| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Departments of Internal Medicine and Endocrinology, Copenhagen University Hospital, DK-2650 Hvidovre, Denmark
ABSTRACT
INTRODUCTION
METHODS AND MATERIALS
RESULTS
DISCUSSION
ACKNOWLEDGEMENTS
FOOTNOTES
REFERENCES
ABSTRACT 
Gotfredsen, Anders, Lene Bæksgaard, and Jannik Hilsted.
Body composition analysis by DEXA by using dynamically changing samarium filtration. J. Appl. Physiol.
82(4): 1200-1209, 1997.
Dual-energy X-ray absorptiometry (DEXA)
has a high accuracy for body composition analysis but is influenced by
beam hardening and other error sources in the extremes of measurement.
To compensate for beam hardening, the Norland XR-36 introduces a
dynamically changing samarium filtration system, which depends on the
current-absorber thickness. With this system we found a good agreement
(r = 0.99) between reference and
measured amounts of tissue or fat percentages in a plastic phantom and
in smaller (~0.5-4 kg) and larger (~5-20 kg) piles of
tissue (ox muscle and lard). Scans of six healthy volunteers covered
with combinations of beef and lard (~5-15 kg) showed a good
agreement (r = 0.99) between reference
and DEXA values of added soft tissue mass and fat percentage. We
conclude that the DEXA method (and, in particular, the Norland XR-36
using dynamic filtration) has a high accuracy for body composition
analysis. It has a potential for gaining status as a reference method
in the future and may presently be used as a supplement to the
traditional methods for body composition analysis.
body composition; dual-energy X-ray absorptiometry; dynamic
samarium filtration; fat-free mass; fat mass
INTRODUCTION THE UNDERSTANDING of the importance of the composition
of the human body [i.e., the division into fat mass (FM),
fat-free mass (FFM), total bone mineral content (TBMC), and other
compartments] has greatly increased during recent years. The
spectrum of research interest spans a range from sports medicine
through cardiology, obesity research, and diabetology to
endocrinological disorders such as adult growth hormone deficiency (19,
22).
Because of the small magnitudes in the changes in body composition
often encountered, high precision and accuracy are needed. However,
even the classic reference methods (underwater weighing, measurement of
total body water and total body potassium) often do not possess the
required accuracy. When these methods are used as single methods, they
are encumbered with the restraints inherent in the two-compartment
models, namely, the assumption of constancy of some central physical or
chemical characteristics of the two compartments (i.e., the specific
densities, the hydration or the potassium concentrations of FM and FFM)
(10). In the extremes of body composition, the assumption of these
constancies fails, which may cause erroneous results (i.e., under- or
overestimation of FM or FFM). A solution to this problem is the use of
three- or four-compartment models that combine several methods, but
these procedures are laborious and ridden with the statistical problem of propagation of the precision errors of the individual methods (10,
11).
Dual-energy X-ray absorptiometry (DEXA) offers an alternative to the
classic criterion methods of body composition measurement. It is
convenient, fast, intermediate in cost, nonhazardous, and low in
radiation. Furthermore, studies have shown that precision errors are
very favorable, not only for bone mineral content (BMC) but also for FM
and FFM (5). DEXA measures three compartments (TBMC, FM, FFM), and it
is not dependent on the assumption of constancy of certain physical or
chemical characteristics of the compartments.
However, DEXA may be influenced by other error sources particular to
the method. The most important of these are
1) beam hardening (i.e., the change
of the beam spectrum due to preferential attenuation of low energies
with increasing absorber thickness);
2) inadequacy of the software
algorithms for discrimination between soft tissue and bone as well as
for interpolation of the soft tissue composition in areas containing
bone; and 3) inadequacy of
calibration algorithms and materials (1, 5). Because of such errors,
values for TBMC, FM, and FFM may be erroneously measured too high or
low. Probably because of the beam-hardening phenomenon, we have earlier found too-high values of FM and FM% in extreme obesity (A. Gotfredsen and J. Hilsted, unpublished observations), and others have found errors
in the fat percentage (6) or in the bone mineral density (7, 8) as a
function of the soft tissue thickness or composition.
To compensate for the beam hardening, the Norland XR-36 DEXA apparatus
introduces a modification of the DEXA method: instead of a constant
filtration of the X-ray beam, it has dynamically changing samarium
filtration. This system is unique to the Norland XR-36 and has not been
previously evaluated.
The purpose of the present study was to investigate the accuracy of the
Norland XR-36 for body composition analysis by measurements in vitro
and in vivo.
METHODS AND MATERIALS Norland XR-36 DEXA Apparatus
The Norland XR-36 uses a complex internal standard composed of a multistep aluminum (7075) wedge and a multistep acrylic wedge placed perpendicularly to each other, giving a multitude of combinations of thicknesses of aluminum and acrylic. This standard is scanned for calibration in the morning whenever measurements are to be made, giving rise to a floating average of the calibration coefficients. From the manufacturer, the calibration standard comes calibrated; that is, each combination of aluminum and acrylic steps (77 points in all) is convertible into amounts of "high-Z" equivalent (hydroxyapatite) and "low-Z" equivalent (plastic), a so-called "basis set decomposition" (14). During the calibration procedure, counts are made at each of the 77 points on the standard, and two-dimensional fourth-degree polynomials are fit by using the high- and low-energy count rates and the high-Z and low-Z equivalents for each step (or point). The result is a set of equations that calculate high-Z equivalent (g/cm2) as a function of high- and low-energy count rates, and another set of equations that calculate low-Z equivalent (g/cm2) as a function of high- and low-energy count rates. The polynomials are downloaded on the scanner. When the actual scans are done, the computer chooses the appropriate polynomials (for the filter used) and processes the high- and low-energy count rates into amounts of high-Z equivalent (bone) in g/cm2 or low-Z equivalent (soft tissue) in g/cm2. This is what is stored in the files and used for all subsequent calculations (17, 18). By the principle of basis set decomposition (14), the equivalent values are further transformed into values of fat or fat-free soft tissue in the pixels not containing bone. The criterion-standard materials used at Norland were calcium phosphate tribasic type IV for bone mineral, 0.6% sodium chloride solution for fat-free soft tissue, and stearic acid for fat.
