PAPER

Validation of dual energy X-ray absorptiometry (DXA)

by comparison with chemical analysis of dogs and cats
JR Speakman
1
*, D Booles
2
and R Butterwick
2
1
Aberdeen Centre for Energy Regulation and Obesity (ACERO), Department of Zoology, University of Aberdeen, Aberdeen, Scotland,
UK; and
2
Waltham Centre for Pet Nutrition, Waltham-on-the Wolds, Melton Mowbray, Leicestershire, England, UK
BACKGROUND: Dual-energy X-ray absorptiometry (DXA) has been used extensively to measure body composition, but has
been validated by comparison to chemical analysis on relatively few occasions. Moreover, these previous validation studies have
ground up entire carcasses prior to chemical analysis, thus potentially obscuring sources of error in the DXA analysis.
OBJECTIVE: The purpose of this study was to validate DXA by comparison to chemical analysis in dogs and cats, performing
chemical analysis on dissected rather than ground carcasses to reveal sources of discrepancy between the two methods.
DESIGN: Sixteen animals (10 cats and 6 dogs weighing between 1.8 and 22.1kg) were scanned by DXA post-mortem using a
Hologic QDR-1000 W pencil beam machine and then dissected into 22 separate components. Individual tissues were dried and
then sub-sampled for analysis of fat content by Soxholet extraction, or ashing in a mufe furnace. Body composition by DXA was
compared to body composition by chemical analysis and discrepancies between the two correlated with chemical composition
of individual tissues. We also explored the capability of the machine to establish the fat contents of mixtures of ground beef, lard
and water.
RESULTS: DXA estimates were strongly correlated with estimates derived from chemical analysis: total body mass (r 1.00),
lean tissue mass (r 0.999), body water content (r 0.992) and fat mass (r 0.982). Across individuals the absolute and
percentage discrepancies were also small: total body mass (13.2g, 1.02%), lean tissue mass (119.4g, 2.64%), water content
(101g, 1.57%) and fat content (28.5 g, 2.04%), where the percentage error is expressed relative to the average mass of that
component across all individuals. Although on average DXA compared very well to chemical analysis, individual errors were
much greater. Individual errors in the lean tissue and fat tissue components were strongly correlated with the fat content of
skeletal muscle and the lean content of mesenteric fat. The error in the DXA estimate of total fat content was related to skeletal
muscle hydration. Experimental studies using mixtures of lean ground beef, water and lard indicated that tissue hydration may
have important effects on the perception of tissue fat content by DXA. Bone mineral content by DXA was approximately 30%
lower than whole body ash content but only 7.7% lower than ash content of the bones.
CONCLUSIONS: On average pencil beam DXA analysis using the Hologic QDR-1000 W machine provides an accurate estimate
of body composition in subjects weighing between 1.8 and 22.1kg. Individual discrepancies, however, can be large and appear
to be related to lean tissue hydration.
International Journal of Obesity (2001) 25, 439 447
Keywords: body composition; dual-energy X-ray absorptiometry; DXA; DEXA; obesity; fat content; lean content
Introduction
Dual-energy X-ray absorptiometry (DXA) or quantitative
density radiography (QDR) was originally developed as a
tool for the assessment of bone mineral contents of subjects
in the context of the diagnosis of osteoporosis. Over the past
decade, however, DXA has become an increasingly impor-
tant tool for the measurement of soft tissue body composi-
tion. Applications of the method include studies of
correlations of fat content to heavy exercise,
1
effects of
dietary intervention on tissue composition
2,3
and the
*Correspondence: JR Speakman, Aberdeen Centre for Energy Regulation
and Obesity (ACERO), Department of Zoology, University of Aberdeen,
Aberdeen AB24 3TZ, UK.
E-mail: J.Speakman@abdn.ac.uk
Received 31 January 2000; revised 4 September 2000;
accepted 22 September 2000
International Journal of Obesity (2001) 25, 439447
2001 Nature Publishing Group All rights reserved 03070565/01 $15.00
www.nature.com/ijo
International Journal of Obesity (2001) 25, 439447
2001 Nature Publishing Group All rights reserved 03070565/01 $15.00
www.nature.com/ijo
impact of various disease states on body composition.
4 7
DXA has several advantages, which include the speed of
measurement and the need for minimal subject instruction.
This has made it a particularly useful method in the pediatric
arena.
8 10
Many studies have assessed the precision of DXA analysis
and have shown very good within-subject repeatability of
the measurements.
11 14
Although there has also generally
proved to be good comparability between DXA and other
indirect methods of assessing body composition, such com-
parability may to an extent only reect the fact that indirect
methods make similar errors. Relatively few comparisons
have been made between DXA estimates of body composi-
tion and evaluations by the `gold standard' method of
chemical analysis, for animals weighing over 1kg,
15 23
all
of which have used pigs as the subject animal. Such studies
have revealed that on average the DXA estimates of fat and
lean components generally compare favorably with the
extracted fat contents and the combined non-fat organic
fraction plus water. Across studies the discrepancies for these
two components average less than 15%,
15 23
although some
studies have revealed greater average discrepancies.
16
Several
of these studies have indicated that DXA estimates of bone
mineral content underestimate the total body mineral con-
tent ( ash) by between 30 and 50%, which is surprising
given the purpose for which the machines were originally
developed.
