Bone Strength at Clinically Relevant Sites Displays Substantial Heterogeneity and Is Best Predicted From Site-Specific Bone Densitometry



In this study we test the hypotheses that mechanical bone strength in elderly individuals displays substantial heterogeneity among clinically relevant skeletal sites, that ex situ dual-energy X-ray absorptiometry (DXA) provides better estimates of bone strength than in situ DXA, but that a site-specific approach of bone densitometry is nevertheless superior for optimal prediction of bone failure under in situ conditions. DXA measurements were obtained of the lumbar spine, the left femur, the left radius, and the total body in 110 human cadavers (age, 80.6 ± 10.5 years; 72 female, 38 male), including the skin and soft tissues. The bones were then excised, spinal and femoral DXA being repeated ex situ. Mechanical failure tests were performed on thoracic vertebra 10 and lumbar vertebra 3 (compressive loading of a functional unit), the left and right femur (side impact and vertical loading configuration), and the left and right distal radius (fall configuration, axial compression, and 3-point-bending). The failure loads displayed only very moderate correlation among sites (r = 0.39 to 0.63). Ex situ DXA displayed slightly higher correlations with failure loads compared with those of in situ DXA, but the differences were not significant and relatively small. Under in situ conditions, DXA predicted 50-60% of the variability in bone failure loads at identical (or closely adjacent) sites, but only around 20-35% at distant sites, advocating a site-specific approach of densitometry. These data suggest that mechanical competence in the elderly is governed by strong regional variation, and that its loss in osteoporosis may not represent a strictly systemic process.


Osteoporotic fractures represent a major health care problem relevant to all geographic areas and populations around the world. They severely affect the morbidity, quality of life, and even mortality of elderly individuals.(1, 2) The costs involved have been estimated at US$14 billion per year in the United States.(3) Optimal noninvasive prediction of bone strength and fracture risk is therefore an important goal in medical diagnostics.

Clinical and experimental studies have shown a high association between bone mineral status, on the one hand, and fracture risk(4) or bone strength,(5) on the other. Dual-energy X-ray absorptiometry (DXA) is currently considered the gold standard for noninvasive bone densitometry.(4) However, it is controversial, whether measurements at one site can reliably predict fracture risk at another location, and which anatomic site is optimal for predicting fracture risk at all relevant sites.(6–8)

Osteoporosis is generally considered a systemic disease, although bone density among different anatomic locations has been shown to display substantial heterogeneity.(9–13) Cheng et al.(9) reported in an experimental study ex situ (without soft tissues) that femoral failure can best be predicted from femoral DXA and vertebral failure from spinal DXA. In contrast, Hassager et al.(14) claimed that substantial errors are involved in spinal DXA under in situ conditions,(15) which are attributable to a nonhomogeneous distribution of fat tissue around the spine, extraskeletal calcification, and the inclusion of the posterior spinal elements. They therefore recommended to estimate spinal fracture risk from measurements at the distal radius, where these errors are smallest.

Moreover, spinal DXA is performed clinically in the lumbar spine, but fractures are most frequent in the lower thoracic spine or thoracolumbar junction.(16, 17) DXA of the lumbar spine may therefore be less predictive of thoracic failure loads than suggested by previous experimental studies in identical vertebrae.(9) Similar considerations also apply to the femur because clinical predictions for both femora are derived from unilateral DXA measurements, although side differences of up to 20% have been reported.(18) Yet, the most important problem with a site-specific approach of estimating bone strength is that the location of potential future fracture is unknown a priori, and that it is problematic to screen all sites in one patient. It is thus of considerable interest to assess which measurement locations (or which combination of measurements) are optimal for predicting bone strength at clinically relevant skeletal sites.

The objective of this study was to experimentally determine bone failure loads at the thoracic and lumbar spine, the proximal femur, and distal radius under different loading configurations in a large sample of elderly human cadavers, and to analyze the predictive ability of bone densitometry under in situ conditions (with normal soft tissues), including an analysis of the total body, spine, proximal femur, and radius.

