Computed Tomography and Dual‐Energy X‐Ray Asorptiometry body composition parameter harmonisation to universalise adipose tissue measurements in a population‐based cross‐sectional study

Abstract

SummaryTo harmonise computed tomography (CT) and dual‐energy x‐ray absorptiometry (DXA) body composition measurements allowing easy conversion in longitudinal assessments and across cohorts to assess cardiometabolic risk and disease. Retrospective cross‐sectional observational study from 1996 to 2008 included participants in the Pennington Center Longitudinal Study (PCLS) (N = 1967; 571 African American/1396 White). Anthropometrics, whole‐body DXA and abdominal CT images were obtained. Multi‐layer segmentation techniques (Analyze; Rochester, MN) quantified visceral adipose tissue (VAT). Clinical biomarkers were obtained from routine blood samples. Linear models were used to predict CT‐VAT from DXA‐VAT and examine the effects of traditional biomarkers on cross‐sectional‐VAT. Predicted CT‐VAT was highly associated with measured CT‐VAT using ordinary least square linear regression analysis and random forest models (R2 = 0.84; 0.94, respectively, p < .0001). Model stratification effects showed low variability between races and sexes. Overall, associations between measured CT‐VAT and DXA‐predicted CT‐VAT were good (R2 > 0.7) or excellent (R2 > 0.8) and improved for all stratification groups except African American men using random forest models. The clinical effects on measured CT‐VAT and DXA‐VAT showed no significant clinical difference in the measured adipose tissue areas (mean difference = 0.22 cm2). Random forest modelling seamlessly predicts CT‐VAT from measured DXA‐VAT to a degree of accuracy that falls within the bounds of universally accepted standard error.