Although these means give us condence that `on aver-
age' the DXA methodology provides useful assessments of
soft tissue body composition, individual discrepancies
between DXA estimates and measurements made by chemi-
cal analysis can be large. For example, although the mean
discrepancy in fat content in the study by Mitchell et al
21
was
only 0.4%, some DXA evaluations of fat content differed by
up to 50% from the actual fat contents by extraction. These
individual discrepancies are important because DXA is often
applied in the clinical setting on an individual basis.
Understanding the nature of such discrepancies however
is obscured by the methodology that is traditionally
employed to perform chemical analysis. During chemical
analysis the entire carcass is generally ground up into a ne
homogenate. Sub-samples of this homogenate can then be
dried, fat extracted and ashed to determine the average
chemical composition of the original subject. This approach
has two important advantages. It obviates the need to dry
out entire carcasses, which would involve prohibitive quan-
tities of material, and the number of analyses is massively
reduced because individual body components are not ana-
lysed separately. However, pooling the component tissues
potentially obscures the sources of error in the DXA assess-
ment. For example, DXA machines operate by quantifying
the attenuation of two X-ray beams of different energies.
Different elements attenuate X-rays by different amounts.
Because two different energies are employed, it is possible to
separate the attenuation due to two different sources. Hence,
in soft tissue, it is possible to assess the relative contribution
of lean and fat tissues, because the elemental composition of
these tissues differs. However, where a third material is
present, such as bone, it is only possible to resolve the
attenuation as that due to bone and that due to soft tissue.
In areas where bone is found the software makes an assump-
tion about the regional distribution of fat by interpolating
the probable lean and fat composition of the soft tissue
component from that of the immediately adjacent tissues.
This may provide a reasonable estimate of the soft tissue
composition, but even this approach does not allow an
assessment of the composition of the brain because it is
encased in the skull, with no adjacent tissues from which
interpolation can be made. Instead the estimated mass of the
brain is assumed to comprise 17% fat.
24
The extent to which
this assumption contributes to errors in the nal fat content
cannot be evaluated by traditional chemical validations
because the brain is pooled with the rest of the body
during the homogenization of the carcass.
In the current study we aimed to compare DXA analysis of
body composition with chemical analysis in domestic dogs
and cats weighing between 1.8 and 22.1kg. This covers the
range of body weights most often employed in pediatric
applications of the DXA methodology to assess body com-
position. Instead of using the traditional method of grinding
up entire carcasses, however, we aimed to dissect the subjects
into their component body parts and perform chemical
analyses for each of these components separately. We
hoped that by taking this approach we might be able not
only to quantify the accuracy of the DXA methodology but
also provide some insight into the sources of individual
discrepancies.
Subjects and methods
Subjects
Ten cats and six dogs varying from 1.8 to 22.1kg (mean
6.4kg) and aged between 8 weeks and 13y were used in the
study. The animals were housed at the Waltham Centre for
Pet Nutrition (Melton Mowbray, England). Animals that had
been scheduled for euthanasia over a period of 4y, for other
clinical reasons, were employed in this protocol. No animals
were deliberately euthanized to take part in this validation
study.
Body composition by DXA
Following euthanasia animals were scanned using a Hologic
QDR-1000 W pencil beam dual-energy X-ray absorptiometer
(Hologic, Waltham, MA). Animals were positioned on their
sides with their fore and hind legs separated and a phantom
calibration block placed immediately adjacent to them.
Scanning took between 10 and 20min. The resultant
images were analysed using the appropriate software for
their body mass (standard analysis, version 6.20A), either
whole body for animals weighing over 10kg or infant whole
body (pediatric) for animals weighing less than 10kg. The
Validation of DXA for body composition
JR Speakman et al
440
International Journal of Obesity
enhanced analysis package was not available for pencil beam
machines when these scans were made. Estimates of total
body mass, lean body mass, fat mass and bone mineral
content were derived. Previous assessments of measurement
repeatability using this particular machine and software on
dog and cat subjects have already been published.
25
Follow-
ing scanning the animals were deep frozen (20

C) and
shipped to Aberdeen on dry ice for chemical analysis.
Body composition by chemical analysis
After arrival in Aberdeen animals were weighed and allowed
to thaw overnight inside plastic bags. Animals were dissected
into 21 separate body components. These comprised: brain,
tongue, pelage, subcutaneous fat, heart, lungs, stomach,
small and large intestines, pancreas, mesenteric fat, pericar-
dial fat, perirenal fat, liver, kidneys, spleen, tail and the
reproductive organs if present. Contents of the alimentary
tract were squeezed out of the tract into a separate vessel, but
the tract was not ushed with ringers and probably con-
tained some residual contents. The skeletal muscle was
dissected off all the bones as much as possible and the
muscle was analysed separately from the remaining carcass,
which consisted of bones and residual bits of muscle that
could not be detached. Dissection was generally performed
by a team of two or three people and took a total of 3 6h for
each animal.
Body components were placed into pre-weighed glass
vessels and reweighed to 0.01g to yield the wet mass of
tissue. Larger components, particularly the carcass, skeletal
muscle and fat components were subdivided into several
vessels. Immediately following weighing the vessels with
the tissues were placed into a fan-assisted drying oven
(Gallenkamp) at 60

C. This temperature is sufciently low

to dry the tissue, avoiding volatization of fatty acids. Samples
were dried to constant mass, which took 14 days. After 14
days samples were removed from the oven and allowed to
cool to room temperature in a desiccator before reweighing.
