Comparison of different bioelectrical impedance analyzers in the prediction of body composition

To investigate the influence of different bioelectrical impedance (BI) analyzers on the prediction of body composition from bioelectrical resistive impedance (R), 146 healthy white adults (73 men; 73 women) were studied at two independent laboratories: The University of Florida (UF) and the USDA, San Francisco. Whole body R was measured on each subject with three different BI analyzers. AT UF analyzers were: Valhalla Scientific model 1990-A (VH), RJL Systems model BIA-101 (RJL), and Medi-Fitness model 1000 (MF). At USDA analyzers were: VH, RJL, and Bioelectrical Sciences model 200Z (BES). The largest difference in R (36 ohms, P ⩽ 0.01) was noted between BES and VH at USDA. When applied to current BI prediction equations, the observed differences among analyzers resulted in differences in predicated % fat of up to 6.3% although most comparisons among mean values (79%) showed differences below 3%. Crossvalidation of the selected BI prediction equations with hydrostatistically determines body composition using the different R values revealed total errors of prediction (E) ranging from 3.6 to 9.8% fat. The prediction equations were most accurate when used with data collected on the same instrument that was to used to develop the equation (E = 3.6 to 5.3% fat). These findings indicate that different analyzers can be a significant source of variation when predicting body composition from R. To minimize this source of variation, it is recommended that BI prediction equations be used with the same type of instrument as that with which they were developed.