Three different software modalities were used for data acquisition: 1) a stationary point measurement program called "hi-point," 2) a multipurpose scan program called "research scan," and 3) the "whole body scan." The hi-point program was used in a mode in which the data acquisition continued until a preset precision error of the count rates of maximum 0.0001 was obtained. The scan time, of course, depended on the absorber thickness but ranged from 1 to 15 min for each measurement. The research scan was used with resolutions (pixel size) of 1.5 × 1.5 or 3.0 × 3.0 mm. The whole body scan was used with a pixel size of 6.5 × 13.0 mm. The scan time of a subject in vivo with use of the whole body scan is ~20 min, and the entry dose of such a scan is <0.5 µSv.
The software version 2.4 was used for all data acquisition and analysis (except for the hi-point program, which was installed separately).
We have earlier investigated the precision errors in vivo in five normal-weight and five obese subjects. They were 2.2% for TBMC, 2.7% for FM, 2.7% for FFM, and 2.6% for FM%, (9a).
Experimental Setup
To investigate the accuracy of the Norland XR-36, four consecutive experiments of increasing complexity were performed: experiments A-D. Experiment A. This experiment comprised measurements on an anthropomorphic phantom constructed by Norland. This phantom has an aluminum (7075) skeleton and, furthermore, is composed of 1.2-cm sheets of acrylic and 0.5-mm sheets of polyvinyl chloride (PVC). It has three segments: head, trunk, and legs. Each segment has 12 sheets of acrylic and 12 sheets of PVC. The bottom sheets have the largest area, with the sheet area successively decreasing toward the top; therefore, the phantom has a stair- (or pyramid)-like appearance. It weighs a maximum of 36.2 kg (without aluminum skeleton) and has a total length of 116 cm. When the phantom is scanned, the plastic parts are placed on top of the aluminum skeleton. The thin PVC sheets are never used alone but always together with their corresponding 1.2-cm acrylic sheet. "Acrylic alone" ("configuration 5") was expected to simulate 70% fat, and acrylic+PVC ("configuration 1"; 1 sheet of each with the same area) was expected to simulate 20% fat. By changing the number of sheets, variation in tissue height was simulated and by changing the number of PVC sheets, variation in fat percentage was simulated. The whole body scan software assumes a constant fat percentage of 20% in the head (because the head has only little fat and because it is impossible to measure the fat percentage in the majority of the pixels where the cranial bones over- or underlie most of the soft tissue); thus the plastic parts of the head region were always used in the 20% fat configuration mode (acrylic+PVC). Therefore, the total phantom with all 12 sheets in the trunk and leg regions in the acrylic-alone configuration mode was nominally not 70% fat but 66.2% fat. The phantom was always changed (for its thickness or its fat percentage) by three sheets at a time (3 sheets = 1 "layer"). The phantom thus had four layers. The maximum height of the phantom was 15 cm, with the use of all 12 acrylic and all 12 PVC sheets, and the minimum height was 3.6 cm, with use of only 3 acrylic sheets. We assumed this to cover the range of tissue heights of a small person in the supine position. To consider an homogenous as possible "fat" distribution when all 12 PVC sheets were not present, the PVC sheets were not clustered together but scattered in the most randomlike manner from bottom to top in the phantom. For the two extreme configurations (configuration 1 = acrylic plus all corresponding PVC sheets = 20% fat; configuration 5 = acrylic alone in the chest and leg regions = 66.2% fat), the thickness was varied from one to four layers (1 layer = 3 sheets). For the intermediate configurations, only the total thickness of the phantom was measured (all 4 layers = all 12 sheets). "Configuration 2" had PVC sheets at sheets 1, 3-6, 8, 9, 11, and 12, and the nominal fat percentage of the total phantom was 31.4%. "Configuration 3" had PVC sheets at sheets 1, 4, 6, 8, 10, and 12, and the nominal fat percentage of the total phantom was 43.4%. "Configuration 4" had PVC sheets at sheets 2, 7, and 10, and the nominal fat percentage of the total phantom was 54.3%. The phantom was scanned by using the whole body scan software (pixel size 6.5 × 13 mm) and the research scan (pixel size 3.0 × 3.0 mm). All measurements were made in triplicate. Furthermore, stationary point measurements were made in the chest region by using the hi-point program. These measurements were made in duplicate. Because the point measurements in the chest region were not influenced by the constant presence of PVC in the head segment or by the random distribution of the PVC sheets, the nominal fat percentages were slightly different for the point measurements: configuration 1, 20.0%; configuration 2, 32.5%; configuration 3, 45.0%; configuration 4, 57.5%; and configuration 5, 70.0%. Experiment B. This experiment, comprised measurements on columns of ox muscle/lard. The basic unit of these soft tissue columns was a "brick" of ox muscle or lard ~4 cm in height, with a ground area of ~15 × 15 cm. Two different sets of bricks were measured. The first set included six bricks of ox muscle with weights from 308 to 640 g and six bricks of lard with weights from 582 to 887 g. In experiment B1, this set was used to investigate the accuracy of measuring amounts of tissue. The bricks of ox muscle or lard were measured separetely, in heights from one brick (~4 cm) to six bricks (~24 cm), with increments of one brick. We assumed this to cover the range of tissue heights of most normal-weight to moderately obese