Validity of dual-energy x-ray absorptiometry for estimation of visceral adipose tissue and visceral adipose tissue change after surgery-induced weight loss in women with severe obesity

Abstract

Objectives:

Reliable and simple methods to quantify visceral adipose tissue (VAT) and VAT changes are needed. We investigated the validity of Dual-energy X-ray Absorptiometry (DXA) compared to Magnetic Resonance Imaging (MRI) for estimating VAT cross-sectionally and longitudinally after surgery-induced weight-loss in women with severe obesity.

Methods:

Women with obesity (n=36; age 43±10 years; 89% Caucasian) with DXA and MRI before bariatric surgery (T0), at (T12) 12- and (T24) 24-months post-surgery were included. CoreScan estimated VAT from 20% of the distance between the top of the iliac crest and base of the skull. MRI-VAT (total VAT) was measured from base of the heart to the sacrum/coccyx on whole-body scan.

Results:

Mean DXA-VAT was 45% of MRI-VAT at T0, 46% at T12 and 68% at T24. DXA underestimated change in MRI-VAT between T0 and T12 by 26.1% (0.81 kg, p=0.03) and by 71.7% (0.43kg, p<.001) between T12 and T24. The relationship between DXA-VAT and MRI-VAT differed between T12 and T24 (interaction p=0.03).

Conclusions:

CoreScan lacks validity for comparing VAT across individuals or estimating the size of changes within individuals, however, within the limits of measurement error, it may provide a useful indicator of whether some VAT change has occurred within an individual.

INTRODUCTION

Adipose tissues differ in their contribution to cardiovascular and metabolic risk. Located deep within the abdominal cavity and surrounding the internal organs, visceral adipose tissue (VAT) has direct access to the liver where it can influence pathological processes (1, 2). The quantification of VAT is of significant clinical interest because it is metabolically active and is linked to various medical conditions including metabolic syndrome, cardiovascular disease, and malignancies such as prostate, breast & colorectal cancers (2).

The standards for quantifying VAT volume are computed tomography (CT) (3) and MRI (4). Both CT and MRI are cost prohibitive and time consuming for patients and post-processing analyses. CT has the added disadvantage of radiation exposure. With rapid scan times and minimal radiation exposure, DXA is increasingly used for total body composition assessments. DXA provides a measure of three body tissues: fat mass (FM); non-bone lean mass (LM) and bone mineral content (5). The CoreScan (GE Healthcare) software estimates VAT volume (converted to mass by a standard factor) within the android or abdominal region, from a whole-body composition scan acquired on GE Lunar iDXA and Prodigy systems. FM in the android region consists of subcutaneous adipose tissue (SAT) and VAT. To accurately estimate VAT, CoreScan must distinguish between the two depots. GE has validated CoreScan against CT cross-sectionally in adults ages 18-90 years with BMI range of 18.5 to 40 kg/m² and found strong agreement between the techniques (6). However, other studies have shown that compared to criterion techniques, CoreScan underestimated (7) or under- or overestimated VAT depending on VAT volume (8). CoreScan had greater variability estimating VAT change in persons with BMI <25kg/m² or VAT mass <500g (9).

Similar automated VAT analysis software is available for other DXA manufacturers. Like CoreScan, Hologic’s InnerCore underestimated VAT cross-sectionally compared to criterion techniques in adults (10, 11), except for one study in which VAT was overestimated (12). InnerCore also overestimated VAT cross-sectionally in children (13). In longitudinal studies, InnerCore underestimated changes in VAT compared to CT (11) and lacked validity for estimating changes in VAT in children (13).

Reliable and simple methods to quantify VAT and VAT changes are also needed for persons who do not fit within the field of view of the MRI bore. VAT measurement by DXA has not been validated in persons with severe obesity, nor longitudinally after rapid weight loss that may result in large decreases in the VAT depot. Roux-en Y gastric bypass (RYGB) has been shown to differentially reduce the VAT depot compared to other bariatric surgeries (14). The objective of this study was to assess the validity of CoreScan for the quantification of VAT compared to MRI, cross-sectionally (before bariatric surgery), and longitudinally after extensive and rapid weight loss through 2 years post Roux-en Y gastric bypass surgery.