The difference in mass between the dry and wet sample
masses was assumed to be water. Inevitably some evapora-
tion of water occurred during the dissection process and
there was a discrepancy between the total mass of the
animal prior to dissection and the total wet weights of the
component tissues. This discrepancy amounted to between
0.6 and 1.2% of the total body weight. We assumed this loss
was proportional to water contents measured for the indivi-
dual components, and we thus reconstructed total water
contents from each tissue by adding back the `loss during
dissection' to each tissue. Dried tissues were ground to a ne
homogenate using a mortar and pestle and the resultant
homogenate was sub-sampled for fat extraction. Two sub
samples of approximately 5g of the tissue were weighed to
0.001 g and placed in lter paper envelopes inside an asbes-
tos thimble. Samples were placed in a Soxholet fat extractor
for 24h using diethyl ether as the solvent. The lter paper
envelopes were then removed and residual ether allowed to
evaporate in air before they were re-dried at 60

C for 24 h
and reweighed. The difference in mass between the pre- and
post-extraction was the extracted fat. We used the average of
the two independent determinations. A further weighed sub-
sample of about 5g of the dry material was burned in a
mufe furnace at 500

C overnight in porcelain crucibles to

establish the organic and ash contents. The lean organic
content was evaluated from the difference between the
estimated total organic fraction and the fat content from
Soxholet.
Total body composition was reconstructed from the indi-
vidual analyses of each tissue. Total fat content was the
summed extracted fat, total body lean content was the sum
of the water and the non-fat organic fraction across all
tissues, and total ash was the summed ash fraction across
all tissues. We also calculated the ash content derived only
from bones by summing across only the vessels containing
bone tissue.
In vitro composition studies
Several previous studies have explored the ability of DXA
machines to detect quantities of lard, beef and bone, either
added to subjects
12,13,17,26
or measured alone.
14,27
We
extended these previous investigations by exploring the
ability of the machine to measure the fat content of
ground beef when lard or water was added to it. Measure-
ments were made on four 2.5 kg samples of lean ground beef.
Two of the samples had 500 g of lard added to them in 25g
aliquots. After each 25g addition the lard was mixed into the
beef and it was formed into a block with a depth of 5cm, and
each block was scanned twice using the infant whole body
package (pediatric). In a second series of manipulations 250 g
of water was added to two further 2.5kg samples of beef in
aliquots of 25 g. After each addition the water was mixed
into the meat, and the blocks were scanned twice.
Statistics
All the variables were log-converted to normalize them prior
to analysis. We examined the relationship between the DXA
and chemical analyses of body composition using regression
analysis and correlation. The mean absolute discrepancy
between the two methods was the difference in mass of
tissue averaged across all 16 individuals. Negative and posi-
tive discrepancies cancel in this calculation, so we also
calculated the mean deviation, which is the average ignoring
the signs of the differences. Individual discrepancies between
the methods were also expressed as an average percentage of
the mass of the component in that individual by chemical
analysis, and this was also calculated either including or
excluding the direction of the differences.
To explore the underlying causes of the discrepancies
between the methods we regressed the difference between
DXA and chemical analysis on the individual components of
the body composition analysis using stepwise multiple
Validation of DXA for body composition
JR Speakman et al
441
International Journal of Obesity
regression with both backward deletion and forward inclu-
sion. In the beef and lard experiment we explored the
relationship between the measured fat content by DXA and
the expected change in measured fat content as different
amounts of fat and water were added to the samples using
regression analysis.
Results
Total body mass
There was a strong correlation (Figure 1a: r 1.000) between
the body mass assessed by DXA and the total mass by
weighing prior to dissection (Figure 1a). The least squares
t regression equation (F
1,14
44774.5, P<0.0001) had an
intercept which did not differ signicantly from 0
(0.00253, P0.951) and a gradient that did not differ
signicantly from 1.0 (1.00032, P0.987). The absolute
difference averaged13.2 g (range 344.3 to 89.5) and in
percentage terms this averaged 1.016% of the gravimetric
estimate of body mass.
Lean tissue
There was a strong correlation between the estimated lean
tissue by DXA and the combined weights of the non-fat
organic fraction and the water content (Figure 1b). The least-
squares t regression equation log
e
lean0.0074 0.9961
log
e
DXA(lean) explained 99.6% of the variation in the
chemical analysis (F
1,14
3123.2). On average the discre-
pancy between the DXA and the chemical analysis averaged
119.4g (s.d. 304.7, s.e. 76.2, range270.1 to 1031.6g). This
was equivalent to an average discrepancy in the lean tissue of
Figure 1 Relationships between (a) whole body mass, (b) lean body mass, (c) fat mass, and (d) bone mineral content of 16 dogs and cats measured by
DXA and by chemical analysis (see text for details of chemical analyses). All the data were log converted to normalize their distributions.
Validation of DXA for body composition
JR Speakman et al
442
International Journal of Obesity
2.64% (s.e. 1.25). However, individual estimates ranged
from2.46% (ie underestimated) to 13.22% (overestimated).
