Racial Differences in the Sums of Skinfolds and Percentage of Body Fat Estimated from Impedance in Black and White Girls, 9 to 19 Years of Age: The National Heart, Lung, and Blood Institute Growth and Health Study
OSB 4, Division of Cardiology, Children's Hospital Medical Center, 3333 Burnet Avenue, Cincinnati, OH 45229. E-mail: firstname.lastname@example.org
Objectives: This National Heart, Lung, and Blood Institute Growth and Health Study report assesses racial differences in fat patterning in black and white girls ages 9 to 19 years, comparing the sum of triceps and subscapular skinfolds (SSFs) and percentage of body fat (%BF) from impedance as two indices of adiposity. It is hypothesized that racial differences in fat patterning manifest during puberty.
Research Methods and Procedures: SSF and %BF were measured annually. Racial differences in SSF and %BF were evaluated by age. Associations between %BF and SSF were evaluated using the Pearson's correlations coefficient. Classification agreement was evaluated using the kappa-statistic. Effects of pubertal stage and race on classification agreement were examined using multivariate models.
Results: White girls had a greater mean %BF at 9 to 12 years of age; black girls had a greater %BF thereafter. Black girls had a greater mean SSF at every age. The correlation coefficient between SSF and %BF was 0.79, and there was good agreement between %BF and SSF in separating high (>85th percentile) from not high (kappa = 0.60 for whites and 0.66 for blacks). SSF associated more with %BF in prepuberty and early puberty than in late puberty.
Discussion: Despite good correlations between %BF and SSF, the two methods indicate different fat patterns in black and white girls.
Adult obesity is associated with increased risk of cardiovascular disease (CVD) morbidity and mortality (1, 2). It is likely that pediatric obesity has similar adverse health consequences that will manifest in adulthood, because the correlations between obesity and CVD risk factors in pediatric populations are significant and similar to those reported in adult populations (3). Moreover, obesity tracks from childhood into early adulthood (4, 5). Thus, evaluating obesity in children and adolescents is important. Ideally, such evaluations should be made using such direct measures as DXA or hydrostatic weighing, but these techniques require sophisticated laboratory settings and are limited because of the required performance of the participants. Consequently, in clinical settings and epidemiological studies, other measures are used, including skinfold thickness measures and bioelectrical impedance.
The National Health and Nutrition Examination Survey II and III used the triceps and subscapular skinfold thickness measures to characterize adiposity (6). There are limits to the application of skinfolds to determine overall adiposity, because skinfolds are site-specific and have limited reproducibility in obese people (7). Bioelectrical impedance has high reliability, is portable, and can be used with anthropometry to obtain estimates of total body water, making possible assessments of fat-free mass (FFM) and fat mass in field studies (8). An important issue is whether such measures yield the same or similar estimates of children's overall adiposity in epidemiological studies, and if there are differences, what these differences imply concerning overall adiposity and fat patterning. The current report compares estimates of percentage of body fat (%BF) derived from impedance-based equations and the sum of triceps and subscapular skinfolds (SSFs) in black and white girls, 9 to 19 years of age, using data from the National Heart, Lung, and Blood Institute Growth and Health Study (NGHS) (9). We hypothesized that different patterns of fat deposition appear during adolescence in black and white girls.
Research Methods and Procedures
The NGHS, its purposes, design, cohort, sampling frames, and study methods have been described in detail (9). Briefly, NGHS is a cohort study of the development of obesity in black and white girls whose ages were 9 and 10 years at enrollment (9). Three clinical centers in Berkeley, California (University of California, Berkeley), Cincinnati, Ohio (University of Cincinnati Medical Center and Children's Hospital Medical Center), and Washington, DC (Westat, Inc.) enrolled a total of 2379 girls (1213 black and 1166 white) in 1987 through 1988. The Berkeley and Cincinnati centers recruited girls from public and parochial schools and Westat recruited girls from a Washington, DC area health maintenance organization. Eligibility was restricted to 9- to 10-year-old girls who declared themselves as being black or white and who lived in racially concordant households (9). Maryland Medical Research Institute serves as the Coordinating Center, and the Project Office is at the National Heart, Lung, and Blood Institute.