METHODS

Participants

Between November 2006 and December 2009, 105 bariatric surgery patients participating in the Longitudinal Assessment of Bariatric Surgery-2 (LABS-2) trial, previously described (15, 16), enrolled in an ancillary body composition study (17, 18, 19) conducted at St. Luke’s-Roosevelt Hospital (n=53) and the University of Pittsburgh (n=52). This ancillary study sought to measure important components of weight change in fat mass and its distribution and FFM, skeletal muscle, and select organs from before surgery to post-surgery 12, 24, 60 and 84 months. LABS-2, a prospective observational cohort trial, included no lifestyle intervention beyond the standard pre- and postoperative recommendations given to patients at its surgical centers. Patients selected the surgical procedure with a LABS-certified surgeon (15). All studies were approved by the Institutional Review Boards of St. Luke’s-Roosevelt Hospital and Columbia University (where body composition of Weill Cornell patients was assessed) and the University of Pittsburgh. All patients provided written informed consent.

Inclusion criteria for the current analyses include having a valid pair of DXA and MRI scans at T0, before bariatric surgery, T12 and T24, approximately 12, and 24, months after surgery, respectively. The sample was further restricted to include only females who underwent RYGB surgery because the small number of males and other surgery types precluded analyses to separate the confounding effects of sex and surgery type. The analytic sample, therefore, include 36 women who underwent RYGB surgery.

Measures

Demographic variables.

Age, race/ethnicity and sex were obtained by self-report.

Body composition measures.

Body weight (Weight Tronix, New York, New York; Scale-Tronix, Wheaton, Illinois), height (Holtain; Crosswell, Wales, New York), and body density (Bod Pod; Cosmed, Rome, Italy; software version 2.3) measurements were obtained. As previously described (17), total body water was assessed by deuterium dilution, in which a ~0.1 g/kg oral dose of D₂O was ingested immediately following a venous blood sample drawn from an antecubital vein. A second blood sample was drawn after 3 hours. A 3 comparent (3C) model was used to estimate FM (20): fat (kilograms) = 2.122 X (BW/d) - 0.779 X TBW - 1.356 X BW, where BW is the body weight in kilograms, d is the body density derived from BodPod, and TBW is the total body water in kilograms. Fat free mass (FFM) was derived as the difference between body weight and FM. Percent body fat by 3C (%BF) was calculated as FM divided by BW X 100.

Magnetic Resonance Imaging.

Acquisition:

As previously described (18, 19), participants whose body size was fully accommodated within the MRI field-of-view from the New York and Pittsburgh sites were placed on a 1.5 T MRI scanner (GE, 6X Horizon, Milwaukee, Wisconsin) table and scanned with arms above their heads. Approximately 40 axial images with 10 mm thickness at 40-mm intervals were acquired across the entire body. Total VAT was measured from the whole-body multi-slice MRI. VAT visible in slices extending from the base of the heart to the tip of the sacrum/coccyx reflects total VAT.

Analysis:

SliceOmatic 4.2 image analysis software (Tomovision, Montreal, CA) was used to analyze images on a PC workstation (Gateway, Madison, WI). MRI-volume estimates were converted to mass using the assumed density of 0.92 kg/L for adipose tissue (21). All scans were read by the same analyst who was masked to time point. The technical errors for 4 repeated readings of the same four whole-body scans by the same observer of MRI-derived VAT volume in our laboratory is 2.3%. Data collected at the University of Pittsburgh were transferred to SLRHC/Columbia University, where all MRI studies were analyzed by the same Image Analyst.

Dual-Energy X-Ray Absorptiometry (DXA).

Acquisition:

Participants were scanned using either a Lunar iDXA or Prodigy machine (GE-Healthcare, Milwaukee, Wisconsin). The scan mode was age and body weight/size appropriate based on the manufacturer’s recommendations.

Analysis:

Whole body scans were analyzed for VAT using GE enCORE software, version 16 with CoreScan by a single trained analyst who was masked to time point. CoreScan VAT region-of-interest (ROI) is defined as the area between the top of the iliac crest and 20% of the distance from the top of the iliac crest to the base of the skull. To assess repeatability of DXA VAT analysis, 20 DXA scans were reanalyzed by the same analyst. The CV was 2.4%. The DXA and MRI scans were acquired at the same visit.