subjects (except for hands and fingers). The second set included five bricks of ox muscle with weights from 359 to 606 g and five bricks of lard with weights from 369 to 491 g. In experiment B2, this set was used to investigate the accuracy of measuring fat percentages. The bricks of ox muscle or lard were measured in successively changing combinations (i.e., 5 ox + 0 lard, 4 ox + 1 lard, 3 ox + 2 lard, and so on) to simulate different fat percentages, whereas, in this experiment, the height was kept constant (always 5 bricks, ~20 cm in height). The fat percentages of ox muscle and lard were measured by chemical fat extraction (Stein's Laboratory, Albertslund, Denmark). Ten samples of ox muscle had a fat percentage of 2.5 ± 1.0%, and 15 samples of lard had a fat percentage of 84.5 ± 5.1%. From these fat percentages and from the amounts of ox muscle and lard, respectively, the "true" fat percentages of the tissue columns in experiment B2 were calculated. They were 2.5, 22.0, 36.8, 54.5, 71.3, and 84.5%. Both the hi-point measurement program and the research scan (pixel size 1.5 × 1.5 cm) were used. All measurements were repeated six times. Because the hi-point measurement program only recognizes tissue heights and not the total amounts of the columns, the true fat percentages for point measurements in experiment B2 were slightly different: 2.5, 18.9, 35.3, 51.7, 68.1, and 84.5%. Each setup of tissue columns was measured together with 5 × 5-cm aluminum (7075) sheets of 4 or 12 cm. Experiment C. This experiment comprised DEXA scans of larger amounts of tissue placed on the scanner tabletop. The tissue was placed on the scanner longitudinally as rectangular piles to simulate the shape of the femoral or truncal region. The tissue was always scanned together with the aluminum spine from the above- mentioned semianthropomorphic phantom. Three lumps of lard were used, weighing 7.0, 8.8, and 7.6 kg, respectively. Three lumps of ox muscle were used, weighing 5.4, 6.4, and 5.4 kg, respectively. In experiment C1, increasing amounts of one tissue type were scanned separately, i.e., three configurations of lard, weighing 7.0, 13.8, and 21.4 kg, respectively, and three configurations of ox muscle weighing 5.4, 11.8, and 17.2 kg, respectively. Each tissue lump had a height of ~10 cm. Because tissue lumps were scanned while on top of each other, the maximum tissue height was ~30 cm, and the minimum tissue height was ~10 cm. We assumed this to cover the range of tissue heights of the trunk and proximal extremities of normal-weight and most obese subjects. The true fat percentages of pure lard and ox muscle for this experiment were determined to be 85 and 10%, respectively, on the basis of earlier chemical measurements (see Experiment B) and actual high-resolution research scans of the individual three tissue lumps. In experiment C2, the lumps of lard and ox muscle were scanned in different combinations, with true fat percentages of 10, 27.5, 37.9, 51.0, 63.9, and 85%, with corresponding total tissue amounts of 17.2, 15.4, 18.8, 11.9, 19.2, and 21.4 kg, respectively. The tissue height in this experiment was ~30 cm. In experiment C, all measurements were made in six repetitions. The whole body scan was used (pixel size = 6.5 × 13 mm). Experiment D. This experiment comprised DEXA scans of six subjects with lard and/or ox muscle placed on top of them. The subjects were two men with body weights of 71.5 and 83.5 kg and corresponding body mass index values of 24.5 and 27.6 kg/m2, respectively, and four women with body weights of 40.3-90.0 kg and corresponding body mass index values of 14.8-31.5 kg/m2. All subjects were healthy, except for one woman, who was anorectic. The additional tissue (lard, ox muscle) was placed on top of the subjects in an approximated anatomic manner, i.e., if possible uniformly over the lower abdomen and over the thighs. Eight different configurations of additional tissue were used, and all eight configurations were used on all six subjects, in addition to the baseline scan without any additional tissue. The tissue configurations were three lumps of lard (true fat percentage = 85%) with weights of 5.5, 10.7, and 15.9 kg, respectively; three lumps of ox muscle (true fat percentage = 10%) with weights of 5.6, 10.5, and 16.2 kg, respectively; and two combinations of lard and ox muscle with true fat percentages of 37 and 59% and weights of 10.0 and 16.4 kg, respectively. The additional tissue heights were 10-15 cm; therefore, we assumed the simulation to represent total tissue heights of moderately to severely obese subjects.Statistics
Comparison of paired variables were made by 1) traditional linear correlation and regression analysis including calculation of the standard error of the estimate (SEE); 2) calculation of the mean differences (MD) and their 95% confidence intervals (CI) based on calculation of the mean square error; and 3) (closely related to the CI of the MD) Student's t-test for paired data (two tailed, significance level P < 0.05). Whenever a difference was encountered and found significant by the t-test for paired data in a situation in which a regression analysis had also been made, the difference was tested for "regression toward the mean" effect (RTM effect) by using a modification of the method of Mee and Chua (16). The test for a positive "intervention" effect (as opposed to the RTM effect) was based on a one-sided t-test, and the t-statistic was calculated as the difference between the intercept and the mean x value divided by the product of the SEE and the square root of 1/n. Significance level was P < 0.05. Multiple comparisons were made by one- or two-way analysis of variance (ANOVA). Pairwise analyses were made by the Student-Newman-Keuls method. The Sigmastat statistical software (version 1.01) was used for these calculations. Significance level was P < 0.05.RESULTS
