| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
ORIGINAL RESEARCH COMMUNICATION |
1 From the Institute of Nutrition of Central America and Panama, Guatemala City, Guatemala (MR-Z and BT), and the Rollins School of Public Health, Emory University, Atlanta, GA (ADS and RM)
2 Supported by grants no. TW-05598 and HD-046125 from the National Institutes of Health and by the Nestlé Foundation (Lausanne, Switzerland).
3 Reprints not available. Address correspondence to M Ramirez-Zea, Institute of Nutrition of Central America and Panama (INCAP), PO Box 1188, Calzada Roosevelt, Zona 11, Guatemala City, Guatemala 01011. E-mail: mramirez{at}incap.ops-oms.org.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Objective: We developed equations to predict %BF from anthropometric measurements in rural and urban Guatemalan adults.
Design: Body density was measured in 123 women and 114 men by using hydrostatic weighing and simultaneous measurement of residual lung volume. Anthropometric measures included weight (in kg), height (in cm), 4 skinfold thicknesses [(STs) in mm], and 6 circumferences (in cm). Sex-specific multiple linear regression models were developed with %BF as the dependent variable and age, residence (rural or urban), and all anthropometric measures as independent variables (the "full" model). A "simplified" model was developed by using age, residence, weight, height, and arm, abdominal, and calf circumferences as independent variables.
Results: The preferred full models were %BF = –80.261 – (weight x 0.623) + (height x 0.214) + (tricipital ST x 0.379) + (abdominal ST x 0.202) + (abdominal circumference x 0.940) + (thigh circumference x 0.316); root mean square error (RMSE) = 3.0; and pure error (PE) = 3.4 for men and %BF = –15.471 + (tricipital ST x 0.332) + (subscapular ST x 0.154) + (abdominal ST x 0.119) + (hip circumference x 0.356); RMSE = 2.4; and PE = 2.9 for women. The preferred simplified models were %BF = –48.472 – (weight x 0.257) + (abdominal circumference x 0.989); RMSE = 3.8; and PE = 3.7 for men and %BF = 19.420 + (weight x 0.385) – (height x 0.215) + (abdominal circumference x 0.265); RMSE = 3.5; and PE = 3.5 for women.
Conclusion: These equations performed better in this developing-country population than did previously published equations.
Key Words: Body composition • anthropometry • adiposity • adults • body fat • obesity • developing country • Guatemala
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Equations to predict body fat from anthropometric measurements allow the estimation of body composition without complex and costly techniques. Most currently used predictive equations were derived from measurements of persons in affluent, industrialized Western societies, and they may be inappropriate for persons with other genotypic and phenotypic characteristics. For example, the equation proposed by Durnin and Womersley (9) tends to overestimate fat mass and percentage body fat (%BF) in populations of developing countries (10).
The current study was conducted to develop and test predictive equations derived from anthropometric measurements of young and middle-aged Guatemalan men and women. Some measurements, such as skinfold thicknesses (STs), require the use of sensitive, calibrated calipers by well-trained anthropometrists whose precision and accuracy in measurements were validated, and thus the use and reliability of those measurements may be inadequate in studies performed with the participation of fieldworkers with little or no anthropometric experience. We therefore developed and tested predictive equations derived from measurements that included weight, height, and several circumferences and STs (the "full" model) or equations that excluded the need for ST measurements (the "simplified" model).
|
SUBJECTS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Written informed consent was obtained from all subjects. All procedures were approved by the institutional review board of INCAP.
