O'CONNOR, DANIEL P.1,2; BRAY, MOLLY S.3,4; MCFARLIN, BRIAN K.1,2; ELLIS, KENNETH J.5; SAILORS, MARY H.3,4; JACKSON, ANDREW S.1,2
The relative distribution of fat and lean mass, particularly the deposition of adipose tissue in the abdominal region, has recently been identified as a risk factor for the development of cardiovascular disease and other comorbid states (3,5). The laboratory methods used to assess abdominal obesity include computed tomography (CT) and dual-energy x-ray absorptiometry (DXA). CT scanning is considered the "gold standard" because of its ability to identify and quantify subcutaneous and visceral body fat depots separately (5,6). DXA has become the most common laboratory method to measure body composition and has the capacity to measure lean mass, fat mass, and bone mineral content of the total body and for body compartments (9).
Anthropometric methods that have been used to assess abdominal adiposity include waist and hip circumference, waist-to-hip ratio, and body mass index (BMI) (22). There is growing evidence of a race/ethnicity group bias associated when using anthropometric measures to assess adiposity of adults (3,5-8,12,13,21,22). In the HERITAGE Family Study (22), hip circumference and WC in combination with BMI provided reasonably accurate estimates of CT-measured abdominal visceral fat (women: r = 0.85, men: r = 0.88), but the anthropometric model systematically overestimated the abdominal visceral fat of African-American (AA) men and women. What is not understood is the race/ethnicity bias of anthropometric variables when evaluating abdominal fat in research and field settings. This study examined the race/ethnicity bias of anthropometric models to measure DXA abdominal fat (DXA Abd-Fat) of diverse young men and women.
Subjects were recruited from the TIGER (Training Intervention and Genetics of Exercise Response) study (20). The TIGER subjects were students enrolled at the University of Houston, Houston, TX. The target subject was a sedentary person under the age of 35 yr who exercised <30 min·wk−1 for the previous 30 d and was not actively limiting caloric intake by dietary modification. Subjects were excluded from the study if they had a physical or physiological contraindication to aerobic exercise, a known metabolic disorder that may alter body composition, or were pregnant or lactating. The study was approved by the University of Houston Committee for the Protection of Human Subjects, and all subjects volunteered to participate by signing informed consent.
The TIGER study subjects engaged in 30 wk (two semesters or nine calendar months) of exercise training, 3 d·wk−1 for 30 min·d−1 at 65%-85% of age- and gender-specific predicted HR reserve (19). The data came from five yearly cohorts. The sample consisted of 771 women and 484 men who ranged in age from 18 to 35 yr. These data included men and women who had from one to three measurement visits during the 9-month exercise training program. The measurements were obtained at baseline (enrollment), and at 15 and 30 wk.
The total number of measurements was 1293 for women and 824 for men. The racial/ethnic group composition of the male and female samples differed slightly. The composition of the men was non-Hispanic white (NHW) or Caucasian, 37%; Hispanic, 25%; AA, 22%; Asian Indian, 5%; and Asian, 9%. The composition of the women was NHW, 29%; Hispanic, 24%; AA, 35%; Asian Indian, 4%; and Asian, 8%. Tables 1 and 2 provide subject characteristics at the baseline measurement stratified by sex and race/ethnicity group.
Height was measured with a stadiometer (seca Road Rod; seca, Hanover, MD), and body weight was measured using a digital scale (seca 770; seca). Waist circumference (WC) was measured by the same investigator at each time point using a Gulick tension-calibrated tape measure (Creative Health Products, Ann Arbor, MI) and a standard technique. BMI was calculated using the standard Quetelet index equation (BMI = weight (kg)/height squared (m2)). Before DXA scanning, all women completed a standard urine pregnancy test. Subjects were asked to lie in a supine position, and whole-body scans were collected using a Hologic Discovery W (Hologic, Bedford, MA). Postacquisition analysis was completed using the adult whole-body software module (QDR v.12.3, Hologic). A "region of interest" box was added to delineate the abdominal compartment (L1-L5).
A random intercepts model fitted using maximum likelihood linear mixed models regression (LMM) with the Stata 11 (StataCorp, College Station, TX) xtmixed program was used for data analysis (18,23). The dependent variable was abdominal fat, defined as the body fat component of the DXA abdominal region of interest. The fixed effects included in all models were WC, height, and race/ethnic group. The NHW group was the referent. All models included a random intercept term to account for the within-subjects dependency due to repeated tests (18).
