Prediction of Vertebral Strength In Vitro by Spinal Bone Densitometry and Calcaneal Ultrasound



Spinal bone mineral density (BMD) measurements and calcaneal ultrasound were compared in terms of their ability to predict the strength of the third lumbar vertebral body using specimens from 62 adult cadavers (28 females, 34 males). BMD was measured using dual X-ray absorptiometry (DXA) in both vertebra and calcaneus. Quantitative computed tomography (QCT) was used to determine trabecular BMD, cortical BMD, cortical area, and total cross-sectional area (CSA) of the vertebral body. Bone velocity (BV) and broadband ultrasonic attenuation (BUA) were measured in the right calcaneus. Vertebral strength was determined by uniaxial compressive testing. Vertebral ultimate load was best correlated with DXA-determined vertebral BMD (r2 = 0.64). Of the QCT parameters, the best correlation with strength was obtained using the product of trabecular BMD and CSA (r2 = 0.61). For vertebral ultimate stress, however, the best correlation was observed with QCT-measured trabecular BMD (r2 = 0.51); the correlation with DXA-determined BMD was slightly poorer (r2 = 0.44). Calcaneal ultrasound correlated only weakly with both ultimate load and stress with correlation coefficients (r2) of 0.10–0.17, as did calcaneal BMD (r2 = 0.18). Both spinal DXA and spinal QCT were significantly (p < 0.001) better predictors of L3 ultimate load and stress than were either calcaneal ultrasound or calcaneal DXA. Multiple regression analysis revealed that calcaneal ultrasound did not significantly improve the predictive ability of either DXA or QCT for L3 ultimate load or stress. Calcaneal DXA BMD, bone velocity, and BUA correlated well with each other (r2 = 0.67–0.76), but were only modestly correlated with the DXA and QCT measurements of the vertebra. These data indicate that spinal DXA and spinal QCT provide comparable prediction of vertebral strength, but that a substantial proportion (typically 40%) of the variability in vertebral strength is unaccounted for by BMD measurements. Ultrasonic measurements at the calcaneus are poor predictors of vertebral strength in vitro, and ultrasound does not add predictive information independently of BMD. These findings contrast with emerging clinical data, suggesting that calcaneal ultrasound may be a valuable predictor of vertebral fracture risk in vivo. A possible explanation for this apparent discrepancy between in vivo and in vitro findings could be that current clinical ultrasound measurements at the calcaneus reflect factors that are related to fracture risk but not associated with bone fragility.


THE HUMAN VERTEBRA is particularly susceptible to osteoporotic fracture. To allow targeting of preventive care, there is an increasing need to predict reliably bone strength and fracture risk in vivo. Dual X-ray absorptiometry (DXA) and quantitative computed tomography (QCT) are commonly used to assess the bone mineral status of the spine and to identify individuals at high risk for osteoporotic fracture. Previous in vitro studies have demonstrated significant relationships between the ultimate compressive strength of the vertebral body and BMD as measured by DXA and/or QCT, with correlations ranging from r2 = 0.59–0.74 for DXA,1,2 and r2 = 0.08–0.61 for QCT.1–7 Most of the previous studies have used QCT alone,3–7 and only two studies1,2 utilized both DXA and QCT.

The mechanical properties of cancellous bone are determined principally by the apparent bone density, the material properties of the solid trabeculae, and the structure of the trabecular framework.8 Bone mineral density (BMD) measurements effectively measure a combination of the first two factors (with an additional bone geometry effect in the case of DXA), but do not provide any information on trabecular structure. In this regard, ultrasound has been promoted as an attractive alternative technique complementary to conventional approaches using ionizing radiation, since it is believed that ultrasound is able to provide information on the mechanical and structural properties of cancellous bone.9 In particular, it has been suggested that ultrasonic velocity is related to both elasticity and density and broadband ultrasound attenuation (BUA) to trabecular structure and density.9 However, on the one hand, despite a number of fundamental studies, the underlying nature of such relationships is still somewhat unclear.10–14 On the other hand, there is growing evidence that ultrasonic measurements can be useful empirical predictors of bone mechanical properties in vitro.15–17

