INTRODUCTION

Every year 8.9 million osteoporosisrelated fractures occur worldwide, representing one fracture every 3 seconds¹, with vertebral being the most common osteoporotic fractures².

Dual-energy X-ray absorptiometry (DXA) is the standard test for diagnosing osteoporosis and evaluating fracture risk³,⁴, as it is a low-radiation, inexpensive technique. The DXA provides two-dimensional (2D) images that measure the bone mineral density of the area (aBMD) projected along the anteroposterior (AP) direction. Various studies show that a low aBMD value, measured in AP DXA explorations, is among the highest fracture risks³-⁵. The decrease of the aBMD standard deviation leads to an increase from 1.5 up to 3.0 times the risk of fracture, depending on its location and its measurement's location⁵. Nevertheless, a low BMD value is not enough to explain every fracture. Recent studies suggest that the risk of fracture is high when the BMD value is low, but this does not mean that fracture risk is negligible when the BMD value is normal³-⁸.

Most osteoporosisrelated vertebral fractures are located in the vertebral body⁹. In AP DXA images of the spinal column, the vertebral body overlaps the posterior vertebral elements, so the BMD in the vertebral body cannot be estimated separately. On the other hand, the risk of fracture depends on the architecture of the trabecular bone and the thickness of the cortical bone¹⁰. However, the trabecular and cortical bone compartments are difficult to assess separately on AP DXA scans.

As an alternative to the DXA, quantitative computed tomography (QCT) provides a three-dimensional (3D) analysis of the bone structures. In QCT imaging, the volumetric BMD (vBMD) can be measured in the vertebral body alone, separate from the posterior vertebral elements, and even trabecular and cortical structures can be evaluated in isolation³,¹¹,¹². Previous studies have appraised the association between vBMD derived from QCT and vertebral fractures⁸,¹³-¹⁷. Finite element models based on QCT have also been analysed to know the mechanical properties of the vertebrae and to predict the risk of vertebral fracture¹⁷-²⁰. However, QCT scans compared to DXA scans involve exposure to a higher dose of radiation, as well as a higher cost. As a consequence, QCT is rarely used in clinical practice for fracture risk assessment.

In order to overcome the limitations of the DXA and QCT scans, various researches suggest the use of 3D modelling methods to determine the shape and distribution of bone density considering a limited number of DXA scans²¹-²⁵. These studies use a threedimensional statistical model of the bone shape and density which is recorded in DXA examinations to obtain a personalized 3D model of such bone (QCT type). The precision of these methods²¹-²⁵ has been evaluated by comparing 3D models and measurements obtained via DXA and QCT. However, as far as we know, no study has been conducted on the association of measurements provided by DXA-based 3D modelling techniques and vertebral fractures.

On another note, AP DXA usually includes only the lumbar region (L1 to L4), since the rib cage overlaps in the projection, avoiding the use of DXA to determine the aBMD in the thoracic spine. Despite this, various studies indicate that the greatest number of vertebral fractures related to osteoporosis occur at the thoracoabdominal junction (T12-L1)¹⁵,²⁶. Although the reason for this higher prevalence is unknown, it has been suggested that thoracic kyphosis and rib cage stiffness predispose this area to fracture as the vertebral load is higher at this location. Even though measurements made at the same fracture location show a greater power of discrimination, Budoff et al.²⁷ found a high correlation between trabecular vBMD in the lumbar vertebrae and trabecular vBMD in the dorsal vertebrae.

The objective of this study was to assess the ability of 3D measurements derived from the DXA to distinguish subjects with incident fractures of the dorsal vertebrae from control subjects. To do this, a retrospective casecontrol study was carried out, which included postmenopausal Caucasian females who experienced a fracture event in the dorsal vertebrae (cases) and control females of the same age without any type of fracture. For each subject, 3D measurements derived from lumbar DXA were obtained at the initial baseline visit (which took place at least one year prior to the vertebral fracture event for subjects in the experimental group) using lumbar DXA AP scans and a DXAbased 3D modelling technique²⁵.

