Inaccuracies Inherent in Dual-Energy X-Ray Absorptiometry In Vivo Bone Mineral Density Can Seriously Mislead Diagnostic/Prognostic Interpretations of Patient-Specific Bone Fragility


  • H. H. Bolotin,

    Corresponding author
    1. Department of Medical Radiations Science, RMIT University, Bundoora, Victoria, Australia
    2. School of Physics, University of Melbourne, Victoria, Australia
    • Address reprint requests to: Professor Emeritus H. H. Bolotin, Ph.D., D.Sc., School of Physics, University of Melbourne, Victoria 3010, Australia
    Search for more papers by this author
  • H. Sievänen

    1. The Bone Research Group, UKK Institute, Tampere, Finland
    Search for more papers by this author


SINCE ITS advent, noninvasive dual-energy X-ray absorptiometric (DXA) in vivo measurement of bone mineral density (BMD) has been accepted almost universally as the methodology of choice in the field of clinical bone fragility. More specifically, DXA generally is considered to be the prime, reliable assessor of the osteopenic/osteoporotic condition(1–6) of bone fracture propensity,(6–9) of correlations of measured BMD with fracture failure load of bones,(10–17) and of the efficacy of remedial bone therapies.(18–26) DXA also is held to be the standard against which newer, emerging alternative methods of bone quality assessment are evaluated.(27, 28)

However, despite the near ubiquitous clinical use of DXA-measured in vivo BMD and the widespread reliance on it, recently attention has been drawn to the prospect that BMD, per se, may not be the primary determinant of bone fracture risk. Reports of large reductions in vertebral fracture rates attributed to antiresorptive therapy, but without commensurate accompanying increases in BMD,(20, 22, 24, 29, 30) have prompted suggestions(7, 31) that BMD may not be homologous with bone strength and that other nondensity particulars may be as, or more, important in this respect.

Notwithstanding the importance of the foregoing, certainly one of the most fundamental issues relevant to all aspects of bone densitometry and fragility studies and resultant in vivo diagnostic/prognostic interpretations relates to the growing number of investigations that have shown DXA-derived in vivo BMD to be subject to sizable inherent systematic inaccuracies that may adversely influence measurement outcomes.(32–39) Clearly, the extent of such in vivo BMD inaccuracies and the dependence of them on soft-tissue anthropometric particulars are of salient import for patient-specific clinical DXA measurements.(38, 39) Such BMD inaccuracies could seriously compromise the integrity of measurements undertaken to diagnose, monitor, and evaluate the osteopenic/osteoporotic condition and predictive bone fragility of any individual patient. Were this the case, reliance on DXA-derived BMD measurement values could lead to misinterpretations and erroneous assessments of the efficacy and/or quantitative effectiveness of drug and other therapeutic regimens intended to ameliorate the osteoporotic condition.(32–41) Moreover, should large in vivo BMD inaccuracies pertain, clinical DXA investigations undertaken to delineate specific anthropometric, dietary, and/or therapeutic factors which may prove biologically causal or remedially effective in altering BMD and bone fracture propensity, could yield tenuous and/or misleading conclusions.(38–41)

These concerns are underscored by the findings of a considerable number of investigations: in situ/in vitro cadaveric DXA studies,(10, 11 17 42) absorptiometrically realistic phantom DXA studies,(40) simulation studies of replicated clinical in vivo bone-site particulars,(38–41) and fundamental quantitative analyses of DXA methodology.(34, 43) Collectively, these observations and results may be seen to constitute a formidable case for reevaluation of the reliability and accuracy of DXA-measured in vivo BMD.

For these reasons, it is important to highlight and evaluate the evidence substantiating the presence of inherent clinical inaccuracies in DXA-measured in vivo BMD, assess the demonstrated extent of these inaccuracies, and draw attention to some of the consequential effects of these inaccuracies on bone densitometry in the clinical context.


Origin of inherent systematic DXA in vivo BMD inaccuracies

Inherent systematic inaccuracies in DXA BMD derive from the known inapplicability of planar DXA methodology to bone sites comprised of more than two absorptiometrically distinguishable components in the entire scan region of interest (ROI); the “two-component DXA limitation.”(32–35, 37–39) Yet, all in vivo bone sites are comprised of bone material, intraosseous soft tissue of some unspecified red/yellow marrow compositional mix, and some combination of lean muscle tissue and fat external to the bone that together constitute at least four absorptiometrically disparate components in the DXA scan ROI. Therefore, it is clear that in vivo bone-site reality does not and cannot strictly conform to or satisfy the two-component DXA restriction.(38) This shortcoming is further exacerbated in any patient-specific case by the inability of DXA to assess the necessary particulars of the bone marrow composition. Further, in vivo DXA can neither determine the fat component (and its degree of inhomogeneity) within the lean extraosseous tissue along any X-ray path traversing bone material nor determine the quantitative extent to which the two-component DXA limitation fails to be satisfied in any given in vivo bone-site scan.