The study was designed to test the following specific hypotheses:

  • (1)Bone strength in elderly individuals (and not only bone mass and density) is site dependent and displays substantial heterogeneity throughout the skeleton.
  • (2)Ex situ DXA provides significantly better estimates of bone strength than in situ DXA because of accuracy errors resulting from the surrounding soft tissues.
  • (3)Site-specific bone densitometry is superior to a non-site-specific approach, although accuracy errors of DXA are spatially variable (e.g., greatest in the spine, smallest in the radius), although relevant errors result from osteoarthritic changes, and although bone properties are variable between spinal segments and between contralateral femora.
  • (4)Bone densitometry from non-site-specific locations can, nevertheless, add significant, independent information to site-specific measurements.
  • (5)The best “global” prediction of bone strength at relevant sites is achieved by determining bone mineral status of the total body rather than by regional analysis.


Study sample

One hundred ten formalin-fixed cadavers with intact skin and soft tissues were examined from a pool of 140 specimens included in a course of macroscopic dissection. Thirty-eight specimens were male (age 76 ± 12 years [mean ± SD]) and 72 female (age 83 ± 9 years). The donors had dedicated their body to the institute several years before death and we therefore assume that they constitute a representative selection of the elderly population resident in Bavaria. Bone biopsy specimens were taken from the left iliac crest (at the site of clinical transiliac biopsy specimens) and prepared for routine histomorphometric assessment (embedding in methylmethacrylate, preparation of 5-μm sections, staining with Goldner, toluidine blue, and von Kossa). Specimens with signs of malignancy (3 cases) and bone disease other than osteopenia (e.g., suspected renal osteopathy, 7 cases) were discarded from the study. Specimens with bilateral hip arthroplasty, bilateral femoral osteosynthesis, or fractures of thoracic vertebra 10 (see below) were also discarded.

Bone densitometry by DXA

The bone mineral content (BMC in grams) and the projected, areal bone mineral density (BMD in grams per square centimeter) of the total skeleton (total body scan), lumbar vertebra number 3 (L3), the left proximal femur (subregions: neck, ward, trochanter, shaft, and total femur), and the left radius (distal 33%) were determined under in situ conditions with a DPX-L scanner (Lunar Corp., Madison, WI, USA), the skin and soft tissues of the cadavers being fully intact. Spinal DXA was performed in an anteroposterior (AP) direction, but not in lateral projection, because the cadavers could not be positioned in the required sideways decubitus position. In 8 cadavers, the spinal DXA measurements were discarded because fractures of L3 were recorded in spinal X-rays (see below). The precision of DXA was tested by measuring a subset of 14 randomly selected subjects (7 women and 7 men) at four different occasions (different days with repositioning of the specimens), and by computing the root mean square (RMS) average CV (CV%) of repeated analyses as a measure of the technical precision.

The spines were then excised from thoracic vertebra 4 (T4) to L5 as well as the left and right femora at midshaft, the left radius, and the right forearm including tendons, ligaments, and the intact hand. Except for the right forearm, the specimens were cleaned of muscles and soft tissues. To exclude prior fracture or other local bone disorders, the femora and radii were radiographed with a Faxitron Cabinet X-ray system (Model 43885A, Faxitron X-ray systems; settings 40-85 kV, 2 mA, time = 120 s; Hewlett-Packard, McMinnville, OR, USA), using an 18 cm × 24 cm film (Agfa Structurix D7DW; Agfa, Leverkusen, Germany). In the spines, four films (two in AP direction; two lateral) were obtained using a Siemens X-ray system (Polyphos 30 M; settings 42-50 kV, 16-20 mA; Siemens, Erlangen, Germany) and a 35 cm × 42 cm film (SR-H, Konica Medical Film; Konica, Hohenbrunn, Germany). The films were evaluated by an experienced skeletal radiologist to exclude bones with fractures.