None of these discrepancies were related to sex, species or
castrate status of the subjects or the version of software used
in the analysis.
Body water
DXA allows a calculation to be made of the expected body
water by multiplying the lean component by a constant
factor. The recommended value is 0.742, implying that
lean tissue component of the body is 74.2% water. From
our chemical analysis the water content of lean tissues
averaged across the individuals was 73.2% (s.e. 2.1,
n16). The water content evaluated by DXA was conse-
quently closely related to the water content evaluated by
chemical analysis. The least-squares t regression log
e
water
0.446 1.0548 log
e
DXA(water) explained 98.7% of the
variation in water content (F
1,14
1096.5, P<0.001). The
mean absolute discrepancy was only 101 g (s.e. 124g) and
the mean error was only 1.57% of the water content by
desiccation. Yet, despite this close average correspondence,
individual values were very discrepant from the mean, and
varied from underestimations of water content by 18.4% to
overestimations by 26.6%. Mean deviation (ignoring sign)
was 6.7% (0.1 26.6%). None of these discrepancies were
related to sex, species or castrate status of the subjects or the
version of software used in the analysis. Rather the large
individual variation occurred because, although the average
tissue content was close to the assumption made in using the
DXA machine to estimate water content, the range of tissue
water contents in the animals was very diverse (59.4
88.4%). This variability was not related to the age of the
animals in the sample (r
2
7.9%, F1.20, P0.292) or other
factors such as the sex, species or castrate status of the
subjects.
Fat content
Fat content by DXA was also very strongly related to the
extracted fat contents by chemical analysis (Figure 1c). The
least squares t regression log
e
fat 0.287 1.0409 log
e
DXA(fat) explained 97.6% of the variation in fat content
(F
1,14
581.24, P<0.0001). The absolute error in fat content
was 28.5g (s.d. 144.3), which was equivalent to 2.04% of
the extracted fat. However, this close average masked a much
larger discrepancy at the individual level, which varied from
underestimates of 20.7% to overestimates of 31.5%. On
average the mean deviation (ignoring the sign of the differ-
ence) averaged 12.2% (s.d. 9.3, range 0.9 31.5%). None of
these discrepancies were related to sex, species or castrate
status of the subjects or the version of software used in the
analysis.
Bone mineral content
There was a strong positive correlation between bone
mineral content (BMC) by DXA and the total body ash
content from the chemical analysis (Figure 1d). However,
the correlation between the two was poorer than for the soft
tissue components (r
2
0.909). Moreover the discrepancy
between BMC and total body ash was much greater than
for the soft tissues. On average DXA underestimated total ash
by on average 23.5% (s.e. 4.5%) and individual discrepan-
cies were as large as 56.7%. This was in part because ash is
derived not only from bones but also from mineral contents
of the other body components. If the ash content of the
bones is considered there was a closer correspondence of the
BMC by DXA to the ash content of the bones, with an
average discrepancy of only 7.4% (s.e. 5.6%). None of
these discrepancies were related to sex, species or castrate
status of the subjects or the version of software used in the
analysis.
Relationships of errors to individual body components
The absolute error in the lean tissue component was strongly
negatively related to the error in the fat component
(r
2
0.68, F
1,14
29.72, P<0.001). Apart from one indivi-
dual, the tradeoff between the two was very close. In some
animals the DXA machine overestimated fat contents, appar-
ently by mistaking lean tissue for fat, yet in other animals
the opposite occurred and the machine underestimated fat
tissue because it mistook it for lean tissue.
The absolute error in the lean tissue component was
related to variation in two of the body components the
amount of fat in muscle, which had a positive effect on the
error, and the amount of water in mesenteric fat, which had
a negative effect. Together these two components explained
91.5% of the error in the DXA evaluation of total lean body
tissue. The regression equation was: error in lean
(g) 16.01.33 muscle fat (g)1.76 mesenteric fat tissue
water (g), F
1,14
69.8, P<0.0001). Fat contents of muscle
and the other major lean components are shown in Table 1.
Muscle was the heaviest component of the lean tissue (aver-
age dry mass 194.4 g) and, apart from the pancreas, which
Table 1 Average total and percentage fat contents of the major lean
tissue components evaluated by Soxholet ether extraction across 16 dogs
and cats, values represent means and (s.e.)
Total fat (g) Percentage fat
Skeletal muscle 194.40 (56.8) 11.38 (2.35)
Alimentary tract 16.95 (3.14) 5.75 (0.51)
Liver 9.05 (0.93) 4.17 (0.90)
Lungs 7.04 (2.18) 6.29 (1.11)
Heart 3.23 (0.83) 7.41 (1.16)
Kidney 2.99 (0.51) 6.99 (0.83)
Spleen 0.54 (0.09) 4.10 (0.86)
Bladder 0.45 (0.12) 2.90 (0.43)
Brain 3.45 (0.60) 8.75 (1.02)
Pancreas 3.26 (0.64) 18.58 (2.12)
Validation of DXA for body composition
JR Speakman et al
443
International Journal of Obesity
was relatively small, also had the highest and most variable
fat content (average 12.5%).