Certified personnel (measurer and recorder) made duplicate measures of height, weight, circumferences, and skinfolds (triceps, subscapular, and suprailiac) as part of the anthropometric assessment, completing the first set in the order listed here before starting the second set. A third measurement was made if the first two measures differed by a set amount. The mean of the two closest measures was used in these analyses. Measurements were taken on the right side of the participant. The fold of skin was firmly grasped between the left thumb and forefinger and then raised. The fold was pinched and raised several times to make certain that no musculature was grasped. The skinfold was held firmly with thumb and forefinger and the calipers placed below the thumb and forefinger. With the fold held, the reading was taken and read to the nearest millimeter, with the grip on the caliper released completely. A senior investigator certified the trainer at each site, who, in turn, trained and certified the local personnel. Quality control duplicate measurements were performed on 150 participants. The interobserver error term accounted for <1.0% of the variance of the skinfold measurements (10).
During the same clinic visit, resistance and reactance readings were made once using an RJL bioelectrical impedance analyzer (model BIA-101; RJL System, Detroit, MI). The resistance and reactance measurements were taken from the right hand to the right foot using a tetrapolar placement of electrodes (Resting ECG Electrodes LMP 3-PG; RJL Systems, Clinton Township, MI) (8). The impedance machine was calibrated before each clinic. For these measurements, subjects were supine with arms away from the trunk, the thighs separated, and metal objects worn on limbs or trunk removed, so there was no metal contact. Pubertal maturation was assessed visually by trained female examiners as described previously (10) and scored using four stages (prepuberty, pubertal but premenarchal, postmenarchal ≤1 year, and postmenarchal >1 year).
Statistical Analysis Methods
All analyses were performed using SAS, version 6.12 (SAS Institute Inc., Cary, NC) (11). Summary statistics were calculated for each race. Racial differences in adiposity between black and white girls were evaluated using analysis of covariance adjusting for age, which was defined as age at last birthday. For this report, two estimates of adiposity were used: SSF measurements and %BF derived from race-specific prediction equations for FFM using height, weight, resistance, and reactance. There are two prediction equations developed by Slaughter et al. (12), using the SSFs: a quadratic equation for girls with a sum of two skinfolds <35 mm and a linear equation for girls >35 mm. We did not convert the skinfolds into %BF using the Slaughter equations for this comparison because the number of variables in our prediction equations (height, resistance, reactance, and weight) was greater than the number in the Slaughter equations (triceps and subscapular). We present the mean and SD %BF from the Slaughter equations by race and single year of age in the Appendix. As expected, the pattern of racial differences was similar to that obtained by skinfolds.
The race-specific equations were derived from an ancillary study (13) and cross-validated as follows. The agreement between the measure of FFM by DXA and the estimate of FFM from the prediction equation was evaluated using the method of Bland and Altman (14). The Bland-Altman plots indicated good agreement between the two methods (DXA and the prediction equations) for girls of each race. In addition, an independent sample of girls from a National Institute of Child Health and Human Development longitudinal study was used to cross-validate the corresponding Children's Hospital Medical Center prediction equations, below (13). The pure error was used to evaluate the cross-validation results from the independent sample. Pure errors for the girls in the National Institute of Child Health and Human Development study were 1.2 kg for black girls and 1.5 kg for white girls. The corresponding coefficients of variation were 5% and 6%, respectively (13). These equations were:
where R is resistance and Xc is reactance.
These equations explained 98.9% and 97.0% of the variability in FFM as measured by DXA in white and black girls, respectively. The root mean square errors for the estimates were 1.1 kg in white girls and 1.9 kg in black girls (13). These equations were used to determine each girl's FFM, and the corresponding value for %BF was calculated. Percent BF was calculated as weight minus FFM divided by weight times 100. Age-specific means and SDs of %BF were calculated.
The cross-sectional association between %BF and SSF using data across all visits was assessed using Pearson's correlation coefficients. To investigate the longitudinal association of %BF with SSF, age, pubertal stage, and race, the following general estimating equation (GEE) model was used:
Here, %BFt was the %BF at aget. The interaction terms tested were race–SSF, race–age, race–stage, and race–age–SSF. Global χ2 tests were performed to determine the overall effect of age, pubertal stage, and their interaction terms, all of which were entered into the model as a series of k-1 dummy (0 and 1) variables to represent the kappa levels of the factor; e.g., for the four levels of stage, three dummy variables were used with prepubertal as the reference group. Nonsignificant interaction terms were dropped from the model.
