Introduction

Obesity is a metabolic disorder characterized by increased body fat (BF)¹ and associated with greater risk of hypertension, diabetes, coronary heart disease, and cancer, all major health concerns in developed and developing countries (1,2,3,4,5,6). The degree of fatness is known to be clinically more useful than body weight (BW) because it is a more accurate measure of adiposity than BW. The use of BF mass in the diagnosis of obesity could potentially reduce the misdiagnosis of overweight and obesity that occurs when using BMI in Asian populations (7,8,9,10).

Percentage BF (%BF) in healthy subjects can be estimated using numerous techniques, including underwater weighing technique, total-body electrical conductivity, bioelectrical impedance, potassium-40 counting, and DXA. These methods are, however, sophisticated and require well-equipped research facilities and trained personnel, which are difficult to apply to large number of subjects, especially in developing countries (11). For routine clinical and epidemiological use, simple and readily available anthropometric measurements are preferable variables for predicting body composition, including %BF (11,12,13,14,15,16,17,18).

In the past, a number of predictive equations of %BF have been developed in largely white populations (12,13,14,15,19,20,21,22,23,24). Few predictive equations have also been specifically developed for Asians (14,25,26,27,28,29,30). However, their validity and applicability to ethnic populations are problematic because body composition varies among and within ethnicities. For instance, for a given BMI level, Asians have a higher fat mass and higher subcutaneous fat mass than whites (8), whereas the opposite was true for blacks who have a higher bone and muscle mass than Asians and whites (31,32,33,34).

The present study's aim was to determine the relationship between %BF and anthropometric measures and to develop and validate a sex-specific predictive equation of %BF for rural Thai adults.

Research Methods and Procedures

Study Design and Participants

This study was designed as a cross-sectional, community-based survey, in which participants were randomly drawn from residents in Koksri and Nongtoom Subdistricts, Muang, Khon Kaen, Thailand. The subjects were 181 men and 255 women between 20 and 84 years old. We excluded participants who had a history of recent acute illness (e.g., myocardial infarction or pneumonia), a chronic condition (e.g., cancer, chronic infection, collagen vascular disease, hepatic or renal impairment, diabetes), a history of taking medication affecting BW (e.g., thyroid hormone, prednisolone, diuretics), and/or who were involved in weight training. The Ethics Committee of Khon Kaen University approved the study protocol, and informed consent was obtained from all participants. The study conformed to the Helsinki Declaration of 1975 as revised in 1983.

Anthropometric Measurements

BW (kilograms) was measured with participants wearing light indoor clothing on an electronic balance accurate to 0.1 kg. Height without shoes was measured using a stadiometer accurate to 0.1 cm. BMI was calculated as the quotient of weight over height squared (kilograms per meter squared). Circumferences (centimeters) were measured with a Harpenden anthropometric tape. The waist circumference (WC) and hip circumference (HC) to 0.1 cm were taken midway between the inferior margin of the last rib and the crest of the ilium in a horizontal plane and around the pelvis at the point of maximal protrusion of the buttocks, respectively. The waist-to-hip ratio (WHR) was then calculated.

Four skinfold measurements were taken with Holtain calipers (Holtain Ltd., Crymych, United Kingdom) at biceps skinfold (BSF), triceps skinfold (TSF), subscapular skinfold (SSF), and suprailiac skinfold (SISF) regions. In brief (35): first, the skin and fat layer were lifted from the underlying tissue by grasping the tissue between the thumb and forefinger; second, calipers were applied 1.0 cm distal from the thumb and forefinger, midway between the apex and base of the skinfold; third, the skinfold continued to be supported with the thumb and forefinger for the duration of the measurement and 2 to 3 seconds afterward; and finally, measurements to the nearest 0.5 mm were done thrice, until readings were within 1.0 mm, then averaged.

Measures of Body Composition

Body compositional parameters (e.g., lean mass and fat mass) were measured by DXA densitometry (model DPX-IQ; Lunar Radiation Corp., Madison, WI). These measurements have an intrasubject coefficient of variation of 1.5% for bone mineral density and 3% to 4% for lean and fat mass (36,37). The software in this device calculates an estimate of weight fat, weight lean, and %BF based on an extrapolation of fatness from the ratio of soft tissue attenuation of two X-ray energies in pixels not containing bone.