Anthropometric model 1 included WC and height, with measurement period as a control variable, and model 2 (full model) added race/ethnicity group and race/ethnicity group × WC interaction terms to the initial model. Both WC and the WC × race/ethnicity interaction terms were centered using the WC grand mean to reduce collinearity effects. Each fixed-effect coefficient was tested with a Wald z-statistic to eliminate variables that were not significantly different from zero (23,26). Likelihood ratio (LR) tests (18,23) were used to compare the initial and final model fit. Model fit was also assessed at each stage of the analysis; model residuals and random intercepts (empirical Bayes best linear unbiased predicted values) were graphically examined with histograms and normal probability plots and were plotted against the respective fixed-effects predicted scores (18,23). Because of the well-documented sex differences in body fatness, the women's and men's data were analyzed using separate models.
The DXA Abd-Fat distributions of both men and women were positively skewed and had curvilinear relations with most of the other study variables, so DXA Abd-Fat was log-transformed (ln(Abd-Fat)) to approximate normality and linear associations. The dependent variable for these analyses was ln(Abd-Fat). Consequently, for the continuous variables, the regression coefficients, β, represented the proportional unit difference in DXA Abd-Fat per unit difference in the respective independent variable; the equation 100(eβ − 1) was used to compute the proportional change (4). For categorical variables, such as race/ethnicity, the regression coefficients represented proportional differences in DXA Abd-Fat of the respective race/ethnic group relative to the NHW reference group after adjusting for other variables in the model. Tables 1 and 2 provide the descriptive statistics for the men and women. Provided are the means and SD for the total samples contrasted by race/ethnic group.
Tables 1 and 2 provide the descriptive statistics for the men and women. Provided are the means and SD for the total samples contrasted by race/ethnic group. Tables 3 and 4 provide the LMM results. For the women, model 1 included WC, WC squared (WC2), and height, all of which were significantly related to ln(Abd-Fat). Adding race/ethnicity and the interaction of race/ethnicity to model 1 to form model 2 significantly improved model fit (P < 0.001). After statistically adjusting for the effects of height, DXA Abd-Fat was proportionately higher for a given WC in the Hispanic (9.0%, P = 0.001) and Asian-Indian (24.5%, P < 0.001) women and lower in the AA women (−7.5%, P = 0.001) compared with the NHW women (Table 3). Furthermore, for every 1-cm increase in WC, DXA Abd-Fat increased at a 0.70% lower rate in the Asian-Indian (P = 0.029) women and a 1.0% lower rate in the Asian (P = 0.043) women compared with the other racial/ethnic groups (Table 3). This differential increase in DXA Abd-Fat and WC attributable to race/ethnicity in women can be seen as differences in the slopes of the respective race/ethnicity regression lines in Figure 1. Also, because proportional differences grow larger in terms of the original units as the values of the independent variable increase, the regression lines tend to diverge as WC increases.
For the men, model 1 included WC, WC2, and height, all of which were significantly related to ln(Abd-Fat) (Table 4). Race/ethnicity and the interactions of race/ethnicity and WC significantly improved model fit (P < 0.001). After statistically adjusting for the effects of height, average DXA Abd-Fat was higher for a given WC in the Asian-Indian (12.6%, P = 0.044) and Asian men (15.4%, P = 0.001) and lower in the AA men (−12.7%, P < 0.001) compared with the NHW men (Table 4). For a 1-cm increase in WC, DXA Abd-Fat increased at a 0.6% higher rate in the AA men (P = 0.004) and a 0.8% lower rate in the Asian-Indian (P = 0.023) men compared with the other racial/ethnic groups (Table 4). This differential increase in DXA Abd-Fat attributable to race/ethnicity in men can be seen as differences in the slopes of the respective race/ethnicity regression lines in Figure 2. Similar to the women's models, the regression lines diverge as WC increases.
FIGURE 2-Relation of...Image Tools
The differential effects of race/ethnicity were most notable as the race/ethnicity-specific regression lines began to diverge at WC values between 90 and 100 cm in the women and about 100 cm in the men (Figs. 1 and 2). For example, the Asian-Indian women have higher DXA Abd-Fat compared with Hispanic women at WC <90 cm, but at WC values >90 cm, the Asian-Indian women have lower DXA Abd-Fat. Similarly, the Asian-Indian men have higher DXA Abd-Fat compared with NHW men at WC <100 cm, but at WC values >100 cm, the Asian-Indian men had lower DXA Abd-Fat than NHW men.
The LMM results showed that using WC to estimate DXA Abd-Fat resulted in biased estimates for the racial/ethnic groups studied unless race/ethnicity is included in the equation. After accounting for height and WC, DXA Abd-Fat was lower in the AA women and higher in the Hispanic and Asian-Indian women compared with the NHW women reference group. In men, DXA Abd-Fat was higher in the Asian-Indian and Asian groups and lower in the AA men relative to the NHW men reference group. We also found several differential racial/ethnic biases. Within women, the difference in DXA Abd-Fat per 1-cm difference in WC was lower in the Asian and Asian-Indian groups compared with the other race/ethnic groups. Within men, the difference in DXA Abd-Fat for a 1-cm difference in WC was lower in the Asian-Indian group and higher in the AA group compared with the other race/ethnic groups.