Technical problems preclude in vivo ultrasonic measurement of the human vertebra, and most of the existing clinical systems measure the calcaneus. In common with the vertebra, the calcaneus is load-bearing and consists primarily of cancellous bone, but it has the additional benefit of being readily accessible for through transmission ultrasonic measurements. Differences in trabecular structure at the two sites18 may affect the relevance of ultrasound measurements at the calcaneus for fracture prediction of the spine. It is by no means clear that measurement of the physical properties (density, structure, or elasticity) of the calcaneus will be predictive for those same properties at the vertebra. Initial clinical reports have suggested that calcaneal ultrasound does predict vertebral fracture in vivo,19 but no in vitro data have yet been presented.

Thus, the aim of this study was to compare the performance of spinal DXA and spinal QCT with calcaneal ultrasound measurements in predicting vertebral strength in a large set of cadaveric specimens.


Matched pairs of the third lumbar vertebral body (L3) without posterior elements and the right calcaneus were removed from 70 cadavers at autopsy within the program of the EC BIOMED1 Concerted Action “Assessment of Bone Quality in Osteoporosis.”20 The surrounding soft tissues were removed, and the specimens were stored at −20°C. High resolution contact radiographs of the vertebrae and the calcanae were taken with a Faxitron X-ray unit (Hewlett-Packed, McMinnville, OR, U.S.A.) to screen for pre-existing fractures and metastases, scoliosis, and other skeletal diseases.

Areal BMD (g/cm2) of the vertebral body and calcaneus was determined with a Hologic 1000-W DXA scanner (Hologic Inc., Waltham, MA, U.S.A.). The reproducibility of this instrument in our hands is typically 0.36% (coefficient of variation, based on 590 daily scans of the manufacturer's phantom). The vertebral body was scanned in the posterior-anterior projection on a 3.5 cm thick Plexiglas sheet to stimulate soft tissue. The calcaneus was scanned laterally, again on Plexiglas, and BMD was measured in a 3 cm wide vertical rectangular region in the posterior part of the calcaneus corresponding to the ultrasonic measurement site.21

QCT measurements were performed with a GE-9800 CT scanner (General Electric Co., Milwaukee, WI, U.S.A.), using scan parameters of 80 Kv, 70 mA, and 2 s. One 10 mm thick axial slice was made in the center of the vertebral body, and the GE phantom was scanned simultaneously for BMD calibration. Trabecular BMD (mg/cc) was determined on the axial slices using an ellipsoidal region of interest as used in clinical applications (Fig. 1A). Cortical BMD (mg/cc) and cortical area (mm2) of the cortical rim were evaluated with the IMAGE MASK software, which highlights and measures all the pixels within user-specified thresholds. A 350 Hu threshold was used, with all pixels higher than 350 Hu being regarded as cortical bone (Fig. 1B).22,23 Using the same software, the total cross-sectional area (CSA, cm2) of the vertebra was measured using a threshold of 50 Hu to separate bone from surrounding soft tissue. The reproducibility of such QCT measurements is typically 2–4%.24

Figure FIG. 1.

Quantitative computed tomography of the vertebra. (A) Ellipsoidal region of interest used for trabecular BMD measurement. (B) Thresholding of cortical bone.

Mechanical testing of L3 was performed with a hydraulic materials testing device. Prior to testing, the vertebrae were thawed to room temperature and kept moist throughout testing. The specimen was positioned between the loading plates of the materials testing instrument, and stiff rubber sheets were inserted between the specimen and the loading plate to simulate the intervertebral disc (Fig. 2). The specimens then were compressed at approximately 8 mm/s until failure. Data from the compressive test were recorded in the form of a load versus deformation curve, and vertebral ultimate load (Newtons, N) was defined as the initial maximal in load achieved after the onset of plastic deformation. In addition, ultimate stress (MPa) was obtained by dividing the ultimate load by the total cross-sectional area (CSA) of the vertebrae measured by QCT as described earlier. The midvertebral cross-sectional area was used because this is the minimum area across which the load is distributed.2 Five vertebrae with pre-existing fractures were excluded, and a further three vertebrae were excluded because of technical problems during testing, leaving a total of 62 specimens in the study.

