Background: Although the BMI is widely used as a measure of adiposity, it is a measure of excess weight, and its association with body fatness may differ across racial or ethnic groups.
Objective: To determine whether differences in body fatness between white, black, Hispanic, and Asian children vary by BMI-for-age, and whether the accuracy of overweight (BMI-for-age ≥ Centers for Disease Control and Prevention (CDC) 95th percentile) as an indicator of excess adiposity varies by race/ethnicity.
Methods and Procedures: Total body dual-energy X-ray absorptiometry (DXA) provided estimates of %body fat among 1,104 healthy 5- to 18-year-olds.
Results: At equivalent levels of BMI-for-age, black children had less (mean, 3%) body fatness than white children, and Asian girls had slightly higher (1%) levels of %body fat than white girls. These differences, however, varied by BMI-for-age, with the excess body fatness of Asians evident only among relatively thin children. The ability of overweight to identify girls with excess body fatness also varied by race/ethnicity. Of the girls with excess body fatness, 89% (24/27) of black girls, but only 50% (8/16) of Asian girls, were overweight (P = 0.03). Furthermore, the proportion of overweight girls who had excess body fatness varied from 62% (8/13) among Asians to 100% (13/13) among whites.
Discussion: There are racial or ethnic differences in body fatness among children, but these differences vary by BMI-for-age. If race/ethnicity differences in body fatness among adults also vary by BMI, it may be difficult to develop race-specific BMI cut points to identify equivalent levels of %body fat.
The prevalence of overweight among children and adolescents in the United States has increased greatly since the 1960s, and 18% of schoolchildren are now overweight (1) based on a BMI (kg/m2) level >95th percentile of the Centers for Disease Control and Prevention (CDC) growth charts (2). There are, however, substantial racial or ethnic differences in overweight. As compared with white children in the United States, the relative prevalence of overweight is ∼50–75% higher among black and Mexican-American children (1,3) and ∼30% lower among Asian children (4).
Although BMI is widely used as a surrogate measure of adiposity, it is a measure of excess weight relative to height, rather than excess body fat, and the changes that occur in body composition during growth and development further complicate its interpretation (5). Several studies have also found that the relation of BMI to body fatness, as determined by dual-energy X-ray absorptiometry (DXA) and densitometry, differs across racial or ethnic groups. In general, Asians have more body fat (6,7,8,9), and blacks have less body fat (10), than do whites; furthermore, these racial or ethnic contrasts are seen even when comparisons are made at equivalent BMI levels (7,8,9). These differences raise the possibility that at a given BMI (e.g., 25 or 30 kg/m2), the risk for obesity-related diseases may vary by race/ethnicity, with Asians having a higher risk (more body fatness) and blacks a lower risk (less body fatness) than whites. To account for these differences in body fatness, it has been suggested that BMI cut points for overweight and obesity be lowered among Asians (11,12) and increased among blacks (13).
Most studies, however, that have examined the relation of BMI to body fatness across racial or ethnic groups have been relatively small and have not considered the possibility that the racial or ethnic differences in body fatness may vary according to the BMI level. For example, differences in the body fatness of blacks and Asians relative to whites may be most pronounced at relatively low BMI levels rather than being constant across the entire BMI distribution. The goals of the current study are to (i) quantify the racial or ethnic differences in body fatness between white, black, Hispanic, and Asian 5- to 18-year-olds, (ii) determine whether these differences vary by BMI-for-age, and (iii) determine whether the accuracy of BMI-for-age as an indicator of excess adiposity varies by race/ethnicity.
Methods and procedures
Between 1995 and 2000, 1,196 healthy volunteers (ages, 5–18 years) were recruited in the New York City area for the Pediatric Rosetta Body Composition Project through local newspaper notices, announcements at schools and activity centers, and by word of mouth. All procedures were in accordance with the ethical standards of St Luke's-Roosevelt Hospital (New York, NY), and approval was obtained from the Institutional Review Board. Consent was obtained from each volunteer's parent or guardian and, when appropriate, from each volunteer.