MATERIALS AND METHODS

RESULTS

Characteristics of the subjects

32 postmenopausal Caucasian women were included in this study: 16 patients with at least one incident osteoporotic fracture of the dorsal vertebrae (experimental group) and 16 subjects of the same age without any type of fracture (control group). The fractured group consisted of 11 subjects with a single fractured dorsal vertebra and 5 subjects with multiple vertebral fractures. A total of 25 incident vertebral fractures were found in the patients in the fracture group: two T4, two T7, one T8, one T9, three T10, one T11, nine T12, five L1 and one L2. It is unknown whether the vertebral fractures presented wedge, biconcave or crush deformities.

Patients in the experimental group had a vertebral fracture event at an average (± standard deviation) of 3.2±2.4 years from the initial baseline visit. The absence of osteoporotic fracture events was ensured for the controls during an average period of 8.4±1.0 years. No significant differences (ρ≥0.05) were observed between the experimental and control groups in terms of age, weight, height and BMI (Table 1).

Table 1. Characteristics of the subjects at the initial baseline visit 

Results expressed as mean ± standard deviation [minimum - maximum];

^*:p values of the unpaired Student t-test;

BMI: body mass index.

2D measurements derived from DXA

In line with WHO classification criteria, 94% of the patients in both groups presented a low aBMD (T-score for L-L4 <-1). The experimental group included 12 women with osteoporosis, 3 with low bone mass and 1 with normal bone mass; while the control group included 5 women with osteoporosis, 10 with low bone mass and 1 with normal bone mass.

The mean aBMD in the L1-L4 segment of the subjects in the experimental group was 8.1% lower compared to those in the control group, although not significant (p=0.11; table 2). There were also no significant differences in BMC and AREA (p>0.05). The aBMD differentiated between the experimental group and the control group with an AUC=0.662. Each decrease in one standard deviation in the aBMD was associated with an almost twofold increase in the probability of presenting an osteoporotic fracture in the dorsal vertebrae (OR=1.862; 95% CI: 0.8624.022).

Table 2. Measurements derived from the DXA of the initial baseline visit in both groups, differences between groups, AUC and OR 

Measurements of the experimental and control groups are expressed as mean ± standard deviation. Differences between groups are expressed as mean (percentage).

^*:p values of the unpaired Student ttest; p values <0.05 are shown in bold;

^a:probability ratio corresponding to a standard deviation of decrease in the measure;

^b:probability ratio corresponding to a standard deviation of increase in the measurement;

AUC: area below the receiver operating characteristic curve; OR: odds ratio; CI: confidence interval; Int.: integral; Trab.: trabecular; Cort.: cortical; aBMD: areal bone mineral density (g/cm²); BMC: bone mineral content (g); area (cm²); vBMD: volumetric bone mineral density (mg/cm³); volume (cm³); Cort. Th .: cortical thickness (mm); BMDs: surface bone mineral density (mg/cm²); Int.: integral bone; Trab.: trabecular bone; Cort.: cortical bone; Total: total vertebra; Body: vertebral body.

3D measurements derived from the DXA

The integral vBMD in the total vertebra in the experimental group was 10.2% lower than in the control group (p<0.05; table 2). In the vertebral body, the differences in vBMD were more pronounced in the trabecular bone (-16.2%, p<0.01) than in the integral bone (-12.8%, p<0.01). Cortical vBMD in the vertebral body was 2.3% lower in the experimental group, as well as non-significant (p=0.477). The cortical vBMD in the vertebral body in the experimental group was 10.0% lower than in the control group (p=0.018). The anatomical distribution of the mean differences between the trabecular vBMD in the vertebral body of the subjects included in the fractured and control groups is shown in figure 1. In it, we can see how the differences in the trabecular vBMD are more pronounced near the end plates of the vertebral body.