Statistical Analyses

The sample were randomized into two groups: a development group that consisted of 98 men and 125 women and a validation group with 83 men and 130 women. Predictive equations of %BF were developed in the development data set, and the performances of the equation were subsequently tested in the validation data set.

In the development data set, bivariate correlations were calculated using the Pearson's product-moment correlation coefficient. To develop predictive equation, the step-wise and backward multiple linear regression analyses were utilized, with the following anthropometric variables being considered potential independent predictors: BMI, WC, HC, BSF, TSF, SSF, and SISF. The analyses were performed in men and women separately. The order of the variables was keep invariant for men and women. In addition, a multiple linear regression model with BMI, age, and sex as independent predictors and %BF as the dependent variable was used.

The performances of these predictive equations were then assessed in the validation data set. The correlation between predicted %BF (by the predictive equations) and measured %BF (by DXA) was determined using the Pearson's product-moment correlation coefficient. The Bland and Altman analysis was used to assess agreement between the methods of estimation. Differences between values were plotted against their mean (38). The sensitivity, specificity, positive predictive values (PPVs), and negative predictive values of the equations were also calculated.

As a comparison and further analysis, published predictive equations were also used to estimate %BF in the same validation data set. These prediction formulas were mainly derived from white populations (Lean et al., for men, 0.353 WC + 0.756 TSF + 0.235 age - 26.4; for women, 0.232 WC + 0.657 TSF + 0.215 age - 5.5; and Deurenberg et al., both sexes, 1.2 BMI + 0.23 age - 10.8 sex - 5.4 (men = 1, women = 0); for men, 1.2 BMI + 0.23 age - 16.2; for women, 1.2 BMI + 0.23 age - 5.4) (12,13) and Chinese populations (Deurenberg et al., both sexes, 1.45 BMI + 0.11 age - 10.4 sex - 5.9 (men = 1, women = 0); for men, 1.45 BMI + 0.11 age - 16.3; for women, 1.45 BMI + 0.11 age - 5.9) (14).

Results

In either development or validation samples, women tended to have higher anthropometric values (except BW) and BF than men, whether expressed as percentages or absolute weight (Table 1). There were no significant differences in any anthropometric variables between the two samples in either men or women.

Table 1. - Demographic, anthropometry, and body composition data stratified by sex.

Full table

Using the World Health Organization (WHO) BMI-based diagnostic criteria (39) for overweight and obesity (BMI 25 and 30 kg/m²), the prevalence of overweight and obesity was 15% and 0%, respectively. In contrast, the WHO abdominal obesity-based criteria (WC 102 cm in men and 88 cm in women) (39) suggested a prevalence of abdominal obesity of 1% and 20%, respectively. However, when the prevalence of obesity was based on %BF (25% in men and 35% in women) (7,39,40), a significantly higher prevalence was observed (i.e., 9.2% and 40% in men and women, respectively).

The correlations among the individual anthropometric variables and %BF are presented in Table 2. The correlations in women were by and large higher than in men. SSF and BMI were highly correlated in men (r = 0.74 and 0.72, p < 0.01), whereas all of the anthropometric variables (except TSF, BSF, and WHR) were highly correlated in women (r = 0.71 to 0.78, p < 0.01).

Table 2. - Correlations between individual anthropometry variables and %BF in men and women.

Full table

Predictive Equations Using Anthropometric Variables

The step-wise multiple linear regression analysis suggested that in men, SSF, BMI, BSF, and WC were significant predictors of %BF, with the equation being: %BF in men = 0.417 SSF + 0.621 BMI - 0.276 BSF + 0.166 WC - 18.466 (R² = 0.68, p < 0.01). However, in women, BMI, HC, and SISF were found to be significant predictors of %BF, with the equation being %BF = 0.417 HC + 0.172 SISF + 0.46 BMI - 23.748 (R² = 0.68, p < 0.01).