A questionnaire was used for establishing race and ethnicity; the criterion was a consistent background (white, black, Hispanic, or Asian) for all four grandparents. Subjects with grandparents of different racial or ethnic groups (n = 92) were not included in these analyses. Most Asian participants were of Chinese or Korean background, and most Hispanic participants were of Dominican or Puerto Rican origin. Some analyses adjusted for differences in Tanner (pubertal) stage, which was assessed by the study physician or nurse in younger subjects, and by self-assessment in subjects over 11 years of age (14,15).
Body weight was measured to the nearest 0.1 kg using an electronic scale (Weight Tronix, New York, NY) and height to the nearest 0.1 cm using a stadiometer (Holtain, Crosswell, UK) with subjects wearing a hospital gown and foam slippers. BMI (kg/m2) was calculated as a measure of relative weight, and BMI-for-age z-scores were calculated from the 2000 CDC Growth Charts (16,17) to account for sex and age differences. These growth charts express BMI levels of children in this study relative to US children of the same sex and age between 1963 and 1980 (6- to 18-year-olds) or between 1963 and 1994 (5-year-olds). (These growth curves were constructed separately for each sex, but combined all racial or ethnic groups.) Overweight was defined as a BMI-for-age z-score ≥1.645 (corresponding to the 95th percentile of a normal distribution) (2,18).
Whole body DXA scans were performed using Lunar models DPX (with pediatric software version 3.8G; Lunar, Madison, WI) and DPX-L (with pediatric software 1.5G) (19). The scan mode was chosen according to the weight guidelines provided by the manufacturer, and each scan provided an estimate of fat mass and fat-free mass. Among adults who had repeated scans, the coefficient of variation for %body fat was 3.3% and the intraclass correlation coefficient was 0.994 (20).
An anthropomorphic spine phantom made of calcium hydroxyapatite embedded in a Lucite block was scanned with both DXA instruments for quality control each morning, and immediately before and after all maintenance visits. Bottles (8 l) of methanol and water, simulating fat and fat-free tissues, respectively, were scanned monthly as quality control markers.
Within each sex, we compared mean levels of %body fat across racial or ethnic groups using regression analyses to assess whether the observed differences were independent of age and BMI-for-age. We used restricted cubic splines in R (21) to model continuous variables; these models do not necessarily have peaks and valleys, and the fit in one region of the data does not influence the fit in other regions (22). The distributions of residuals from these least-squares models were not markedly nonnormal; furthermore, coefficients and P values were similar to those from several models fit with robust or resistant methods (23).
To determine whether racial or ethnic contrasts in %body fat differed by BMI-for-age, we included a race/ethnicity × BMI-for-age term in these models, and we show predicted levels of %body fat by BMI-for-age for each racial or ethnic group. Because group differences can be difficult to interpret when slopes (%body fat vs. BMI-for-age) change, we also present these differences relative to levels among white children. At each BMI-for-age, the predicted level of %body fat among white children was subtracted from the predicted level in other racial or ethnic groups; the differences were then plotted.
We also assessed the ability of overweight (BMI-for-age ≥ CDC 95th percentile) to identify children with excess body fatness correctly. There is no widely accepted definition of excess adiposity in childhood and adolescence (24), but the prevalence of overweight in the current sample was 16% among boys and 15% among girls. We therefore defined “excess body fatness” to be a %body fat level at or above the 84th (boys) or 85th (girls) percentile at a given age. These cut points did not vary by race/ethnicity, and were based on the distribution of %body fat in the current sample. Levels of %body fat vary substantially by age (24), and we used quantile regression (25,26) to estimate these cut points (25,26). The resulting cut points varied by each value of age (which was measured to the nearest day), with the 84th percentile for %body fat among boys varying from 18% (at age 6.0 years) to 36% (11.0 years) to 23% (17.0 years).
We calculated positive predictive values (the proportion of overweight children who had excess body fatness) and sensitivities (the proportion of children with excess body fatness who were overweight) within each racial or ethnic group. We also calculated the area under the receiver operating characteristic curve as an overall measure of the screening ability of BMI-for-age. This area represents the probability that the BMI-for-age of a randomly selected child with excess body fatness would be higher than that of a child without excess body fatness.
