Full text of DOI:10.1359/jbmr.2001.16.2.417

JOURNAL OF BONE AND MINERAL RESEARCH
Volume 16, Number 2, 2001
© 2001 American Society for Bone and Mineral Research
Inaccuracies Inherent in Patient-Specific Dual-Energy X-
Ray Absorptiometry Bone Mineral Density Measurements:
Comprehensive Phantom-Based Evaluation
1
2
3
4
5
¨
¨
H.H. BOLOTIN, H. SIEVANEN, J.L. GRASHUIS, J.W. KUIPER, and T.L.N. JARVINEN
ABSTRACT
An extensive series of dual-energy X-ray absorptiometry (DXA) scans and dual polyenergetic X-ray simulation
studies of 150 different phantom arrays were carried out to evaluate quantitatively the extent of systematic
inaccuracies inherent in DXA in vivo bone mineral density (BMD). These measurements are particularly
relevant to lumbar vertebral and proximal femoral sites. The phantoms were specially fabricated near perfect
absorptiometric representations of bone material, red marrow (RM) and yellow marrow (YM), and extraosse-
ous mixtures of fat (F) and lean muscle that spanned the full range of soft tissue anthropometrics encountered
clinically. In each case, the DXA-measured BMD values obtained using Hologic, Lunar, and Norland
densitometers were found to be virtually the same and to be in excellent agreement with the corresponding
quantitative simulation study BMD results. Comparisons of the known phantom BMD values and DXA-
measured BMD in each case allowed the BMD inaccuracies to be evaluated. These present findings show that
these ubiquitous inaccuracies in DXA BMD methodology are of in vivo soft tissue anthropometric genesis. It
is found that patient-specific DXA-measured in vivo BMD inaccuracies as high as 20% or more can be readily
anticipated clinically, particularly in cases of osteopenic, osteoporotic, and elderly patients. As these inaccu-
racies exceed considerably DXA precision errors, they may compromise patient-specific evaluations of
fracture risk and, in prospective studies, mask or exaggerate clinically significant true changes in BMD. It is
concluded that the magnitudes and variability of inherent inaccuracies in DXA-measured in vivo BMD
underscore the need for prudence and circumspection in interpretations and assessments of DXA-based
clinical studies. (J Bone Miner Res 2001;16:417–426)
Key words: bone mineral density inaccuracies, marrow, dual-energy X-ray absorptiometry, soft tissue
anthropometrics, phantoms
projected to increase dramatically as the proportion of the
population at the osteoporosis-prone age increases sharply
in most developed countries over the next decades.(2,3)
INTRODUCTION
STEOPOROSIS AND associated bone fractures among post-
O
menopausal women and the elderly are a significant Current evaluation of osteoporosis and bone mineral status
public health problem worldwide,(1) the extent of which is of the skeleton is based mainly on planar dual-energy X-ray
1School of Physics, University of Melbourne, Victoria, Australia.
2The Bone Research Group, UKK Institute, Tampere, Finland.
3Departments of Medical Informatics and Radiology, Erasmus University Rotterdam, Rotterdam, The Netherlands.
4Department of Radiology, Academical Hospital Dijkzigt, Rotterdam, The Netherlands.
5Department of Surgery, Tampere University Hospital, and Medical School and Institute of Medical Technology, University of
Tampere, Tampere, Finland.
417

418
BOLOTIN ET AL.
absorptiometry (DXA), the present noninvasive method of complete definition or coherent understanding of these in-
choice for clinical practice and research.(4) DXA-derived
accuracies.
To evaluate the prior quantitative analyses and simulation
studies(14–16) against actual DXA-measured BMD values,
we carried out a comprehensive set of measurements using
Hologic (Waltham, MA, USA), Lunar (Madison, WI,
USA), and Norland (Fort Atkinson, WI, USA) DXA in-
struments to scan 150 different phantom arrays. These
phantom materials were specially formulated and fabricated
to replicate the complete ranges of lean muscle tissue
(LMT)/F tissue, YM/RM mixes, and BMD encountered
clinically.(25–28) As well, they reproduced the absorptiomet-
ric properties of each of these separate tissues across the full
DXA X-ray energy range.(29) These DXA results are com-
pared with those of our simulation studies of the same
phantom arrays.
areal bone mineral density (BMD) generally is considered
to have an acceptable ability to predict fracture risk(5) and
provide a reasonable determinant of the mechanical integ-
rity and fragility(6,7) of given bone, as well. However, not-
withstanding its excellent measurement precision and the
apparent advances accompanying the use of DXA, concern
persists that in vivo DXA measurements may be subject to
considerable systematic inaccuracies.(8–16) Of particular
disquiet are those in vivo BMD inaccuracies arising from
absorptiometric disparities between local extraosseous
body composition (fat [F]/lean tissue mix) and the specific
bone marrow composition (yellow marrow/red marrow
[YM/RM]ratio) within the bone site scan region of interest
(14–16)
(ROI).
It is well known(8–10,14,15) that DXA methodology does
not extend to bone sites comprised of more than two
absorptiometrically distinguishable components (the “two-
component DXA limitation”). Taking bone material as one
component, DXA methodology is restricted by the necessity
for strict satisfaction of its presumption that the composition
and distribution of all extra- and intraosseous soft tissues
and other body constituents present within the scan ROI
effectively constitute an absorptiometrically homogeneous
second “component.”(9,14,15) Because it is clear that this
two-component composition does not conform to anatomi-
cal reality, it follows that DXA in vivo BMD measurements
must be inherently inaccurate to some degree, either under-
or overestimating the true BMD to some indeterminate
extent in any given patient. These issues also have been
enunciated in the review by Webber.(8)
MATERIALS AND METHODS
Phantom array details
The composition and volumetric densities of the phantom
materials used in the present studies were such as to repli-
cate the linear attenuation coefficients (␮, cmϪ1) of each of
the relevant tissues present in the ROIs of in vivo lumbar
vertebral and proximal femoral bone sites at all energies
spanning the dual-energy X-ray range of DXA instrumen-
tation (20–140 keV): bone material, LMT, F, RM, and YM.