The present study was planned to examine the accuracy of DEXA for body composition analysis through study levels (or setups) of increasing complexity. If the results from a certain level were to disclose insufficient accuracy, the study was planned to halt until supposedly necessary changes in hardware or software had been made by the manufacturer. Having completed experiment A, we found unsatisfactory variability of BMC values between different phantom heights and unsatisfactory variability of fat percentage values with the point measurement software within the same setup (in some cases, coefficients of variation were >5%). We discovered, in cooperation with Norland, that the imprecision was caused by table variations at the location of the calibration standard. The problem was solved by changing the orientation of the calibration standard during calibration (perpendicular to the transversal scan direction instead of longitudinal to it). Furthermore, an error in the software for the calculation of the soft tissue mass and fat percentage was discovered and corrected. After soft tissue changes had been made, experiment A was redone (results appear in this study). The remaining part of the study (experiments B-D) was completed without further changes because the results showed sufficient accuracy.
The results of experiment A
(measurements of the plastic phantom) are shown in Tables
1-5.
Table 1 shows the nominal weights of the plastic phantom and the
weights measured by DEXA. The agreement was good, as documented by
correlation and regression analysis and MD and their CI. For
configuration 1, DEXA (whole body
scan) vs. nominal weight r = 0.999, SEE = 209.7 g, y = 1.03x 
18.10, MD = 725.4
g, and CI = 1261.5 to 189.2 g; DEXA (research scan) vs.
nominal weight r = 1.000, SEE = 20.4 g, y = 1.02x + 20.99, MD = 582.0 g,
and CI = 932.3 to 231.6 g. For
configuration 5, DEXA (whole body
scan) vs. nominal weight r = 1.000, SEE = 8.6 g, y = 1.03x + 88.24, MD = 791.1 g,
and CI = 1,229.9 to 352.3 g; DEXA (research scan) vs.
nominal weight r = 1.000, SEE = 15.6 g, y = 1.03x + 20.11, MD = 839.4 g, CI = 1,348.9 to 329.8 g. All four comparisons of measured
with nominal weights showed significant differences by using the
t-test, and the differences were not
due to the RTM effect. The tendency was slightly higher than nominal
weights by using of DEXA (see MD above).
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||||
Table 2 shows the nominal fat percentages of the different
configurations of the plastic phantom and the corresponding fat percentages measured by DEXA. The agreement was good, but it should be
noted that for configurations
1-3, which represent the clinically most common
fat percentages, DEXA showed values that were ~4% (fat) higher than
nominal. Correlation and regression analysis and MD and their CI were
as follows: whole body scan r = 0.999, SEE = 0.49%, y = 1.08x 6.54, MD = 2.86%, and
CI = 1.104.62%; research scan
r = 0.999, SEE = 0.56%,
y = 1.04x 6.36, MD = 4.48%, and
CI = 3.335.63%; point measurement
r = 0.999, SEE = 0.22%, y = 1.14x 8.09, MD = 1.64%, CI = 1.35 to 4.63%. The comparisons of measured with nominal fat
percentages involving the two scan modes showed significant differences
by using the t-test, and the
differences were not due to the RTM effect. The comparison of the point
measurement was not significantly different.
Table 3 shows the influence of the plastic thickness (number of layers) of the plastic phantom on the DEXA-measured values of fat percentages of the plastic phantom itself. The absolute differences between the layers were small. The largest difference between any two fat percentages within any fat percentage (plastic configuration) or scan mode was 3.8% fat (5.6%) (acrylic alone, whole body scan, difference between 4 layers and 1 layer). A two-way ANOVA was made for each configuration (configuration 1 or 5), including the comparisons between the effect of changing the number of layers and the effect of the scan mode (whole body scan or research scan). Because the nominal fat percentages for the point measurements were different from those for the two scan modes, the point measurements were analyzed separately by a one-way ANOVA. Regarding configuration 1, significant differences were found between the number of layers but not between the scan modes, allowing for the effect of the layers. There was a significant interaction between the effects of layer thickness and scan mode, i.e., the effect of changing the number of layers depended on which scan mode was used. Pairwise comparisons showed that the fat percentage of the whole body scan measurement of one layer was significantly different (~2.5% more fat percentage) from the whole body scan measurements of all the other layer thicknesses as well as from the research scan measurements of two- and three-layer thicknesses. No other statistically significant pairwise comparisons were found for configuration 1. With regard to configuration 5, significant differences were found between both the number of layers and between the two scan modes (whole body scan ~4.0% less fat percentage than the research scan), but there was no significant interaction between the effects of number of layers and scan mode. Pairwise comparisons showed significant differences between the fat percentages of one layer and those of the other three layer thicknesses (the fat percentage of one layer was ~3.0% less). No other statistically significant pairwise comparisons were found for configuration 5. No significant differences were found in the one-way ANOVAs of the point measurements.