When the above equations were validated in the validation sample, the correlation coefficient of the predicted and DXA-measured %BF was r = 0.82 (p < 0.01) (Figure 1). There was no significant difference between the predicted and DXA-measured %BF. The Bland and Altman analysis indicated that in men, the mean difference was 0.03%, with the limit of agreement being between -8.2% and 8.3%; in women, the mean difference was -0.01%, with the limit of agreement being between -9.0% and 9.0% (Figure 2).

Figure 1.

Scatter plots between %BF from prediction equation and DXA in (A) men and (B) women.

Full figure and legend (45K)

Figure 2.

Agreement assessment between %BF estimated by prediction equations and DXA in (A) men and (B) women. Means are plotted against the difference between the two methods. The center line represents the mean difference between the two methods, and the other two lines represent two SDs from the mean.

Full figure and legend (60K)

When the published predictive equations (12,13,14) were applied to the same validation data set, the mean differences between predicted and DXA-measured %BF were greater than those derived from the equations developed in the present study (Table 3; Figure 3).

Figure 3.

Comparison of predicted %BF for prediction equations A to C and DXA. Means are plotted against the difference between the two methods. The center line represents the mean difference between the two methods, and the other two lines represent two SDs from the mean.

Full figure and legend (81K)

Table 3. - Correlation coefficients (r), mean difference, and limit of agreement of %BF predicted by different equations compared with %BF measured by DXA.

Full table

Predictive Equations Using BMI and Age

As a further analysis, we were also interested in deriving predictive equations of %BF based on simple and readily available measures such as BMI and age because these equations can also be very helpful in a community setting. The multiple linear regression analysis was used to estimate regression parameters associated with these variables. The predictive equation was: %BF = 1.648 BMI + 0.065 age - 15.3 sex - 10.672, where in sex man was coded 1 and woman 0. The equation accounted for 83% variance of the measured %BF.

When the equation was validated in the validation data set, the correlation between the equation-based predicted and DXA-measured %BF was 0.93 (p < 0.01) (Figure 4). The mean difference was 0.16%, with 95% confidence interval ranging from -0.75% to 0.44%. The limits of agreement (Bland and Altman analysis) ranged from -8.9% to 8.6%. When published predictive equations (13,14) were applied to the same data set, they tended to overestimate the %BF; as a result, the limits of agreement (or differences between DXAmeasured and equation-predicted %BF) were higher than those derived from the present study's equations (Table 4; Figure 5).

Figure 4.

Scatter plots between %BF from prediction equation and DXA in (A) men and (B) women.

Full figure and legend (45K)

Figure 5.

Agreement assessment between %BF estimated by prediction equations and DXA in (A) the Thai population, (B) white population, and (C) Chinese population. Means are plotted against the difference between the two methods. The center line represents the mean difference between the two methods, and the other two lines represent two SDs from the mean.

Full figure and legend (62K)

Table 4. - Correlation coefficients (r), mean difference, and limit of agreement of %BF predicted by different equations based on BMI variable compared with %BF measured by DXA.

Full table

Diagnostic Performance

The sensitivity, specificity, PPVs, and negative predictive values of the predictive equations in terms of assessing the probability of obesity (defined as measured %BF being >25% in men and 35% in women) are shown in Table 5. In women, the prevalence of obesity in the validation sample was 35.4%, and if the anthropometry-based predictive equations were used to diagnose obesity, the sensitivity was 0.70, but the PPV was 0.91; these values were virtually identical to those values derived from the BMI-based equation. In men, because the prevalence of obesity was low (3.6%), the diagnostic measures were not reliable, despite the perfect PPV value. The diagnostic performance of the present study's equations was consistently better than Lean et al.'s or Deurensberg et al.'s equations.

Table 5. - Diagnostic performance of various predictors of %BF in the validation sample.

Full table

Discussion

BF mass is probably an appropriate and important measure for defining obesity in Asian populations (7,8,9,10). Indeed, in the present study population, BMI-based definition of obesity appeared to underestimate the prevalence of obesity compared with percentage fat mass-based definition. Despite many techniques that have been developed to assess body composition, few are applicable or suited to clinical practice; that is, requiring only simple equipment and relatively inexpensive, noninvasive, and nontime-consuming. Results of the present study suggest that it is possible to use simple, noninvasive anthropometric variables to predict %BF, which can, in turn, be used to assess the degree of obesity in an individual subject.