The fabrication of these phantom tissue substitutes was
based on the essentials of the loaded epoxy resin method of
White et al.,(30) except for that of YM, which was replicated
equally well(31) (and more simply) by high-density polyeth-
ylene slabs of the proper volumetric density. The mass
attenuation coefficient (␭, cm2 gϪ1) of each of these tissues
was matched across the full DXA energy range using
Araldite M and hardener HY5162 (Ciba Geigy Australia
Ltd., Thomastown, Victoria, Australia) and specifically for-
mulated loadings of powdered polyethylene, calcium car-
bonate, and hydroxyapatite (Sigma-Aldrich, Castle Hill,
NSW, Australia), as required. The volumetric density ␳
(g/cm3) of each phantom tissue substitute also was matched
by incorporating in the epoxy resin mix, the requisite spe-
cific quantity of low-density (ϳ0.25 g/cm3) phenolic micro-
spheres (PMS) BJO-0930 (Asia-Pacific Microspheres SDN
BHD, Selangor, W. Malaysia), so that the linear attenuation
coefficient ␮ ϭ ␭␳ of each tissue was reproduced. The
specifications of each of these phantom materials are given
in Table 1.
Inaccuracies in DXA-measured in vivo BMD, particu-
larly of the lumbar vertebrae and proximal femoral sites,
have been observed to be sizable in typical patient-specific
cadaveric studies.(12,13) Recent quantitative simulation stud-
ies have shown these inaccuracies may be of particular
concern for osteoporotic and elderly patients with low
BMD.(14,15) Particularly for these patients, DXA-measured
in vivo BMD inaccuracies exceeding 2 SDs of the observed
population-based, age-specific BMD normative data can be
manifest,(14–16) even in cases in which extraosseous soft
tissues are as near uniform in composition as possible.
Given the important role patient-specific BMD values serve
in the multiple clinical objectives of screening, diagnosis,
monitoring, and in evaluations of the efficacy of drug or
other bone-therapeutic regimens, it is vital that the analytic
quantitative simulation studies(14–16) that have exposed
these systematic inherent DXA BMD inaccuracies are eval-
uated critically to confirm their actual magnitude, trends,
and extent.
It is clear that neither the full range of inherent inaccu-
racies affecting DXA in vivo BMD measurements nor the
quantification of the requisite particulars of soft tissue com-
position and distribution throughout the scan ROI can be
exposed successfully by means of clinical in vivo or cadav-
eric in situ/in vitro DXA investigations. As a consequence,
despite a number of such pertinent studies,(12,13,17–24) some
of which have displayed sizable specimen-specific BMD
After curing at room temperature for about 12 h, the
epoxy resin-based materials were heat-treated at 80°C for 3
hours.(30) The resultant materials were then machined to
close tolerances (Ϯ0.02 mm) in the form of parallelepipeds
of required dimensions to allow assembly of them into 150
different arrays, all 20 cm thick, as represented schemati-
cally in Fig. 1. A fixed 20-cm phantom “torso” or “hip”
thickness, representing a typical clinical case, was selected
for consistency with the dual monochromatic simulation
studies reported earlier.(14,15)
Each assembly of constituent phantom slabs was clamped
tightly between two thin, L-shaped aluminum cover-sheets
inaccuracies, these investigations have failed to provide to form ready arrays for DXA scans. In each phantom array,

INACCURACIES IN DXA BMD MEASUREMENTS
419
TABLE 1. DESIGN SPECIFICATIONS OF FABRICATED
PHANTOM TISSUE MATERIALS
Constituents
(percentage weights)
Tissue
␳ (g/cm3)
Bone material
aE (36.0), hydroxyapatite
(17.0), CaCO3 (47.0)
1.85
b
LMT
F
aE (69.8), polyethylene
1.03
0.92
(21.9), CaCO3 (5.8),
cPMS (2.5)
b
aE (43.2), polyethylene
c
(53.4), PMS (3.4)
YM
RM
dPolyethylene (100.0)
0.96
1.04
b
aE (68.4), polyethylene
(21.3), CaCO3 (7.7),
cPMS (2.6)
a E ϭ Araldite M (71.4) ϩ Hardener HY5162 (28.6)
b Powdered polyethylene (density ϭ 0.938 g/cm3).
c PMS density ϭ 0.25 g/cm3
d Solid high-density polyethylene slabs.
FIG. 1. Schematic representation of a typical phantom assembly
employed in the present work indicating phantom bone position and its
composition, the direction of incident DXA scan X-rays (actual and
simulated), and a symbolic indication of the scan raster. Each ex-
traosseous phantom block was a 4-cm cube; the overall thickness of the
torso containing the bone was 20 cm. The q designates the composition
of the extraosseous soft tissue in terms of the ratio of the areal density
of F to that of LMT. YM and RM were in the form of slabs of requisite
thicknesses and 4-cm ϫ 4-cm cross-sectional area normal to the X-ray
beam.
