Comparison of DXA and CT in the Assessment of Body Composition in Premenopausal Women With Obesity and Anorexia Nervosa

Authors


(mbredella@partners.org)

Abstract

Accurate methods for assessing body composition in subjects with obesity and anorexia nervosa (AN) are important for determination of metabolic and cardiovascular risk factors and to monitor therapeutic interventions. The purpose of our study was to assess the accuracy of dual-energy X-ray absorptiometry (DXA) for measuring abdominal and thigh fat, and thigh muscle mass in premenopausal women with obesity, AN, and normal weight compared to computed tomography (CT). In addition, we wanted to assess the impact of hydration on DXA-derived measures of body composition by using bioelectrical impedance analysis (BIA). We studied a total of 91 premenopausal women (34 obese, 39 with AN, and 18 lean controls). Our results demonstrate strong correlations between DXA- and CT-derived body composition measurements in AN, obese, and lean controls (r = 0.77–0.95, P < 0.0001). After controlling for total body water (TBW), the correlation coefficients were comparable. DXA trunk fat correlated with CT visceral fat (r = 0.51–0.70, P < 0.0001). DXA underestimated trunk and thigh fat and overestimated thigh muscle mass and this error increased with increasing weight. Our study showed that DXA is a useful method for assessing body composition in premenopausal women within the phenotypic spectrum ranging from obesity to AN. However, it is important to recognize that DXA may not accurately assess body composition in markedly obese women. The level of hydration does not significantly affect most DXA body composition measurements, with the exceptions of thigh fat.

Introduction

Accurate methods for assessing body composition in subjects with obesity and anorexia nervosa (AN) are important for determination of metabolic and cardiovascular risk factors and to monitor the effects of therapeutic interventions on fat and muscle compartments (1,2). Several methods for assessing body composition exist such as anthropometry, bioelectrical impedance analysis (BIA), dual-energy X-ray absorptiometry (DXA), magnetic resonance imaging (MRI) and computed tomography (CT) (3,4,5,6,7,8,9,10,11). CT and MRI have shown excellent accuracy in assessing muscle and fat areas in cadaveric studies (12,13). However, because these methods are expensive, time-consuming and/or require radiation, and may have limited availability, they are impractical in clinical settings and for large research studies. DXA has been widely used for osteoporosis screening and diagnosis (14,15). It is readily available, relatively inexpensive, and requires minimal radiation exposure. DXA is also used to measure body composition, and studies have shown strong correlations between body composition parameters obtained by DXA and those obtained by CT or MRI in adults and adolescents of normal weight (6,7,9,16,17,18). However, obesity and AN can cause changes in body composition that may impact the assessment of fat mass and lean soft tissue mass by DXA. For example, it has been shown that the level of hydration can alter the validity of DXA-derived estimates of body composition (19,20,21,22). Subjects with AN can have marked variability in hydration depending on the stage of the disorder, and the percent of water may decrease with increased body fat in subjects with obesity. Given the need for reliable methods to assess body composition in subjects with obesity and AN, the purpose of this study was to determine the use of DXA in subjects of different weights using CT as a standard of reference. We investigated the agreement between CT and DXA for measuring abdominal fat, thigh muscle mass, and thigh fat in three groups of premenopausal women: normal weight, obese, and AN. In addition, we assessed the impact of hydration on DXA-derived body composition measures using BIA.

Materials and Procedures

The study was approved by the Partners Healthcare institutional review board and was Health Insurance Portability and Accountability Act compliant. Written informed consent was obtained from all subjects before the study.

Subjects

We studied a total of 91 premenopausal women. Of the 91 women, 34 were overweight or obese, 39 had AN, and 18 were of normal weight. Obese and lean subjects were recruited from the community through advertisements. AN subjects were referred to our study by eating disorders care providers or recruited through advertisements. Inclusion criteria for all three groups were: age 18–45 years and female gender. AN subjects fulfilled all Diagnostic and Statistical Manual IV criteria for AN. Overweight or obese subjects had a BMI ≥25 kg/m2, and lean controls had a BMI ≥19 kg/m2 and <25 kg/m2. Exclusion criteria for all three groups included pregnancy and weight >280 pounds (due to the limitations of the DXA and CT scanners).