To evaluate differences in the classification of participants for adiposity, participants were separated into two categories using the age-specific 85th percentile level from the NGHS cohort of SSF and%BF. We arbitrarily selected the 85th percentile to ensure sufficient participants in each category to permit statistical comparisons. The classification agreement between SSF and %BF was evaluated using the kappa-statistic. kappa values >0.75 represent excellent agreement, values ranging from 0.40 to 0.75 represent fair to good agreement, and values <0.40 represent poor agreement (15).
The NGHS cohort consists of 2379 girls, 1213 black and 1166 white, seen up to 10 times over 10 years. Of a possible 23,790 usable participant visits, data were obtained on 19,058 (80%). The bioelectrical impedance equations resulted in 69 negative %BF values (0.4%). These impossible values could have been the result of measurement errors at the clinic visit, resulting in the recorded weight being less than actual weight and less than the estimated FFM, or of limits to the fit of the prediction model itself. Compared with girls with positive %BF values, participants with negative %BF were significantly shorter (mean height: 151.3 cm vs. 157.4 cm; p < 0.0001), significantly lighter (mean weight: 42.1 kg vs. 56.3 kg; p < 0.0001), and had a significantly smaller mean SSF (21.4 vs. 31.8; p < 0.0001). These 69 negative %BF values were dropped from the analyses.
Comparisons of Adiposity by Age, within and between Races
Means and SDs for SSF and %BF are presented for each race at each age from 9 to 19 years (Tables 1 and 2). Mean SSF increased for every age from 9 to 15 years of age for both black and white girls. In black girls, mean SSF decreased between ages 14 and 16 years, before continuing to increase; in white girls, it decreased from 15 to 16 years of age, before continuing to increase. Mean %BF increased for every age from 9 to 19 years in black girls but not in white girls. In white girls, the mean %BF changed little from ages 9 to 13 years, and there were minor fluctuations from year to year. At age 13 years, mean %BF was slightly less than at age 9 years (25.5% vs. 26.0%). Mean %BF in white girls increased every year from 14 to 19 years of age. During the 10-year study, mean %BF increased in black girls from 22.2% to 34.2% and in white girls from 26.0% to 30.5% (increases of 54% and 17%, respectively). Mean SSF increased from 24.4 mm to 45.7 mm in black girls and from 23.1 mm to 37.0 mm in white girls (increases of 87.3% and 60.2%, respectively). Black girls had significantly larger SSF than did white girls at every age, but white girls had significantly greater mean %BF than did black girls at ages 9 to 11 years. At age 12 years, the difference was of borderline significance (whites higher, p = 0.086); at age 13 years, the difference was of borderline significance (p = 0.092), but black girls were marginally higher. At every age from 14 to 19 years, black girls had higher %BF. The racial differences in SSF and %BF at the race-specific 85th percentiles indicated that black girls had more body fat than did white girls at every age. Racial differences in SSF and %BF at the 15th percentile were not consistent. White girls had greater SSF at the 15th percentile at ages 9, 10, and 18 years only, with equal SSF at age 16 years. White girls had greater%BF at the 15th percentile at every age but age 19 years, where %BF was equal.
Table 1. Means and SDs of SSFs* in black and white girls ages 9 to 19 years by race and age
Table 2. Means and SDs of %BF estimated from prediction equations* in black and white girls ages 9 to 19 years by race and age
Agreement between Methods
The overall correlation coefficient between %BF and SSF was r = 0.79 (p < 0.001). Table 3 presents the classification agreement by age, race, and classification group. Black girls above the 85th percentile by one method were above the 85th percentile by the other method ∼70% of the time before age 15 years, but <60% of cases after age 15 years, showing a trend to less agreement with age. The percentage of agreement in white girls showed less effect of age, being slightly lower than in black girls in early puberty and slightly higher in ages 17 to 19 years. Over 90% of the girls in each race classified as not high by one method were classified similarly by the other method. The overall kappa-statistic indicated good agreement between %BF and SSF (0.60 for whites and 0.66 for blacks; Table 3).