Statistics

Data were tabulated and histograms were examined for unusual distributions, relationships, or outliers. In regression analyses, residuals were examined for violations of model assumptions. Univariate linear regressions with MRI VAT as the dependent variable were used to model the relationship between MRI and DXA VAT at T0, T12 and T24. Multiple regression equations with height, weight, BMI, FFM, age, %BF or FM as independent predictors assessed whether these covariates improved DXA's estimate of VAT. At T12 and T24, and multiple regression analyses were used to analyze the DXA/MRI relation to VAT mass separately for participants who fit into the MRI at baseline (T0) from those who did not (fitters vs non-fitters). The relationship of DXA VAT to MRI VAT was similar for fitters and non-fitters, therefore the samples were combined for cross-sectional analyses at these time points. Longitudinal analyses with time as a repeated measure, DXA VAT as a time-varying predictor and the interaction of time and DXA VAT were used to test whether the relation of DXA VAT to MRI VAT differed over time.

All statistical analyses were conducted using SPSS Statistics v27 (IBM, NY, USA) and SAS v9.4 (SAS Institute, Cary, NC).

RESULTS

Participant characteristics.

Data from 36 female participants were included in the analyses (Table 1). At T0 (before RYGB bariatric surgery), 18 participants had a set of paired MRI and DXA scans, 36 at T12 (12 months post-surgery) and 26 at T24 (24 months post-surgery). All participants were women who had undergone RYGB surgery. The average age at baseline was 43 years; 89% were Caucasian. Figure 1A,B shows images of MRI scans before bariatric surgery and 1 year after bariatric surgery and images of a DXA scan.

Table 1.

Characteristics of participants before surgery and at post-surgery visits

	T0		T12		T24
	Did not fit MRI at baseline n=19	Fit MRI at baseline n=18	Did not fit MRI at baseline n=19	Fit MRI at baseline n=17	Did not fit MRI at baseline n=13	Fit MRI at baseline n=13
Race/ethnicity
Caucasian	11 (57.9)	16 (88.9)	11 (57.9)	15 (88.2)	8 (61.5)	12 (92.3)
African American	5 (26.3)	1 (5.6)	5 (26.3)	1 (5.9)	3 (23.1)	1 (7.7)
Hispanic	3 (15.8)	1 (5.6)	3 (15.8)	1 (5.9)	2 (15.4)	0
Age (years)	40±10	43±10	41±10	43±10	44±9	44±8
Height (cm)	165.3±5.5	163.5±7.4	165.7±5.5	163.5±8.1	165.3±6.7	164.8±7.7
Weight (kg)	130.3±16.4	110.6±12.6	80.2±14.5	71.9±13.1	75.8±11.9	74±16.2
BMI	47.65±5.1	41.4±3.9	29.1±4.5	26.8±3.7	27.8±4.2	27.1±5.1
Fat mass (kg) by 3C	70.87±12.1	59.1±10.1	28.6±11.5	25.5±10.3	27.2±8.8	27.4±13.3
Body fat % by 3C	54.1±3.9	53±4.7	34.6±8.1	34.3±8.2	35±7.7	35.3±9.1
MRI VAT (kg)		4.6±1.7	1.3±0.8	1.6±1.6	0.7±0.5	1.2±1.3
DXA VAT (kg)		2.1±0.8	0.6±0.4	0.7±0.5	0.5±0.3	0.8±0.7

Note: Continuous variables are shown as mean ± SD. Categorical variables are shown as frequency (%). Fat mass and Body fat % were calculated using a 3-compartment model. Group means (fit MRI at baseline vs. did not fit MRI at baseline) were not different at each time point (p > 0.05). Participants who did not fit MRI were excluded from analyses at baseline.

Abbreviations: 3C, three-compartment model; BF%, body fat percentage; DXA, dual-energy x-ray absorptiometry; FM, fat mass; MRI, magnetic resonance imaging; T0, before bariatric surgery, T12, 12 months after bariatric surgery; T24, 24 months after bariatric surgery; VAT, visceral adipose tissue.

Figure 1:

Imaging output and region of interest (ROI) of DXA & MRI scans.