the extraosseous material was of a given absorptiometrically
homogeneous F-to-LMT areal density ratio q ϭ (␳FxF/
␳
LMTxLMT), where x is the linear dimension (cm) of the
given tissue along the path of incident X-rays and q is the
extraosseous soft tissue descriptor relevant to DXA
scans.(14,15) The values of q represented, q ϭ 0.0, 0.1, 0.3,
0.6, and 1.0, spanned the full range (up to 50% F fraction)
encountered clinically(25) [q/(q ϩ 1) the F fraction by
weight]. A 4-cm3 “bone site” was positioned centrally
within each array. Bone material slabs were inserted sym-
metrically within this site in such a way (Fig. 1) that the
relative trabecular bone volume (TBV; that is, the fraction
of the total bone volume occupied by bone material) took on
values of TBV ϭ 0.05, 0.10, 0.15, 0.20, and 0.35, encom-
passing the full clinical TBV range.(26) The remaining bone-
site space between the bone slabs was filled with a particular
combination of RM and YM phantom slabs, so that the
fraction of the marrow volume, which was yellow, g ϭ
(YM)/(YM ϩ RM), took on values of g ϭ 0.0, 0.2, 0.4, 0.6,
0.8, and 1.0 (ranging from purely RM to wholly YM,
respectively). In this way, 150 different phantom arrays
replicating as many soft tissue anthropometrics and TBV
combinations were assembled and scanned.
Expressed in terms of such averaged absorptiometric val-
ues, the phantom arrays used here constituted BMD values
ranging grossly from ϳ0.40 to ϳ1.43 g/cm2.
Because all phantom arrays were regular in shape with
the bone site having abrupt, sharp edges parallel to the
direction of the incident DXA X-ray flux, conditions were
ideal for the operation of the bone edge algorithms incor-
porated in each of the three DXA instruments used. Simi-
larly, to establish conditions as near to ideal as possible for
DXA in vivo BMD measurements, all phantom arrays had
identical overall dimensions, and both the intra- and the
extraosseous phantom materials were separately absorptio-
metrically homogeneous throughout their respective do-
mains in the scan ROI.
It should be noted that under ideal circumstances DXA-
derived BMD represents the areal density of bone material
thicknesses (␳x, g/cm2) averaged over all of the X-ray paths
through the bone volume(29,32) in which the trabeculae
within the thin cortical wall may not be distributed uni-
formly in the scan ROI or may not be present at all (i.e., in
purely cortical bone segments, such as the femur midshaft).
DXA scanning of phantom arrays
The DXA measurements of these 150 different phantom
(Note that TBV is analogous to volumetric bone mineral arrays were carried out with three different DXA instru-
apparent density [BMAD], that is, the total bone material ments: Hologic QDR-1000 (software version 6.10), Lunar
mass divided by the total volume within the cortical enve- DPX-␣ (software version 1.1z), and Norland XR-26 (soft-
lope.) Naturally, because DXA cannot resolve the structural ware version 2.2.5). All instruments were calibrated daily
details of a given bone, to constitute full absorptiometric and routine quality control procedures were performed ac-
bone site replicas, the phantom assemblies used need only cording to recommendations of the manufacturers and the
represent lumped measures of the actual bone structures. particular DXA center.

420
The scanning procedures for each of the DXA instru-
BOLOTIN ET AL.
ments were similar. Each phantom array was positioned at
a fixed scanning location in the middle of the scan table. The
scan modes used in this study conformed to the standard
applications in each DXA center. The Hologic scans were
carried out with the performance scan mode [70 kilovolt
(peak)/140 kV(p), 2.0 mA], the Lunar scans with the me-
dium scan mode [76 kV(p), 750 ␮A], and the Norland scans
at the scan speed of 45 mm/s [100 kV(p), 1.0 mA]. The
width and length of the scanned region of the phantom array
were 90 mm and 24 mm, respectively. In this way, the
aluminum coversheets did not intrude on the scan ROI. The
pixel sizes were 1.003 mm ϫ 0.965 mm (Hologic), 1.2
mm ϫ 1.2 mm (Lunar), and 1.0 mm ϫ 1.0 mm (Norland).
Further, to avoid any potential influence of brand-specific
bone edge detection algorithms on the BMD measurements,
the ROI used for BMD analysis was located in the middle of
the detected bone region of each phantom array. The total
area analyzed of each scan ROI was approximately 400
mm2, the exact measurement area dependent on the partic-
ular pixel size of the DXA instrument used.
To evaluate the reproducibility of the study procedures,
repeated scans of some 60 different phantom arrays were
carried out with each DXA instrument with disassembly,
reassembly, and repositioning in each case.
Polyenergetic dual-energy simulation studies of
phantom arrays
Each of the dual polyenergetic spectral domains of a
given DXA instrument was replicated differentially (rela-
tive X-ray intensities within narrow [1 keV], contiguous
energy bins), as shown in Fig. 2. The simulated dual-energy
spectrum of each densitometer was obtained by introducing
the requisite critical absorber or other filter substance into
the incident Bremsstrahlung spectrum emitted from the
X-ray tube anode at the high voltage(s) stipulated for the
X-ray generator of the given DXA instrument. As an ex-
ample, the high voltage potential applied to the X-ray gen-
erator in the Hologic QDR 1000 alternates between 70
kV(p) and 140 kV(p) with synchronously introduced filters
of Al and (Al ϩ Cu), respectively. In this manner, the
resultant simulated low- and high-energy X-ray domains of
the Hologic densitometer were formulated separately. The
dual-energy X-ray groupings of the Lunar [fixed 76 kV(p)]
and Norland [fixed 100 kV(p)] instruments were similarly
replicated (Fig. 2) using the Ce and Sm K-edge filtration
stipulated by their respective manufacturers.(29)
The full simulation was carried out as per the ideal
standard DXA scanning mode of the lumbar vertebrae and
proximal femoral bone sites.(29) The known linear attenua-
tion coefficient of each of the absorptiometrically distin-
guishable bone and soft tissue phantom substitutes pertain-
ing to each 1-keV energy bin within the span of X-ray
energies of each of the three dual polychromatic Hologic,
Lunar, and Norland DXA photon spectral regions was used
in the present set of simulation studies.(31,33,34) For each
1-keV energy bin within each of the low- and high-energy
FIG. 2. Simulated dual-energy X-ray spectra used in the simulation
study phases of the present work for (a) the Hologic, (b) Lunar, and (c)
Norland DXA densitometers. The higher energy component of each
dual-energy spectrum is shown dashed. For the Lunar and Norland
instruments, both the low- and the high-energy X-ray spectral regions
are incident on the scanned object simultaneously; in the Hologic case,
these regions are produced alternately.