Table 1 shows mean levels of various characteristics within each racial or ethnic group. Mean ages were similar across the four racial or ethnic groups, but there were statistically significant differences in levels of Tanner stage, weight, height, BMI, BMI-for-age, and %body fat (boys only). Among both boys and girls, black and Hispanic children had the highest BMI-for-age and the highest prevalence of overweight. However, despite the high BMI-for-age of black children (mean z-scores of +0.6 to +0.7), black boys had the lowest mean level of %body fat (17%), and the mean body fatness of black girls was similar to that among white and Asian girls (25–27%). In contrast, Hispanic children had the highest mean BMI-for-age and %body fat.
Table 1. . Descriptive characteristics by race and sex
We then examined racial/ethnic differences in body fatness after accounting for age and BMI-for-age (Table 2). Levels (mean ± s.d.) of %body fat are shown for white children, and mean levels in other groups are expressed relative to those among white children. Whereas racial/ethnic differences in %body fat were only slightly altered by adjustment for age, they were strongly influenced by adjustment for BMI-for-age. At similar levels of age and BMI-for-age (3rd row), white boys had 3.5% more body fat than black boys and 0.5% more body fatness than Hispanic boys (P < 0.001 for difference across the four groups). The mean adjusted body fatness of white girls was higher than that of black girls (2.6%) and Hispanic girls (0.5%), but 0.7% lower than that of Asian girls (P < 0.001). Further adjustment for height and Tanner stage only slightly altered these results (data not shown), and the combination of racial/ethnic group, age and BMI-for-age accounted for > 80% of the variability in body fatness in each sex.
Table 2. . Mean levels of %body fat by sex and race
Stratified analyses, however, suggested that these racial/ethnic differences varied by BMI-for-age (bottom of Table 2). For example, the mean body fatness of Asian boys was ∼1–2% higher than that of white boys between the 50th and 95th percentiles of BMI-for-age, but the mean %body fat of overweight Asian boys was 2.2% lower than that of overweight white boys. Fairly similar results were seen among girls, with relatively thin (BMI-for-age < 85th percentile) Asians having more (1.2–1.8%) body fat than whites, but overweight Asian girls having a mean body fatness that was 3% lower than that of white girls. As assessed in regression analyses, levels of %body fat did not significantly differ by race/ethnicity among overweight children or among girls with a BMI-for-age between the 85th and 94th percentiles. There were, however, fewer (n = 170) overweight children than in the other three groups (185–413).
Figure 1 shows predicted levels of %body fat from sex-specific regression models that included age, race/ethnicity, BMI-for-age, and the interaction between BMI-for-age and race/ethnicity. The relation of BMI-for-age to %body fat differed by race/ethnicity among both boys (P = 0.02) and girls (P = 0.01). In agreement with the stratified results, the relatively high body fatness of Asian girls (Figure 1b) was most evident among thinner (BMI-for-age z-score <1.0) children, while among overweight girls, levels of body fatness were highest among whites.
Because differences across groups can be difficult to assess when slopes change, these racial/ethnic contrasts in %body fat are expressed relative to white children (Figure 2). As compared with white children (represented by the horizontal line at y = 0), black children showed the largest differences in %body fat, with levels that were 2–4% lower across the BMI-for-age distribution. Patterns among Asian and Hispanic children were more complex, but at the 50th percentile of BMI-for-age, Asian boys had a higher %body fat level than did white (by ∼1%), Hispanic (2%), and black (5%) boys. Fairly similar differences were also seen at the 50th percentile among girls, with Asians having a higher level of %body fat than other groups. In contrast, at the 95th percentile of BMI-for-age, white, Asian and Hispanic boys had similar levels of %body fat. At the 95th percentile among girls, whites had the highest level of body fatness—∼1.5% more than Hispanics and Asians, and ∼2.5% more than blacks. Differences in body fatness among overweight girls continued to increase with increasing levels of BMI-for-age, so that at a z-score of 1.96 (the 97.5 percentile), the mean body fatness of white girls was ∼3% higher than among other girls. Additional age-stratified analyses (data not shown) suggested that the white excess in body fatness among overweight children was more pronounced among 12- to 18-year-olds than among 6- to 11-year-olds.