Figure 1a: (i) a single cross-sectional image from an MRI scan at the L4-L5 vertebrae showing VAT mass in purple; (ii) Three-dimensional volume rendering in a single participant of total VAT mass (brown) before surgery (left) and 12 months post-surgery (right). Participant is positioned in the MRI scanner with arms extended above head.

Figure 1b: (i) whole body DXA scan showing bone (left) and soft tissue (right); (ii) visceral and subcutaneous tissue within the DXA VAT region of interest.

Cross-sectional analyses

DXA-estimated VAT is only about half of MRI-measured total VAT at T0 and T12, and slightly more than half of MRI-measured total VAT at T24 (Table 1). Table 2 and Figure 2 present univariate and multiple regression models describing the relationship between DXA VAT and MRI VAT at each time point. DXA VAT alone accounted for 77% of the variance in MRI VAT at T0 and T24 and 65% at T12. At T0, FFM was negatively related to MRI VAT (p=0.038), but marginally increased the variance explained to 79%. No other variable (height, weight, BMI, FM, %BF, age) contributed significantly to the estimation of MRI VAT. FM helped explain more of the variance in MRI VAT at T12 (p=0.024), while age did so at T24 (p=0.004). The relationship between DXA VAT and MRI VAT appeared to differ at each time point as the regression coefficient relating DXA VAT to MRI VAT differed, however the significance of these differences can only be tested in the longitudinal analyses.

Table 2.

Univariate and multiple linear regression models estimating MRI VAT from DXA VAT at T0, T12 and T24

	Outcome	Predictors	Regression Coefficient	r	Adj. R2	SE	p-value
T0				0.86	0.77	0.80	<.001
Model 1	MRI VAT (kg)	(Constant)	0.55				0.35
		DXA VAT (kg)	1.95				<.001
T12				0.81	0.65	0.49	<.001
Model 2	MRI VAT (kg)	(Constant)	0.19				0.26
		DXA VAT (kg)	1.74				<.001
				0.84	0.69	0.46	<.001
Model 3	MRI VAT (kg)	(Constant)	−0.17				0.43
		DXA VAT (kg)	1.43				<.001
		Fat mass (kg)	0.02				0.02
T24				0.88	0.77	0.31	<.001
Model 4	MRI VAT (kg)	(Constant)	0.03				0.81
		DXA VAT (kg)	1.36				<.001
				0.92	0.84	0.26	<.001
Model 5	MRI VAT (kg)	(Constant)	−0.88				<.01
		DXA VAT (kg)	1.33				<.001
		Age (years)	0.02				0.004

Note: Fat mass was calculated using a 3-compartment model. Models 1, 2, and 4 describe linear regression results at the respective time points where MRI VAT was entered as the dependent variable and DXA VAT was entered as the independent variable. Models 3 and 5 describe multiple regression results of the models adjusting for one covariate. Only models with significant covariates are presented.

Abbreviations: DXA, dual-energy x-ray absorptiometry; MRI, magnetic resonance imaging; SE, standard error; T0, before bariatric surgery; T12, 12 months after bariatric surgery; T24, 24 months after bariatric surgery; VAT, visceral adipose tissue.

Figure 2.

Scatter plot showing the relationship between MRI VAT (kg) and DXA VAT (kg) at T0, T12, and T24. The slope at T12 is significantly different from the slope at T24 (p=.03). DXA, dual-energy x-ray absorptiometry; MRI, magnetic resonance imaging; T0, before bariatric surgery; T12, 12 months after bariatric surgery; T24, 24 months after bariatric surgery; VAT, visceral adipose tissue.

Longitudinal analyses

Although DXA VAT severely underestimates MRI VAT, if the relation of MRI VAT to DXA VAT is relatively stable over time, then DXA VAT measures may still be useful to identify within-patient change for clinical purposes. Longitudinal models were developed solely to test whether the relationship between DXA VAT and MRI VAT differs over time (Table 3). The equations are not intended for predicting MRI VAT from DXA VAT in other data sets. Only cases that had both measures at both time points were used in these analyses. Between T0 and T12, the interaction of DXA VAT by time was not significant (p=0.556). As such, within the limited power of this analysis, the hypothesis that DXA VAT and MRI VAT are related in the same manner at both time points cannot be rejected. However, significant amounts of change in MRI VAT were not accounted for by DXA changes (0.81 kg, p=0.03) (equation 1). The model relating MRI VAT to DXA VAT from T0 to T12 is as follows:

$$\mathbf{MRI}\mathbf{VAT}(\mathbf{kg})=1.61{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg})+0.81{\mathbf{kg}}^{\ast }\mathbf{time}+0.28\mathbf{kg}$$ (1)

(in which time is a dummy variable equal to 1 at T0 and 0 at T12.