Anthropometry

Body weight was measured with a standard balance beam scale to the nearest 0.1 kg in triplicate and averaged. Height was measured with a stadiometer to the nearest 0.1 cm in triplicate and averaged. BMI was calculated as weight divided by height squared (kg/m2).

BIA

BIA was used to measure % total body water (TBW) (precision <3%) using Bioelectrical Analyzer model BIA 101 (RJL Systems, Clinton Township, MI). Impedance was measured between the right wrist and ipsilateral ankle using a tetrapolar arrangement with two drive and two measurement electrodes using a standard protocol (10). Sex, age, and racial-ethnic specific BIA prediction equations provided by the manufacturer were used. Standard error of the estimate for TBW using the equations provided by the manufacturer is 8.1% (ref. 23).

DXA

DXA measurements of body composition were performed in all patients (Hologic, Waltham, MA). The following parameters were obtained: trunk fat, fat and lean soft tissue mass of the legs (in kg), % trunk fat, % fat, and % lean soft tissue mass of the legs. Coefficients of variation of DXA in our laboratory are 1.7% for body fat and 2.4% for lean soft tissue mass (24).

Computed tomography

Each subject underwent cross-sectional CT scan of the abdomen at the level of L4 and of the left mid-thigh. Patients were placed supine in the CT scanner and a lateral and frontal scout image was obtained to identify the level of L4 and the mid-thigh (equidistant from the femoral head and medial femoral condyle), which served as landmarks for the single-slice image. Scan parameters were standardized: 144 table height, 80 kV (abdomen), 120 kV (thigh), 70 mA (abdomen), 170 mA (thigh), scan time 2 s, 1 cm slice thickness, 48 field of view. Fat attenuation coefficients were set at −50 to −250 HU as described by Borkan et al. (3). Whole-body cross-sectional area was computed by outlining the outer contour of the abdomen. A second outline of the back and abdominal wall musculature (inner contour) was used to define the subcutaneous fat area. Visceral abdominal fat was defined as the area within the inner contour comprising all pixels with attenuation coefficients between −50 and −250 HU. The total fat area was then calculated as the sum of subcutaneous and visceral abdominal fat. Subcutaneous, visceral, and total adipose tissue areas were determined. The same technique was employed to the left thigh where the total, fat, and muscle areas were identified. Coefficients of variation of CT in our laboratory are 2.5% for fat and 1.1% for muscle area. The error for CT in the assessment of body composition compared to cadaveric studies has been reported as 1.3% (ref. 25).

The % abdominal fat was calculated by dividing the total abdominal fat area by the total abdominal area × 100, % thigh fat was calculated by dividing the thigh fat area by the total thigh area × 100, and % thigh muscle was calculated by dividing the thigh muscle area by the total thigh area × 100. Total, fat, and muscle areas were determined using image processing software (Alice, version 4.3.9; Parexel, Waltham, MA and Accuview, version 3.130; AccuImage Diagnostics, San Francisco, CA).

Statistical analysis

Statistical analysis was performed using JMP (version 5.0.1a; SAS Institute, Cary, NC) and MedCalc (version 9.2.1.0; MedCalc, Mariakerke, Belgium) software. Variables were compared using the Student's t-test. The Tukey-Kramer test was used to adjust for multiple comparisons. Correlation analysis between trunk fat, left thigh fat, and left thigh fat-free mass determined by DXA and abdominal fat, left thigh fat, and left thigh muscle area determined by CT was performed in obese, AN, and lean controls. Partial correlation coefficients are also reported after controlling for TBW. In order to determine agreement between the two methods, the % fat of abdomen/trunk and % fat and % lean soft tissue/muscle mass of the thigh from CT and DXA were compared using Bland-Altman analysis (26).