We then assessed whether the ability of BMI-for-age to identify children with excess fatness correctly varied by race/ethnicity (Table 3). As assessed by the area under the curves, BMI-for-age was a good predictor of excess body fatness, with values ranging from 0.88 (white boys) to 0.99 (white girls). Among boys, neither the positive predictive values (probability that an overweight child actually has excess body fatness) nor the sensitivities (probability that a child with excess body fat is overweight) varied significantly by race/ethnicity; values ranged from 65 to 83%. Among girls, however, there tended to be differences in the positive predictive values (P = 0.12) and sensitivities (P = 0.03) across racial/ethnic groups. The positive predictive value was highest among whites, with all 13 overweight girls having excess body fat, and was the lowest (62%) among Asians. (As shown in Figure 2, overweight white girls had more body fatness than did other overweight girls). In contrast, the sensitivity was highest (89%) among black girls and lowest (50%) among Asian girls; only 8 of 16 Asian girls with excess body fatness were overweight.
Table 3. . Classification of excess body fatness by BMI-for-age
Our results indicate that at similar levels of age and BMI-for-age, the body fatness of children and adolescents can differ by up to 5% across racial/ethnic groups. Overall, we found that black children had substantially less (∼3%) body fatness than white children, and that Asian girls tended to have slightly higher (1%) levels of %body fat than white girls. These racial/ethnicity differences, however, varied by BMI-for-age: the excess body fatness of Asian girls and the decreased body fatness of black boys were most evident among relatively thin children. In contrast, the mean body fatness of overweight white children was 2–3% higher than that of overweight Asian children, black children, and Hispanic girls. Although a high BMI-for-age was a good indicator of excess body fatness, the ability of the CDC 95th percentile to identify girls with excess body fatness varied by race/ethnicity. Of the girls with excess body fatness, 89% of blacks, but only 50% (8/16) of Asians, were overweight.
Several studies have found differences in body composition between whites and blacks, with black adults and children having a higher fat-free body density due to greater mineral bone density and body protein content (13). Similar differences have also been noted among children, with black children having less body fat (as assessed by densitometry) (10) and greater bone mineral content (DXA) (27,28,29) than white children. Although not all of these studies have controlled for differences in weight and height, adjustment for BMI-for-age in the current study substantially increased the white excess in body fatness. Both genetic and hormonal explanations for these differences in body composition have been suggested (13,30).
Fewer studies of body composition have been conducted among Hispanic children, but a study in Los Angeles found that they had more body fat (measured by DXA) than expected based on an equation (including weight and triceps skinfold thickness) developed in a biracial (1/3 black) sample (31). Furthermore, at similar weights and heights, Mexican-American children have been found to have more fat mass than white children (27,28). However, it is possible that the relation of BMI-for-age to body fatness differs across Hispanic subgroups, and that studies conducted among Mexican-American children may not be relevant to Hispanic children in this study who were primarily of Dominican or Puerto Rican origin. Although we found little difference in the mean body fatness of Hispanic and white children after controlling for BMI-for-age, the body fatness of overweight Hispanic girls was almost 3% lower than that of overweight white girls.
Despite their lower BMIs, Asian adults in the United States generally have more body fat (as determined by skinfold thicknesses (32) and DXA (6)) than white adults. More recent studies have found that at the same BMI, adults from China, Malaysia, India, Japan, and Indonesia (7,33,34), as well as Korean women, (35) and children from Singapore (9), have more body fat (as determined by DXA or four-compartment models) than do Europeans. However, “Asians” comprise many diverse subpopulations, and it is possible that not all of these groups have increased body fatness. As compared with Europeans, for example, levels of %body fat appear to be similar among children in Beijing (9) and lower among Polynesian adults (36).