In 62 femora and lumbar segments DXA was repeated under ex situ conditions within a homogeneous aqueous environment and in a standardized position. The proximal femur was scanned in the AP direction, and the spine in the lateral direction. No ex situ scans were obtained in the radius because soft tissue error is less important here than at the other sites.(15)

Mechanical testing

The vertebral bodies T10 (n = 110) and L3 (n = 102) were tested to failure as a functional spinal unit with intact ligaments and intervertebral discs in an axial compressive loading configuration. Adjacent vertebral bodies were embedded parallel to the endplates (three-segment method)(19, 20) and tested to failure without posterior elements, using a uniaxial material testing machine (Zwick 1445, Ulm, Germany). The load to failure was determined at a rate of 6.5 mm/s, identifying the first peak, followed by a drop of >10% from the load-displacement curve. To ensure fracture, the segments were compressed to 25% of their original height.

One of both proximal femora in each individual was tested by simulating a sideways fall on the greater trochanter. This side impact configuration was adapted from Courtney et al.(21, 22) and Bouxsein et al.,(23) the femoral head and shaft being placed on the testing table (shaft 10° from horizontal; neck in 15° internal rotation), and the load being applied to the greater trochanter. The load was applied to the greater trochanter through a pad, which simulated the soft tissue cover. The contralateral femur was tested in a vertical loading configuration, by orientating the shaft vertically and applying a load to the femoral head parallel to the shaft.(24, 25) Pairs of femora were sorted according to age, the left and right side being assigned in alternating order to the two loading configurations. In cases in which only one femur was available (unilateral hip arthroplasty or osteosynthesis; n = 11), the specimens were assigned in alternating order to both tests, so that 104 were tested in a side impact and 105 in vertical loading. The failure load was defined as the peak of the load-displacement curve, applying a loading rate of 6.5 mm/s. The femora were X-rayed after testing and the fracture patterns were classified. In the side impact configuration, 48 specimens displayed cervical fractures (27 subcapital and 21 transcervical; 10 type I [<30°], 17 type II [30–70°], and 21 type III [>70°] according to the Pauwels' classification). Forty specimens showed trochanteric fractures (18 pertrochanteric, 9 intertrochanteric), 13 displaying crush fractures of the trochanter. Specimens that revealed clinically untypical fractures (crush fractures of the trochanter [n = 13] or fractures of the head [n = 10] or shaft [n = 6]) were treated separately. In vertical loading, 95 specimens displayed fractures through the femoral neck and 10 through the greater trochanter. In this configuration, only cervical fractures were considered in further analyses.

On the left radius, a 3-point-bending test was performed, 33% proximal to the wrist. Five specimens could not be tested because of osteosynthetic material or because of being suspected to have sustained a previous fracture. To test identical anatomic regions in all specimens (independent of the individual bone size), the shaft was fixed at 17 and 50% of the individual bone length, using two elements that were able to rotate and translate freely. From the failure load at 6.5 mm/s and from the individual moment arms, the failure moment (Nm) was determined. Five specimens were excluded because of fracture at a site other than the 33% location. From the distal end of the same radius, a 10-mm plano-parallel axial slice was obtained adjacent to the joint surface, using a diamond band saw (Exact Trennschleifsystem, Norderstedt, Germany). An axial compression test was performed using a 0.8 mm/s loading rate, recording the initial peak (failure load) of the load-displacement curve with a subsequent drop of >10%. On the right forearm, a more complex failure test was adopted from Myers et al.(26) and Spadaro et al.,(27) simulating a fall on the outstretched hand. This test was performed by means of an Instron testing machine (Model 4302; Instron, High Wycombe, Bucks, UK). The ulna and radius were embedded in plaster (True Rock; Benzer Dental AG, Zurich, Switzerland) with the hand being positioned in 70° dorsiflexion, 10° radial abduction, and with the forearm in pronation. The load was applied to the hand through a wedge, the loading rate being 3.3 mm/s. Failure loads were recorded as the peaks of the load-displacement curves and were considered for further analyses only if a typical Colles' fracture was produced (n = 85). The type of fracture was determined from two X-rays (two planes) and from surgical preparation of the fracture zone.