Comparisons between DEXA values and nominal fat percentages. With regard to the low fat-percentage configuration (configuration 1), DEXA values were ~5% (fat) (~25%) higher than nominal. MD between nominal fat percentages and DEXA and their CI were as follows: whole body scan MD = 5.10%, CI = 3.266.94%; research scan MD = 5.05%, CI = 3.796.31%; point measurement MD = 4.80%, CI = 3.825.78%. The three comparisons showed significant differences by using the t-test. With regard to
the high fat percentage configuration (configuration
5), DEXA values were ~0.5% (fat) higher than
nominal. MD between nominal fat percentages and DEXA and their CI were as follows: whole body scan MD = 0.25%, CI = 2.16 to 2.66%;
research scan MD = 4.20%, CI = 1.966.44%; point measurement MD = 2.68%, CI = 0.30 to 5.05%. The comparisons of
the research scan and the point measurement showed significant
differences from nominal fat percentages by using the
t-test, whereas the comparison of the
whole body scan showed no significant difference.
Table 4 shows the influence of plastic thickness of the phantom on the
DEXA-measured BMC values of its aluminum skeleton. It can be seen that
the influence of the absorber thickness on the BMC values was quite
small regardless of the fat percentage and the software used. A two-way
ANOVA was made for each configuration (1 or
5), including the effect of changing
the number of layers and the effect of the scan mode (whole body scan
or research scan). With regard to configuration
1, significant differences were found among the BMC
values measured with different layer thicknesses of soft tissue
equivalent and between the two scan modes, but no significant
interaction was found between the effects of layer thickness and scan
mode. Pairwise comparisons showed significant differences among the BMC
values measured with all different soft tissue-equivalent layer
thicknesses, except the comparison between three and four layers. The
tendency was increasing BMC values with increasing soft tissue-
absorber thickness and higher BMC values with the whole body scan mode
compared with the research scan mode. The relative differences were
such that the largest difference in BMC between two layer thicknesses
was 1.6%, and the average difference between the whole body scan and
the research scan modes was 5.0%. With regard to
configuration 5, there were no
significant differences among the BMC values measured with different
layer thicknesses of soft tissue equivalent. There was a significant
difference between BMC values measured with the whole body scan and the
research scan, the measurements of the whole body scan being ~3.5%
higher. There was no significant interaction between layer thickness
and scan mode in this configuration.
Table 5 shows the influence of fat percentage (plastic configuration)
of the plastic phantom on the DEXA-measured BMC values of its aluminum
skeleton. Again, the influence of the plastic on the BMC values was
small. With use of the whole body scan software, there was a decline in
BMC values of 18.9 g (2.1%) from the "leanest" to the
"fattest" plastic configuration. With use of the research scan
software, there was no systematic shift in BMC values but a difference
of 47.3 g (5.3%) between the highest and lowest BMC values. On
average, the whole body scan BMC values were 26.2 g (3.0%) higher than
those measured by the research scan. A two-way ANOVA showed significant
differences in the BMC values, both between the configurations and
between the scan modes, and there was a significant interaction between
configuration and scan mode, i.e., the effect of changing the
configuration depended on which scan mode used. Pairwise comparisons
showed that all configurations differed significantly, except for the
BMC values of configuration 3 vs.
those of configurations 2 and
4.
The results of experiment B (small
columns of tissue) are given in Tables 6
and 7. The results shown are those of
measurements of the tissue columns together with 4-mm aluminum. Use of
12-mm aluminum gave identical results (not shown).
|
||||||||||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||
27.0 g, CI = 49.0 to 4.9 g; muscle r = 1.000, SEE = 14.1 g, y = 1.0x + 37.6, MD = 18.8 g, CI = 35.3 to 2.4 g. The two comparisons showed significant differences by using the t-test, and
the differences were not due to RTM effect.
There was also, in the columns of tissue, a very good agreement between
chemically determined fat percentage and fat percentages determined by
DEXA (experiment B2, Table 7).
Correlation and regression analysis and MD and CI were as follows:
research scan r = 0.999, SEE = 1.8%,
y = 1.0x 2.6, MD = 3.5%, and
CI = 1.75.2%; point measurement
r = 1.000, SEE = 0.9%,
y = 1.0x 0.8, MD = 1.2%, and CI = 3.3 to 0.9%. The comparison of the research scan showed
significant difference by using the
t-test, and the difference was not due
to RTM effect. The comparison of the point measurement showed no
significant difference.
The results of experiment C (larger
lumps of tissue together with aluminum spine from plastic phantom) are
shown in Fig. 1, A-D. The agreement between total soft
tissue placed on the scanner, the true fat percentage, and the
calculated amounts of total fat-free and fat tissues on the one hand
and the DEXA measurements of the corresponding variables on the other
was very good. Correlation coefficients, SEE, and the regression
equations are given in the figure. The MD and their CI
were the following: total soft tissue MD = 187.7 g, CI = 325.3 to 50.1 g; fat percentage MD = 0.38%, CI = 1.37 to 0.61%; total fat free tissue MD = 33.5 g,
CI = 204.3 to 137.4 g; and total fat tissue MD = 154.3 g,
CI = 363.1 to 54.6 g. The comparison of the total soft tissue
showed significant difference by using the
t-test, and the difference was not due to RTM effect. The remaining three comparisons showed no significant differences.
