AbstractAimsSingle-frequency segmental Bioelectrical Impedance Analysis (BIA) is commonly used to estimate body composition. To enhance the value of information derived from BIA, especially for use in large-scale epidemiological studies, we developed and validated equations to predict total and regional (arms, legs, trunk, android, gynoid, visceral) body composition parameters (lean mass and fat mass) from anthropometry and single-frequency (50 kHz) segmental BIA variables, using Dual Energy X-ray Absorptiometry (DEXA) as the criterion method.MethodsThe 11,559 adults (age 30 to 65) from the UK population-based Fenland Study with data on DEXA, BIA and anthropometry were randomly assigned to a Derivation sample (4,827 men; 5,732 women) or a Validation sample (500 men; 500 women). Prediction equations based on anthropometry and BIA variables were derived using forward stepwise multiple linear regression in the Fenland Derivation sample. These were validated in the Fenland Validation sample and also in the UK Biobank Imaging Study (2,392 men; 2,606 women) using Pearson correlations and Bland–Altman models.Results and ConclusionsBland Altman analyses revealed no significant mean bias for any predicted DEXA parameter (all P>0.05) for the fenland population. Bias expressed as % of the mean was between -0.6% and 0.5% for all parameters in both men and women, except for visceral FM and subcutaneous abdominal FM (range -3.6 to 1.1%). However, in UK Biobank most predicted parameters showed significant bias: % mean bias was <2% in both sexes only for total fat mass and total lean mass, and was >10% for leg and visceral fat mass in both sexes. In conclusion, new equations based on anthropometry and BIA variables predicted DEXA parameters with sufficient accuracy to assess relative differences between individuals, and were sufficiently accurate to predict absolute values for total body but not regional fat and lean mass.