Statistical analysis

The descriptive data of failure loads and DXA are given as the means (± SEM) and range of values. A paired t-test was used to identify systematic differences in failure loads among different sites. Correlation coefficients were computed by linear regression analysis and were considered significant at a 1% level. Fisher z transformation was used to assess whether certain coefficients were significantly higher than others. Stepwise multiple regression models (forward mode) were used to determine whether non-site-specific densitometry added significant, independent information to site-specific DXA. To obtain a quantitative measure of “global” failure strength, we computed the percentage deviation of the individual failure loads from the mean at T10, the femur (side impact test), and the radius (fall configuration). The mean of these three percentage deviations was then correlated with bone densitometry at different sites.


The bone failure loads and the number of specimens tested after applying exclusion criteria are displayed in Table 1. The correlation between mechanical failure loads of the vertebral bodies, proximal femur, and distal radius sites were statistically significant (p < 0.01), but displayed only moderate correlation coefficients (Figs. 1A-1C; Table 2). Different test configurations at the same skeletal element produced higher coefficients than those at different anatomic sites (Table 2). Quantitative results for the in situ DXA measurements and their precision are given in Table 3. The precision errors were substantially lower than the intersubject variability.

Table Table 1. Descriptive Statistics of Mechanical Failure Loads at Different Skeletal Sites and Different Loading Conditions
original image
Table Table 2. Correlation (r) Among Mechanical Failure Loads at Different Skeletal Sites and Loading Conditions
original image
Table Table 3. Bone Densitometry Values (In Situ DXA) in Men and Women, Interindividual Variability for Entire Sample, and Measurement Precision
original image
Figure FIG. 1..

(A-C) Correlations between failure loads at different sites and (D-F) correlation between failure loads and site-specific bone densitometry. (A) Failure load of thoracic vertebral body 10 versus femur in side impact configuration. (B) Failure load of thoracic vertebral body 10 versus distal radius in fall configuration. (C) Failure load of femur in side impact configuration versus distal radius in fall configuration. (D) Failure load of thoracic vertebral body 10 versus in situ DXA (BMC) of lumbar vertebra 3. (E) Failure load of femur in side impact configuration versus in situ DXA (BMD) of femoral neck. (F) Failure load of distal radius in fall configuration versus in situ DXA (BMC) of distal radius.

Site-specific in situ DXA displayed a significantly (p < 0.05) higher association with bone failure loads (Figs. 1D-1F) than non-site-specific measurements in all cases (Table 4). Total body analysis showed a trend for lower coefficients, although the difference from site-specific analysis reached significance only in the distal radius.

Table Table 4. Correlation Between In Situ Bone Densitometry (Including Skin and Soft Tissues) and Mechanical Failure Loads at Different Sites
original image

When comparing the predictive ability of in situ DXA versus ex situ DXA, the correlation coefficients were generally higher for ex situ conditions at the same site. However, the difference in the coefficients was only statistically significant (p < 0.05) for failure in the lumbar spine with the lateral ex situ BMC of L3 (r = 0.86) versus the AP in situ BMC of L3 (r = 0.75). Ex situ DXA did not generally display a higher association with bone strength at other sites, compared with in situ DXA (Table 5).

Table Table 5. Correlation of Mechanical Failure Loads at Various Sites for In Situ Versus Ex Situ DXA, and With DXA at the Same (Left) Side Only
original image

When combining in situ densitometric data from several sites in multiple regression models, thoracic spinal failure loads were best predicted by L3 BMC, total body BMD, and femoral BMD as independent variables (multiple r = 0.73, adjusted r2 = 0.47). Failure loads of L3 were best predicted by L3 BMC and total body BMC (multiple r = 0.71, adjusted r2 = 0.50; Table 4). In the radius and femur, densitometric data from other sites did not add independent, predictive information, except for 3-point-bending of the radius (Table 4).

The highest correlation with estimated “global” failure strength was by total body BMC (r = 0.75) and radial BMC (r = 0.76), but prediction from spinal BMC (r = 0.71) and femoral BMC (r = 0.62 to 0.69) was only marginally lower.


We selected an experimental (biomechanical) study design because bone strength cannot be objectively determined in vivo, and because the effect of soft tissues cannot be explicitly analyzed under clinical conditions. The longitudinal study of association of fracture rates at different sites(28) is confounded by extraskeletal factors, such as a higher propensity to fall in individuals with low visual acuity, neuromuscular deficits, cardiovascular insufficiency, and so forth.