Results

Clinical characteristics and body composition of study subjects

Subject characteristics and body composition by CT and DXA of the three groups are shown in Table 1. Study participants ranged from 18 to 45 years, mean age: 29.5 ± 7.7 years (s.d.) and in BMI from 13.3 to 42.0 kg/m2, mean BMI 24.4 ± 8.3 kg/m2 (s.d.). Subjects with AN were younger than obese subjects and lean controls. As expected, there was a significant difference in weight and BMI between the three groups. AN subjects had lower and obese subjects had higher abdominal/trunk and thigh/leg fat as determined by CT and DXA compared to lean controls. AN subjects had lower muscle mass and obese subjects had higher muscle mass compared to lean controls; however, the % muscle mass was lowest in obese subjects and highest in AN subjects.

Table 1.  Clinical characteristics and body composition of study subjects (values are means ± s.d.)
inline image

Correlation between DXA and CT body composition parameters

There was a strong correlation between DXA and CT measurements of abdominal/trunk fat, thigh/leg fat, and muscle/lean soft tissue mass in AN subjects, obese subjects, and lean controls (Figure 1). Correlation coefficients were highest in the obese and lean controls, whereas the strength of the correlation between CT abdominal fat area and DXA trunk fat mass was slightly lower in the AN group. After controlling for % TBW, determined by BIA, partial correlation coefficients for the relationship between CT and DXA body composition parameters were comparable to the original coefficients in all groups (Table 2), except for thigh/leg fat assessment in the AN group implying a negligible contribution of body water to the DXA measurements in the normal-weight and obese populations and to DXA measurement of abdominal fat and thigh muscle in AN subjects. As visceral fat is a strong determinant of metabolic and cardiovascular risk factors and DXA trunk fat mass is often used as an estimate for visceral adipose tissue, we performed correlation analyses of visceral fat area determined by CT on trunk fat determined by DXA. CT visceral fat area and DXA trunk fat mass correlated in subjects with AN (r = 0.62, P < 0.001), obesity (r = 0.70, P < 0.001), and lean controls (r = 0.51, P < 0.001).

Figure 1.

Correlation analysis between CT and DXA measures of (a) abdominal/trunk fat, (b) leg/thigh fat, and (c) leg/thigh muscle mass in obese subjects (closed triangles), AN subjects (closed circles), and lean controls (open squares). AN, anorexia nervosa; CT, computed tomography; DXA, dual-energy X-ray absorptiometry.

Table 2.  Regression analysis of body composition parameters derived from CT and DXA
inline image

Agreement between DXA and CT body composition parameters

The CT % abdominal fat and DXA % trunk fat, CT % thigh fat and DXA % leg fat, and CT % thigh muscle and DXA % leg lean soft tissue were used to determine agreement between CT and DXA body composition measurements (Table 3). DXA underestimated abdominal fat in all three groups. The difference between CT and DXA measurements became larger with increasing weight. This trend was also observed within groups in AN and lean controls but not within the obese group (Figure 2). DXA also underestimated thigh fat in all three groups; however, the mean difference in measurements was smaller than the abdominal fat measurements. As with the abdominal fat measurements, the difference became larger with increasing weight. DXA overestimated muscle mass in all three groups, with larger mean differences in subjects with higher weights.

Table 3.  Agreement between body composition determined by DXA and CT using Bland-Altman analysis
inline image
Figure 2.

Bland-Altman analysis of CT- and DXA-derived abdominal fat in (a) AN, (b) lean control, and (c) obese subjects. DXA underestimates abdominal fat in all three groups compared to CT with the error becoming larger with increasing weight. This trend is also visible within the AN and lean control group but not within the obese group. AN, anorexia nervosa; CT, computed tomography; DXA, dual-energy X-ray absorptiometry.

