OBESITY | VOLUME 17 NUMBER 5 | MAY 2009 935 PERSPECTIVES LETTERS TO THE EDITOR of weight loss in order to determine the utility of peripheral mononuclear cells for future human biomarker studies in obesity research. DISCLOSURE The authors declared no conflict of interest. © 2009 The Obesity Society REFERENCES 1. Kant P, Hull M. Letter to the editor: effect of weight loss on proinflammatory state of mononuclear cells in obese women. Obesity, this issue. 2. Sheu WHH, Chang TM, Lee WJ et al. Effect of weight loss on proinflammatory state of mononuclear cells in obese women. Obesity (Silver Spring) 2008;16:1033–1038. 3. Rankin JW, Turpyn AD. Low carbohydrate, high fat diet increases C-reactive protein during weight loss. J Am Coll Nutr 2007;26:163–169. 4. Laimer M, Ebenbichler CF, Kaser S et al. Markers of chronic inflammation and obesity: a prospective study on the reversibility of this association in middle-aged women undergoing weight loss by surgical intervention. Int J Obes Relat Metab Disord 2002;26:659–662. 1 Division of Endocrinology and Metabolism, Department of Internal Medicine, Taichung Veterans General Hospital, Taichung, Taiwan; 2 Department of Education and Medical Research, Institute of Medical Technology, National Chung- Hsing University, Taichung, Taiwan; 3 College of Medicine, Chung Shan Medical University, Taichung, Taiwan; 4 School of Medicine, National Yang-Ming University, Taipei, Taiwan. Correspondence: Wayne H.-H. Sheu (whhsheu@vghtc.gov.tw) doi:10.1038/oby.2009.7 Predictive Equations for Body Fat in Asian Indians Pranav C. Yajnik 1 and Chittaranjan S. Yajnik 1 TO THE EDITOR: We read with interest the article by Goel et al., titled “Predictive Equations for Body Fat and Abdominal Fat With DXA and MRI as Reference in Asian Indians,” which offers predictive equations for body fat and abdominal fat in Asian Indians as functions of simple anthropometric measures in 171 apparently healthy North Indian respondents to a local advertisement (1). We noticed a possible error in the recommended equation for body fat percent. Correspondence with authors gave us the correct equations which we used for further analysis (%Body Fat = 42.42 + 0.003 × age (years) + 7.04 × gender (M = 1, F = 2) + 0.42 × triceps skinfold (mm) + 0.29 × waist circumference (cm) + 0.22 × weight (kg) - 0.42 × height (cm)). We tested Goel’s recommended equation to calculate body fat percent in subjects of the Pune Maternal Nutrition Study (PMNS), a community-based study in six villages near Pune, Maharashtra, India (2). Detailed anthropometric measurements and body fat measurements (dual-energy X-ray absorptiometry (DXA), Lunar DPX-IQ) were available in 645 men (mean age 34 years, triceps skinfold 8.7 mm, waist circumference 80.2 cm, weight 57.1 kg, height 165.7 cm) and 681 women (27 years, triceps skinfold 9.7 mm, waist circumference 65.9 cm, weight 44.5 kg, height 152.9 cm), with a mean body fat percent of 17.7% (s.d. 8.44) and 25.6% (8.13), respectively. Mean error (predicted—DXA) for fathers (1.71 percentage points) was significantly different from zero and also from mean error for mothers (-0.23% points) (P < 0.001 for both). Importantly, the Bland–Altman method indicates that Goel’s equation produced a systematic bias in prediction of body fat percent such that there was overestimation at lower values and underestimation at higher values. Moreover, the bias for men and women were different (estimated by linear regression; slope for men -0.33 (99% CI -0.27, -0.39) and for women -0.48 (-0.42, -0.55); intercept for men 7.80 (6.56, 9.03) and for women 12.04 (10.30, 13.79)). When used as a classifier, the equation becomes progressively less sensitive at higher cut points of adiposity (Table 1). The equation fares poorly in children (at 6 years of age R 2 = 0.15, at 12 years R 2 = 0.48 with considerably stronger biases compared to adults). Although statistical models are an attractive and potentially useful surrogate for measurements that are difficult to make, most statistical models fail to capture the complex relationships that exist between the dependent and predictor variables. This often leads to biases which vary across populations, i.e., populations differing because of gender (see Figure 1), race, lifestyle (rural–urban) etc. The relationship between anthropometric measures (BMI and waist circumference) and adiposity (body fat percent) is known to vary between populations of different ethnicities (3). Importantly, our analysis shows that this is also true of different populations from the same ethnic group. Our sample consists exclusively of rural Indians while Goel’s presumably contains mostly urban volunteers. There is also a substantial difference in the body fat percent between PMNS subjects and Goel’s volunteers. All of these factors (along with many unmeasured/ unknown factors) may contribute to the biases that Goel’s equation produces. SPC 100x Sample S2 S2 S1 + - - Figure 1 Figure indicates the electrophoretic mobility shift assay (EMSA) band specificity testing. The p65/p50 DNA binding activity from nuclear extracts of different samples is shown (lanes 1 and 2). In the competitive assay, 100× specific cold competitor (SPC) was added to the reaction mixture (lane 3). Table 1 Sensitivity and specificity of Goel’s equation when used as a classifier (for %Body Fat) on PMNS adults Males Females Sensitivity Specificity Sensitivity Specificity >25% 0.71 0.97 0.77 0.85 >30% 0.47 0.98 0.57 0.99 >35% 0.27 0.99 0.33 1