DXA spectral groupings, the fraction of incident photon flux site ROI was obtained and the true simulation of ideal dual
transmitted through the “torso” in the posterior-anterior polyenergetic Hologic, Norland, and Lunar DXA phantom
direction along each of the photon paths through the bone in vivo scans affected.

INACCURACIES IN DXA BMD MEASUREMENTS
421
Each such simulated BMD measurement was corrected (g ϭ 0) to wholly yellow (g ϭ 1.0), for true TBV values
for beam hardening in a standardized manner. DXA beam- of 0.10, 0.15. 0.20, and 0.35, and for q ϭ 0.1, 0.3, 0.6,
hardening corrections are predicated on the hardening of the and 1.0.
incident X-ray beam by the soft tissue immediately lateral to
Prior investigations(25–28) have reported the following
the bone representing identically that produced by all soft clinically pertinent ranges of these soft tissue (g and q) and
tissues present along all X-ray paths passing through the TBV parameters: marrow,(27,28) ϳ0.15 Ͻ g Ͻ ϳ0.6 for
bone. Because the beam-hardening corrective procedures normal subjects and ϳ0.35 Ͻ g Ͻ ϳ0.85 for osteoporotics
and algorithms of the Hologic, Lunar, and Norland instru- (fairly independent of age) with considerable overlap be-
ments have been shown to yield effectively equivalent re- tween normals and osteoporotics; extraosseous tissue,(25)
sults,(35,36) we elected to carry out simulated beam- ϳ0.07 Ͻ q Ͻ ϳ0.9; vertebral trabecular bone volume,
hardening calibration procedures for all three instruments ϳ0.13 Ͻ TBV Ͻ ϳ0.33 for normal subjects with some
following the Hologic method,(29) because it appears to be decrease with age, but the decrease of the same order as its
the more conceptually transparent. In this mode, a thin, variability(26); and iliac crest,(28) TBV decreases from
standardized bone specimen is inserted into the X-ray beam ϳ0.26 to ϳ0.16 as age increases from 20 to 65 years for
passing through the extraosseous soft tissue laterally adja- normal subjects, with osteoporotics in the range of ϳ0.06 Ͻ
cent to the bone segment being scanned during a portion of TBV Ͻ ϳ0.16 with no discernible age trend, and normals
the time attenuation of the X-ray beam along that trajectory and osteoporotics virtually indistinguishable above the age
is being registered. By this means, the inserted bone spec- of 65 years. Thus, based on these observations(25,28) and
imen is subjected to the X-ray flux that has been beam- others cited by Webber,(8) close to the full ranges of q, g,
hardened by the full extent of the soft tissue in the “base- and TBV shown in Figs. 3–5 must be considered clinically
line” region. Using the simulated photon attenuation of the relevant.
unobstructed lateral soft tissue as a baseline reference mea-
The present results for all three DXA instruments and
surement, the simulated DXA measurement of the BMD of their simulated representations are virtually identical. We
this inserted bone specimen was obtained. The ratio of the find that the inherent DXA inaccuracies in BMD of a given
known, true BMD value of the thin bone insert to that of its patient may differ from the true value, either under- or
measured value defines the DXA beam-hardening correc- overestimating it by a magnitude corresponding to several
tion factor to be applied to the measured BMD value of the SDs of the normative, age-specific, population-based BMD
actual bone segment in the ROI of that given scan.
data. This is particularly so for the lower ranges of TBV, the
A complete series of dual monochromatic simulation higher values of g, and the lower q values.
studies also was carried out for the same 150 phantom
It is to be noted that when the marrow mix parameter (g)
arrays for each of the DXA instruments used. The same is such that the linear attenuation coefficient of marrow is
procedures described earlier(14,15) were followed in these identical to that of the specific extraosseous F-to-LMT
simulated DXA studies using dual energies of 43 keV and mixture (q), the DXA-derived BMD value is accurate. It
110 keV (Hologic), 38 keV and 70 keV (Lunar), and 46 keV also is seen (Figs. 3–5) that such (q,g) designations are
and 80 keV (Norland).(29)
independent of TBV, as must be so, and that the particular
g value for which this is so is sensitively dependent on the
patient q value. However, we point out that DXA method-
ology cannot determine that (q,g) value in any standard scan
case. Further, there is no suggestion in prior anthropometric
RESULTS
The 95% limits of agreement(37) as obtained from the studies that the values of q and g are correlated(25–28); that is,
repeated phantom BMD measurements were Ϯ0.027 g/cm2 the (q,g) parameter-pair value is not known in any given
for Hologic scans, Ϯ0.054 g/cm2 for Lunar scans, and patient-specific DXA scan.