A most serious consequence of the violation of this intrinsic two-component DXA limitation is the under- or overestimation of BMD. This is so because DXA methodology erroneously and unavoidably attributes to bone material any difference between the X-ray absorptiometric characteristics of the specific bone marrow composition and the particular extraosseous soft-tissue composition within the particular bone-scan ROI of a given patient.(38–40) This is the case even when these soft tissues are homogeneously constituted throughout their separate, respective, intra- and extraosseous domains (e.g., even when fat is distributed uniformly throughout the lean muscle tissue in the ROI external to the bone). For this reason, the DXA in vivo BMD measurement must be inherently inaccurate to some indeterminate extent for any given patient, and, in general, the DXA scan will necessarily result in a measured value of BMD that differs from the true value.(38) (In the present context, the true value of BMD denotes that which would have been measured were there no DXA BMD measurement inaccuracies.)

Results of anatomically realistic simulation studies of vertebral and femoral in vivo DXA scans

Based on the comprehensively developed analytic underpinnings of DXA,(38) an extensive series of quantitative simulation studies of typical DXA in vivo lumbar vertebral and proximal femoral sites have been reported(38–40) in which the full ranges of anatomically realistic BMD and soft-tissue anthropometrics encountered clinically within the scan ROI were represented. These studies showed that, even for cases in which the extraosseous tissues in the bone-scan ROI are homogeneously constituted and uniformly distributed, patient-specific in vivo BMD inaccuracies readily exceeding ±20% (i.e., some two SDs of age-moderated, population-based, normative BMD data) can be anticipated. This is particularly so for postmenopausal women, the elderly, and the osteopenic/osteoporotic—the very groups for which it is most important that in vivo DXA gauge BMD accurately. In addition, these studies showed, both quantitatively and qualitatively, that the well-documented correlations between percent body fat mass/body mass/body mass index (BMI) and DXA-measured in vivo BMD(5, 43–52) seem unlikely to be of bone-biological origin, but, instead, appear fully consistent with being manifestations of inherent DXA in vivo BMD inaccuracies unaccompanied by any true BMD changes.(39) At the same time, and again based on DXA in vivo inaccuracies induced by changes in soft-tissue habitus, these simulation studies also provide a realistically credible explanation for the apparent lack of expected BMD increases with remedial antiresorptive therapy regimens.(41)

DXA scans of absorptiometrically realistic phantoms replicating bone material, marrow, fat, and lean muscle tissue compositions

Norland XR-26 (Norland Corp., Fort Atkinson, WI, USA), Lunar DPX-α (Lunar Corp., Madison, WI, USA), and Hologic QDR-1000 (Hologic, Waltham, MA, USA) DXA instruments were each used to carry out the same extensive set of BMD scans of 150 different phantom arrays. The phantom assemblies were comprised of materials specially formulated and fabricated to span the anthropometric ranges of BMD and intra- and extraosseous soft-tissue compositions encountered clinically.(40) Additionally, the X-ray attenuation coefficients of all relevant in vivo tissues were matched virtually exactly by their respective phantom representations across the full DXA X-ray energy range. To establish conditions most favorable for DXA BMD measurements, all phantom arrays had identical overall dimensions, all intra- and extraosseous phantom materials were separately absorptiometrically homogeneous throughout their respective domains in the scan ROI, and the geometry of the phantom arrays were effectively ideal for bone-edge detection algorithms incorporated in each of the DXA instruments used.(40) The results of these studies—effectively identical for all three of these widely used DXA instruments—corroborated in every respect the inherent systematic DXA BMD inaccuracies found in the extensive quantitative simulation studies described previously(38, 39) and those carried out for each of these phantom arrays.(40)

