Development and Cross-Validation of Prediction Equation of Body-Fatness from NHANES

Background/Purpose: Obesity is one of the world’s leading health concerns, requiring field researchers to develop accurate measures of body fatness. The use of prediction equations based on anthropometric measures has been practically advocated in large-scale epidemiological studies; however, few have been developed using a nationally representative sample of U.S. adults. Therefore, the aim of this study is to develop and cross-validate a body fat percentage (BFP) prediction equation from a U.S. national representative sample of adults.

Method: Data from the 1999-2006 National Health and Nutrition Examination Survey (NHANES) were analyzed for this study. A preliminary prediction equation was developed using the 1999-2004 NHANES data which included 10,684 adults (≥20 years, male = 5,450). Body mass index (BMI) measured by height (cm) and weight (kg), waist circumference and demographic characteristics such as age, gender, and race/ethnicity were considered as predictors. BFP was measured by Dual-energy X-ray absorptiometry (DXA), the criterion measure. The preliminary equation was then cross-validated using data from the 2005-2006 NHANES which included 2,845 adults (≥20 years, male = 1,472). The final equation was developed from the total sample. Regression analyses were applied to develop and cross-validate the preliminary and final prediction equations. The accuracy of prediction equations was evaluated by R², root mean square error (RMSE), and mean signed difference (MSD) against BFP.

Analysis/Results: The average age, BMI, and BFP in the total sample were 45.45 ± .26 years, 27.90 ± .09 kg/m², and 33.72 ± .12%, respectively. Inversed body mass index (1/BMI), age, gender, and race/ethnicity were retained in the preliminary prediction equation. R², RMSE, and MSD of the preliminary equation were .84, 3.50%, and -.01%, respectively. In a cross-validation sample, R², RMSE, and MSD were .84, 3.49%, and -.57%, respectively. The final equation based on total sample was 61.93-680.06×(1/BMI)-12.05×(gender)+.08×(age)+[-1.95(African American (AA)) or .36(Mexican  American (MA))×(ethnicity)] where female = 0 and male = 1 for gender, age in years, and Caucasian American = 0, AA = 1, and MA = 2 for ethnicity, explaining 84% of variance in BFP with 3.50% and -.15% of RMSE and MSD, respectively.

Conclusions: The developed equation to predict BFP had strong R² and moderate RMSE in a nationally representative sample. This equation can be a useful and practical method for estimating BFP with reasonable accuracy for U.S. adults.