The hypotheses tested are answered as follows:

  • (1)Mechanical failure loads display significant, but only moderate correlations among different sites. This suggests that mechanical competence in the elderly is governed by substantial regional variation and that loss of bone strength in osteoporosis may not represent a strictly systemic process.
  • (2)Ex situ DXA displays slightly higher associations with mechanical failure loads at the same site compared with in situ analyses, but the difference is significant only for lateral ex situ versus AP in situ scans in the lumbar spine. Ex situ DXA appears to involve little advantage when estimating bone strength at distant skeletal sites.
  • (3)In situ DXA analysis at the site of interest (spine, femur, radius) is more predictive of bone strength than that at other sites, even when accounting for the spatially variable accuracy errors. This also applies when DXA and failure loads are determined either in another segment of the spine or in the contralateral femur.
  • (4)Non-site-specific measurements add independent predictive information to spinal strength and radial strength in 3-point-bending, but the gain is relatively small. No significant contributions of non-site-specific data are found in the proximal femur.
  • (5)Although total body analysis provides the most balanced estimate, it is not significantly superior to regional analysis in estimating global bone strength. No particular site displays obvious advantages in estimating the failure loads throughout all sites.

Potential limitations of the present study include the lack of detailed medical history and the use of fixed cadavers. In contrast to specimens obtained from pathological dissection, however, this sample does not include a preselection of highly pathological cases. By obtaining X-rays of all regions of interest before mechanical testing and by performing a histomorphometric analysis of the iliac crest, we were able to apply exclusion criteria similar to those of clinical studies. Edmonston et al.(29) and Augat et al.(24) concluded from a review of literature and their own experiments that mechanical strength of entire bones is only minimally affected by fixation, and that the correlation between BMC and mechanical strength is unchanged. We have shown that prolonged (10 month) formalin fixation has no significant effect on DXA, even under in situ conditions with surrounding soft tissues.(30) In the current study, we find the DXA measurements are in the range of those reported in vivo,(31) and that failure loads are similar to those reported in fresh specimens of comparable age.(23) The correlations between failure loads and ex situ DXA were also in the range of those in fresh specimens.(5) We therefore believe that the results of this study are not critically affected by fixation.

Only few studies have previously investigated the association between failure loads at different skeletal locations in the same subjects.(9, 24, 32) Our current study extends these analyses by reporting the association between all clinically relevant skeletal sites under various testing configurations and in a large sample. The relatively low correlation of bone strength at different anatomic sites may question the concept that osteoporotic loss of bone strength is a strictly systemic process. It is currently unclear which factors lead to preferential loss of bone strength at certain anatomic locations but not in others. A potential explanation may be a regional variation of loss of muscle strength, to which the bones adapt. This pattern may depend on the individual lifestyle and vary from person to person. Another potential explanation is that not the loss, but the acquisition, of bone strength during maturation is site dependent, and that the high degree of heterogeneity is thus not specific to elderly subjects.

In the spine, other authors(9, 19, 20, 33) reported similar or lower correlations between ex situ DXA and strength. Ebbesen et al.(34) reported a somewhat higher association, but tested vertebrae without endplates, eliminating variability originating from the intervertebral discs. In our study, ex situ lateral DXA displayed a significantly higher correlation with failure of L3 compared with in situ AP DXA, but surprisingly the association of both types of measurement was equal for T10.

In the femur, the relationship between ex situ DXA and strength in the vertical loading configuration was similar to that of previous studies.(35–37) In the side impact test, we found similar correlations to those observed by some investigators,(21, 38) but lower correlations than those observed by other authors.(9, 23) We observed very high correlations for trochanteric, but not for cervical fractures. Previous studies comprised small sample sizes and CIs were therefore large. The fracture patterns observed in our study (e.g., distribution between cervical and trochanteric fractures) resembled those observed clinically, and the proportion of nontypical fractures (25%) was similar to that obtained in other experimental studies.(23) Under in situ conditions, we found the highest association of femoral failure loads with cervical BMD, irrespective of whether failure occurred through the neck or the trochanter. Under ex situ conditions, however, the highest association was with trochanteric BMD, as reported in previous ex situ studies. This indicates that soft tissue accuracy errors of DXA(15) may have a higher impact on trochanteric than on cervical measurements.