The error in the fat tissue component was also related to
the fat content of muscle and lean component of the
mesenteric fat depot, but the effects were in the opposite
directions to the effects on lean tissue. The least squares t
regression, error in fat (g) 1.080.827 muscle fat (g) 18.1
mesenteric fat (lean) (g), explained 85.5% of the variation in
error in the fat tissue estimate (F
1,14
38.18, P<0.001). Fat
contents and water contents of the major fat depots are
shown in Table 2. Mesenteric fat was the second heaviest
fat store in the animals and, apart from pericardial fat, had
a high percentage water content (average 35%). Overall
there was a strong relationship between the error in the
DXA estimate of total fat content and the hydration status
of the skeletal muscle tissue (Figure 2).
In vitro studies
When we added lard to the lean beef the DXA machine
detected the increased fat content in the sample. There was a
strong relationship between the detected quantity of fat in
the sample and the amount of added lard (Figure 3a). The
tted relationship matched closely the expected line if all
the added lard was 100% fat. When we added water to the
beef there was no effect on the estimated fat content of the
sample until approximately 150 g had been added. There-
after there was a strong increase in the estimated fat content
until at 250 g of added water (10% of the original beef mass)
the estimated fat content was double that in the same
sample prior to addition of the water (Figure 3b).
Discussion
The small discrepancy between the DXA calculated body
mass and that measured gravimetrically prior to dissection
indicated that mass loss due to desiccation during storage
between DXA scanning and dissection=chemical analysis
was trivial. The discrepancy (1.0%) was similar to that
reported in previous validation studies, 1.2%,
23
1.4%,
21
Table 2 Average total fat contents by Soxholet ether extraction and
percentage water contents of the major fat depots across 16 dogs and
cats, values are means and (s.e.)
Fat depot Total fat (g) Percentage water
Mesenteric 108.5 (18.0) 35.0 (4.31)
Pericardial 6.73 (1.70) 47.8 (4.10)
Abdominal=renal 63.0 (18.6) 26.3 (3.09)
Subcutaneous 425.0 (121.0) 25.9 (3.06)
Figure 2 Error in the DXA estimate of fat content (relative to chemical
analysis) plotted against the skeletal muscle water content by chemical
analysis.
Figure 3 Estimated fat contents by DXA of 2.5 kg blocks of ground beef
to which 25g aliquots of (a) lard and (b) water were added. The thin line
in both cases follows the expected relationship and the thick line follows
the tted least squares regression. In (b) the best t was provided by a
polynomial t.
Validation of DXA for body composition
JR Speakman et al
444
International Journal of Obesity
0.2 1.5%
16
and conrms the widely reported nding that
DXA can make accurate estimates of total body mass.
Our measurements indicate that on average the Hologic
pencil beam DXA machine (QDR 1000 W) provides accepta-
ble estimates of the lean tissue component (mean error 2.6%),
body water component (mean error 1.6%) and fat content
(mean error 2.04%) for subjects in the range 1.8 22.1kg. The
analysis package (pediatric or adult) did not affect the accu-
racy of the estimates. These errors compare favourably with
the mean percentage errors between DXA and chemical
analysis, which have been reported previously using pigs as
the experimental subjects. For example, in previous chemical
validations the mean errors in lean tissue averaged 8%,
18
5%
19
, 1.3 5.9%
16
and 4.6%
23
and in fat content averaged
23%,
18
12%,
23
8%,
19
0.4%
21
and 35 234%.
16
In accord with most of the other validation studies which
have compared total body ash to DXA estimates of bone
mineral content we found that the DXA estimate was sub-
stantially lower than the total ash estimate. Fusch et al
23
and
Picaud et al
17
both found a shortfall of 42% by DXA in small
piglets and Brunton et al
16
reported discrepancies of on
average 30% in small pigs, but not in larger animals. Mitchell
et al
21
also reported no signicant difference in larger ani-
mals. Our discrepancy was slightly lower than the maxima
reported elsewhere, at 23.5%. Individual variation in the
discrepancy was however large (s.e. 4.5%), indicating that
a constant conversion factor from bone mineral content by
DXA to total mineral content would be unlikely to provide
an accurate measure of total body mineral content for
hydrodensitometry.
By performing the dissected analysis we could account for
at least part of this overestimate. Our analysis indicates that
approximately 15% of total inorganic ash, in dogs and cats,
originates in tissues that are not bone. This contributes to
ash but is not perceived as bone mineral by the DXA
machine. Interspecic variation in this `none bone-derived
ash' may account for the discrepancies reported in other
studies. Accounting for this contribution to total ash, how-
ever, still left a discrepancy of on average 7.7% to ash derived
from bone, indicating other problems in the comparison of
chemically derived ash to BMC by DXA.
Although the mean discrepancies in the estimated soft
tissue components across a group of subjects were relatively
small, the variations between individuals were much larger.
This effect has also been observed in all the previous valida-
tion studies in comparison to chemical analysis. In adult
pigs, for example, although the average discrepancy between
fat content by DXA and chemical analysis was only 0.4%, the
individual discrepancies were occasionally over 40%. Simi-
larly, when water contents by DXA were compared with
water contents by hydrogen isotope dilution in dogs,
28
the
individual differences varied between 0.4 and 39.7%, which
is consistent with the discrepancies between water content
by chemical analysis and by DXA reported here.