Ϯ0.039 g/cm2 for Norland scans. These values compare
The BMD inaccuracies just considered are derived from
well with the in vitro differences observed among the dif- differences between the linear attenuation coefficients of the
ferent scanners of the same manufacturer, the limits of extraosseous F-LMT and the intraosseous RM-YM mixes,
agreement ranging from Ϯ0.02 to Ϯ0.05 g/cm2, dependent with these soft tissue mixes uniformly constituted and ho-
on BMD level.(38)
mogeneously distributed throughout their respective do-
Representative results obtained in the present set of quan- mains. Although not without some similarity, these BMD
titative simulation studies and actual DXA scans of 150 inaccuracies stem from anthropometric and absorptiometric
phantom arrays, each with a different combination of BMD underpinnings distinctly different from those of some pre-
(TBV), RM/YM mix (g), and extraosseous F-to-LMT areal viously reported studies(18,21,39) in which F was not homo-
density ratio (q), are displayed in Figs. 3–5. The present geneously distributed through the extraosseous scan ROI.
results for the Hologic, Lunar, and Norland DXA scans are
shown separately; each is overlaid with the results of sim-
DISCUSSION
ulation studies obtained using faithful replication of the
dual-energy polychromatic X-ray spectral regions of that
instrument for corresponding phantom array representa-
Given the established 2-fold increase in population-based
tions. The inaccuracies in the measured BMD, expressed as fracture risk per each SD decrement in BMD(5) and the
the percent difference between the measured value and the demonstrated sizable inaccuracy manifest in actual DXA
known true phantom value, are plotted versus the RM/YM scans, this study can be seen to have considerable relevance
parameter value (g) spanning the range from purely red to clinical assessment of patient-specific fracture risk. The

422
BOLOTIN ET AL.
FIG. 3. Results of the present quantitative simulation studies (solid lines) and actual DXA scans (individual data point symbols) of a series of
different phantom arrays depicted in Fig. 1 for the Hologic (QDR-1000) DXA instrument. The fractional inaccuracies (in percent) inherent in
DXA-measured in vivo BMD are plotted as a function of the fraction of the marrow mix that is yellow, g (see text). These are shown for
representative F-to-LMT areal density ratios, (a) q ϭ 0.1, (b) 0.3, (c) 0.6, and (d) 1.0, of the extraosseous soft tissues in the scan ROI of the bone
segment under study for trabecular bone volumes TBV designated A (solid circles) ϭ 0.10, B (solid triangles) ϭ 0.15, C (solid squares) ϭ 0.20,
and D(asterisks) ϭ 0.35. Note that the ordinate scales are not all the same. Zero percent inaccuracy represents a BMD measurement yielding the
true BMD. Positive (negative) inaccuracies correspond to DXA-measured BMD overestimates (underestimates) of the true BMD value. The
clinically relevant regions of these data representations are ϳ0.13 Ͻ TBV Ͻ ϳ0.33, ϳ0.07 Ͻ q Ͻ ϳ0.9, and ϳ0.15 Ͻ g Ͻ ϳ0.85.
present finding opens for serious consideration the prospect fracture risk. This is strengthened by the meta analysis of
that a substantial part of the variability in DXA-measured Marshall et al.(5) of the evidence gleaned from a 90,000
patient-specific BMD may arise more from natural varia- person-year observation of 2000 fractures published be-
tions in soft tissue anthropometrics than to natural variations tween 1985 and 1994 in which DXA-measured in vivo
in true BMD in the normal population; certainly, this pos- BMD was assessed as a predictor of fracture risk. Although
sibility cannot be ruled out or ignored. When cast in terms these authors found a general population association be-
of clinical considerations, the critical TBV, g, and q ranges tween BMD and fracture risk of about the same degree as
that are particularly susceptible to poor accuracy pertain to that observed between blood pressure and stroke, they con-
subjects with truly osteopenic or osteoporotic bones (low cluded that too wide an overlap exists in the measured BMD
TBV), postmenopausal or elderly subjects (high g). Of note, of patients who develop a fracture and those who do not to
the error (underestimation) is accentuated for those with be able to identify with any certainty individuals who will
lean (frail) body build (low q). These anthropometric fea- develop a future fracture.(5) Because the present work has
tures tend to characterize that segment of the general pop- been confined to inherent inaccuracies associated with
ulation mainly targeted in DXA investigations.
patient-specific DXA in vivo BMD measurements and not
In this context, the foregoing suggests that a considerably directly to overall population screening scans, there appears
cautious clinical view be taken with respect to patient- to be no inconsistency between the findings of the present
(5)
specific in vivo DXA BMD values as predictors of bone study and the observations of Marshall et al.

INACCURACIES IN DXA BMD MEASUREMENTS
423
FIG. 4. Results corresponding to those of Fig. 3, but for the Lunar (DPX-␣) DXA instrument.
These inherent BMD inaccuracies also may serve to make hematopoietic marrow activity (increases in RM content) in
equivocal assessments and prognostic interpretations of pro- any given patient.(44)
spective DXA studies and/or monitoring of individual pa-
As a concomitant of the intrinsic two-component DXA
tients, as well. This is so because the primary purpose of limitation, any disparities that may actually exist in any
such prospective scans often is to monitor changes and given case between the meted X-ray attenuation character-
establish rates of change of BMD (usually some few percent istics of soft tissues along photon paths laterally adjacent to
in magnitude over a reasonable follow-up time) to gauge the the bone and those on paths intersecting the bone segment
rate of bone loss or to assess the efficacy of any prescribed (not measurable via DXA scans) are unavoidably assigned
treatment over a period of time. Yet, the present work has by DXA to bone material. Thus, should the RM/YM mix
shown, as did the earlier works,(14–16) that changes in soft along any given X-ray trajectory be such that its linear
tissue anthropometrics, increases or decreases in F fraction attenuation coefficient is less (greater) than that of the
(usually caused by body weight changes), or alterations in particular extraosseous F/LMT mix laterally adjacent to the
marrow composition between two prospective DXA scans bone, this absorptivity deficit (excess) is attributed by DXA
can equally well translate into a marked apparent, but illu- methodology to BMD along that path, thereby under- or
sory, change in BMD value between these measurements, overestimating the true BMD value. Consequently, the
without any true change in bone material having oc- greater the ratio of the actual thickness of the marrow (for
curred.(14)
any given red/yellow mix) to that of the bone material
Because bone marrow is one the most labile of body present along any given X-ray path traversing the bone (i.e.,
tissues, age,(40–43) aging per se, illness, drug treatment, the smaller the TBV), the greater is the magnitude of the
immobilization, alterations in physical activity regimens, BMD inaccuracy, irrespective of the true BMD value being
space flight, infection, etc., may readily alter marrow com- over- or underevaluated (linear attenuation coefficient ␮ of
position toward the more yellow or to more pronounced marrow greater or smaller than that of the extraosseous soft

424
BOLOTIN ET AL.