These foregoing findings can be summarized fairly as follows. DXA-measured in vivo BMD inaccuracies are very much patient specific, because the magnitude of the BMD value extracted from a given DXA measurement depends on: (i) the exact specifications of the bone marrow composition, (ii) the detailed composition of the extraosseous soft tissue, and (iii) the true (not measured) value of BMD (i.e., proportional to the average thickness of bone material along all X-ray paths traversing the given bone-site) actually pertaining within the specific scan ROI of each given patient.(40) It is the case that (a) for any given extraosseous soft-tissue composition in the scan ROI, the more yellow the bone marrow the more DXA underestimates true BMD; (b) for any given bone marrow composition, the smaller the proportion of fat in the extraosseous lean tissue the more the DXA-measured BMD underestimates the true value; and (c) for any given marrow and extraosseous soft-tissue compositions, the smaller the true BMD at any particular bone site, the greater is the DXA over- or underestimate of it. Thus, for any individual with low true BMD, more yellowish bone marrow and leaner soft-tissue habitus (postmenopausal, osteopenic, osteoporotic, and elderly persons), these inherent inaccuracies in DXA could lead to measured BMD values that are sizable over- or underestimates of the true BMD value, the extent of which depends most particularly on the patient-specific soft-tissue anthropometrics in the scan ROI of the given bone sites(s) that were interrogated.

This being the case, it must be noted that bone marrow is one of the most labile of soft tissues(53–56) and that the extraosseous soft-tissue composition within the DXA scan ROI varies considerably from bone site to bone site and for any selected bone site may vary over time in any given patient and from patient to patient. For these reasons, it can be expected that the inaccuracies inherent in DXA in vivo methodology may result in seemingly arbitrary (if not capricious) BMD values being extracted from the measurements. Thus, for postmenopausal, osteopenic, osteoporotic, and elderly individuals (generally lower true BMD, more yellowish marrow, and often leaner than the normal population), these inaccuracies may readily give rise to underestimates of true BMD as large as 20-30%.(38–40)

Results of in situ/in vitro cadaveric studies

A number of carefully detailed cadaveric studies(10, 11, 17, 36), 42) have provided incisive evidence of large inaccuracies in specimen-specific (patient-specific) DXA BMD measurements. Of these, the in vitro investigation of Kuiper et al.(36) showed quite conclusively that the BMD values obtained from DXA measurements of excised human cadaveric femoral neck specimens, immersed in a water bath after removal of all external soft tissues (“denuded”), were consistently smaller when the given bone specimen was scanned with the intraosseous marrow intact than when measured again after the marrow was removed and replaced by water. The latter arrangement constitutes a reasonably close approximation to a two-component DXA scan ROI. Further, the differences in these paired, carefully standardized DXA-measured BMD values tended to be greater the higher the percent of fat in the original bone marrow (determined by chemical analysis of the extracted marrow). These observations are in excellent agreement with the findings of the analytic simulation and phantom studies described previously(38–41); they display (and confirm) the established(41) trend and extent of inherent DXA BMD inaccuracies induced by the different absorptiometric properties of the various pertaining bone marrow compositions.

Comparison of the findings of the cadaveric bone fragility studies of Lochmüller et al.(10) and Bouxsein et al.(11) serves to exemplify and illustrate sizable DXA-measured BMD inaccuracies directly attributable to variations in specimen-specific soft-tissue composition particulars within the scan ROI of their respective studies. The former workers scanned vertebrae L2-L4 only in situ, the latter obtained BMD values of the femoral trochanteric region only ex situ (DXA-scanned in a water bath with marrow intact). In both studies, the fracture failure load of the corresponding excised and denuded cadaveric bone (marrow intact) was determined after the DXA measurement. The overall BMD versus fracture failure load correlation coefficient found in the in situ study(10) was small (r2 = 0.23), which was somewhat lower than reported by others,(28)57) but marginally higher than that reported by Bjarnason et al.(17) This is to be compared with the femoral in vitro work of Bouxsein et al.(11) in which the much higher value of r2 = 0.90 pertained. A number of other similar studies(12–16) of fracture failure load versus DXA in vitro BMD measurements of denuded cadaveric lumbar vertebral and proximal femoral bone specimens also yielded generally higher correlation coefficients (r2 values) than did the in situ works of Lochmüller et al.(10) and Bjarnason et al.(17) In the context of the analytical underpinnings of DXA,(38) the generally smaller r2 values found in the in situ studies can be seen as due principally to the more sizable and varied DXA BMD inaccuracies expected in these cases. This is anticipated as generally broader absorptiometric disparities between the compositions of the extraosseous fat/lean muscle tissue and the intraosseous red/yellow marrow combinations would pertain in situ (and in vivo) than would be the case between the various bone marrow compositions within the excised, denuded specimens and the standardized water bath in the in vitro scans.(12) Thus, because the DXA in vitro BMD measurements of denuded cadaveric bone specimens in the works of Bouxsein et al.(11) and others(12–17, 42) tend to approach two-component scan ROIs to a greater extent than do DXA in situ (in vivo) BMD situations, the in vitro BMD values are tacitly taken in these studies as the better approximations to true BMD values.