These discrepancies at the individual level pose serious
problems for the use of DXA on an individual basis. Since the
discrepancies far exceed the individual repeatability of mea-
surements made on this machine,
25
this suggests the errors
reect intrinsic aspects of the body composition of indivi-
dual subjects rather than precision errors of performing a
single scan. Our analyses of dissected components of the
body suggest why the errors occur in the DXA analysis.
Although the brain consisted only of 8.35% fat and the
DXA machine assumed 17% fat for the brain, this discre-
pancy did not appear to be important. This was probably
because the total amount of fat in the brain was small
(averaging only 3.4 g across individuals) and thus overesti-
mating this minor component did not contribute much to
the total error.
It appeared that the machine had problems identifying fat
in skeletal muscle tissue. In fact the errors in the lean and fat
components were linked almost gram for gram with the
amounts of fat in muscle. Why fat in muscle is particularly
important, rather than the fat contents of other lean tissues
is unclear. Perhaps this error is related in part to the fact that
skeletal muscle is generally located close to areas of bone and
hence the error may be linked to errors in the interpolation
process by which soft tissue composition is reconstructed for
those pixels in which bone is also detected. Alternatively,
skeletal muscle was the largest component of the lean tissue
and, because it also had a high percentage fat content (12.5%
compared to approximately 5% in the other tissues, except
brain and pancreas), the amount of fat in skeletal muscle was
10 30 greater than other lean tissues and comprised
over 85% of the total fat in lean tissue. Consequently, in
regression analysis, this component would be much more
likely to be correlated with errors in total fat and lean
than the relatively minor variation in the other compo-
nents. This suggests there probably isn't anything special
about fat in skeletal muscle.
To an extent this error was offset by a contrasting prob-
lem. The machine appeared to get confused by components
of the mesenteric fat tissue. In particular, increased water
content of the mesenteric fat tissue was negatively correlated
with the error in the total lean component and increased
lean component of the tissue was positively associated with
the error in fat content. This might suggest that, as the water
content of this tissue increased, it was more likely to be
classied overall as lean tissue, but as its lean component
increased, it was more likely to be classed overall as fat. These
effects would offset the errors in total lean and fat compo-
nents that occur because of failure to detect fat in muscle.
Why the machine should get confused only by mesenteric
tissue is unclear. It was unlikely to be a consequence of the
errors in the corrections for areas where bone is present,
because these animals were lying on their sides and hence
mesenteric fat was not in an area overlying bone. When
humans are scanned they normally lie dorsally, but we could
not scan our animals in this manner because of the con-
sequences of dorsal placement for the positions of the limbs.
Apart from two studies which explicitly stated that animals
were positioned dorsally,
16,21
previous comparisons of DXA
Validation of DXA for body composition
JR Speakman et al
445
International Journal of Obesity
to chemical analysis have probably all used animals lying on
their sides, rather than dorsally, because of the convenience
of lateral placement. Lateral placement will force the
machine to make different corrections for areas where bone
is detected. Whether lateral placement generates greater or
lower discrepancies to chemical analysis is an interesting
issue requiring further study, but it is unlikely to be a factor
inuencing our detection that mesenteric fat can be mis-
interpreted by the machine. Probably of more signicance is
the fact that mesenteric fat is a relatively large fat depot and
had a relatively large water content (35% compared to 25%
in the two other major fat depots, subcutaneous and peri-
renal). Consistent with our ndings, recent studies have also
identied the extent of truncal adiposity as a factor inuen-
cing the ability of DXA to adequately resolve fat free mass of
elderly subjects.
26
The overall relationship between skeletal muscle tissue
hydration and the error in fat content points to the impor-
tance of tissue hydration as a factor underpinning errors in
the DXA method. This interpretation is supported by the
measurements we made with mixtures of ground beef, lard
and water. Like several other studies,
12 14,25,26
we found that
when lard was added to the tissue being scanned the
machine was able to accurately detect it. However, we
found that increasing hydration of the tissue led the
machine to systematically overestimate fat content. Increas-
ing hydration by 10% increased the perceived fat content of
the sample by approximately 100%. This effect is consider-
ably greater than the expectation based on the perception of
fat by DXA in pure deionized water, which is approximately
8.6%.
29
On this basis adding 250g of water should only have
increased the perceived fat content of the meat block by 25g
(ca 12%). The reasons for this discrepancy are unknown, but
it is a repeatable phenomenon on the machine we used.
Other studies point to an important role for tissue hydration
in the error in estimates of fat content by DXA.
23,30
In
particular, Fusch et al
23
found that tissue hydration declined
as a function of increasing body mass in their experimental
subjects, and this decrease in hydration was linked to an
increasing underestimate of the fat content of the animals.
Although Going et al
31
manipulated hydration status and
found no effect on fat content by DXA, in a sample of 19
adults, the extent of their manipulations (less than 2%) was
small relative to the individual variability in tissue hydra-
tion, and in the range where our own in vitro manipulations
detected no effect (Figure 3b).
On average, in a relatively large sample of individuals, the
errors associated with variations in tissue hydration appear
to cancel, but in any particular individual this may not be
the case. Thus an individual that has a low lean tissue
hydration (high lean tissue fat content) will be more likely
to have an underestimated total fat content and an over-
estimated lean content when analysed by DXA. In contrast
an individual with relatively high tissue hydration would be
likely to have an overestimated total fat content and an
underestimated total lean component. These effects would
be consistent across scans on the same individual leading
to high scan to scan repeatability as has been previously
observed on this machine
25
and by other.