The results of experiment D (tissue on subjects) are shown in Fig. 2, A-D. There was a very good agreement between true values and DEXA measurements regarding total amount of soft tissue added (Fig. 2A), fat percentage of soft tissue added (Fig. 2B), amount of fat added (Fig. 2C), and amount of fat-free soft tissue added (Fig. 2D). The correlation coefficients, SEE, and the regression equations are given in the figure. MD and their CI were as follows: total soft tissue MD = 620.9 g, CI = 457.7
784.1 g; fat percentage MD = 3.14%, CI = 2.61 to 8.89%; fat MD = 544.3 g, CI = 405.6 to 1494.2 g; and fat-free soft tissue MD = 83.7 g, CI = 999.8 to
1,167.2 g. The comparison of the total soft tissue showed significant difference by using the t-test, and
the difference was not due to RTM effect. The remaining three
comparisons showed no significant differences.
Figure 3 shows the mean total BMC values of the subjects in experiment D as a function of the amount and fat composition of the soft tissue placed on them. There were no significant differences between the TBMC values of the different configurations (one-way ANOVA) but a tendency toward a small overestimation of TBMC with a large cover of very fat tissue.
DISCUSSION
The present data demonstrate that DEXA has a high accuracy for body composition measurements. It is obvious that, in a strict sense, this conclusion is specific to the Norland XR-36 because there are substantial technical differences among the different DEXA-scanner brands. They use different hardware, X-ray sources, X-ray spectra, detectors, calibration standards, and software algorithms. However, the basic DEXA principle is the same and, therefore, some extrapolation from the present data may be made to cover other DEXA brands, such as the Lunar and the Hologic scanners.
Many of the comparisons in the present study turned out to show statistically significant differences. These results serve, of course, to stress the fact that soft tissue thickness and composition have significant influence on DEXA measurements of soft tissue thickness (or mass) and composition themselves, as well as some influence on BMC measurements. It is obvious that care should be taken in the use and interpretation of DEXA measurements of body composition. However, although statistically significant, most of the observed differences were small (5% or less). The reason why such small differences turn out to be statistically significant is, of course, in part, the very high precision of the DEXA measurements. However, the DEXA method has technical limitations that expectedly may influence measurements of body composition. The most important of these is the fact that DEXA can only determine the amount, but not the composition, of the soft tissues concealed by bone, which therefore has to be extrapolated from the composition of the adjacent soft tissues. Other error sources are atypical fat distribution and very high thicknesses of tissue (in very obese subjects), which may give rise to the so-called beam-hardening effects.
Most studies regarding the validity of DEXA for body composition
analysis have dealt with the comparison in vivo between DEXA and other
body composition determination methods (2, 3, 13). The knowledge that
can be summarized from the in vivo studies is that, in general, there
is a good agreement between DEXA and classic reference methods (total
body water, underwater weighing, total body potassium) in
the different populations studied. The comparisons have seldom been
unfavorable for DEXA because the agreement between DEXA and any other
method has most often been as good or better than the agreements among
the classic reference methods themselves (2, 3, 13). Fuller et al. (4)
have found a bias of 6% for FM% by DEXA compared with FM% by
total body water
(D2O) in elderly men.
There are only a few truly accurate studies on DEXA for body
composition analysis. Haarbo et al. (6) found a perfect agreement between DEXA fat percentages and chemically measured fat percentages in
15-cm columns of mixtures of ox muscle and lard by using the Lunar
apparatus. They also found, however, that the fat percentage of an
ethanol-water mixture was seen by DEXA to decrease from 28 to 22% when
the fluid height was increased from 8 to 16 cm and again to increase to
27% when the fluid height was further increased to 24 cm. Svendsen et
al. (20) found a very good agreement between DEXA fat percentages and
chemically measured fat percentages (after total body grinding and
homogenization) in seven pigs (35-95 kg;
r = 0.98, SEE = 2.9%,
y = 0.9x 1.1%) also by using the
Lunar scanner. In the same paper, Svendsen et al. presented data on six
healthy human subjects scanned with and without 8.8 kg of lard placed
on their ventral body side. No significant difference was seen between
measured (DEXA) and expected change in FFM or FM. TBMC changed ~7%,
which was, however, compensated for by the number of pixels because the
total body mineral density did not change significantly. Our data using
the Norland XR-36 were quite comparable to these results because we
found good agreement between measured (DEXA) and expected change in
FFM, FM, and fat percentage in our six subjects with added tissue
combinations of up to 16.2 kg. DEXA accurately predicted both the
amount of soft tissue added (SEE = 0.21 kg; Fig.
2A) and the fat percentage of the
added tissue (SEE = 5.6%; Fig. 2B).
The slightly higher SEE in this study, compared with the pig study by
Svendsen et al. (20), reflects the physiologically very extreme
situation of placing 16 kg of lard or ox muscle on top of a human. In
contrast to the Svendsen et al. study, we did not find any significant
differences between TBMC measurements after addition of soft tissues.