The illuminating study by Svendsen et al.(42) on the impact of soft-tissue composition on in vivo accuracy of DXA-measured BMD is of particular importance in this regard. These investigators compared the DXA-measured BMD values of in situ and in vitro lumbar vertebral specimens (L2-L3 and L2-L4 in lateral and anteroposterior [AP] projections, respectively), forearm, and five proximal femoral regions of 14 cadavers. Exactly the same vertebral bone specimens were DXA-measured in a standardized orientation in both the in situ and the in vitro facets of their investigation, the in vitro scans being of excised, denuded specimens immersed in a 71.4% wt/wt water/ethanol bath. The linear regression analysis of their corresponding in situ versus in vitro BMD values yielded overall accuracy in the standard estimate of errors (SEE%) for the L2-L4 vertebrae of 5.3% and 9.7% for the AP and lateral scans, respectively. The analogous SEE% found for the various femoral sites ranged from about 3% to about 11%. Nevertheless, it is important in the present context to note that their data displayed several individual, specimen-specific inaccuracies exceeding 20% in these 14 cadaveric cases, some of the largest of which pertained to vertebral specimens displaying DXA in vitro BMD values in the lower ranges. The trend and extent of these specimen-specific BMD inaccuracies are fully consistent with and complementary to other observations and findings already summarized above.

Cadaveric studies by Bjarnason et al.,(17) in which biomechanical measurements were made of 32 individual vertebrae from the same 14 cadavers used in the earlier work of Svendsen et al.,(42) clearly display large specimen-specific inaccuracies in DXA-derived BMD. These later workers(17) compared the measured fracture failure load of each of the 32 individual vertebrae with BMD values obtained from separate AP and lateral DXA scans of the same vertebra both in situ and in vitro. Although their analyses and discussion focused on overall linear regression-derived dependencies of DXA-measured BMD and bone fragility, the format of their published data provides a one-to-one correspondence between the in situ and in vitro (denuded specimens in a 71.4% water/ethanol, wt/wt, bath) BMD values of each of these vertebral specimens (see in vitro vs. in situ data in Fig. 2 of Bjarnason et al.(17)). The present Fig. 1 displays these in situ/in vitro comparisons extracted from their lateral and AP scan results,(17) respectively. Despite the limited number of different combinations of body soft tissue and bone marrow compositions, which only 14 in situ cadavers unavoidably represent, Fig. 1A, nevertheless, clearly shows that the DXA-measured BMD values obtained from lateral scans of these vertebrae exhibited measurable, sizable differences between the in situ and in vitro measurements of the same vertebra in 30 of these 32 cases. More than half of these BMD disparities (inaccuracies) exceeded 10%, 5 of the 30 reflected inaccuracies greater than 20%, and one displayed a measured in situ versus in vitro BMD difference of ∼53%. It is of particular interest that in their lateral scan cases (Fig. 1A), 11 of the inaccuracies constituted overestimates of BMD and 19 were underestimations, whereas in their AP scans of the same vertebrae (Fig. 1B), almost the reverse pertained for these same vertebrae, with 20 overestimates and 11 underestimates of BMD. The results of the analytic, simulation, and phantom studies(34, 38–41) strongly suggest that the observed reversal in the in situ versus in vitro over- and underestimates in the lateral and AP DXA-measured BMD of these same 32 vertebrae arose from in situ differences in: (i) the ratio of the areal densities of extraosseous fat and lean muscle tissue (the particular body soft-tissue composition external to the bone) and (ii) any inhomogeneities in the distribution of fat through the lean muscle tissue within the ROI of each of these two alternative scans.

Figure FIG. 1..

Percentage differences between DXA measured in situ and in vitro BMD values for (A) lateral scans and (B) AP scans for each of the 32 vertebrae excised from 14 cadavers, as extracted from the results reported by Bjarnason et al.(17) Displayed in groupings of 5% BMD differences are the number of cases out of 32 that the in situ DXA BMD values separately overestimate (positive inaccuracy) and underestimate (negative inaccuracy) the BMD values meted in vitro in the two alternative scans.


First, it should be noted that the term “true BMD” is used here to designate that value of BMD that would have been extracted from a standardized DXA scan were no BMD inaccuracy associated with the measurement, while “measured BMD” is that value extracted from an actual standardized DXA scan of the same bone site.