The WC values at which the racial/ethnic equations begin to diverge, 90-100 cm for women and 100 cm for men, are consistent with the definition of abdominal obesity currently used in the United States for women (88 cm) and men (102 cm) (16), although the International Diabetes Federation recommends several race/ethnicity-specific values that are 5-10 cm lower than these US definitions (10). Abdominal obesity is strongly associated with the risk of various chronic diseases, including cardiovascular disease and diabetes (10,15). These results suggest that the risk attributable to abdominal obesity as defined by WC may differ among different racial/ethnic groups.
The finding that race/ethnicity affects the relationship between WC and abdominal fat is consistent with previous reports (1,2,14,22,24). For example, the HERITAGE family study used CT scanning to measure abdominal compartment body fat in NHW and AA men and women, ages ranging from 17 to 65 yr. Stanforth et al. (22) reported that for both men and women, with age, BMI, and WC statistically controlled, the abdominal visceral fat of AA men and women was significantly lower than that of NHW men and women, similar to our results, as well as those of others (1,2). The observation of higher fat in the abdominal region among Asian men and women relative to NHW and other races/ethnicities has also been reported previously (14), although our analysis is unique in that it evaluated nonlinear trends and interaction effects and found a complex effect of race/ethnicity on the relation of WC to DXA Abd-Fat. These TIGER results allow the generalization of race/ethnicity effects on the relation between WC and abdominal fat to NHW, AA, Hispanic, Asian, and Asian-Indian young adults.
Whereas the TIGER results are consistent with prior studies, there are important differences. First, abdominal visceral fat of the TIGER subjects was measured with DXA, whereas most of the prior studies used CT. The major limitation of DXA is that it does not differentiate subcutaneous and visceral abdominal fat stores. Whereas CT is considered the gold standard for visceral fat mass, DXA now is the most common method used to measure body composition (9). Second, the data of the prior studies were cross-sectional, whereas the TIGER data were longitudinal with subjects having from one to three measurements during 30 wk. The statistical model used to evaluate the data in this study provided a means of examining the relation of DXA Abd-Fat and WC within an individual subject and accounting for various individual characteristics such as height and race/ethnicity (18,25).
We explored random coefficient models including WC as a random effect to evaluate individual differences in the relation of WC with DXA Abd-Fat, but none of those approaches improved model fit. After accounting for race/ethnicity in the relation of WC with DXA Abd-Fat, within-subject effects on the WC to DXA Abd-Fat relation were minimal. We also explored BMI models for both men and women, and these models included significant race/ethnicity and race/ethnicity × BMI interaction effects, indicating that the bias in estimating DXA Abd-Fat using BMI without accounting for race/ethnicity introduces biases similar to those in our WC models. This has been shown previously with these TIGER data using BMI to estimate total body DXA percent fat (11).
The strengths of the current study are the use of large samples of men and women, the inclusion of multiple tests during 30 wk, and the inclusion of multiple racial/ethnic groups. A limitation of this study is that the TIGER subjects are not a random sample of the US population. The TIGER subjects were self-selected college-aged students who elected to participate in the study for which they received college course credit. Nevertheless, the TIGER men and women are a reasonable representation of the population of young adults, with overweight and obesity prevalence similar to national norms (17).
In summary, the present study confirms the generalizability of racial/ethnic body composition differences. There is a racial/ethnic bias when estimating DXA Abd-Fat with WC. This bias includes both a constant difference between some of the race/ethnicity groups and a differential race/ethnicity effect for some groups that is a function of the value of WC. The race/ethnicity group bias increases when WC is greater than about 90 cm in women and 100 cm in men. These data show that WC is associated with DXA Abd-Fat of both men and women, but using WC to estimate DXA Abd-Fat without accounting for the race/ethnicity group results in biased estimates of DXA Abd-Fat with diverse young adults.
The project was supported by National Institutes of Health grant R01DK062148 (MSB).
None of the authors has professional relationships with companies or manufacturers who will benefit from the results of the present study.
The authors thank Ian Turpin for his management of the TIGER study cohorts and the many graduate and undergraduate students who worked to ensure the TIGER study was a success.
The results of the present study do not constitute endorsement by the American College of Sports Medicine.