Similarly, Jensen et al. (12) found only a minor effect on TBMC in nine
healthy subjects scanned by a Hologic QDR-1000W DEXA scanner with up to 23 kg of lard placed anteriorly. Tothill et al. (21) have compared TBMC
and fat percentage of three different DEXA scanners (Norland XR-36,
Hologic QDR-1000W, Lunar DPX) by using the same semianthropomorphic plastic phantom as in the present study and measurements in 11 volunteers. There were high correlations among the three
scanners, but there were significant differences. For TBMC
measurements, they found Lunar > Hologic > Norland. For fat
percentage, they found Norland > Lunar > Hologic. Hassager et al.
(9) placed 14 kg of lard on the trunks of five lean men and measured
them in a dual-photon absorptiometry scanner. They found no changes in
TBMC and a good agreement between measured and expected changes in FM
and FFM. It must be remembered, however, that dual-photon absorptiometry, the predecessor of DEXA, is quite different from DEXA;
therefore, an extrapolation of these results to DEXA may not be made.
It should also be pointed out that caution should be practiced in
comparisons with previous publications because most of the previous
studies were made by using different software versions than are now
used on commercial densitometers.
The nominal fat percentages of the plastic phantom were given by the manufacturer on the basis of theoretical considerations of the molecular density of the materials. The nominal fat percentages were 70% for one 1.2-cm acrylic sheet and 20% for one 1.2-cm acrylic sheet plus one 0.5-mm PVC sheet. From these percentages and the actual configuration of the phantom, the nominal fat percentages of any configuration were calculated. From Tables 2 and 3, the impression is that these nominal fat percentages were not the true fat percentages. For configuration 5, the DEXA and the nominal fat percentages were quite close (slightly higher for DEXA scans and slightly lower for DEXA point measurements). For configuration 1, however, the DEXA fat percentages were consistently ~5% higher than nominal. There were also differences between DEXA-derived fat percentages and chemically derived fat percentage values in tissue (Table 7). However, these differences (or errors) were smaller: on average, 3.5% fat overestimation by the DEXA research scan and 1.2% fat underestimation by the DEXA point measurement. We suggest that the nominal fat percentages of the plastic phantom may be reevaluated for further future use.
In conclusion, the Norland XR-36 DEXA scanner is accurate for FM, FFM, and TBMC measurements in both small objects of mammal tissue and in humans spanning a wide range of body configurations. The accuracy errors in vivo for soft tissue changes are ~200 g for total soft tissue mass and ~1,000 g for changes in FM and FFM. These figures must be remembered when clinical studies are planned or when the method is used in daily clinical work.
ACKNOWLEDGEMENTS
We are indebted to Todd Leroy and Mike Grman of the Norland Corporation for technical assistance.
FOOTNOTES
Address for reprint requests: A. Gotfredsen, Dept. of Internal Medicine, Slagelse Hospital, Ingemannsvej 18, DK-4200 Slagelse, Denmark.
Received 15 May 1995; accepted in final form 13 November 1996.
REFERENCES
| 1. | Blake, G. M., D. B. McKeeney, S. C. Chhaya, P. J. Ryan, and I. Fogelman. Dual energy X-ray absorptiometry: the effects of beam hardening on bone density measurements. Med. Phys. 19: 459-465, 1992. [Medline] . |
| 2. |
Friedl, K. E.,
J. P. DeLuca,
L. J. Marchitelli,
and
J. A. Vogel.
Reliability of body-fat estimations from a four-compartment model by using density, body water, and bone mineral measurements.
Am. J. Clin. Nutr.
55:
764-770,
1992.
|
| 3. | Fuller, N. J., S. A. Jebb, M. A. Laskey, W. A. Coward, and M. Elia. Four-component model for the assessment of body composition in humans: comparison with alternative methods, and evaluation of the density and hydration of the fat free mass. Clin. Sci. Lond. 82: 687-693, 1992. [Medline] . |
| 4. | Fuller, N. J., M. B. Sawyer, M. A. Laskey, P. Paxton, and M. Elia. Bias between two reference methods of body composition assessment in elderly men (Abstract). Clin. Nutr. 13, Suppl. 1: P25, 1994. |
| 5. | Goodsitt, M. M. Evaluation of a new set of calibration standards for the measurement of fat via DPA and DEXA. Med. Phys. 19: 35-44, 1992. [Medline] . |
| 6. | Haarbo, J., A. Gotfredsen, C. Hassager, and C. Christiansen. Validation of body composition by dual energy X-ray absorptiometry (DEXA). Clin. Physiol. Oxf. 11: 331-341, 1991. . |
| 7. | Hangartner, T. N., and C. C. Johnston. Influence of fat on bone measurements with dual-energy absorptiometry. Bone Min. 9: 71-81, 1990. . |
| 8. |
Hansen, M. A.,
C. Hassager,
K. Overgaard,
U. Marslew,
B. J. Riis,
and
C. Christiansen.
Dual-energy X-ray absorptiometry: a precise method of measuring bone mineral density in the lumbar spine.
J. Nucl. Med.
31:
1156-1162,
1990.
|
| 9. | Hassager, C., S. S. Sørensen, B. Nielsen, and C. Christiansen. Body composition measurement by dual photon absorptiometry: comparison with body density and total body potassium measurements. Clin. Physiol. Oxf. 9: 353-360, 1989. . |
| 9a. | Hendel, H. W., A. Gotfredsen, A. Andersen, L. Højgaard, and J. Hilsted. Body composition during weight loss in obese patients estimated by dual energy X-ray absorptiometry and by total body potassium. Int. J. Obes. 20: 1111-1119, 1996. . |
| 10. | Heymsfield, S. B., and M. Waki. Body composition in humans: advances in the development of multicompartment models. Nutr. Rev. 49: 97-108, 1991. [Medline] . |
| 11. |
Heymsfield, S. B.,
J. Wang,
J. J. Kehayias,
S. Heshka,
S. Lichtman,
and
R. N. Pierson, Jr.
Chemical determination of human body density in vivo: relevance to hydrodensitometry.