Figure 1. Anatomic distribution of the mean differences in trabecular vBMD, between the subjects included in the experimental group and those in the control group. The differences are shown in the mid-coronal plane (centre) and the median lateral plane (right) of the vertebral body. The image on the left shows the plans that were used. The red-yellow areas indicate the regions where the difference in trabecular vBMD between the subjects with vertebral fracture and the control subjects is on average lower (in blue-green areas this difference is higher). Non-significant changes (unpaired Student t-test) were marked in black. The pink outline indicates the periosteal surface of the vertebral body 

The trabecular vBMD in the vertebral body was related to higher values of AUC (0.801) and OR (5.060; 95% CI: 1.406-18.208), compared to other measurements derived from DXA (Table 2). Slightly lower values were found for the integral vBMD in the vertebral body (AUC=0.793 and OR=4.557; 95% CI: 1.411-14.718). The AUC map associated with the trabecular vBMD values calculated at each voxel of the volumetric images of the subjects included in the experimental and control groups is shown in figure 2. Only the AUC of the 90^th percentile (AUC>0.720) are represented. A maximum AUC value of 0.815 was reached. Trabecular vBMD measurements show a higher AUC value near the end plates.

Figure 2. Map of the AUC calculated by using the trabecular vBMD in each voxel of the volumetric images of the subjects included in the experimental group and the control group. Only AUCs greater than the 90^th percentile are represented (AUC >0.778). The maximum AUC registered is 0.930 

Cortical sBMD in the vertebral body was connected to higher values of AUC (0.734) and OR (2.649; 95% CI: 1.111-6.313), compared to other measurements made on cortical bone (Table 2). The anatomical distribution of the mean differences in cortical sBMD between the subjects included in the experimental and control groups, is shown in figure 3 (top). More pronounced differences (magenta-coloured) were found in the end plates of the L1, L2 and L4 vertebrae. The cortical sBMD in the lower end plate was the measure of sBMD with the highest AUC (0.754) and OR (2.793; 95% CI: 1.060-7.358) values. Figure 3 (bottom) shows the AUC value calculated using cortical sBMD at each vertex of the vertebral body surface. AUC values of the 90th percentile (i.e. in the range of 0.777-0.836) are circled in red and were found mainly on the end plates.

Figure 3. Top: anatomical distribution of the mean differences in the cortical sBMD of the vertebral body between the subjects included in the experimental group and the control group. Non-significant changes (unpaired Student t-test) are shown in grey. Bottom: AUC was calculated by using the cortical sBMD at each vertex of the vertebral body surface of the subjects included in the experimental group and the control group. The regions where the differences in the cortical sBMD were not significant (unpaired Student t-test), in the region of interest of the total vertebra, are shown in grey. The regions listed at an AUC greater than the 90^th percentile (that is, an AUC>0.777) are circled in red. The maximum AUC was 0.836 

DISCUSSION

In the present study, the ability of 3D measurements derived from lumbar DXA, to discriminate between postmenopausal women with and without osteoporotic fractures in the dorsal vertebrae, was evaluated. 3D measurements derived from the DXA were performed at the initial baseline visit (at least one year before the vertebral fracture event), using standard DXA scans and a 3D modelling technique²⁵.

Age, gender, and BMI are independent risk factors for osteoporosisrelated fractures³,⁴. In this study, a database of postmenopausal females paired by age was used to eliminate the possible effect of age and gender on the results. Although inclusion criteria related to height or weight were not used to recruit subjects, no significant differences were found between the groups in terms of height, weight, and BMI at the initial baseline visit.