Table 3.

Longitudinal Models of MRI VAT Regressed on DXA VAT from T0 and T12 and from T12 and T24

	Outcome	Predictors	Regression Coefficient	t	SE	p-value
T0 and T12
Model 1	MRI VAT (kg)	Intercept	0.16	0.55	0.28	0.59
		DXA VAT	1.80	4.82	0.37	<.01
		Time (T0)	1.07	1.99	0.54	0.07
		Time (T12)	0	0	0	0
		DXA VAT x Time (T0)	−0.26	−0.60	0.43	0.56
		DXA VAT x Time (T12)	0	0	0	0
Model 2	MRI VAT (kg)	Intercept	0.28	1.47	0.19	0.16
		DXA VAT	1.61	8.52	0.19	<.0001
		Time (T0)	0.81	2.5	0.32	0.03
		Time (T12)	0	0	0	0
Model 3	MRI VAT (kg)	Intercept	−1.17	−2.81	0.42	0.01
		DXA VAT	1.55	8.59	0.18	<.0001
		Body fat % by 3C	0.05	3.43	0.01	<.01
T12 and T24
Model 4	MRI VAT (kg)	Intercept	0.07	0.52	0.14	0.61
		DXA VAT	1.28	6.48	0.20	<.0001
		Time (T12)	0.08	0.49	0.17	0.63
		Time (T24)	0	0	0	0
		DXA VAT x Time (T12)	0.59	2.34	0.25	0.03
		DXA VAT x Time (T24)	0	0	0	0

Note: Fat mass and Body fat % by 3C were calculated using a 3C model. Time is a dummy variable. For comparisons between T0 and T12, time is equal to 1 at T0, and 0 at T12. For comparisons between T12 and T24, time is equal to 1 at T12 and 0 at T24. In each model, MRI VAT is entered as the dependent variable and DXA VAT is entered as the independent variable. Models 1 and 4 test whether the relationship between MRI VAT and DXA VAT differ at each interval (significance of the DXA VAT-by-time interaction). Model 2 shows results for model 1 after the nonsignificant DXA VAT-by-time interaction was removed. Model 3 shows results for model 2 after adjusting for BF% and removal of the time variable, which was no longer significant after BF% was added.

Abbreviations: 3C, three-compartment; BF%, body fat percentage; DXA, dual-energy x-ray absorptiometry; MRI, magnetic resonance imaging; T0, before bariatric surgery, T12, 12 months after bariatric surgery; T24, 24 months after bariatric surgery; VAT, visceral adipose tissue.

This simplifies to the following

$$\mathbf{T}0\mathbf{MRI}\mathbf{VAT}(\mathbf{kg})=1.61{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg})+1.09\mathbf{kg}$$ (2)

$$\mathbf{T}12\mathbf{MRI}\mathbf{VAT}(\mathbf{kg})=1.61{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg})+0.28\mathbf{kg}$$ (3)

Weight, BMI, FM by 3C, FFM and %BF were tested as covariates. With any covariate included in the model, there was no overall effect of time. Only FFM and %BF made contributions that approached or exceeded p=0.05. For example, between T0 and T12, for each additional 1% increase in %BF, MRI VAT increases by 0.05 kg (equation 4 below) beyond that predicted by DXA VAT. Substituting FFM for %BF in the model, the FFM contribution approaches significance with a coefficient of −0.04, p=0.051 (model not shown).

$$\mathbf{MRI}\mathbf{VAT}(\mathbf{kg})=1.54{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg})+0.05{}^{\ast }\%\mathbf{BF}-1.17\mathbf{kg}$$ (4)