11 14
At the indi-
vidual level DXA is clearly a method with high precision but
low accuracy.
This suggestion that the DXA machine may miss some
components of body fat, but also misinterpret some fat stores
as lean tissue may have important implications if the
machines are used to make individual diagnoses of fat con-
tent. Individual errors reported here (mean deviations ignor-
ing the sign of the difference) averaged 6.5% for lean tissue
and 12.2% for fat tissue, but were occasionally greater than
20%. This level of accuracy would clearly be unacceptable for
many applications of the methodology to the assessment of
body fat content. Moreover, intervention studies that affect
the regional distribution of fat, or longitudinal studies of
changes in fatness which accompany growth and ageing
may also be misinterpreted if DXA analyses alone are used
to assess body fatness changes. For example, mobilization of
fat away from muscle tissue and its redeposition into the
subcutaneous fat depot might be interpreted as an overall
increase in body fatness and a reduction in total lean tissue,
when no such change had actually occurred. Conversely
shifting fat from the peripheral store into muscle might
erroneously be interpreted as a reduction in fatness and an
increase in lean tissue. We suggest this problem may be
minimized if studies employ a number of indirect methods
simultaneously, rather than relying solely on DXA to evalu-
ate body composition changes. This has in any case been a
routine procedure in many studies.
2 10
Although the pencil beam DXA machine validated here
has been superceded by models which perform fan-beam
analyses, many pencil beam machines are still in clinical
practice and research papers based on their use are still being
published.
32
Our data on the capability of this machine
to assess soft tissue composition, therefore, still has consid-
erable clinical relevance. Moreover, studies comparing
fan-beam and pencil beam machines suggest close
correspondence in their estimates of body fatness across
subjects.
33
Nevertheless, it would be interesting to know if
the effects we have detected using a pencil beam machine at
pediatric weights are also found at larger body masses, and
also with the fan-beam technology, and indeed in pencil
beam machines manufactured by other companies, since
comparative data indicate that machines made by different
companies do not produce identical results.
34
Acknowledgements
We are grateful to Pat Nugent, Phil Anderson and Helen
Munday for technical assistance with the DXA scanning and
to Maria Johnson, Diane Jackson, Susan Thomson, Richard
Delahay, Emma Williams and Eric Swanepoel for assistance
with the dissection and chemical composition work.
Validation of DXA for body composition
JR Speakman et al
446
International Journal of Obesity
References
1 Pichard C, Kyle UG, Gremion G, Gerbase M, Slosman DO. Body
composition by X-ray absorptiometry and bioelectrical impe-
dance in female runners. Med Sci Sports Exercise 1997; 29: 1527
1534.
2 Hendel HW, Gotfriedsen A, Hojgaard L, Andersen T,
Hilsted J. Change in fat-free mass assessed by bioelectrical impe-
dance, total body potassium and dual energy x-ray absorptiome-
try during prolonged weight loss. Scand J Clin lab Invest 1996; 56:
671 679.
3 Podenphant J, Gotfriedsen A, Engelhart M, Andersen V, Heit-
mann BL, Kondrup J. Comparison of body composition by dual
energy x-ray absorptiometry to other estimates of body composi-
tion during weight loss in obese patients with rheumatoid arthri-
tis. Scand J Clin Lab Invest 1996; 56: 615 625.
4 Nau KL, Dick AR, Peters K, Schloerb PR. Relative validity of
clinical techniques for measuring the body composition of per-
sons with amyotrophic lateral sclerosis. J Neurol Sci 1997; 152:
S36 42.
5 Pichard C, Kyle UG, Janssens JP, Burdet L, Rochat T, Slosman DO,
Fitting JW, Theiband D, Roulet M, Tschopp JM, Landry M, Schutz
Y. Body composition by X-ray absorptiometry and bioelectrical
impedance in chronic respiratory insufciency patients. Nutrition
1997; 13: 11 12.
6 Paton NIJ, MacAllan DC, Jebb SA, Noble C, Baldwin C, Pazianias
M, Grifn GE. Longitudinal changes in body composition
measured with a variety of methods in patients with AIDS.
J AIDS Syn Hum retrovirol 1997; 14: 119 127.
7 Royall D, Greenberg GR, Allard JP, Baker JP, Harrison JE,
Jeejeebhoy KN. Critical assessment of body composition measure-
ments in malnourished subjects with Crohns-disease the role
of bio-
electrical impedance analysis. Am J Clin Nutr 1994; 59: 325 330.
8 Sievanan H, Backstrom MC, Kuusela AL, Ikonen RS Maki M. Dual
energy X-ray absorptiometry of the forearm in preterm and term
infants: evaluation of the methodology. Pediat Res 1999; 45: 100
105.
9 Lapillonne A, Braillon PM, Delmas PD, Salle BL. Dual-energy
X-ray absorptiometry in early life. Horm Res 1997; 48:
43 49.
10 Cochat P, Braillon P, Feber J, HadjAissa A, Dubourg L, Liponski I,
Said MH, Glastre C, Meunier PJ, David L. Body composition in
children with renal disease: use of dual-energy X-ray absorptio-
metry. Pediat Nephrol 1996; 10: 264 268.