Figure FIG. 2.

Configuration of the mechanical test on the vertebra.

Ultrasonic measurements of the calcaneus were made with a laboratory-based system that has been described in detail elsewhere.25 Two 1/2-in diameter 1 MHz broadband flat-focus transducers (Harisonics I3-0108-S, Staveley NDT, Slough, U.K.) were mounted in a water bath 60 mm apart. One transducer was driven by an electrical pulse generator to produce broadband ultrasonic pulses that were detected by the second transducer and subsequently digitized by digital storage oscilloscope interfaced to a personal computer. Bone velocity (BV, m/s) was derived from transit times using the standard substitution formula.26 Transit time measurements were based on the first zero-crossing criterion to define pulse arrival.25 Broadband ultrasound attenuation (BUA, dB/MHz) was determined from a least-squares linear regression of attenuation against frequency from 0.2–0.6 MHz.27 Prior to ultrasonic measurement, a site was marked in ink at a standardized position on the medial side of calcaneus as described elsewhere,25 allowing the ultrasonic transducers to be aligned with a clearly defined landmark. Calcaneal thickness was determined as a mean of five measurements made at the marked site with vernier callipers. Calcanae were thawed and degassed overnight in water under a partial vacuum before ultrasound measurements. Reproducibility with this system was determined by performing 10 repeat measurements on one calcaneus, with the transducers being removed and repositioned between the measurements, yielding a coefficient of variation (standard deviation/mean × 100%) of 0.5% for BV and 2.5% for BUA.

Statistical comparisons between groups were assessed using Student's t-test. Simple linear regression analysis was used to assess the relationships between two variables. The standard error of the estimate (SEE) for the regression was calculated as the root of the sum of squares of the residuals divided by the sample size, a measure of the scatter about the regression line.28 For purposes of comparison, SEE was expressed in percentage terms as 100 × SEE/mean. Differences between two correlation coefficients were assessed using Fisher's Z-transformation.28 Stepwise multiple regression analysis was performed to test whether combinations of related variables improved the correlations with vertebral strength. Probabilities were considered significant at the 5% level.


Table 1 gives summary statistics for the studied variables. Although there was no statistically significant difference in the mean ages of males and females, all other parameters were significantly higher in males compared with females. The mean vertebral ultimate load was 57% higher in males, and ultimate stress was approximately 37% higher.

original image

Table 2 gives the bivariate correlations between L3 strength (both ultimate load and ultimate stress) and the ultrasound, DXA and QCT measurements. For ultimate load, the strongest correlation was observed with DXA-determined vertebral BMD (r2 = 0.64; Fig. 3A). Of the QCT parameters, the product of vertebral trabecular BMD and CSA gave the best correlation with strength (r2 = 0.61; Fig. 3B). For L3 ultimate stress, however, the correlation with DXA-determined vertebral BMD was moderate (r2 = 0.44; Fig. 3C), and the best correlation was observed with QCT-determined trabecular BMD (r2 = 0.51; Fig. 3D). In contrast, the correlations between calcaneal ultrasonic measurements and vertebral strength (both ultimate load and ultimate stress) were significantly poorer (p < 0.001) than those described above for spinal BMD (r2 = 0.10–0.17; Figs. 4A, 4B, 4C, 4D, Table 2). Similarly, calcaneal DXA measurements were relatively poor predictors of vertebral strength (Table 2).

original image
Figure FIG. 4.

Relationships between vertebral strength and calcaneal ultrasound: (A) ultimate load and bone velocity; (B) ultimate load and BUA; (C) ultimate stress and bone velocity; and (D) ultimate stress and BUA. (Open circles for males, solid circles for females; SEE, standard error of the estimate.)

Figure FIG. 3.