The extensive evidence of sizable inherent systematic inaccuracies in DXA-measured in vivo BMD displayed in the results of a considerable number of relevant and pertinent investigations makes it clear that DXA-measured BMD is not necessarily synonymous with true BMD. As such, it would appear necessary and prudent that the distinction between measured and true BMD values be circumspectively considered before definitive conclusions are educed from any DXA-based assessments of BMD, bone fragility, the osteopenic/osteoporotic condition, and/or the effectiveness of antiresorptive drugs or other remedial therapies in any given patient case. These cautionary qualifications are wholly consistent with but extend considerably beyond the conclusion of Marshall et al.(8) that the DXA measurement of BMD should not be relied on to identify those individuals who will develop a future fracture.

This stricture should not be ignored in any given study in which regression analysis is relied on to interrogate trends, relationships, and/or correlations between DXA-derived BMD values and any other given measured parameter (e.g., bone fracture propensity, BMI, body fat mass, body lean tissue mass, therapeutic regimen effectiveness, etc.). The same wariness may be attached justifiably to those DXA-based findings relating BMD to fracture failure load, which have indicated the causative dependence of measured BMD on bone fragility to be less positive than expected from particular antiresorptive drug therapies. Such circumspection appears warranted in light of the realistic prospect that outcomes may, in fact, be distorted artifactually because of measured rather than true BMD being the associated factor in each of these studies. For example, the lack of observed BMD changes expected from antiresorptive therapy may not necessarily imply that BMD is not a good measure of bone fragility or strength, but only that measured BMD, due to the inherent and sizable DXA in vivo inaccuracies, may not be.(41) Similarly, the relationship between BMI, body soft-tissue composition, etc., and DXA in vivo BMD measurements(5, 43–52) can be attributed more to these anthropometric features affecting the measured BMD values than to some related biologically causal mechanism affecting true BMD.(39, 41)

From the evidence presented, it is clear that patient-specific DXA-measured in vivo BMD inaccuracies can readily exceed ±20% or more in individual cases without any available clinical means to identify these cases. From this perspective, it is therefore reasonable to question the reliability of DXA-measured changes in vertebral BMD to gauge the effectiveness of therapeutic drugs for any given person. Conversely, but not without similar underpinnings, such treatment regimens as hormone replacement therapy, antiresorptive drugs, etc. may, if evaluated primarily on DXA in vivo BMD measurements, be judged erroneously to be either of exaggerated efficacy or of unexpected inadequacy. This follows from the very real prospect that one or another of these therapies might induce relatively modest changes in (i) bone marrow composition(41, 53, 54) and/or (ii) the particular distribution and proportions of fat within the lean extraosseous tissue in the local bone-site vicinity sufficient to cause patient-specific DXA-measured BMD to either over- or underestimate true BMD. Changes in both bone marrow and extraosseous body composition accompanying aging(55, 56) also can lead to analogous uncertainties and ambiguities.

Indeed, even some of instances of weaker than expected correlation coefficients relating fracture failure loads(10, 12, 14, 16), (17, 28) with DXA-measured in vitro BMD of cleansed, excised cadaveric vertebrae might very well have a related origin. From the present perspective, these DXA measurements may be seen as having been influenced as much or more by the absorptiometric disparity between the in vitro bath (in which each specimen was scanned) and that of the particular intact or residual bone marrow within the given vertebra than to a weak relationship between true BMD and fracture failure load. Going yet a step further, one also might question the extent to which “biological” variability (± ∼20% about the mean) in the population-based, age-matched normative DXA BMD data is due to natural variations in true BMD values among individuals and to what degree it may be due more to person-to-person variations in the soft-tissue anthropometrics of the normal population. If traced to the latter, it might largely help explain the number of false-negative osteoporotic diagnoses of those persons known by other means to be osteoporotic, but whose measured BMD values fall within the ±2 SD band of the DXA-based in vivo BMD normative data. Thus, if true BMD were a reasonably good gauge of bone fragility (as it might well be), evidence in support of it might actually be obscured by the very DXA methodology now used to mete it.

The ramifications of inherent systematic DXA in vivo BMD inaccuracies are clearly wide-ranging and potentially serious. For this reason, it would not be unwarranted to view patient-specific bone fragility interpretations based on this methodology with considerable circumspection and/or to advocate reassessment of much of the existing “conventional wisdom,” which now rests on DXA in vivo bone densitometry.