1. Camhi SM, Bray GA, Bouchard C, et al. The relationship of waist circumference and BMI to visceral, subcutaneous, and total body fat: sex and race differences. Obesity (Silver Spring)
2. Carroll JF, Chiapa AL, Rodriquez M, et al. Visceral fat, waist circumference, and BMI: impact of race/ethnicity. Obesity (Silver Spring)
3. Chang CJ, Wu CH, Chang CS, et al. Low body mass index but high percent body fat in Taiwanese subjects: implications of obesity cutoffs. Int J Obes Relat Metab Disord
4. Cohen J, Cohen P, West SG, Aiken LS. Quantitative Scales, Curvilinear Relationships, and Transformations. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences
. Mahwah (NJ): Lawrence Erlbaum Associates; 2003. p. 193-254.
5. Deurenberg P. Universal cut-off BMI points for obesity are not appropriate. Br J Nutr
6. Deurenberg P, Weststrate JA, Seidell JC. Body mass index as a measure of body fatness: age- and sex-specific prediction formulas. Br J Nutr
7. Deurenberg P, Yap M, van Staveren WA. Body mass index and percent body fat: a meta analysis among different ethnic groups. Int J Obes Relat Metab Disord
8. Deurenberg-Yap M, Schmidt G, van Staveren WA, Deurenberg P. The paradox of low body mass index and high body fat percentage among Chinese, Malays and Indians in Singapore. Int J Obes Relat Metab Disord
9. Heymsfield S. Human Body Composition
. 2nd ed. Champaign (IL): Human Kinetics; 2005. p. 523.
10. International Diabetes Federation. IDF worldwide definition of the metabolic syndrome [Internet]. Brussels (Belgium): International Diabetes Foundation; 2011 [cited 2010 April 30]. Available from: http://www.idf.org/metabolic_syndrome
11. Jackson AS, Ellis KJ, McFarlin BK, Sailors MH, Bray MS. Body mass index bias in defining obesity of diverse young adults: the Training Intervention and Genetics of Exercise Response (TIGER) study. Br J Nutr
12. Jackson AS, Ellis KJ, McFarlin BK, Sailors MH, Bray MS. Cross-validation of generalised body composition equations with diverse young men and women: the Training Intervention and Genetics of Exercise Response (TIGER) Study. Br J Nutr
13. Jackson AS, Stanforth PR, Gagnon J, et al. The effect of sex, age and race on estimating percentage body fat from body mass index: The Heritage Family Study. Int J Obes Relat Metab Disord
14. Lear SA, Humphries KH, Kohli S, Birmingham CL. The use of BMI and waist circumference as surrogates of body fat differs by ethnicity. Obesity (Silver Spring)
15. National Cholesterol Education Program, National Heart, Lung, and Blood Institute. Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III): Final Report
. Bethesda (MD): National Institutes of Health; 2002. p. 284. Available at: http://www.nhlbi.nih.gov/guidelines/cholesterol/atp3full.pdf
16. National Institutes of Health. Clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults-the evidence report. Obes Res
. 1998;6(2 suppl):51-209S.
17. Ogden CL, Carroll MD, Curtin LR, McDowell MA, Tabak CJ, Flegal KM. Prevalence of overweight and obesity in the United States, 1999-2004. JAMA
18. Rabe-Hesketh S, Skrondal A. Multilevel and longitudinal modeling using Stata. College Station (TX): Stata Press; 2005. p. 317.
19. Ross RM, Jackson AS. Exercise Concepts, Calculations, and Computer Applications
. Carmel (IN): Benchmark Press; 1990. p. 312.
20. Sailors MH, Jackson AS, McFarlin BK, et al. Exposing college students to exercise: the Training Interventions and Genetics of Exercise Response (TIGER) study. J Am Coll Health
21. Schutte JE, Townsend EJ, Hugg J, Shoup RF, Malina RM, Blomqvist CG. Density of lean body mass is greater in blacks than in whites. J Appl Physiol
22. Stanforth PR, Jackson AS, Green JS, et al. Generalized abdominal visceral fat prediction models for black and white adults aged 17-65 yr: the HERITAGE Family Study. Int J Obes Relat Metab Disord
23. Stata. User's Guide: Release 10
. College Station (TX): StataCorp LP; 2007. p. 356.
24. Sumner AE, Micklesfield LK, Ricks M, et al. Waist circumference, BMI, and visceral adipose tissue in white women and women of African descent. Obesity (Silver Spring)
25. Twisk JWR. Applied Longitudinal Data Analysis for Epidemiology: A Practical Guide
. Cambridge (UK): Cambridge University Press; 2003. p. 301.
26. West B, Welch KB, Galecki AT. Linear Mixed Models: A Practical Guide Using Statistical Software
. Boca Raton (FL): Chapman & Hall/CRC; 2007. p. 353.