The combination of BMI, SSF, BSF, and WC could provide a reasonably good estimate of %BF in men, whereas the combination of BMI, HC, and SISF could also yield similar predictive performance in women. The predicted %BF derived from these equations was highly concordant with the DXA-measured values. The equations also yielded no significant bias and favorable limits of agreement. This suggests that the equations can be used in a community or clinical setting when DXA instruments are not available.

In the past, the Lean and Deurenberg equations have been widely used for predicting %BF in Asian populations (12,13,14). However, the present study suggested that in the Thai population, the performance of these equations was not as good as the predictive equations developed in the Thai population. This may suggest that the predictive equations have to be ethnicity-specific to be useful. This is likely due to the difference in fat mass and fat distribution among ethnicities that have been regarded as a potential caveat in applying various predictive equations (7,8,9,26,27).

BMI is usually considered a surrogate marker of excess BW, especially from excess adiposity. However, it is questionable whether BMI can be used to determine %BF in the population. This study attempted to develop the prediction equation for %BF using BMI as a tool for obesity study in epidemiological research and in public health applications. Multiple linear regression analysis suggested that BMI, age, and sex could explain 83% of the variation in %BF. The performance of this BMI- and age-based equation was also good, and this suggests that this simpler equation could be used as an alternative for predicting percentage fat mass in the Thai population.

A recurrent issue of prediction is the applicability of predictive equations to an individual subject. If the %BF-based WHO criteria is used to diagnose obesity, the predictive equations developed in this study have good PPVs (above 90%), sensitivity (>70%), and specificity (>90%). These figures are encouraging, and they suggest that the equations can be used to aid the diagnosis of obesity in an individual subject.

Nevertheless, the present findings must be interpreted within the context of a number of potential strengths and weaknesses. A major strength of this study lies in its validity and sampling scheme. The measurement of BF and fat-free mass in this study was based on the DXA instrument, which is considered to be one of the most accurate and valid methods of measurement. The sample size was reasonably large to allow for a stable estimation of relations between BF and BMI. Despite that the subjects in this study were randomly selected and well characterized, the study subjects were Thai, among whom body size, lifestyles, cultural backgrounds, and environmental living conditions are different from other populations. Thus, care should be taken when extrapolating these results to other populations. The measurement error of BF could result in misclassification of obesity, and BW was measured at a single time-point that may not reflect a true long-term weight of a subject. These two sources of measurement errors, albeit inevitable, could have affected the result, despite the fact that such a limitation is present in any study of this type.

In summary, we have developed and validated a set of equations based on anthropometric variables for predicting %BF in the Thai population. The performance of these equations was good such that they can be used in clinical setting where DXA is not available.

Notes

¹ Nonstandard abbreviations: BF, body fat; %BF, percentage BF; BW, body weight; WC, waist circumference; HC, hip circumference; BSF, biceps skinfold; TSF, triceps skinfold; SSF, subscapular skinfold; SISF, suprailiac skinfold; PPV, positive predictive value; WHO, World Health Organization.