Relationships between vertebral strength and vertebral BMD: (A) ultimate load and DXA BMD; (B) ultimate load and (QCT trabecular BMD × cross-sectional area); (C) ultimate stress and DXA BMD; and (D) ultimate stress and QCT trabecular BMD. (Open circles for male, solid circles for females; SEE, standard error of the estimate.)

Stepwise multiple regression analysis was performed to test whether combinations of different measurements improved the prediction of vertebral strength. In multiple regression models, including all of the measured parameters, only DXA vertebral BMD and the product of QCT-determined trabecular BMD and CSA were significant predictors of L3 ultimate load, with a resulting adjusted r2 of 0.65. When ultimate load was replaced by ultimate stress in the model, only vertebral DXA BMD and QCT-determined trabecular BMD were associated with L3 ultimate stress, giving an adjusted r2 of 0.54. Ultrasonic measurements at the calcaneus did not significantly improve the predictive ability of either DXA or QCT for L3 ultimate load or ultimate stress. Combining ultrasonic velocity with BUA did not significantly improve the prediction of strength compared with the performance of a single ultrasonic parameter alone.

DXA BMD, bone velocity, and BUA of the calcaneus correlated well with each other (r2 = 0.67–0.76), but were only modestly correlated with the DXA and QCT measurements of the vertebra (Table 3). Combining calcaneal DXA BMD with calcaneal ultrasound in multiple regression models did not significantly improve the correlations with vertebral strength.

original image


Our data indicate that spinal DXA and spinal QCT have a similar ability to predict the strength of the vertebral body in vitro. Trabecular BMD as measured by QCT explained approximately 44% of the variance in vertebral ultimate load, but when cross-sectional area was incorporated, the predicted variability increased to 61%. These findings are consistent with earlier studies.2,4,6 This indicates that the cross-sectional area of the vertebral body, a parameter easily measured using CT, provides useful information in addition to trabecular BMD. In line with previous work,2 DXA was better than QCT in predicting ultimate load, but QCT was better than DXA in predicting ultimate stress (although these differences were not statistically significant). These observations are explained by the fact that ultimate load depends on both bone material properties and bone size, whereas ultimate stress is, in theory, independent of bone size. DXA measures an areal bone density which integrates contributions from both cortical and cancellous bone and which is partially related to bone size, and this tends to improve the statistical relationships between DXA BMD and ultimate load. However, QCT measures a volumetric BMD of cancellous bone alone and is therefore more strongly related to intrinsic bone material properties as reflected by ultimate stress. It should be noted that approximately 40% of the variance in vertebral strength is unaccounted for by the parameters that have been studied. This indicates that other factors may play a substantial role in determining the strength of the lumbar vertebra. Trabecular structure may contribute to vertebral strength,29–31 but it is likely that the overall vertebral geometry is an even more important determinant. More sophisticated approaches to combining BMD and geometrical measurements, such as in finite element models,32,33 could be expected to improve the prediction of vertebral strength.

To our knowledge, this is the first study to investigate the relationship between calcaneal ultrasound and vertebral strength in vitro. Ultrasonic measurements at the calcaneus were only weakly correlated with vertebral strength, and multiple regression analysis revealed that ultrasound did not add to the ability of DXA or QCT to predict vertebral strength in vitro. One explanation for the poor performance of ultrasound in this context could be that the physical properties of the calcaneus are not closely correlated with those of the vertebra. Both the calcaneus and the vertebral body consist primarily of trabecular bone, but both sites are subject to different patterns of mechanical loading. The calcaneus lies at a distal appendicular site and is subjected to a complex loading regime. In contrast, the L3 vertebra lies in the axial skeleton and experiences predominantly compressive loading. Previous studies have shown differences in trabecular structure18 and bone composition34 between the calcaneus and the vertebra and only a moderate correlation (r2 = 0.59) between BMD as measured by DXA at the two sites.35 In the present study, an even weaker association (r2 = 0.30) between calcaneal and vertebral DXA BMD was observed.

