RESPONSE

Abstract

Dear Editor-in-Chief: We thank Dr. Kwon for his letter regarding our paper (1) and welcome the opportunity to respond. Dr. Kwon's major concern is whether our Asian sample was representative of the US population. He states that "…Asians represent 45% of the study participants…" and "…the mean height and BMI of Asians were 0.5 cm and 0.6 kg·m−2higher for men, and 0.2 cm lower and 1.9 kg·m−2higher for women…." These statements are inaccurate. Asians represented 13% (181/1394) of the TIGER sample but were excluded from the analyses used to develop the prediction equations (p. 1960). We reported no descriptive statistics for Asians and do not know the source of the data to which Dr. Kwon refers. He further states that "their study data indicate that given BMI level, Asians had a lower BF% than non-Hispanic whites," which is also inaccurate. Our BMI regression models included both skinfold thicknesses (∑3S) and BMI (Table 3). Thus, the correct interpretation is that, for a given BMI and ∑3S, Asian men had equivalent BF% and Asian women had a slightly higher BF% than non-Hispanic whites-opposite of Dr. Kwon's statement. With respect to representativeness, we used a large sample with subjects from the three major ethnic groups in the United States. These subjects' stature and overweight/obesity prevalence were within sampling error of US population values (p. 1964). Skinfold thicknesses, BMI, and DXA percent body fat were widely distributed (Table 1). Dr. Kwon's second concern is "whether race/ethnicity-specific equations are necessary." He presents several points, which were all addressed in our article. The differences in accuracy between equations were not "minimal." Using the non-Hispanic whites' BMI model for Hispanic women would result in a nearly 2% underestimation of DXA percent body fat (Tables 2 and 3), a substantial bias. The men's non-Hispanic white and Hispanic equations did not differ appreciably-which is why we presented a single equation for both groups (Table 3). We demonstrated that "the equations for Hispanic women can be accurately applied to Asian women" on page 1963 and in Table 4. The inclusion of race/ethnicity did "provide a better fit than the reduced model," seen in the P values of the ethnicity regression coefficients (Table 2), which are exactly equivalent to comparing models including and excluding these coefficients. The ethnicity-specific equations did "increase the accuracy of estimated BF%" if one considers total error (bias + random error). We described these bias components (p. 1962-3): "race/ethnic group bias is well documented forseveral BMI prediction models…" with nine references. Dr. Kwon's questions whether the "estimates of ∑3S, ∑3S2, and BMI may differ by race/ethnicity." No: the interactions of each measure with ethnicity failed to attain statistical significance. Most body composition regression equations have been developed in non-Hispanic white subjects. Our large sample was ethnically diverse. Our concurrent validity study has standard errors among the lowest reported in the literature and reports ethnicity-specific error estimates. As we stated (p. 1963), these equations should be cross-validated with other populations, including Asians, Asian-Indians, and subjects older than 35 years. Daniel P. O'Connor, PhD Andrew S. Jackson, PED Brian K. McFarlin, PhD Department of Health and Human Performance University of Houston Houston, TX Molly S. Bray, PhD Mary H. Sailors, PhD Department of Epidemiology University of Alabama-Birmingham Birmingham, AL Kenneth J. Ellis, PhD USDA/ARS Children's Nutrition Research Center Baylor College of Medicine