References

1. Hodge, A. M., Zimmet, P. Z. (1994) The epidemiology of obesity. Baillieres Clin Endocrinol Metab 8: 577–599. | Article | PubMed | ISI | ChemPort |
2. Manson, J. E., Willet, W. C., Stampfer, M. I. (1995) Body weight and mortality among women. N Eng J Med 333: 677–685.
3. Bray, G. A. (1996) Health hazards of obesity. Endocrinol Metab Clin North Am 25: 907–919. | Article | PubMed | ChemPort |
4. Troiano, R. P., Frongillo, E. A., Jr, Sobal, J., Levitsky D. A. (1996) The relationship between body weight and mortality: a quantitative analysis of combined information from existing studies. Int J Obes Relat Metab Disord 20: 63–75. | PubMed | ChemPort |
5. Must, A., Spadano, J., Coakley, E. H., Field, A. E., Colditz, G., Dietz, W. H. (1999) The disease burden associated with overweight and obesity. JAMA 282: 1523–1529. | Article | PubMed | ISI | ChemPort |
6. Allison, D. B., Frontaine, K. R., Manson, J. E., Stevens, J., VanItallie, T. B. (1999) Annual deaths attributable to obesity in the United States. JAMA 282: 1530–1538. | Article | PubMed | ISI | ChemPort |
7. Deurenberg, P., Deurenberg-Yap, M., van Staveren, W. A. (1998) Body mass index and percent body fat: a meta analysis among different ethnic groups. Int J Obes Relat Metab Disord 22: 1164–1171. | Article | PubMed | ChemPort |
8. Wang, J., Thorton, J. C., Russell, M., Burastero, S., Heymsfield, S. B., Pierson, R. N. Jr (1994) Asians have lower body mass index (BMI) but higher percent body fat than do whites: comparisons of anthropometric measurements. Am J Clin Nutr 60: 23–28. | PubMed | ISI | ChemPort |
9. Guricci, S., Hartriyanti, Y., Hautvast, JGAJ, Deurenberg, P. (1998) Relationship between body fat and body mass index: differences between Indonesians and Dutch Caucasians. Eur J Clin Nutr 52: 779–783. | Article | PubMed | ChemPort |
10. WHO Expert Consultation (2004) Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies. Lancet 363: 157–163. | Article | PubMed | ISI |
11. Ellis, K. J. (2001) Selected body composition methods can be used in field studies. J Nutr 131: 1589–1595.
12. Lean, MEJ, Han, T. S., Deurenberg, P. (1996) Predicting body composition by densitometry from simple anthropometric measurements. Am J Clin Nutr 63: 4–14. | PubMed | ISI | ChemPort |
13. Deurenberg, P., Westrtrate, J. A., Seidell, J. C. (1991) Body mass index as a measure of body fatness: age and sex-specific prediction formulas. Br J Nutr 65: 105–114. | Article | PubMed | ISI | ChemPort |
14. Deurenberg, P., Keyou, G. e., Hautvast, JGAJ, Jingzhong, W. (1997) Body mass index as predictor for body fat: comparison between Chinese and Dutch adult subjects. Asia Pacific J Clin Nutr 6: 102–105.
15. Durnin, JVGA, Womersley, J. (1974) Body fat assessed from total body density and its estimation from skinfold thickness: measurement on 481 men and women aged from 16 to 72 years. Br J Nutr 32: 77–79. | Article | PubMed | ISI | ChemPort |
16. Svendsen, O. L., Haarbo, J., Hassager, C., Christiansen, C. (1993) Accuracy of measurements of body composition by dual energy x-ray absorptiometry in vivo. Am J Clin Nutr 57: 605–608. | PubMed | ISI | ChemPort |
17. Deurenberg, P., van de Kooij, K., Evers, P., Hulshof, T. (1990) Assessment of body composition by bioelectrical impedance in a population aged >60 y. Am J Clin Nutr 51: 3–6. | PubMed | ISI | ChemPort |
18. Teran, J. C., Sparks, K. E., Quinn, L. M., Fernandez, B. S., Krey, S. H., Steffee, W. P. (1991) Percent body fat in obese white female predicted by anthropometric measurements. Am J Clin Nutr 53: 7–13.
19. Durnin, JVGA, Rahaman, M. M. (1967) The assessment of the amount of fat in the human body from measurements of skinfold thickness. Br J Nutr 21: 681–689. | Article | PubMed | ISI | ChemPort |
20. Jackson, A. S., Pollock, M. L., Ward, A. (1980) Generalized equations for predicting body density of women. Med Sci Sports Exerc 12: 175–182. | PubMed | ChemPort |