Longitudinally, for T12 and T24, models were developed to predict MRI VAT from DXA VAT, time, fitters vs non-fitters, and all 2-way interactions (the 3-way was not significant). Fit did not contribute to the model and was excluded. There was no overall effect of time (p=.63). The DXA VAT by time interaction was significant, indicating that DXA VAT is related differently to MRI VAT at T12 and T24 (equation 5). The model relating MRI VAT to DXA VAT from T12 to T24 is:

$$\begin{gathered} \mathbf{MRI}\mathbf{VAT}(\mathbf{kg})=1.28{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg})+0.08(\mathbf{kg}){}^{\ast }\mathbf{Time}+0.59{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg}){}^{\ast } \\ \mathbf{Time}+0.07\mathbf{kg} \end{gathered}$$ (5)

(in which time = 1 at T12 and 0 at T24. This simplifies to the following:

$$\mathbf{T}12\mathbf{MRI}\mathbf{VAT}(\mathbf{kg})=1.87{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg})+0.15\mathbf{kg}$$ (6)

$$\mathbf{T}24\mathbf{MRI}\mathbf{VAT}(\mathbf{kg})=1.28{}^{\ast }\mathbf{DXA}\mathbf{VAT}(\mathbf{kg})+0.07(\mathbf{kg})$$ (7)

A 0.42 kg difference of MRI VAT between visits remains unexplained by these variables. BMI, weight, and height did not contribute significantly to the model. Two measures of body composition, FM by 3C (p=0.031) and %BF (p=0.030) made statistically significant additions to the model and affected the DXA-VAT regression coefficients, but did not improve the prediction of MRI VAT at each time point. In exploratory analyses, DXA VAT-by-race interaction showed a trend (p=0.071) suggesting that the association of DXA VAT with MRI VAT may differ by ethnicity/race groups.

DISCUSSION

This study investigates the validity of the CoreScan software, compared to the criterion MRI, for estimating VAT and VAT change after rapid weight loss, one and two years post bariatric surgery. DXA values underestimated VAT by about half (Table 1). Results from cross-sectional regression analyses suggest different relationships of DXA VAT to MRI VAT at each time point (different coefficients). Results from longitudinal analyses provided further evidence of different relationships between DXA VAT and MRI VAT between T12, one year post-surgery and T24, two years post-surgery (significant DXA VAT by time interaction), but not between pre-surgery and T12. As such, it was not possible to develop a single conversion factor to relate or convert DXA estimates to MRI measurements. The absence of a consistent relationship between DXA and MRI preclude DXA’s usefulness as a tool for accurate cross-sectional or longitudinal assessments. CoreScan substantially underestimated change in VAT from T0 to T12 by 26% and to a greater extent from T12 and T24 (72%).

These results extend the findings by Dias (13), Fourman (11) and Taylor (12) who compared DXA to MRI cross-sectionally and longitudinally. Both Dias and Taylor found an overestimation of VAT by DXA cross-sectionally while Fourman found an underestimation. Dias also found no association between longitudinal changes in DXA- and MRI-VAT. Taylor found DXA underestimated changes in VAT volume at 3 months by 33% and by 47% at 12 months. One major difference between our study and the Dias and Taylor studies is that they used a 5.2 cm slice by DXA and a 5.0 cm MRI window to closely approximate the measurement site of DXA-VAT while our study used a larger DXA scan area (approximately 10 cm) (6) and all VAT by MRI.

As our results confirm, a large part of MRI VAT is not included in the DXA ROI. The estimate of VAT by DXA is from a small region only, which does not encompass the entire anatomical area that contains VAT, and this sub-region is being compared to total VAT volume by MRI. Perhaps, as weight loss occurs, and as the VAT comparent shrinks due to the effects of RYGB surgery or more generally due to any lifestyle intervention, the remaining VAT falls within or outside of the DXA ROI and the final resting place differs by individual. Figure 1 illustrates the dramatic change in the VAT depot following RYGB surgery. Support for this hypothesis is provided by Shuster et al. who caution against the use of single slice CT for longitudinal assessment of VAT because soft tissue structures continuously move and may alter the location of VAT in a given slice (2). The same applies to the use of single and multi-slice MRI approaches.