In the radius, we observed the highest correlation between in situ radial BMD and failure in 3-point-bending (r = 0.89). This is not surprising, in view of the controlled mechanical conditions under which failure occurs at this purely cortical location. The association between radial BMC and failure in a fall configuration was similar to that reported by Myers et al.(26) under ex situ conditions, confirming that accuracy errors from soft tissues(15) are not a critical problem at the distal radius.

With respect to correlations between failure load versus BMC, and failure load versus BMD, it must be borne in mind that bone size is a potential confounder of areal BMD (grams per square centimeter).(39) From an engineering point of view, bone mass (BMC) is more highly correlated with failure load than with volumetric density,(5) and this has also been shown empirically both in the proximal femur(39) and in the lumbar spine.(40) The reason for the areal BMD to be highly correlated with failure load (as documented in this and other experimental studies) is that some information on bone size (depth) is retained in the measurements,(39) and that, in contrast to BMC, determination of areal BMD avoids some of the problems involved in accurately defining anatomical borders of certain regions of interest (for instance in the femoral neck). The complex interaction of volumetric density and bone size in determining areal BMD, however, must be borne in mind when interpreting the failure load to BMD relationships in this and other experimental studies.

Under in situ conditions, about 50-60% of the variability of mechanical failure loads was explained by bone densitometry at the site of interest (also accounting for other segments of the spine and the contralateral femur), 30-50% by total body analysis, and only 20-35% by regional DXA measurements at distant sites. These findings also apply to the spine and are in contrast to reports of Hassager et al.(14) and Bjanarson et al.(41) These authors reasoned that it may be preferable to assess bone mineral status at the radius or femur to predict fracture risk in the spine, avoiding accuracy errors from soft tissue and posterior spinal elements. Our finding suggests, however, that measurements should be performed directly at the spine, even in elderly individuals with some degree of osteoarthritic changes. Nevertheless, it is interesting to note that only in the spine (but not in the femur or radius) measurements from other sites added significant, independent information to the prediction of strength. We found no optimal measurement site to predict mechanical strength at all sites of interest. If only one regional measurement is obtained in a patient, it must be borne in mind that this does not provide optimal prediction of future fractures at other sites.

Prospective clinical studies have indicated that the risk of hip fracture is better predicted from DXA at the proximal femur than from measurements at other sites, but clinical results in the spine and radius have been less clear.(6–18) Our present study provides experimental evidence that site-specific measurements are superior to non-site-specific analyses at all relevant skeletal locations.

In conclusion, this experimental study shows that ex situ DXA provides slightly superior prediction of site-specific bone strength compared with in situ DXA. Non-site-specific data add significant information in the spine, but not in the femur or distal radius (except for 3-point-bending). In estimating “global” strength throughout the skeleton, total body scans display no significant advantage over regional analysis, and no particular site is identified as being optimal. A site-specific approach of bone densitometry is found to be superior to non-site-specific analysis, even when accounting for spatially variable in situ accuracy errors. The data suggest that mechanical competence in the elderly is governed by strong regional variation and that its loss in osteoporosis may not represent a strictly systemic process.


This work was funded by the Deutsche Forschungsgemeinschaft (DFG) under grant LO 730/L2-1. We are grateful to Jan Grimm (Medizinische Physik, Klinik für Diagnostische Radiologie, Kiel, Germany) for reading the X-rays; Klaus-Peter Maag (Institut für Medizinische Informations verarbeitung, Biometrie und Epidemiologie, LMU München) for his help with the statistical analysis; and Gudrun Goldmann, Nadine Krefting, Stephanie Kranz, Dominik Bürklein, and Markus Bachmeier (Forschungsgruppe Muskuloskelettales System, Anatomische Anstalt München) for their help with radiography, DXA scanning, and biomechanical testing.