Another relevant factor could be that the currently used ultrasonic measurements may not be optimal in terms of acquiring information about bone mechanical properties. Previous studies have demonstrated that ultrasonic measurements correlate (r2 = 0.56–0.93) with mechanical properties when the ultrasonic and mechanical measurements are performed on the same specimens along the same measurement axis, for bovine bone,15,36 human vertebral bone,12 and human calcaneal bone.16,17 However, these relationships may largely represent the underlying influence of bone apparent density as a common factor determining both ultrasonic and mechanical properties. To date, the expected theoretical relationships between mechanical properties and ultrasound velocity have been confirmed experimentally only in small defatted specimens of well-defined geometry and at much lower ultrasonic frequencies than used currently in vivo.36 Furthermore, it should be noted that there is growing evidence for some important underlying errors in the current ultrasonic methodology used in bone.25,37,38

Bouxsein and colleagues39 demonstrated that ultrasonic measurements at the calcaneus were less well correlated with femoral strength than were DXA BMD measurements, and that ultrasound did not add predictive information independently of BMD. We confirmed and extended these findings using femoral specimens from the same cadavers as the present work.21,40 The present study indicates that calcaneal ultrasound measurements do not add predictive ability when combined with DXA or QCT. These in vitro results stand in marked contrast to in vivo studies which suggest that calcaneal ultrasound is a valuable indicator of fracture risk at both femur41 and vertebra,19 with a predictive ability similar to and essentially independent of BMD. This apparent contradiction needs to be addressed urgently. One explanatory factor could be that BMD measurements are inherently more accurate in vitro than in vivo, due to the improved positioning and the absence of any soft tissue variations. The performance of BMD in vivo may be impaired such that the predictive ability of BMD becomes comparable to that of calcaneal ultrasound. Another possibility is that ultrasound measurements in vivo provide information that is not related to bone fragility but still reflects fracture risk. For example, it is possible that a number of factors such as soft tissue thickness, tissue temperatures, marrow fat composition, and blood flow may affect in vivo ultrasound measurements, and these factors could in turn be related, either directly or indirectly, to fracture risk at both the femur and spine. Indeed, a very recent study42 has shown that the presence of ankle edema caused a mean reduction of 23.9 m/s in velocity and of 5.5 dB/MHz in BUA.

The current study has several limitations. First, the subjects recruited may not be representative of the overall elderly population, since they were obtained at postmortem and did not represent a genuinely randomized sample. Second, axial compression tests cannot fully simulate the mechanical conditions of a real-life fracture event. We tested isolated vertebral bodies with a rubber sheet simulating the intervertebral disc. The healthy intervertebral disc has a fluid nucleus pulposus which helps to distribute forces evenly over the vertebral endplates, whereas the rubber sheets had only a limited ability to deform and distribute loads. Also, a combination of compressive, shearing, and torsional forces can be expected in vivo, whereas we applied compressive forces only. As a consequence, the fractures we observed may not have been representative of those seen in cases of spinal osteoporosis. Finally, it should be noted that densitometry and mechanical testing were performed on the vertebral body alone as opposed to the intact vertebra, including posterior elements, and therefore the measurements may not be fully representative of the in vivo situation. However, previous work has demonstrated that while the posterior elements contain approximately 50% of the bone mineral content (BMC) of the vertebra, BMC measurements with and without the posterior elements are highly intercorrelated.43

In conclusion, this study demonstrates that spinal DXA and spinal QCT have a similar ability to predict vertebral strength in vitro. However, calcaneal ultrasound is a poor predictor of vertebral strength in vitro, and ultrasound does not add predictive information independently of BMD measurements by DXA or QCT at the spine. These findings contrast markedly with emerging clinical data, suggesting a promising role for calcaneal ultrasound in vertebral fracture risk prediction. We suggest that further investigation of this apparent contradiction is essential.


This work was performed within the framework of the EC BIOMED1 Concerted Action Assessment of Quality of Bone in Osteoporosis, contract no. BMH1-CT92–0296. The authors thank Mr. J. Nijs and Mr. H. Borghs for the DXA measurements, and Mrs. J. Cartois and Ms. A. Vandereijcken for expert secretarial help.