Reference method matters: DXA system differences influence agreement with 3D optical imaging body composition estimates

BACKGROUND: Dual-energy X-ray absorptiometry (DXA) is often used as the reference method for calibrating and validating other clinically useful body composition devices, although previous studies report DXA system differences in component estimates. This study evaluated the hypothesis that body composition estimates, specifically percent body fat (BF%), acquired with a method such as 3D optical (3DO) imaging might lead to varying conclusions when judged against different DXA systems and software versions. METHODS: BF% was evaluated in healthy adult participants with a 3DO imaging system (Size Stream, Mobile Fit) and three DXA systems (Hologic Horizon A [Apex 4.0.2] and Discovery A [Apex 5.6.1.3]), and GE Lunar iDXA [enCore 13.60.033] on the same day. Agreement between BF% estimates by 3DO and each DXA system was evaluated with multiple statistical procedures. RESULTS: s), Bland-Altman analyses, mean differences, mean absolute errors, root-mean square errors, and concordance correlations differed between %BF by 3DO and each of the DXA devices. Relative to BF% by 3DO, agreement measures were best for Discovery A compared to the other two DXA scanners. A linear mixed-effects model with method as a fixed effect and participant as a random intercept revealed a statistically significant effect of assessment method on estimated BF%: Horizon A DXA (X ± SD, 36.4 ± 9.3%) differed significantly from 3DO (33.3 ± 8.8%), iDXA (35.2 ± 10.9%), and Discovery A (34.4 ± 10.0%); and iDXA and Discovery A both differed significantly from 3DO and Horizon A DXA. Significant interactions between method and sex and method and BMI category were also observed, indicating the differences between measurement methods were not uniform across males and females or BMI categories. CONCLUSIONS: Our findings indicate that conclusions related to a clinical body composition method's accuracy depend on the specific DXA device and associated software version chosen to serve as the reference method.