Discussion

Our study showed that DXA is a useful technique for assessing body composition in premenopausal women across a weight spectrum ranging from obesity to AN, as demonstrated by the strong correlation with body composition parameters determined by the gold standard CT. However, it should also be noted that our Bland-Altman agreement analyses showed that DXA underestimated abdominal fat mass and overestimated extremity muscle mass, and that this bias increased with increasing weight. The underestimation by DXA of abdominal fat exceeded 30% in some obese women. Therefore, caution should be used when interpreting such DXA results, especially in a cohort with a wide range of weights.

With the increased incidence of obesity and the increased awareness of eating disorders, simple and reliable methods for assessing body composition are needed. DXA is a noninvasive method for assessment of body composition, is widely available, and relatively inexpensive. DXA uses two X-ray beams of different energies that are attenuated differently as they pass through various tissues in the body. Assumptions regarding body thickness and the level of hydration are used to convert the attenuation data into mass values for bone, fat, and lean soft tissue for the whole body or subregions (27,28). Several studies have shown a strong correlation with body composition determined by DXA and CT or MRI in subjects of normal weight (6,7,9).

Given the influence of body thickness and level of hydration on DXA measurements, we wanted to determine the accuracy of DXA in assessing body composition in subjects with obesity and AN. We found a strong correlation between abdominal and thigh fat and thigh muscle areas assessed by CT and trunk and leg fat and leg lean soft tissue mass assessed by DXA in subjects with AN, obesity, and normal weight. There was a significant difference in % TBW, as determined by BIA, between the three groups, with AN subjects having the highest % TBW and obese subjects having the lowest % TBW. As abnormal levels of hydration can alter the attenuation coefficient of soft tissues, we controlled for body water and determined partial correlation coefficients. After controlling for body water, the correlation coefficients between DXA and CT measurements were similar for most parameters, except for thigh fat in the AN population, where the correlation coefficient decreased from 0.92 to 0.80 after adjusting for TBW In the normal-weight and obese subjects and for abdominal fat and thigh muscle in AN subjects, body water had a negligible influence on DXA-derived body composition measurements in our study. This is in concordance with other studies showing no significant effect of hydration on DXA measurements in normal-weight subjects and ex vivo (29,30).

Visceral adiposity, independent of total adiposity, has been associated with cardiovascular disease, insulin resistance, and type 2 diabetes (31,32). Therefore, simple, noninvasive methods to accurately determine visceral adiposity are important for early detection of individuals at risk for metabolic and cardiovascular disease. CT or MRI are able to distinguish visceral from subcutaneous fat with a high level of precision and are considered the gold standard for assessment of visceral adiposity; however, they are expensive, often not readily available, and CT involves radiation exposure (3). In our study there was a correlation between visceral fat determined by CT and trunk fat determined by DXA in the obese group. However, our Bland-Altman agreement analysis demonstrated that DXA may underestimate abdominal fat mass. This is of clinical importance as subjects with obesity are at highest risk for cardiovascular mortality. Importantly, our data suggest that DXA trunk fat may not be able to be used as a surrogate for visceral adipose tissue in subjects with obesity.

Body composition in AN can be affected by several factors including the level of physical exercise, amount of fluid intake, vomiting, or laxative abuse. Body composition in AN subjects with extreme weight loss caused by starvation might be different from AN subjects who perform excessive exercise (28,33), requiring accurate methods for assessment of fat and lean soft tissue mass. Also, loss of muscle mass is associated with reduced muscle strength and decreased bone mineral density in this population (34). In our study, DXA showed a strong correlation with body composition determined by CT in AN, independent of the level of hydration, except for thigh fat. In the AN population, hydration status may be more variable, depending on the stage of illness and thus may have a greater or lesser impact on the accuracy of DXA. This could be significant when using DXA to assess changes in body composition during treatment and additional studies are needed to evaluate treatment effects on TBW and DXA measurements in this population

