External cross-validation of bioelectrical impedance analysis for the assessment of body composition in Korean adults

Bioelectrical impedance analysis (BIA) models must be validated against a reference method in a representative population sample before they can be accepted as accurate and applicable. The purpose of this study was to compare the eight-electrode BIA method with DEXA as a reference method in the assessment of body composition in Korean adults and to investigate the predictive accuracy and applicability of the eight-electrode BIA model. A total of 174 apparently healthy adults participated. The study was designed as a cross-sectional study. FM, %fat, and FFM were estimated by an eight-electrode BIA model and were measured by DEXA. Correlations between BIA_%fat and DEXA_%fat were 0.956 for men and 0.960 for women with a total error of 2.1%fat in men and 2.3%fat in women. The mean difference between BIA_%fat and DEXA_%fat was small but significant (P < 0.05), which resulted in an overestimation of 1.2 ± 2.2%fat (95% CI: -3.2-6.2%fat) in men and an underestimation of -2.0 ± 2.4%fat (95% CI: -2.3-7.1%fat) in women. In the Bland-Altman analysis, the %fat of 86.3% of men was accurately estimated and the %fat of 66.0% of women was accurately estimated to within 3.5%fat. The BIA had good agreement for prediction of %fat in Korean adults. However, the eight-electrode BIA had small, but systemic, errors of %fat in the predictive accuracy for individual estimation. The total errors led to an overestimation of %fat in lean men and an underestimation of %fat in obese women.