Hydrostatic weighing
Body density was determined at INCAP’s body-composition laboratory by using hydrostatic weighing that was corrected for residual lung volume (11). All measurements were performed in the morning; subjects were either in a fasting state or had eaten a light breakfast. Subjects were asked to abstain for 24 h from dairy products, legumes, and other foods usually associated with flatulence. They urinated and defecated or expelled flatus shortly before the underwater weighing. After being weighed on a digital scale (Toledo Scale, Division of Reliance Electronic, Worthington, OH), each subject sat on a metal and plastic-mesh chair suspended from a hanging scale with a 10-kg capacity and 10-g sensitivity (Detecto Scales Inc, Brooklyn, NY). The chair was lowered with an electric pulley into a 150 x 150 x 180-cm tank of water at 36–37 °C until only the subject’s head was above the surface. Each subject wore a tight-fitting swimming suit that was rubbed underwater to expel any trapped air bubbles. Subjects used a nose clip and breathed through a mouthpiece connected to a 3-way valve. After the subject’s head was submerged by reclining the back of the chair, the subject exhaled completely and held his or her breath for several seconds. Underwater weight was recorded after the forced exhalation, and the chair was straightened to raise the subject’s head above the water surface. This procedure was repeated until 3 weights were recorded within 50 g. When the third weight replicate was obtained within that range, the 3-way valve was switched to connect the submerged subject’s mouthpiece to a 9-L respirometer with a helium catharometer (W.E. Collins Inc, Boston, MA). The subject’s head was raised above the water surface, and he or she breathed quietly through the mouthpiece for 5–7 min until residual lung volume was calculated by helium dilution (12). Water temperature was recorded to 0.1 °C. Body volume was calculated from the difference between the subject’s dry and underwater weights (mean of the 3 measurements) and from water density at the measured temperature. Residual lung volume plus 100 mL to compensate for intestinal gas was subtracted from body volume (11). Body density was then calculated and %BF was ascertained by using the equation of Siri (13).
Anthropometric measurements
Two trained anthropometrists whose skills had been validated obtained all measurements in triplicate by using standard techniques (14, 15). Body weight was recorded to the nearest 0.01 kg on a calibrated digital balance (Toledo Scale). Height was measured to the nearest 1.0 mm with the use of a measuring tape while the subject was standing on bare feet with head, shoulders, buttocks, and heels leaning against a surface that was at a 90° angle to the floor. STs were measured to the nearest 0.1 mm with the use of a Holtain caliper (Holtain Ltd, Crosswell, United Kingdom) at tricipital, subscapular, suprailiac, and abdominal sites on the nondominant side of the body (14). A flexible metal measuring tape was used to measure circumferences to the nearest 1.0 mm at the midupper arm, midthigh, and calf on the nondominant side of the body and at the natural waist (smallest midabdominal circumference), abdomen at the umbilical level, and hip at the greater trochanter level (15). If the difference between the 3 measurements was > 0.5 kg for body weight, > 0.5 cm for height and circumferences, and > 0.5 mm for STs, a fourth measurement was obtained, and the mean of the 3 closest measurements was used. In 1%, 2%, 5%, 6%, 7%, and 27% of cases, a fourth measurement was required for midupper-arm circumference; body weight, height, and calf circumference; abdominal and thigh circumferences; tricipital ST and abdominal circumference; subscapular ST and natural waist circumference; and suprailiac and abdominal STs, respectively.
Statistical analysis
Means and SDs were calculated for continuous variables. Differences between urban and rural men and between urban and rural women were compared by using Student’s t test. Predictive equations were derived separately for men and women. The %BF was the dependent variable because it is the only major component of variation in body composition at all anthropometric sites measured. A full predictive model was explored by using age, weight, height, 4 STs, and 6 circumferences as potential predictors. A simplified model that included only age, weight, height, and 3 circumferences that were considered easy to measure under field conditions (ie, midupper arm, abdomen, and calf) was also explored.
The men and women were divided into separate model-building (
70%) and validation (30%) subsamples. Assignment to each subsample was random and was stratified by tertiles of %BF within each sex. We identified the subset of regressors that best predicted %BF by examining the exhaustive set of multiple linear regression models. Multiple linear regression models were applied to the model-building subsample and evaluated by using SAS/STAT statistical software (version 8.02; SAS Institute Inc, Cary, NC). The most parsimonious set of regressors was ascertained by identifying the models that resulted in either the lowest values for Schwartz Bayesian criterion (SBC; 16) or the lowest values for the absolute difference between the Cp statistic and the number of regressors, represented by p (17). These 2 approaches identify optimal models from nonnested subsets of potential regressors. Initially, we selected the 3 regression models with the smallest values for SBC and the 3 models with the smallest absolute values for Cp –p. The root mean square error (RMSE) was used as a measure of precision of the predictive equation (18). The selected equations were then applied to calculate the %BF of the subjects in the validation subsample, and the 3 models for each sex with the lowest mean squared prediction error (MSPR) in the validation subsample were identified. The MSPR is a better indicator than is SEE of how well the regression model predicts the results from another data set (18). The pure error (PE) was calculated as the square root of the MSPR (19).