FIG. 5. Results corresponding to those of Fig. 3, but for the Norland (XR-26) DXA instrument.
tissue, respectively). It is clear, therefore, that TBV is the extraosseous F infused in or covering lean tissue, no pres-
more pertinent bone parameter relevant to proper analysis of ence or variability in the thickness of intraluminal fluid, gas,
inaccuracies in DXA in vivo BMD scans than is BMD for or solids in the gastrointestinal tract, etc. present within the
bone sites largely trabecular in nature (lumbar vertebrae, entire DXA scan ROI), the present results should best be
proximal femoral sites, os calcis, etc.). Although BMD is considered only indicative lower limits of DXA in vivo
that quantity measured by DXA and TBV is not, it is BMD inaccuracies to be anticipated for individual clinical
nevertheless the case that the measured value of BMD is patients.(14)
also a parameter very much dependent on bone size and
shape,(45) whereas the parameter TBV, by definition, is actual DXA scans of all phantom arrays to be fully consis-
independent of overall bone dimensions. tent with the present simulation study findings in each
The data in Figs. 3–5 show the present results for the
For the reasons mentioned previously and to allow the corresponding case. Shown, too, is that the present actual
present findings to be examined and assessed on a more Hologic, Lunar, and Norland DXA scans of each of these
fundamental basis, the results of the present series of inves- phantom arrays display virtually the same DXA-measured
tigations have been described in terms of the most relevant BMD values, in accord with the multicenter clinical find-
bone site–specific parameterizations of bone and soft tissue ings of earlier investigators,(46) the measured BMD inaccu-
quantities and excludes vagaries of bone size (cross- racies clearly exceeding the precision of the associated in
sectional area) and whole body characterizations. Thus, in vivo measurements. It is significant, as well, that the per-
all cases, the percentage inaccuracies in BMD pertaining are centage inaccuracies in BMD found for both the simulation
presented as functions of TBV, q, and g.
studies and the actual DXA scans of the same phantom
As absorptiometric disparities between the extraosseous arrays display the same trends, magnitudes, and dependen-
tissues present along X-ray trajectories lateral to and pass- cies as a function of TBV and soft tissue anthropometrics
ing through the interrogated bone were specifically ex- (q,g), and have the same zero inaccuracy intersection points
cluded in the present studies (i.e., no inhomogeneities in (a sensitive cross-evaluation criterion). These findings also

INACCURACIES IN DXA BMD MEASUREMENTS
425
serve to confirm the analytic underpinnings(14,15) on which Research Laboratory in Rotterdam, especially recognizing
these simulations are based. Henk Mulder, Jeanette Kamman, Margreet Vlastuin, Jac-
The validity of the means by which the dual-energy queline Baden, and Irene Hogerheide for their generous
polyenergetic DXA spectral domains have been replicated support in making available to us the Lunar and Hologic
also is established, with the present work representing a instruments. We are pleased to acknowledge receipt of
self-consistent, reliable, and reproducible system of analy- contributions from Chemcor Australia Ltd. (Altona, Victo-
sis, simulation, and DXA operation. Also, we note that the ria, Australia), Courtenay Polymers (Dandenong, Victoria,
results of the present simulation study using the replicated Australia), and Cadillac Plastics Australia (Seaford, Victo-
dual polyenergetic DXA spectral regions (including beam- ria, Australia) in the form of powdered and solid polyeth-
hardening effect corrections) and those of the prior(14,15) ylene used in the phantom fabrication facet of this work.
dual monoenergetic DXA simulations are essentially iden- Likewise, the contributions from ChemPlast Specialties
tical. This is an important point, because it confirms the S.p.A (Milano, Italy), Asia-Pacific Microspheres Ltd.
equivalence of the two seemingly disparate simulation bases (Selangor, Malaysia), and RevChem Plastics Inc. (Red-
and establishes the validity of DXA simulation studies lands, CA, USA) in the form of PMS are gratefully ac-
carried out using the more simply executed and consider- knowledged. The contributions from Cemic Oy (Espoo,
ably less involved dual monoenergetic simulation regimen Finland) and Plastiroll Oy (Ylojarvi, Finland) in the form of
for assessment, prediction, evaluation, and formulation of polyethylene and those from Ciba Specialty Chemicals Ltd.,
DXA equivalent results.
Helsinki, Finland, in the form of epoxy samples are ac-
The present series of DXA scans of specially formulated knowledged as well. The skill and cooperation of the work-
phantom material arrays and dual polyenergetic simulated shop staff of the School of Physics (University of Mel-
replications of the corresponding representations of near bourne) in machining the phantom materials to close
ideal DXA in vivo vertebral scans have quantitatively con- tolerances is sincerely appreciated. This study was finan-
firmed and validated the specifics of underlying soft tissue cially supported by the Research Fund of Tampere Univer-
anthropometrics that can insinuate substantial and unavoid- sity Hospital.
able inherent inaccuracies into DXA-measured in vivo
BMD outcomes.