Table 3. Agreement between classifications by SSFs and by %BF in black and white girls ages 9 to 19 years by age and race for two-way classification
Effects of Race, Age, and Pubertal Stage on Relationship between SSF and %BF
Because initial results of the GEE models investigating the relationship between %BF and SSF, age, pubertal stage, and race showed that the race–SSF and race–age interactions were highly significant (p < 0.001), separate models were fit by race. Table 4 presents the results of the GEE model for black girls. The SSF term was significant (p < 0.01), indicating a strong positive relationship between SSF and %BF. There was a 0.55% absolute increase in %BF for each millimeter increase in SSF. The global test for differences among the SSF–age interaction terms was highly significant (p < 0.001), but individual terms reached borderline significance for only three age groups. The terms had no clear trend, although all were negative. By comparison, the global test for differences among the SSF–maturation stage interaction terms was highly significant (p < 0.001) and all of the individual terms were significantly different from zero. The increasingly negative individual terms indicate that the association between SSF and %BF becomes weaker with puberty.
Table 4. Parameter estimates and 95% confidence interval from GEE model with %BF as outcome for black girls ages 9 to 19 years
|Intercept||8.96§||7.90/10.02||SSF × age interactions|| |
|SSFs||0.55§||0.50/0.60||β estimate||95% confidence interval|
|Age 10 years*||0.67||−0.31/1.66||−0.009||−0.047/0.030|
|Age 11 years||0.71||−0.43/1.85||−0.028||−0.072/0.016|
|Age 12 years||0.92||−0.38/2.22||−0.043†||−0.089/0.003|
|Age 13 years||0.86||−0.72/2.44||−0.032||−0.086/0.021|
|Age 14 years||1.61||−0.31/3.54||−0.054†||−0.115/0.007|
|Age 15 years||2.26†||−0.02/4.54||−0.045||−0.113/0.023|
|Age 16 years||2.46†||−0.15/5.08||−0.021||−0.096/0.054|
|Age 17 years||3.41‡||0.48/6.34||−0.029||−0.111/0.053|
|Age 18 years||3.65‡||0.38/6.92||−0.037||−0.125/0.051|
|Age 19 years||5.78§||2.17/9.40||−0.083†||−0.178/0.012|
|Overall tests for age effect and interaction effect|| ||p = 0.018|| ||p < 0.001|
|Postmenarcheal < 1 year||2.71§||1.20/4.22||−0.100§||−0.156/−0.041|
|Postmenarcheal 1–2 years||4.44§||2.67/6.21||−0.129§||−0.193/−0.066|
|Postmenarcheal 2–3 years||5.37§||3.29/7.45||−0.146§||−0.215/−0.077|
|Postmenarcheal 3–4 years||6.03§||3.54/8.52||−0.160§||−0.237/−0.082|
|Postmenarcheal 4–5 years||5.82§||3.03/8.61||−0.156§||−0.239/−0.073|
|Postmenarcheal 5–6 years||6.52§||3.40/9.63||−0.178§||−0.268/−0.088|
|Postmenarcheal 6–7 years||6.79§||3.35/10.23||−0.183§||−0.278/−0.087|
|Postmenarcheal 7+ years||7.71§||3.79/11.63||−0.211§||−0.314/−0.107|
|Overall test for stage effect and interaction effect|| ||p < 0.001|| ||p = 0.003|
Table 5 presents the results of the GEE model for white girls. The SSF term is significant (p < 0.01) and indicates that there is a 0.34% absolute increase in %BF for every millimeter increase in SSF. The association between SSF and %BF is less in white girls than in black girls (0.55% vs. 0.34%; p < 0.001). The global test for differences among the SSF–age interaction terms was of borderline significance (p = 0.08), but none of the individual terms was significant. The global test for differences among the SSF–maturation stage interaction terms was highly significant (p < 0.001), and most of the individual terms were significant. The individual terms became increasingly negative, indicating that the association between SSF and %BF becomes weaker with increasing pubertal maturation.
