y-old groups studied at the US Department of Agriculture facility in Houston (21, 22), with amounts of body fat that were almost the same. The children studied by Houtkooper et al (26, 39) were heavier, whereas those studied by Hewit et al (27) and Slaughter et al (11) were lighter.

BODY COMPOSITION IN CHILDREN 697 TABLE 7 Estimates of percentage body fat in the present study with use of skinfold-thickness equations published by several authors 1 Sex Race Body fat Author All children Girls Boys African Americans Whites Leaner Fatter Siri (6) 24.60 ± 0.88 [127] 2 24.07 ± 1.21 [63] 25.13 ± 1.28 [64] 24.31 ± 1.35 [62] 24.88 ± 1.15 [65] 17.22 ± 0.78 [65] 32.35 ± 0.83 [62] Lohman et al (7) 19.16 ± 0.94 [127] 2 18.59 ± 1.30 [63] 19.73 ± 1.37 [64] 18.85 ± 1.44 [62] 19.46 ± 1.23 [65] 11.26 ± 0.84 [65] 27.45 ± 0.89 [62] Weststrate and 19.47 ± 0.95 [123] 2 18.92 ± 1.29 [61] 20.01 ± 1.39 [62] 19.18 ± 1.48 [59] 19.74 ± 1.22 [64] 11.92 ± 0.83 [65] 27.94 ± 0.90 [58] Deurenberg (8) Schutte et al (9) 24.05 ± 0.79 [127] 2 23.57 ± 1.09 [63] 24.52 ± 1.15 [64] 23.86 ± 1.18 [62] 24.23 ± 1.06 [65] 17.42 ± 0.70 [65] 31.00 ± 0.75 [62] Brozek et al (5) 23.97 ± 0.81 [127] 2 23.48 ± 1.12 [63] 24.46 ± 1.19 [64] 23.70 ± 1.25 [62] 24.23 ± 1.06 [65] 17.15 ± 0.72 [65] 31.12 ± 0.77 [62] DXA 26.63 ± 0.90 [129] 26.13 ± 1.26 [64] 27.12 ± 1.29 [65] 25.63 ± 1.31 [64] 27.61 ± 1.23 [65] 17.89 ± 0.44 [65] 35.50 ± 0.79 [64] 1 x ± SE; n in brackets. Body density was the criterion variable from which Siri (6), Lohman et al (7), Schutte et al (9), and Brozek et al (5) derived their models for estimating fat from skinfold thicknesses. The model of Weststrate and Deurenberg (8) added age. For Schutte et al, a race-specific formula was used. DXA, dual-energy X-ray absorptiometry. 2 Significantly different from DXA, P < 0.001. Comparison of anthropometric measures in our study with those of other large surveys also provides a gauge of the degree to which the present sample was representative. The Bogalusa Heart Study, a long-standing total community survey of cardiovascular disease risk factors in children and young adults living 90 miles from Baton Rouge, LA, provides a valuable frame of reference (48, 49). Overall, our sample from Baton Rouge was slightly taller (145 compared with 139 cm) and heavier (40.8 compared with 37.6 kg) than the Bogalusa one (48). This was especially true for the boys in our sample, who were 6 cm taller and 3 kg heavier than their Bogalusa counterparts. In the girls, the differences were much smaller: 1 cm in height and 3.2 kg in weight. Correspondingly, BMI values were higher in the Baton Rouge group than in the Bogalusa group: 20.0 compared with 16.6 for the boys and 18.4 compared with 17.3 for the girls. In most cases, skinfold-thickness measurements were higher in the Bogalusa sample. At the triceps site, the Baton Rouge boys averaged 15.2 mm compared with 11.4 mm for the Bogalusa boys. For the girls, the situation was reversed: the Bogalusa girls had an average value of 13.9 mm and the Baton Rouge girls averaged 12.2 mm. At the subscapular site, the Bogalusa means exceeded the Baton Rouge means by 1.2 mm in boys and 4.1 mm in girls. Our data on African Americans and whites were similar to those found for 10-y-olds in the National Heart, Lung, and Blood Institute Growth and Health Study (50). We conclude that, when DXA is used as the reference method, body fat can be estimated from skinfold thicknesses more reliably in fatter children than in leaner ones. A single, experienced FIGURE 4. Percentage of body fat as measured by dual-energy X-ray absorptiometry (DXA fat) in whites ( )and African Americans (*) regressed on percentage of body fat from densitometry calculated by using the formulas of Siri (6), Brozek et al (5), Schutte et al (9), Lohman et al (7), and Weststrate and Deurenberg (8).