21. Sloan, A. W., Burt, J. J., Blyth, C. S. (1962) Estimation of body fat in young women. J Appl Physiol 17: 967–970. | PubMed | ISI |
22. Wilmore, J. H., Behnke, A. R. (1970) An anthropometric estimation of body density and lean body weight in young women. Am J Clin Nutr 23: 267–274. | PubMed | ISI | ChemPort |
23. Katch, F. I., McArdle, W. D. (1973) Prediction of body density from simple anthropometric measurements in college-age men and women. Hum Biol 45: 445–454.
24. Ortiz, O., Russell, M., Daley, T. L., et al. (1992) Differences in skeletal muscle and bone mineral mass between black and white females and their relevance to estimates of body composition. Am J Clin Nutr 55: 8–13. | PubMed | ISI | ChemPort |
25. Wang, J., Deurenberg, P. (1996) The validity of predicted body composition in Chinese adults from anthropometry and bioelectrical impedance in comparison with densitometry. Br J Nutr 76: 175–182. | PubMed |
26. Dierkes, J., Schultink, J. W., Gross, R., Praestowo, SMB, Pietrzik, K. (1993) Body composition of Indonesian adults assessed by skinfold thickness and bioelectrical impedance measurements and by a body mass index equation. Asia Pacific J Clin Nutr 2: 171–176.
27. Küpper, J., Bartz, M., Schultink, J. W., Lukito, W., Deurenberg, P. (1998) Measurements of body fat in Indonesian adults: Comparison between a three-compartment model and widely use methods. Asian Pacific J Clin Nutr 7: 49–54.
28. Lanham, D. A., Stead, M. A., Tsang, K., Davies, P. S. (2001) The prediction of body composition in Chinese Australian females. Int J Obes Relat Metab Disord 25: 286–291. | Article | PubMed | ChemPort |
29. Kwok, T., Woo, J., Lau, E. (2001) Prediction of body fat by anthropometry in older Chinese people. Obes Res 9: 97–101.
30. Demura, S., Yamaji, S., Goshi, F., Kobayashi, H., Sato, S., Nagasawa, Y. (2002) The validity and reliability of relative body fat estimates and the construction of new prediction equations for young Japanese adult males. J Sports Sci 20: 153–164.
31. Cohn, S. H. (1977) Body elemental composition: comparison between black and white adults. Am J Physiol 232: 419–422.
32. Gasperino, J. A., Wang, J., Pierson, R. N., Heymsfield, S. B. (1995) Age-related changes in musculoskeletal mass between black and white women. Metabolism 44: 30–34. | PubMed |
33. Robson, JRK, Bazin, M., Soderstrom, R. (1971) Ethnic differences in skinfold thickness. Am J Clin Nutr 24: 864–868.
34. Zillikens, M. C., Conway, J. M. (1990) Anthropometry in blacks: applicability of generalized skinfold equations and differences in fat patterning between black and whites. Am J Clin Nutr 52: 45–51. | PubMed |
35. Heymsfield, S. B., Baugartner, R. N., Pan, S. F. (1999) Nutrition assessment of malnutrition by anthropometric methods. Shils, ME eds. Modern Nutrition in Health and Disease 903–922.Williams and Wilkins Baltimore, MD.
36. Mazess, R. B., Barden, H. S., Bisek, J. P., Hanson, J. (1990) Dual-energy x-ray absorptiometry for total-body and regional bone-mineral and soft tissue composition. Am J Clin Nutr 51: 1106–1112. | PubMed | ISI | ChemPort |
37. Russell-Aulet, M., Wang, J., Thornton, J., Pierson, R. N. Jr (1991) Comparison of dual-photon absorptiometry systems for total-body bone and soft tissue measurements: dual-energy X-rays versus gadolinium 153. J Bone Miner Res 6: 411–415. | PubMed |
38. Bland, J. M., Altman, D. G. (1986) Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 307–310. | Article | PubMed | ISI | ChemPort |
39. WHO (1995) Physical Status: The Use and Interpretation of Anthropometry. Technical Report Series 854 WHO Geneva, Switzerland.
40. AACE/ACE Obesity Task Force (1998) AACE/ACE position statement on the prevention, diagnosis, and treatment of obesity. Endocr Pract 4: 297–330.

Acknowledgments

The study was supported by Thailand Research Fund. The authors thank Bryan Roderick Hamman for assistance with the English language presentation of the manuscript.