Although DXA-measured fat is strongly correlated with MRI measured adipose tissue, our main finding was that associations were biased in both HIV-infected and control populations. As the average amount of fat increases, the difference between DXA and MRI tends to increase, with DXA giving larger estimates of fat, particularly for limb fat. Since controls have more limb fat than HIV-infected subjects and controls showed a greater upward shift in DXA fat, DXA estimated a higher prevalence of peripheral lipoatrophy compared with MRI in HIV-infected subjects. Although there is no accepted cutoff that defines HIV-associated lipoatrophy, we compared the prevalence of subjects having leg SAT below the 10th percentile of controls, finding a higher prevalence of this definition of lipoatrophy by DXA compared with MRI. In contrast, differences in DXA and MRI trunk fat estimates were much smaller in all subgroups, and there were gender- and race-related differences.
DXA and MRI measure distinct, but overlapping compartments. DXA estimates fat content by tissue density, while MRI measures adipose tissue volume. In addition to cellular lipid, adipose tissue contains extracellular water (~12% of total volume in analyses of excised specimens(22
)), a small amount of intracellular water, other types of cells besides adipocytes, and extracellular solids. While these relationships may be affected as a result of fat depletion and composition changes related to lipoatrophy, this is not the complete answer, since bias was also seen in controls. Furthermore, if the inclusion of the non-lipid component in MRI analysis was the cause of the difference, one would expect MRI to give higher results than DXA, while the opposite was true.
Our finding that DXA and MRI are highly correlated but have important biases is supported by previous work in smaller studies of HIV-uninfected subjects. A positive bias was found comparing limb fat in DXA to MRI in a small study of 16 healthy men and women(11
). A study of 13 healthy, premenopausal women found high correlation but poor agreement between DXA, MRI and underwater-measured adiposity, with differences between DXA and MRI atributed to fat calibration errors (9
). Investigators concluded that no method can yet be regarded as a satisfactory reference technique.
Our finding that bias is proportional to the average amount of fat is similar to findings in the general population of more error in DXA in healthy men with higher adiposity and body thickness (23
). They felt DXA trunk fat precision and accuracy may be decreased by several factors, such as observer error in delineating specific regions due to X-ray beam inability to detect the small amount of soft tissue mass.
For trunk, we found positive bias in men, but negative bias in women. This may be due to the fact that DXA estimates do not differentiate between intra-abdominal and subcutaneous fat, and the women in our study have less VAT but more upper and lower trunk SAT compared with men, both in HIV and controls (6
). A small 12-16 week study of HIV-infected subjects(8
) found that DXA and MRI estimates of changes in SAT and VAT were strongly associated (R2
=0.70, p<0.001), although DXA estimated larger changes in total body fat compared with MRI.
A possible contributor to these differences is that DXA measured fat also includes fat that MRI measured adipose tissue cannot detect. For example, in the trunk region, fat in the liver, intestine and all other viscera are not included in MRI-measured AT. Likewise, intra-muscular fat cannot be detected by MRI. In addition, small adipose tissue depots below the resolution of MRI are not included in MRI. These small adipose tissue depots include some of the VAT and inter-muscular adipose tissue depot in both trunk and limb regions. These differences may partially explain why DXA measured fat is higher than MRI measured adipose tissue. Additionally, in our MRI and DXA protocol, the cut-off between limbs and trunk in MRI and DXA are not identical. DXA limb fat may include more hip fat than MRI measured limb adipose tissue. The bias identified in the Bland-Altman analysis may also be due to more fat in the hip region in heavier subjects and women have more fat in the hip region than men do. However, these latter issues do not apply to arm fat, which shows similar trends.
What is the significance of these differences between DXA and MRI? Both DXA and MRI have been used in previous studies to estimate regional and total adiposity in HIV infection, but results from studies utilizing DXA may not be able to be directly extrapolated to studies in which MRI or other methods are used. Consequently, comparisons of HIV and controls and the prevalence or amount of lipoatrophy will differ depending on which method is used to quantify regional adipose tissue as well as how lipoatrophy is defined. However, when certain guidelines are followed (24
), DXA has been found to have adequate internal validity for measuring body composition changes.
One limitation of our study was the lack of an absolute reference standard for estimating regional fat quantities. Another limitation was that several different DXA machine models were utilized in this study. Prior work (25
) has found that although fan- and pencil-beam models are highly correlated, small but significant differences exist between the instruments. However, sensitivity analysis including only Hologic machines, admittedly including both fan- and pencil-beam models, did not change our key finding that DXA is more likely than MRI to find lipoatrophy in HIV-infected subjects. DXA and MRI estimates of regional fat also differ because the cuts are slightly different: DXA cuts are at an angle perpendicular to the femoral neck, while MRI cuts are perpendicular to the longitudinal axis of the body. Although the vast majority of subjects had DXA and MRI scans performed on the same day, we also examined the association of time between scans with difference between DXA and MRI measured fat, finding little association (rho ≈ −0.02 or less, p>0.70). Finally, direct comparison is limited by the fact that DXA measures fat by attenuation of x-ray, while MRI directly measures AT volume (3
Further study in other populations is required to characterize the differences between DXA and MRI measurements of adipose tissue, including those testing differences in clinical outcomes between the two techniques. Comparison of DXA and MRI should also be made among HIV-infected subjects with other methods such as CT, since small studies of HIV-uninfected subjects have found important differences in variability and accuracy between these three methods (26
). For example, the slice traditionally used in CT studies of visceral adipose tissue is not the best marker of visceral adiposity compared to MRI measures (28
). In the current study, we found that although DXA-measured adipose tissue correlates strongly with MRI-measures in both HIV-infected subjects and controls, the difference between MRI and DXA increases as the average fat increases, particularly for limb fat. DXA may therefore estimate a higher peripheral lipoatrophy prevalence in HIV-infected subjects. Both leg and arm fat were higher for DXA, but the DXA-MRI differences vary among important subgroups such as leg fat in HIV vs. controls and trunk fat in men vs. women. Therefore, caution must be used when comparing study results using different methods of fat measurement.