Because most studies have reported that the body fatness of Asians is greater than that of whites, it is not surprising that the sensitivity of overweight among Asian children is relatively low. Many “normal-weight” Asian children have high levels of %body fat, and of the Asian girls with excess body fatness, only 50% (vs. 89% of black girls) were overweight. However, the positive predictive value for overweight was also low among Asians: only 67% (22/33, boys) and 62% (8/13, girls) of overweight Asian children had excess body fatness. Although positive predictive values can be influenced by the prevalence of excess body fatness, we do not think these differences account for our findings. The prevalence of excess body fatness among Asian girls (10%) and white girls (13%) were fairly similar in this study.
Our results indicate that the low positive predictive value among Asian children arises because overweight Asian children have less (∼2–3%) body fatness than do overweight white children (Table 2, Figure 2). It is the “crossover” in body fatness, with normal-weight Asian children have more body fatness, but overweight Asian children having less body fatness than white children (at a similar BMI-for-age) that leads to low levels of both sensitivity and positive predictive value among Asian children. Although most investigators have examined only mean (overall) differences in body fatness, some previous results are in agreement with these findings. The excess body fatness of Japanese adults was most evident at lower BMI levels (33), and the decreased body fatness of Polynesian adults was most pronounced at high BMIs (36).
Rather than using BMIs of 25 and 30 kg/m2 for overweight and obesity among adults, it has been suggested (11,12,13) that these cut points should vary by race/ethnicity to account for differences in (i) the relation of BMI to body fatness, (ii) susceptibility to obesity-related diseases, and (iii) fat distribution (37). Our data cannot address the latter two possibilities, but if racial/ethnic differences in body fatness among adults also vary by BMI (as we observed among children and adolescents), it may be difficult to derive BMI cut points that result in equivalent levels of body fatness among whites and Asians. Although our results were based on relatively small numbers of overweight children within each race/ethnicity, we found that overweight Asian children had lower levels of %body fat than overweight white children. A lower BMI-for-age cut point among Asian children, such as the CDC 85th percentile, substantially increased the sensitivity, but decreased the positive predictive value from ∼65 to 40% (data not shown). In attempting to derive race-specific BMI cut points, the possibility that mean differences in body fatness may not be applicable to overweight and obese persons should be considered.
There are several limitations to this study. Although the overall sample was large (n = 1,104), there were relatively few overweight children within each race/ethnic group. Furthermore, the sample was not randomly selected and is not representative of the general population; subjects were recruited through newspaper notices, announcements, and word of mouth. However, we have found previously (38) that the BMIs of children in this sample differ only slightly from levels in the National Health and Nutrition Examination Survey. In addition, although DXA estimates of adiposity are highly correlated with those from the criterion four-compartment model, there can be substantial differences in estimates of body fatness (39,40,41,42). However, whereas the use of densitometry and other techniques that assume a constant fat-free mass density can lead to be biased estimates of %body fat among blacks and Asians (43,44,45), the errors associated with DXA do not systematically differ by race/ethnicity (40,41). In addition, there is no widely accepted classification of “excess adiposity” among children, and our internal cut points, based on the distribution of body fatness in our sample, contrasts with the external criteria used to classify overweight (CDC 95th percentile of BMI-for-age). It should also be realized that although BMI-for-age (along with age and race) accounts for much of the variability in body fatness, there can be large errors in the prediction of the %body fat of an individual.
In summary, the relation of BMI-for-age to body fatness among children and adolescents differs by race/ethnicity. These racial/ethnic differences, however, vary substantially by BMI, and we found that at the equivalent levels of BMI-for-age, overweight Asian children had less body fatness than did overweight white children. Although further study is needed, our results suggest that the application of “average” racial/ethnic differences in body fatness across the BMI distribution may not be appropriate. If race/ethnicity differences in body fatness among adults also vary by BMI, it may be difficult to develop race-specific BMI cut points to identify equivalent levels of %body fat.
The authors declared no conflict of interest.
This work was supported in part by DK 37352. D.S.F. was responsible for data analysis and writing of this article. J.W., M.H., and J.C.T. were responsible for subject recruitment, data collection, and quality control. R.N.P., J.W., M.H., and J.C.T. were responsible for study design. R.N.P. was responsible for overall project supervision. Z.M. was involved in data analysis and revision of this article. W.H.D. and M.H. provided advice and were involved in extensive revision of this article.