11 Coleman RJ, Hudson JC, Barden HS, Kemnitz JW. A comparison
of dual-energy X-ray absorptiometry and somatometrics for
determining body fat in rhesus macaques. Obes Res 1999; 7:
90 96.
12 Madsen OR, Jemsen JEB, Sorensen OH. Validation of a dual-
energy X-ray absorptiometer: measurement of bone mass and
soft-tissue composition. Eur J Appl Phys Occup Phys 1997; 75:
554 558.
13 Jensen MB, Hermann AP, Hessov I, Moskilde L. Components of
variance when assessing the reproducibility of body composition
measurements using bio-impedance and the Hologic QDR-2000
DXA scanner. Clin Nutr 1997; 16: 61 65.
14 Haarbo J, Gotfredsen A, Hassager C, Christiansen C. Validation of
body composition by dual-energy x-ray absorptiometry (DEXA).
Clin Phys 1991; 11: 331 341.
15 Pintuaro SJ, Nagy TR, Duthie CM, Goran MI. Cross calibration of
fat and lean measurements by dual-energy X-ray absorptiometry
to pig carcass analysis in the pediatric body weight range. Am J
Clin Nutr 1996; 63: 293 298.
16 Brunton JA, Bayley HS, Atkinson SA. Validation and application of
dual-energy X-ray absorptiometry to measure bone mass and body
composition in small infants. Am J Clin Nutr, 1993; 58: 839 845.
17 Picaud JC, Rigo J, Nyamugabo K, Milet J, Senterre J. Evaluation of
dual-energy X-ray absorptiometry for body-composition assess-
ment in piglets and term human neonates. Am J Clin Nutr 1996;
63: 157 163.
18 Ellis KJ, Shypailo RJ, Pratt JA, Pond WG. Accuracy of dual-energy
X-ray absorptiometry for body composition measurements in
children. Am J Clin Nutr 1994; 60: 660 665.
19 Svedsen OL, Haarbo J, Hassager C, Christiansen C. Accuracy of
measurements of body composition by dual-energy x-ray absorp-
tiometry in vivo. Am J Clin Nutr 1993; 57: 605 608.
20 Brunton JA, Weiler HA, Atkinson SA. Improvement in the accu-
racy of dual energy x-ray absorptiometry for whole body and
regional analysis of body composition: validation using piglets
and methodologic considerations in infants. Pediat Res 1997; 41:
590 596.
21 Mitchell AD, Conway JM, Potts WJE. Body composition of pigs by
dual-energy X-ray absorptiometry. J Anim Sci 1996; 74: 2663
2671.
22 Mitchell AD, Conway JM, Scholz AM. Incremental changes in
total and regional body composition of growing pigs measured by
dual-energy x-ray absorptiometry. Growth Dev Aging 1996; 60:
95 105.
23 Fusch C, Slotboom J, Fuehrer U, Schumacher R, Keisker A,
Zimmermann W, Moessinger A, Boesch C, Blum J. Neonatal
body composition: dual-energy X-ray absorptiometry, magnetic
resonance imaging, and three-dimensional chemical shift ima-
ging versus chemical analysis in piglets. Pediat Res 1999; 46: 465
473.
24 Hologic QBD1000 Instruction manual.
25 Munday HS. Assessment of body compositionin cats and dogs. Int
J Obes Relat Metab Disord 1994; 18(Suppl 1): S14 S21.
26 Snead DB, Birge SJ, Kohrt WM. Age-related differences in body
composition by hydrodensitometry and dual x-ray absorptio-
metry. J Appl Physiol 1993; 74: 770 775.
27 Lapillonne A, Braillon PM, Claris O, Ho PS, Delmas PD, Salle BL.
Use of dual-energy X-ray absorptiometry for measurements of
small quantities of mineral. Biol Neonate 1997; 71: 198 201.
28 Son HR, Avignon A, La Flamme DP. Comparison of dual-energy
X-ray absorptiometry and measurement of total body water
content by deuterium oxide dilution for estimating body com-
position in dogs. Am J Vet Res 1998; 59: 529 532.
29 Nord RH, Payne RK. A new equation set for converting body
density to percent body fat. Asia Pacic J Clin Nutr 1995; 4:
177 179.
30 Pietrobelli A, Wang ZM, Formica C, Heymseld SB. Am J Physiol
1998; 37: E808 E816.
31 Going SB, Massett MP, Hall MC et al. Detection of small changes
in body composition by dual x-ray absorptiometry. Am J Clin Nutr
1993; 57: 845 850.
32 Barbeau P, Gutin B, Litaker M, Owens S, Riggs S, Okuyama T.
Correlates of individual differences in body-composition changes
resulting from physical training in obese children. Am J Clin Nutr
1999; 69: 705 711.
33 Kelly TL, Steiger P, Shepherd JA. Body composition determina-
tions using fan beam technology. Meeting abstract, International
Congress on Obesity, Paris; August 1998.
34 Guo SS, Wisemandle WA, Tyleshevski FE, Roche AF, Chumlea
WC, Siervogel RM, Speckler B, Heubi J. Inter-machine and inter-
method differences in body composition measures from dual
energy X-ray absorptiometry. Nutr Health Aging 1997; 1: 29 37.
Validation of DXA for body composition
JR Speakman et al
447
International Journal of Obesity