The present work has shown that these inaccuracies in
REFERENCES
DXA in vivo BMD methodology can be ubiquitous with
measurement inaccuracies as high as 20% or more readily
anticipated clinically. This is particularly so in the cases of
osteopenic, osteoporotic, and elderly patients, the very
groups it is most important for in vivo DXA to gauge BMD
accurately.
1. Riggs BL, Melton LJ III 1992 The prevention and treatment of
osteoporosis. N Engl J Med 327:620–627.
2. Cooper C, Campion C, Melton LJ III 1992 Hip fractures in the
elderly: A worldwide projection. Osteoporos Int 2:285–289.
3. Kannus P, Niemi S, Parkkari J, Palvanen M, Vuori I, Jarvinen
M 1999 Hip fractures in Finland between 1970 and 1997 and
predictions for the future. Lancet 353:802–805.
The sizable inherent systematic inaccuracies in DXA-
measured in vivo BMD shown and corroborated by the
present findings serve to call into question certain aspects of
the reliability of definitive patient-specific osteopenic and/or
osteoporotic categorizations based on DXA screening and
diagnostic scans. So too, the present results tend to render
tenuous the dependability of prospective DXA measure-
ments carried out on a given individual to distinguish un-
equivocally between changes in true BMD and those appar-
ent BMD increases or decreases arising from changes in soft
tissue anthropometrics and/or altered bone marrow compo-
sition. The intrinsic shortcomings of DXA giving rise to
these ubiquitous BMD inaccuracies also call attention to the
need for the exercise of prudence and circumspection in
clinical interpretations and assessments of DXA-based clin-
ical studies of the effectiveness or efficacy of particular
drug, diet, or exercise therapies introduced to increase or
maintain bone mass or to prevent bone loss in any given
patient.
4. Genant HK, Engelke K, Fuerst T, Glueer C-C, Grampp S,
Harris S, Jergas M, Lang T, Lu Y, Majumdar S, Mathur A,
Takada M 1996 Noninvasive assessment of bone mineral and
structure: State of the art. J Bone Miner Res 11:707–730.
5. Marshall D, Johnell O, Wedel H 1996 Meta-analysis of how
well measures of bone mineral density predict occurrence of
osteoporotic fractures. Br J Med 312:1254–1259.
6. Courtney AC, Wachtel EF, Myers ER, Hayes WC 1995 Age-
related reductions in the strength of the femur tested in a fall-
loading configuration. J Bone Joint Surg Am 77A:387–395.
7. Bouxsein ML, Coan BS, Lee SC 1999 Prediction of the
strength of the elderly primal femur by bone mineral density
and quantitative ultrasound measurements of the heel and tibia.
Bone 25:49–54.
8. Webber CE 1987 The effect of fat on bone mineral measure-
ments in normal subjects with recommended values of bone,
muscle, and fat attenuation coefficients. Clin Phys Physiol
Meas 8:143–148.
9. Sorenson JA 1990 Effects of nonmineral tissues on measure-
ments of bone mineral content by dual-photon absorptiometry.
Med Phys 17:905–912.
10. Tothill P, Pye DW 1992 Errors due to non-uniform distribution
of fat in dual X-ray absorptiometry of the lumbar spine. Br J
Radiol 65:807–813.
11. Tothill P, Avenell A, Reid RM 1994 Precision and accuracy of
measurements of whole-body bone mineral: Comparisons be-
tween Hologic, Lunar and Norland dual-energy absorptiom-
eters. Br J Radiol 67:1210–1217.
12. Svendsen OL, Hassager C, Skodt V, Christiansen C 1995
Impact of soft tissue on in vivo accuracy of bone mineral
measurements in the spine, hip, and forearm: A human cadaver
study. J Bone Miner Res 10:868–875.
ACKNOWLEDGMENTS
The authors thank Markku Jarvinen M.D., Tampere Uni-
versity Hospital, for his early recognition of the potential
significance of the research undertaken and his vital, prac-
tical support of it; Ilkka Vuori, M.D., UKK Institute, for his
contribution; and Good Clinical Practice and the Image

426
BOLOTIN ET AL.
13. Kuiper JW, Van Kuijk C, Grashuis JL, Ederveen AGH, 30. White DR, Martin RJ, Darlison R 1977 Epoxy resin tissue
Schutte HE 1996 Accuracy and influence of marrow fat on substitutes. Br J Radiol 50:814–821.
quantitative CT and dual-energy X-ray absorptiometry mea- 31. Goodsitt MM, Hoover P, Veldee MS, Hsueh SL 1994 The
surements of the femoral neck in vitro. Osteoporos Int 6:25–
30.
14. Bolotin HH 1998 Analytic and quantitative exposition of
composition of bone marrow for a dual-energy quantitative
computed tomography technique: A cadaver and computer
simulation study. Invest Radiol 29:696–704.
patient-specific systematic inaccuracies inherent in planar 32. Wahner HW, Steiger P, von Stetten E 1994 Instruments and
DXA-derived in vivo BMD measurements. Med Phys 25:139–
151.
measurements techniques. In: Wahner HW, Fogelman I (eds.)
The Evaluation of Osteoporosis: Dual Energy X-ray Absorp-
tiometry in Clinical Practice. Martin Dunitz, Ltd., London,
UK, pp. 14–34.
15. Bolotin HH 1998 A new perspective on the causal influence of
soft tissue composition on DXA-measured in vivo bone min-
eral density. J Bone Miner Res 13:1739–1746.