We performed Bland-Altman Analysis between DXA and CT body composition measurements to determine agreement between the two methods. However, as we used a standard DXA and CT protocol, which will likely be used in clinical practice or in clinical trials, we only had the DXA trunk fat measurement and DXA leg fat and lean soft tissue measurements available. Similarly, our standard CT protocol for assessment of body composition consists of a single abdominal slice through the level of L4, the region of highest abdominal fat, and a single slice through the mid-thigh, the region of highest muscle mass. Abdominal fat determined by single-slice has been shown to be closely correlated with total body abdominal fat volumes (35,36). In order to determine agreement between the two methods, we used % abdominal/trunk fat and % thigh/leg fat and muscle mass from CT and DXA. Using these ratios, DXA underestimated the abdominal and thigh fat and overestimated the muscle mass. Although the absolute differences in ratios could be a result of our technique, the finding of an increased bias with increasing weight is not likely to be so. As the % trunk fat includes the thorax and abdomen, the % fat of the trunk is likely lower than the % fat from a single CT slice though the abdomen, at point of highest fat content. Similarly, there is a higher % lean soft tissue mass in the entire leg compared to the single slice of the mid-thigh, explaining overestimation of muscle mass by DXA. Several studies using manually defined subregions on DXA, which corresponded to the CT region of the abdomen have demonstrated that DXA underestimates abdominal fat ranging from 10 to 26% (refs. 7,17,37), underestimates thigh adipose tissue by 10% (ref. 9) and overestimates thigh muscle mass by 4.4–12% (refs. 9,18,38). These values are similar to our results. In our study, DXA underestimated mean trunk fat by 10.7% in AN but by 27.5% in obese subjects. Similarly, DXA underestimated thigh fat by 3.8% in the AN group and 11.1% in the obese group. As in previous studies, DXA overestimated thigh muscle mass. In our study, thigh muscle mass was overestimated by 3.5% in AN subjects and 9% in obese subjects. A reason for the overestimation of thigh muscle mass could be the inclusion of skin and lean portions of adipose tissue in the DXA measurement. In our study, the mean differences in measurements of abdominal and thigh adipose tissue and thigh muscle/lean soft tissue mass increased with increasing weight, implying decreased accuracy of DXA body composition measurements with increasing weight.

Our study had several limitations. First, our DXA and CT protocols did not allow exact determination of agreement between the two methods because we performed single-slice CT of the abdomen and thigh and compared body composition measurements with those obtained by DXA of the entire trunk and leg, respectively. However, the purpose of our study was to use a simple DXA protocol, which would likely be used in clinical practice or in large clinical trials. Using our standard protocol, agreement between CT and DXA was similar compared to studies using specific subregions to closely correlate DXA and CT measurements (7,9,17,37,38,39). Using this standard protocol, we saw excellent correlations between DXA- and CT-derived body composition measurements, though the agreement analysis demonstrated some bias. Second, we only studied premenopausal women. There are sex- and age-related differences in DXA measurements, and studies in men, children, or the elderly might yield different results. Third, we did not include a severely obese cohort because of weight constraints of both DXA and CT technologies. However, as DXA may underestimate abdominal fat mass, this may represent a problem in severely obese subjects should scanning equipment be developed that can accommodate higher weight patients.

In conclusion, DXA provides a reasonable estimate of body composition in premenopausal women with obesity and AN for many purposes. However, it is important to note that DXA underestimated trunk and thigh fat and overestimated thigh muscle mass, a bias which increased with increasing weight and was as high as 30% for abdominal fat in markedly obese women and therefore should be used with caution. The level of hydration was not a significant factor affecting body composition measurements with DXA in most women with the exception being thigh fat in women with AN. Our data suggest that although DXA can be used to assess body composition in groups of different weights it is significantly inaccurate when used at the higher end of the weight spectrum in premenopausal women.

ACKNOWLEDGEMENT

This work was supported in part by the following grants: UL1 RR025758, HL077674, MO1 RR01066, M01 RR01066-27S1, and K23RR23090.

DISCLOSURE

The authors declared no conflict of interest.

Ancillary