Because rural-to-urban migration is associated with changes in lifestyle that predispose to obesity (20), we tested whether the same models might be used for persons living in urban and rural settings by adding location of residence as an independent variable in all regression models and by testing the significance of this variable. We also tested the significance of the interaction between the coefficients for anthropometric measures and urban or rural residence, by using a partial F test with the number of df accounted for by the interaction terms and the change in explanatory power. Statistical significance was declared at P < 0.05.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Selected characteristics of the subjects are shown in Table 1
. Fifty-seven men and 58 women lived in 7 rural communities located 20–60 km from Guatemala City, and 57 men and 65 women lived in Guatemala City. Urban men were older, taller, heavier, and fatter than rural men (Table 1). Urban women were older and taller than their rural counterparts, but their body composition and measurements were similar, except that rural women tended to have smaller tricipital STs and smaller arm and calf circumferences. Men and women had a broad range of %BF (4.2%–35.9% and 13.2%–52.2%, respectively).
|
. RMSE was 3.0 and 2.4 for the men and the women, respectively. Six predictors of body fat—weight, height, tricipital and abdominal STs, and abdominal and thigh circumferences—appeared in the men in all models. Rural or urban residence appeared in 4 of the 6 models. In the women, 3 STs (tricipital, abdominal, and subscapular) and the hip circumference appeared as predictors in 5 of the 6 models. Rural or urban residence was not a factor in the women in any model. The predictive parameters for the equations shown in Table 2 accounted for 86% and 88% of the variance (adjusted r2) in %BF in the men and women, respectively. To test the accuracy of the 6 predictive regressions for each sex, we calculated %BF in the validation subsample. As Table 2 shows, the PE ranged from 3.3 to 3.5 in men and from 2.9 to 3.1 in women.
|
. In men, one model had SBC and Cp –p that were among the lowest 3 for both criteria; hence only 5 equations are presented. RMSE was somewhat greater in the simplified models than in the full models (3.7 for men and 3.4 for women), and the predictive parameters accounted for 80% and 74% of variance in %BF in men and women, respectively. In men, all predictive equations involved weight and abdominal circumference, and 4 of 5 equations also involved calf circumference. All equations for women involved abdominal circumference, and 4 of 6 involved weight or arm circumference or both. When applied to the validation subsample, PE and MSPR were larger with the simplified than with the full model, and this difference was greater in women than in men.
|
The final sets of the preferred 3 full and 3 simplified predictive equations for each sex living in an urban or rural setting are shown in Table 4. The individual predicted and measured %BF values in the men and the women in the validation subsample were compared by using the first of the full and simplified regression models shown in Table 4 (Figure 1). Comparisons of the predicted and measured values using either the second or third sets of models gave similar results. Taking into account these similarities and the small differences in PE, RMSE, and adjusted r2 shown in Table 4 for the 3 full or simplified models for each sex, it is reasonable to select as the first option the equations with the lowest number of predictors. For men, the preferred full model included body weight, height, tricipital and abdominal skinfold thickneses, and abdominal and thigh circumferences, and the simplified model included only body weight and abdominal circumference. For women, the preferred full model included tricipital and subscapular STs and abdominal and hip circumferences, and the simplified model included weight, height, and abdominal circumference.
|
|
are shown in Table 5. The similarity in results supports the decision to propose the use of the equations that involve the smallest number of anthropometric predictors.
|
compares our equations with other, previously published equations (9, 21-23). One of those equations was derived from measurements in marginally undernourished Guatemalan adolescents and young adults aged 11–25 y (23). The others were derived from measurements in white adults in industrialized, affluent societies. In men, our equations and those of Lean et al (22) and Conlisk et al (23) gave mean values similar to the actual measurements of %BF, but our full model gave smaller prediction errors (MSPR and PE) and similarly or better adjusted r2. The use of the simplified model resulted in an error that is similar to that resulting from the use of the equation of Lean et al and smaller to that resulting from the use of the equation of Conlisk et al. In women, our equations were the only ones that coincided with the measured mean values, and the errors of prediction were markedly smaller than those of the other equations. |
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The first set of equations uses several STs, body segment circumferences and, in men, weight and height. Those equations have an acceptably small predictive error of 3%, and the predictors account for 86%–88% of the variance in %BF in both men and women. We suggest using the sex-specific equations that involve the smallest number of anthropometric predictors, 5 in men and 4 in women, for measurements performed by adequately trained personnel in either urban or rural persons.
