Objectives: Assessment of the precision and accuracy of dual-energy X-ray absorptiometers (DXA) from three manufacturers, used for measuring soft-tissue composition, and comparability with each other and other techniques.
Design: Measurements of an anthropomorphic model of variable composition and thickness. 11 volunteers measured with each instrument and by underwater weighing (UWW) and three brands of bioelectric impedance analysis (BIA) apparatus.
Results: New software, introduced by each manufacturer during the course of the investigation, led to changes in measured fat proportion. The precision of determination of fat proportion by DXA in the (small) model was 3-4% (coefficient of variation), with little difference between brands. In vivo precision was 2-3%. In the model, measurements of % fat differed from the nominal values, but variation with thickness was small. There were significant mean differences of total fat proportion in the volunteers between pairs of DXA instruments of 2.6-6.3% fat. The SDs of the differences were 1.8-2.9% fat. Regional differences were greater, with trunk % fat being particularly underestimated by Hologic relative to Lunar and Norland. Compared with UWW, mean % fat was the same for Hologic, but higher for Lunar and Norland. SDs of 4% demonstrated inadequate agreement. The differences varied with proportion of bone in lean tissue, questioning the assumption of constant density of lean tissue in UWW. There were no mean differences of % fat between the BIA instruments and DXA and UWW, but SDs of 3-6% suggest that BIA using these instruments does not offer an acceptable accuracy in estimating fat proportion.
Conclusions: UWW has limitations as a reference method. DXA is a useful technique, but its limitations, particularly regarding assumptions about fat distribution, must be borne in mind. The differences of fat proportion recorded by the three DXA instruments are such as to preclude interchangeability in measurements of individual subjects or in clinical trials.