Am. J. Clin. Nutr.
50:
1282-1289,
1989.
|
| 12. | Jensen, L. B., F. Quaade, and O. H. Sørensen. Bone loss accompanying voluntary weight loss in obese humans. J. Bone Min. Res. 9: 459-463, 1994. [Medline] . |
| 13. |
Johansson, A. G.,
A. Forslund,
A. Sjödin,
H. Mallmin,
L. Hambraeus,
and
S. Ljunghall.
Determination of body compositiona comparison of dual-energy X-ray absorptiometry and hydrodensitometry.
Am. J. Clin. Nutr.
57:
323-326,
1993.
|
| 14. | Lehmann, L. A., R. E. Alvarez, A. Macovski, and W. R. Brody. Generalized image combinations in dual KVP digital radiography. Med. Phys. 8: 659-667, 1981. [Medline] . |
| 15. |
Mazess, R. B.,
H. S. Barden,
J. P. Bisek,
and
J. Hanson.
Dual-energy X-ray absorptiometry for total-body and regional bone-mineral and soft-tissue composition.
Am. J. Clin. Nutr.
51:
1106-1112,
1990.
|
| 16. | Mee, R. W., and T. C. Chua. Regression toward the mean and the paired sample t-test. Am. Statistician 45: 39-42, 1991. . |
| 17. | Nord, R. H., and R. K. Payne. Body
composition by DEXAa review of the technology. Asia
Pac. J. Clin. Nutr. 3, Suppl. 1: S1-S6, 1994.
|
| 18. |
Nord, R. H.,
and
R. K. Payne.
Computation of bone density from dual energy X-ray dataan empirical method (Abstract).
Calcif. Tissue Int. Suppl.
44:
5-62,
1989. .
|
| 19. | Roubenoff, R., and J. J. Kehayias. The meaning and measurement of lean body mass. Nutr. Rev. 49: 163-175, 1991. [Medline] . |
| 20. |
Svendsen, O. L.,
J. Haarbo,
C. Hassager,
and
C. Christiansen.
Accuracy of measurements of body composition by dual-energy X-ray absorptiometry in vivo.
Am. J. Clin. Nutr.
57:
605-608,
1993.
|
| 21. | Tothill, P., A. Avenell, J. Love, and D. M. Reid. Comparisons between Hologic, Lunar and Norland dual-energy X-ray absorptiometers and other techniques used for whole-body soft tissue measurements. Eur. J. Clin. Nutr. 48: 781-794, 1994. [Medline] . |
| 22. | Vogel, J. A., and K. E. Friedl. Body fat assessment in women. Special considerations. Sports Med. 13: 245-269, 1992. [Medline] . |
| 23. | Wahner, H. W., W. L. Dunn, M. L. Brown, R. L. Morin, and B. L. Riggs. Comparison of dual-energy X-ray absorptiometry and dual photon absorptiometry for bone mineral measurements of the lumbar spine. Mayo Clin. Proc. 63: 1075-1084, 1988. [Medline] . |
This article has been cited by other articles:
|
|
 |
|
  A. Larkin, N. Sheahan, U. O'Connor, L. Gray, A. Dowling, E. Vano, P. Torbica, D. Salat, A. Schreiner, V. Neofotistou, et al. QA/ACCEPTANCE TESTING OF DEXA X-RAY SYSTEMS USED IN BONE MINERAL DENSITOMETRY Radiat Prot Dosimetry, April 1, 2008; (2008) ncn086v1. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G J Slater, G M Duthie, D B Pyne, and W G Hopkins Validation of a skinfold based index for tracking proportional changes in lean mass. Br. J. Sports Med., March 1, 2006; 40(3): 208 - 213. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Storgaard, C. B. Jensen, M. Bjornholm, X. M. Song, S. Madsbad, J. R. Zierath, and A. A. Vaag Dissociation between Fat-Induced in Vivo Insulin Resistance and Proximal Insulin Signaling in Skeletal Muscle in Men at Risk for Type 2 Diabetes J. Clin. Endocrinol. Metab., March 1, 2004; 89(3): 1301 - 1311. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Storgaard, X. M. Song, C. B. Jensen, S. Madsbad, M. Bjornholm, A. Vaag, and J. R. Zierath Insulin Signal Transduction in Skeletal Muscle From Glucose-Intolerant Relatives With Type 2 Diabetes Diabetes, December 1, 2001; 50(12): 2770 - 2778. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. M. Rosenfalck, S. Maghsoudi, S. Fisker, J. O. L. Jørgensen, J. S. Christiansen, J. Hilsted, A. Vølund, and S. Madsbad The Effect of 30 Months of Low-Dose Replacement Therapy with Recombinant Human Growth Hormone (rhGH) on Insulin and C-Peptide Kinetics, Insulin Secretion, Insulin Sensitivity, Glucose Effectiveness, and Body Composition in GH-Deficient Adults J. Clin. Endocrinol. Metab., November 1, 2000; 85(11): 4173 - 4181. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
