Validation of bio-impedance spectroscopy: Effects of degree of obesity and ways of calculating volumes from measured resistance values

BACKGROUND: Bioelectrical-impedance spectroscopy (BIS) is a very attractive method for body composition measurements in a clinical setting. However, validation studies often yield different results. This can partly be explained by the different approaches used to transform measured resistance values into body compartments. OBJECTIVE: The aim of this study was to compare the linear regression (LR) method with the Hanai Mixture theory (HM). Secondly, the effect of degree of overweight on the accuracy of BIS was analysed. DESIGN: In 90 people (10 M, 80 F; body mass index (BMI) 23 - 62 kg/m(2)) total body water (TBW) and extracellular water (ECW) were measured by deuterium and NaBr dilution methods, respectively, and by BIS. Resistance values of ECW (R-ECW) and TBW (R-TBW) were used for volume calculations. Data of half the group were used for LR based on L-2/R (L = length, R = resistance) to predict TBW and ECW and to calculate the constants used in the HM (k(ECW)), k(p)) Prediction equations and constants were cross-validated in Group 2. RESULTS: Bland and Altman analysis showed that the LR method underestimated TBW by 1.1 I (P < 0.005) and ECW by 1.11 (P < 0,005). The HM approach underestimated ECW by 0.81 (P < 0,005), The correlations with the dilution methods and the SEEs for TBW and ECW were comparable for the two approaches. The prediction error of BIS for TBW and ECW correlated with BMI. The constant kECW, and the specific resistivities of the ECW and intracellular water (ICW) P-ICW and P-ICW were also correlated with BMI. CONCLUSIONS: The mixture approach is slightly more accurate than linear regression, but not sensitive enough for clinical use. The constants used in the HM model are not constants in a population with a wide variation in degree of overweight. The physical causes of the correlation between BMI and constants used in the model should be studied further in order to optimize the mixture model.