STs are not obtained routinely in epidemiologic studies because of the lack of high-quality calibrated skinfold-thickness calipers. Other studies yielded questionable results because the persons who measure the STs were not carefully trained and tested in the use of such calipers. Even in the context of our carefully conducted study with highly trained anthropometrists, a high proportion of measurements exceeded our tolerable error and had to be repeated. Therefore, we decided to develop simplified predictive equations that would not require such measurements. The predictive error increased to 3.7%, which is still reasonably acceptable, and the predictors accounted for a variance of 74%–80% in %BF. In studies that involve fieldworkers with minimum training in physical anthropometry or that do not include the use of appropriate ST calipers, we suggest using the simplified sex-specific equations that rely only on body weight and abdominal circumference (and height in women).
The simplified models proposed in this study may also be useful for estimating changes in body composition during weight loss by obese persons. It has been suggested that STs are less precise in overweight persons than are circumferences during follow-up of weight loss (24, 25). Furthermore, abdominal circumference has been regarded as a good predictor of total fat in other populations (22).
Differences in the association between anthropometric measurements and body fat content have been reported across populations, and those differences alter the relation between reference values and actual measures of body composition. These differences have been related to variations in the distribution of subcutaneous fat and in body proportions between various ethnic groups (26-28). For example, American Indians, Asians, blacks, and Hispanics tend to deposit less fat on their extremities than on their trunk and tend to have more subcutaneous fat in the upper body than do whites (27-29). Although the predictive equations proposed in this report were derived among Guatemalan urban and rural subjects, they may be appropriate for adults of Amerindian-European descent in other Latin American countries. Nevertheless, they should be validated in other such populations, especially those with greater stature or different phenotypic characteristics than the participants in this study (Table 1). They also may be more appropriate than other published equations for the calculation of body fat among Amerindians from several countries, but this conjecture should be validated.
Predictive equations for body composition are population-specific because of the high degree of colinearity of anthropometric measures. This multicolinearity can make regression coefficients unstable, which results in inflation of the variance of the least-squares estimators for the regression coefficients. Our measure of optimal model selection, Cp –p, reduces this instability (17). Equations with minimum Cp values have maximum r2 values, minimum RMSE values, and a minimum bias attributable to multicolinearity. The models recommended in this study, which were based on both the Cp –p and SBC statistics, were precise (ie, they yield small RMSE values in the model-building group) and accurate (ie, the PE in the validation group was small, and the values did not differ significantly from those obtained with RMSE).
Another issue that should be resolved is whether ethnicity affects the composition of fat-free mass and consequently its density, which would cast doubt on the use of Siri’s equations for all populations (30-32). It would be desirable to clarify this issue by such means as assessing body composition with the use of techniques involving a 4-compartment model that may be less affected by ethnic differences (33).
In conclusion, %BF can be predicted with reasonable accuracy from anthropometric measurements in adults of Amerindian-European descent. Two sets of equations are proposed for application under laboratory or field conditions. These equations give better results in this type of population than do previously published equations derived from other ethnic groups.
|
ACKNOWLEDGMENTS |
|---|
MR-Z and BT devised the study and the experimental design. MR-Z directed the experimental work, participated in data analysis and interpretation, and prepared the manuscript draft. BT participated in the interpretation of the results and in manuscript review. RM was instrumental in obtaining funding, supervised the project, and participated in data interpretation and manuscript review. ADS participated in data analysis and interpretation and manuscript review. None of the authors had a personal or financial conflict of interest.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  C Corvalan, C. Gregory, M Ramirez-Zea, R Martorell, and A. Stein Size at birth, infant, early and later childhood growth and adult body composition: a prospective study in a stunted population Int. J. Epidemiol., June 1, 2007; 36(3): 550 - 557. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. O. Gregory, J. Dai, M. Ramirez-Zea, and A. D. Stein Occupation Is More Important than Rural or Urban Residence in Explaining the Prevalence of Metabolic and Cardiovascular Disease Risk in Guatemalan Adults J. Nutr., May 1, 2007; 137(5): 1314 - 1319. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
