Bioelectrical impedance analysis as an alternative to dual-energy x-ray absorptiometry in the assessment of fat mass and appendicular lean mass in patients with obesity

Obesity is a challenge for bioelectrical impedance analysis (BIA) estimations of skeletal muscle and fat mass (FM), and none of the equations used for appendicular lean mass (ALM) have been developed for people with obesity. By using different equations and proposing a new equation, this study aimed to assess the estimation of FM and ALM using BIA compared with dual-energy x-ray absorptiometry (DXA) as a reference method in a cohort of people with severe obesity.This cross-sectional study compared a multifrequency BIA (TANITA MC-780A) versus DXA for body composition assessment in adult patients with severe obesity (body mass index [BMI] of >35 kg/m2). Comparisons between measured (DXA) and predicted (BIA) data for FM and ALM were performed using the original proprietary equations of the device and the equations proposed by Kyle, Sergi, and Yamada. Bland-Altman plots were drawn to evaluate the agreement between DXA and BIA, calculating bias and limits of agreement (LOA). Reliability was analyzed using intraclass correlation coefficient (ICC). Stepwise multiple regression analysis was used to derive a new equation to predict ALM in patients with obesity and was validated in a subsample of our cohort.In this study, 115 patients (72.4% women) with severe obesity (mean BMI of 46.1 [5.2] kg/m2) were included (mean age 43.5 [8.6] y). FMDXA was 61.4 (10.1) kg, FMBIA was 57.9 (10.3) kg, and ICC was 0.925 (P < 0.001). Bias was -3.4 (4.4) kg (-5.2%), and LOA was -14.0, +7.3 kg. Using the proprietary equations, ALMDXA was 21.8 (4.7) kg and ALMBIA was 29.0 (6.8) kg with an ICC 0.868, bias +7.3 (4.0) kg (+34.1%) and LOA -0.5, +15.1. When applying other equations for ALM, the ICC for Sergi, et al. was 0.880, the ICC for Kyle, et al. was 0.891, and the best ICC estimation for Yamada, et al. was 0.914 (P < 0.001). Bias was +2.8 (2.8), +4.1 (2.9), and +2.7 (2.8) kg, respectively. The best-fitting regression equation to predict ALMDXA in our population derived from a development cohort (n = 77) was: ALM = 13.861 + (0.259 x H2/Z) - (0.085 x age) - (3.983 x sex [0 = men; 1 = women]). When applied to our validation cohort (n = 38), the ICC was 0.864, and the bias was the lowest compared with the rest of the equations +0.3 (+0.5) kg (+2.7%) LOA -5.4, +6.0 kg.BIA using multifrequency BIA in people with obesity is reliable enough for the estimation of FM, with good correlation and low bias to DXA. Regarding the estimation of ALM, BIA showed a good correlation with DXA, although it overestimated ALM, especially when proprietary equations were used. The use of equations developed using the same device improved the prediction, and our new equation showed a low bias for ALM.