Table 5. Parameter estimates and 95% confidence interval from GEE model with %BF as outcome for white girls ages 9 to 19 years
|Intercept||18.30§||17.54/19.06||SSF × age interactions|
|SSFs||0.34§||0.30/0.37|| || |
|Age 10 years*||−0.61||−1.39/0.17||0.007||−0.023/0.038|
|Age 11 years||−0.57||−1.55/0.41||−0.004||−0.040/0.033|
|Age 12 years||−0.67||−1.70/0.36||−0.009||−0.045/0.027|
|Age 13 years||−0.91||−2.09/0.28||−0.012||−0.053/0.030|
|Age 14 years||−1.43‡||−2.80/−0.05||−0.008||−0.053/0.038|
|Age 15 years||−0.91||−2.46/0.64||−0.002||−0.053/0.049|
|Age 16 years||−1.05||−2.79/0.70||0.017||−0.038/0.073|
|Age 17 years||−1.51||−3.46/0.43||0.041||−0.020/0.100|
|Age 18 years||−1.16||−3.33/1.02||0.037||−0.028/0.102|
|Age 19 years||−1.43||−3.80/0.94||0.042||−0.027/0.110|
|Overall tests for age effect and interaction effect|| ||p = 0.44|| ||p = 0.08|
|Postmenarcheal < 1 year||0.68||−0.43/1.80||−0.049‡||−0.088/−0.010|
|Postmenarcheal 1–2 years||2.07§||0.78/3.37||−0.069§||−0.112/−0.026|
|Postmenarcheal 2–3 years||2.97§||1.50/4.43||−0.091§||−0.139/−0.043|
|Postmenarcheal 3–4 years||3.68§||1.99/5.37||−0.110§||−0.162/−0.057|
|Postmenarcheal 4–5 years||3.89§||1.97/5.81||−0.113§||−0.172/−0.053|
|Postmenarcheal 5–6 years||4.88§||2.75/7.00||−0.142§||−0.207/−0.078|
|Postmenarcheal 6–7 years||5.18§||2.86/7.51||−0.149§||−0.216/−0.081|
|Postmenarcheal 7+ years||6.14§||3.58/8.67||−0.171§||−0.243/−0.099|
|Overall test for stage effect and interaction effect|| ||p < 0.001|| ||p < 0.001|
These race-specific models show that the association between SSF and%BF is very strong in both black and white girls at younger ages but becomes weaker as the girls mature. Increasing age by itself has only a minimal impact on this association.
Results of this study indicated that SSF associated with%BF differently between the races and with age for each race. Although results indicated a highly significant correlation between %BF and SSF (r = 0.79; p < 0.001) and good agreement in classifying participants, the two methods produced different epidemiologic pictures of obesity in black and white girls this age. That is, white girls had significantly greater%BF by impedance at ages 9 to 11 years, and moderately greater %BF at age 12 years (p = 0.086), but black girls had greater mean SSF at every age, including ages 9 to 12 years. Thus, there were clear racial differences in fat patterning between the races at these ages. Black girls had significantly greater %BF from 14 to 19 years of age and moderately higher adiposity at age 13 years (p = 0.09). The shift in total adiposity as determined using impedance seems to occur at ages 12 and 13 years. In addition, in multivariate analysis, an increase of 10 mm in the SSF was associated with a larger increase in total fat in black girls than in white girls. This finding suggests that subcutaneous fat accounts for a larger portion of total fat in black girls and a smaller proportion of the total fat in white girls. Conversely, this finding suggests that visceral adiposity accounts for a larger proportion of total adiposity in white girls. If white girls have more visceral fat than do black girls, this could partly explain why white girls have a greater %BF by impedance at ages 9 to 12 years, despite lower SSF.
Selecting a single best or most accurate measure of obesity is difficult. No measure optimally describes obesity in all age–sex–race groups because of differences in bone density, muscle, and adipose tissue. The body mass index (BMI, [kg/m2]), a commonly used index of nutritional status, is easy to measure and calculate, but it includes measures of bone mass and FFM as well as fat mass. Skinfolds have been used to assess levels of adiposity in national surveys (7), but they have poor reliability, especially between studies, and calipers are prone to problems. The quadratic equation of Slaughter et al. (12) has been reported to be superior to other skinfold-based equations in comparison to four-component model but to need “further refinement” for use in minority populations (16). Bioelectric impedance measures have been used in combination with anthropometric variables in epidemiological surveys to estimate body composition and levels of body hydration of children and adults (17, 18, 19). The reliability of bioelectrical impedance is excellent (20). Theoretical limitations to impedance include possible differences in the hydration of FFM in extremely underweight (lower) and extremely obese individuals (higher), leading to overestimation and underestimation of %BF, respectively (21). Practical limitations to impedance include the effects of recent meals (increased hydration and absorption of fluid into blood space) and recent physical activity (increasing blood flow to the muscles or decreasing body fluids through perspiration). The practical limitations can be controlled for by taking measurements in the morning under fasting conditions and before exercise (22). Impedance considers the stature of the participant to be the length of the conductor and assumes that height is a good correlate for the length of the participant's arms and legs. Although this assumption may not be met in all cases, we derived our equations on a sample of girls with a wide range of heights, weights, and (presumably) limb lengths, and received good results, which were cross-validated with an external sample (12).