In previous studies comparing DXA to MRI (7, 8, 9, 10, 13, 22), or DXA to CT (6, 11), measures of VAT by MRI or CT were obtained from a specific sub-region, typically at the level of lumbar vertebrae L4 and/or L5 with the rationale that this ROI is representative of total VAT and anatomically close in size to the DXA ROI. We believe that the over- and underestimation of VAT by DXA seen in previous studies may be due to the differences in the size of the DXA vs MRI ROI being compared (23). Some studies have attempted to correct for the difference in the size of the DXA and MRI ROIs. In one study, a procedure was employed to develop scaling factors to normalize the differences in size of these two ROIs (22). In the current study, due to our finding of different relationships at the different time points, it is not possible to employ such normalizing procedures.

Studies using smaller or ‘representative’ regions of interest may miss important differences or changes in VAT thereby bringing into question the validity of DXA-VAT assessments. For example, the menopausal transition period is associated with greater visceral fat accumulation (24). Our participants ages ranged from 23 to 56 years at baseline. Some of these women may have been perimenopausal and could have transitioned over the two-year investigation period. Menopausal transition, although not assessed in this study, could possibly lead to differential changes in VAT based on whether the measure is by DXA vs. MRI depending on the location of VAT deposits relative to the DXA ROI. If greater VAT accumulates within the DXA ROI, then DXA’s estimates would be larger whereas if the excess VAT falls outside the ROI, then DXA’s estimates would be smaller. We do not believe that menopausal transition affected our finding because our estimate of VAT by DXA 12 months post-surgery matches that reported in other studies of women of perimenopausal age (25). Future studies should assess menopausal status when evaluating changes in VAT by DXA. If this hypothesis is confirmed, it would limit the applicability of DXA-VAT estimation in menopausal and perimenopausal women.

In the current study we investigated what other variables would help improve DXA estimates of VAT. Our results show that DXA estimates are affected by the amount of total body FM present. When predicting MRI VAT from DXA VAT, FM is positively related to MRI VAT independent of its association with DXA VAT. For a given DXA VAT estimate, MRI VAT will be larger as FM increases. This finding agrees with several studies (13, 22) and helps explain DXA’s inability to accurately predict MRI VAT absent other measures. Taken together, our results suggest that improving DXA’s estimation of VAT may involve increasing the DXA ROI.

Study Strengths and Limitations.

Strengths of the study include MRI-measured VAT acquired contemporaneously with DXA VAT studies on a sample of participants with extreme obesity with repeated measures over 2 years. The study is limited by the smaller sample size in the T0 group; hence, it has lower power for cross-sectional comparisons at T0 and for detecting longitudinal changes from T0 to T12. Menopausal status was not assessed in this study. The analytic sample was intentionally limited to women who have undergone RYGB surgery which allowed for control of confounding effects of sex and surgery type, but limits generalizability.

Conclusion.

DXA estimates approximately fifty to seventy percent of total MRI-measured VAT mass. In regressions, DXA VAT accounts for a large amount of the variance in MRI VAT cross-sectionally. When using DXA to estimate change in VAT, the coefficients relating changes in DXA VAT to MRI VAT differ for the interval between T12 and T24 thereby making it unusable for estimating longitudinal changes. DXA VAT as measured by CoreScan, within the limits of error, may be useful as an indicator of whether VAT has increased or decreased or shifted within the same individual (not addressable in this study), however DXA lacks accuracy for comparing VAT mass across individuals or for estimating the size of changes within an individual.

What is already known?

• Excess VAT is an established risk factor for type 2 diabetes, cardiovascular disease, and several malignancies.

• The standards for measuring VAT (MRI and CT) are cost prohibitive, especially for repeated measures.

• CoreScan software (GE) estimates VAT from whole-body DXA scans and VAT estimated by DXA is correlated with VAT measured by MRI at L4-L5 vertebrae, however, these regions of interest are not identical.

What does this study add?

• This is the first study to compare total VAT mass by MRI (criterion) with VAT mass by DXA (CoreScan).

• VAT mass estimated by DXA is only 45-68% of VAT measured by MRI

• The relationship between VAT estimated by DXA and VAT measured by MRI is not constant but changes over time.

• DXA lacks validity for comparing VAT across individuals or estimating size of changes within individuals, however, within the limits of measurement error, it may provide a useful indicator of some VAT change within a individual.

How might these results change the direction of research or the focus of clinical practice?

• Improving CoreScan estimation of total VAT may require increasing the DXA region of interest.