16. Bolotin HH 1998 A new perspective on the effects of body
weight and body fat mass on DXA-measured in vivo bone
mineral density. Osteoporos Int 8:514.
33. Wapstra AH, Nijgh GJ, Van Lieshout R 1959 Gamma rays. In:
Nuclear Spectroscopy Tables. North-Holland, Amsterdam,
The Netherlands, pp. 66–74.
34. Goldstein GW 1957 U.S. Nat Bur Standards Circular 583.
17. Farrell TJ, Webber CE 1989 The error due to fat inhomoge- 35. Nord RH 1987 Technical considerations in DPA. In: Genant
neity in lumbar spine bone mineral measurements. Clin Phys
Physiol Meas 10:57–64.
18. Cullum ID, Ell PJ, Ryder JP 1989 X-ray dual photon absorp-
tiometry: A new method for measurement of bone density. Br J
Radiol 62:587–592.
HK (ed.) Osteoporosis Update 1987, pp. 203–212.
36. Nord RH, Payne RK 1995 Body composition by dual-energy
X-ray absorptiometry–a review of the technology. Asia Pac
J Clin Nutr 4:167–171.
37. Bland JM, Altman DG 1986 Statistical methods assessing
agreement between two methods of clinical measurement.
Lancet 1:307–310.
19. Gotfredsen A, Podenphant J, Norgaard H, Nilas L, Nielsen
V-AH, Christiansen C 1988 Accuracy of lumbar spine bone
mineral content by dual photon absorptiometry. J Nucl Med 38. Kolta S, Ravaud P, Fechtenbaum J, Dougados M, Roux C
29:248–254.
1999 Accuracy and precision of 62 bone densitometers using
20. Ho CP, Kim RW, Schaffler MB, Sartoris DJ 1990 Accuracy of
a European Spine Phantom. Ostoporos Int 10:14–19.
dual-energy radiographic absorptiometry of the lumbar spine: 39. Hansen MA, Hassager C, Overgaard K, Marslew U, Riis BJ,
Cadaver study. Radiology 176:171–173.
Christiansen C 1990 Dual-energy x-ray absorptiometry: A
precise method of measuring bone mineral density in the
lumbar spine. J Nucl Med 31:1156–1161.
21. Hangartner TN, Johnston CC 1990 Influence of fat on bone
measurements with dual-energy absorptiometry. Bone Miner
9:71–81.
40. Custer RP 1974 An Atlas of the Blood and Bone Marrow.
Saunders, Philadelphia, PA, USA, pp. 33–42.
22. Hassager C, Jensen SB, Gotfredsen A, Christiansen C 1991
The impact of measurement errors on the diagnostic value of 41. Minaire P, Edouard C, Arlot M, Meunier PJ 1984 Marrow
bone mass measurements: Theoretical consideration. Osteopo-
ros Int 1:250–256.
23. Sabin MA, Blake GM, MacLaughlin-Black SM, Fogelman I
changes in paraplegic patients. Calcif Tissue Int 36:338–340.
42. Glowacki J 1995 Influence of age on human marrow. Calcif
Tissue Int 35(Suppl 1):S50–S51.
1995 The accuracy of volumetric bone density measurements 43. Ricci C, Cova M, Kang YS, Yang A, Rahmouni A, Scott WW,
in dual X-ray absorptiometry. Calcif Tissue Int 56:210–214.
24. Antonacci MD, Hanson DS, Hegeness MH 1996 Pitfalls in the
measurement of bone mineral density by dual energy X-ray
absorptiometry. Spine 21:87–91.
25. Wang J, Heymsfield SB, Aulet M, Thornton JC, Peirson JRN
1989 Body fat from body density: Underwater weighing vs.
dual-photon absorptiometry. Am J Physiol 256:E829–E834.
26. Dunnill MS, Anderson JA, Whitehead R 1967 Quantitative
Zerhouni EA 1990 Normal age-related pattern of cellular and
fatty bone marrow distribution in the axial skeleton: MR
imaging study. Radiology 177:83–88.
44. Mazess RB 1983 Errors in measuring trabecular bone by
computed tomography due to marrow and bone composition.
Calcif Tissue Int 35:148–152.
45. Sievanen H 2000 Physical model for DXA-derived BMD.
Invest Radiol 35:325–330.
histological studies on age change in bone. J Pathol Bacteriol 46. Genant HK, Grampp S, Gluer CS, Faulkner KG, Jergas M,
94:275–291.
Engelke K, Hargiwara S, Van Kuijk C 1994 Universal stan-
dardization for dual X-ray absorptiometry: Patient and phan-
tom cross-calibration results. J Bone Miner Res 9:1503–1514.
27. Meunier PJ, Aaron J, Edouard C, Vignon G 1971 Osteoporosis
and the replacement of cell populations of the marrow by
adipose tissue. Clin Orthop 80:147–154.
28. Hartsock RJ, Smith EB, Petty CS 1965 Normal variations with
aging of the amount of hematopoietic tissue in bone marrow
from the anterior iliac crest: A study made from 177 cases of
sudden death examined by necropsy. Am J Clin Pathol 43:
326–331.
29. Dunn WL, Wahner HW 1994 Commercial instruments avail-
able for bone mineral measures. In: Wahner HW, Fogelman I
(eds.) The Evaluation of Osteoporosis: Dual Energy X-ray
Address reprint requests to:
Prof. Emeritus H.H. Bolotin, D.Sc., Ph.D.
School of Physics
University of Melbourne
Victoria 3010, Australia
Absorptiometry in Clinical Practice. Martin Dunitz, Ltd., Lon- Received in original form March 20, 2000; in revised form August
don, UK, pp. 35–48. 5, 2000; accepted August 24, 2000.