Differences in the percentages of black and white girls with inconsistent classifications may reflect racial differences in fat patterning and the relation between subcutaneous fat and total fat. Possible reasons for these differences need to be examined further. Skinfolds measure subcutaneous fat at one or more sites to characterize total adiposity; they are restricted to the selected sites and do not measure internal fat. Bioelectrical impedance estimates total body water to determine (total) FFM. Total fat mass is calculated by subtracting total FFM from total body weight. It, therefore, does not target either subcutaneous or internal fat specifically but may provide better estimates of total fat than do skinfolds. Although estimates of%BF from impedance may be better than estimates from skinfolds for these reasons, such judgments must await comparisons to magnetic resonance imaging.
A greater proportion of total adiposity in the viscera could help explain, in part, why white girls have lower concentrations of high-density lipoprotein cholesterol and lower concentrations of triacylglycerol, despite the smaller SSF. Visceral adiposity has been shown to be more strongly associated with CVD risk factors than weight, BMI, or SSF (23, 24, 25), but the waist-to-hip ratio may not be a useful estimate of this fat deposit in children and visceral adiposity varies widely in children (26). An earlier report from NGHS showed that overweight white girls with increased central adiposity had a greater increase in the clustering of CVD risk factors than did overweight white girls with less central adiposity. Visceral adiposity was not measured in NGHS.
Conway et al. (27) reported that adult black women had significantly less visceral fat as measured by computerized tomography than did white women after controlling for BMI, despite having greater SSFs. In a subsequent study at the National Institute of Child Health and Human Development, Yanovski et al. (28) showed in 20 black and 20 white girls, 7 to 10 years of age, that black girls had less total visceral and subcutaneous adipose tissue by magnetic resonance imaging. In the current study, black girls had less total fat up to age 12 years, despite greater SSF at ages 10 to 12 years (the racial difference at age 9 years was marginally significant; p = 0.10). It is unlikely that black girls would have greater triceps and subscapular skinfolds but less subcutaneous adiposity overall. The most likely explanation for the differences between this study and the study by Yanovski et al. (28) is the ages of the participants. Black girls may have less total fat (and smaller subcutaneous and visceral fat compartments) early in life but accrete more fat, starting with subcutaneous deposits in the peripubertal period. The findings of Conway et al. (27) suggest that adult black women have a lower proportion of total fat as visceral fat than do white counterparts but greater adiposity overall. Additional research is needed to determine when this racial difference in fat compartments arises.
This research was supported by National Institutes of Health/National Heart, Lung, and Blood Institute Contracts HC-55023-26 and HL 48941. We thank the members of the NGHS Steering Committee: Clinical Centers: University of California, Berkeley (Zak Sabry, principal investigator; Patricia B. Crawford; and Frank Falkner), Children's Hospital Medical Center (John A. Morrison, principal investigator years 1 through 9; Stephen R. Daniels, principal investigator years 9 and 10; Frank M. Biro; and Dennis L. Sprecher), and Westat, Inc. (George B. Schreiber, principal investigator, and Ruth Striegel-Moore); Coordinating Center: Maryland Medical Research Center (Bruce A. Barton, principal investigator; Sue Y. S. Kimm; and Robert P. McMahon); and the National Institutes of Health Program Office: Division of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, Bethesda, MD (Eva Obarzanek, project officer, and Gerald Payne). We also thank Myron A. Waclawiw from the National Heart, Lung, and Blood Institute for help with the statistical analyses.