No significant differences were observed in the aBMD (-8.1%, p=0.110), but in the integral vBMD (-10.2%, p=0.044). On the other hand, higher ORs were found for the vBMD measurements obtained via DXA in the vertebral body (OR=4.557; 95% CI: 1.411-14.718 in the integral bone, and OR=5.060; 95% CI: 1.406-18.208 in the trabecular bone) in comparison with the aBMD (OR=1.862; 95% CI: 0.862-4.022). These results are consistent with various studies present in the literature, where it was found that the OR for QCT-derived measurements of vBMD are similar or higher, if compared with aBMD measurements¹⁴,¹⁶. Melton et al.¹⁴ reported a slightly higher OR for vBMD in the L1-L3 segment (OR=2.2; 95% CI: 1.1-4.3) in the integral bone and OR=1.9; 95% CI: 1.0-3.6 in the trabecular bone), compared to aBMD (OR=0.7; 95% CI: 0.4-1.2). Anderson et al.¹⁵ reported a higher OR for vBMD in L3 (OR=5.3; 95% CI: 1.3-21 in the integral bone and OR=5.6; 95% CI: 1.3-23.4 in the trabecular bone), compared to aBMD (OR=2.8; 95% CI: 1.0-8.0). Grampp et al.¹⁶ reported a higher OR for vBMD in the L1-L4 segment (OR=3.0; 95% CI: 1.5-6.1 in the integral bone and OR=4.3; 95% CI: 1.8-10.1 in the trabecular bone), compared to aBMD (OR=2.4; 95% CI: 1.4-4.2).

The trabecular vBMD in the vertebral body was the measure that best discriminated between the experimental and the control groups, with an AUC of 0.801, compared to 0.662 for aBMD. Similar findings have been found in the literature in studies based on QCT³,⁴,¹¹-¹⁷,³⁰. Chalhoub et al.¹³ reported an AUC of 0.79 for trabecular vBMD, compared to 0.72 for aBMD. Melton et al.¹⁴ reported an AUC of 0.78 for trabecular vBMD, compared to 0.75 for aBMD. Grampp et al.¹⁶ reported an AUC of 0.82 for trabecular vBMD, compared to 0.78 for aBMD. Imai et al.²⁰ reported an AUC of 0.77 for trabecular vBMD, compared to 0.71 for aBMD.

Degenerative diseases of the vertebral spine, abdominal aortic calcification, and other sclerotic lesions artificially increase the aBMD measure obtained in the AP DXA³,⁴,¹¹,³⁰, despite the fact that patients with such pathologies have a higher risk of fracture. Trabecular vBMD in the vertebral body may be less sensitive to artifacts produced by these diseases, which are often found on the vertebral surface (cortical bone) or in the posterior arch. This could explain the higher AUC values found for the trabecular vBMD in our study. In this sense, 3D measurements derived from the DXA of the trabecular bone in the vertebral body could provide an alternative measure, overcoming the limitation of the diagnosis based on the aBMD by ruling out bone spurs, local deformations in the periosteal surface or in the posterior vertebral processes³⁰.

In the present study, the differences were less pronounced in the cortical bone (cortical sBMD: -10.1%, AUC=0.734) than in the trabecular bone (trabecular BMD: -16.2%, AUC=0.801). Biomedical studies demonstrated that the contribution of the cortical bone to the vertebral force is usually low in normal subjects, but it could be considerable in subjects with osteoporosis³⁰,³¹. The precision of measurements derived from the DXA in the trabecular and the cortical bones was evaluated in previous works²⁷. However, the cortex of the vertebral body is very thin (from 180 to 600 µm with an average thickness of 380 µm)³², and DXAbased 3D modelling methods can hardly model local deformities, which could affect the precision of the cortical. Cortical sBMD is considered a stronger measure of cortical bone than cortical vBMD, since generally it is easier to measure in lowresolution images³¹,³².

Local differences between the experimental and the control groups were analyzed using color-coded images. The mean differences in the trabecular vBMD between the subjects included in the experimental groups and those in the control group and their respective AUC were greater near the end plates and smaller in the centre of the vertebral body. Experimental studies of vertebral fractures in specimens show how the end plates of the vertebral body are the regions where failure at a tissue level begins³³-³⁶. These findings are consistent with biomechanical studies that show that the maximum load fraction on the trabecular bone normally occurs near the end plates²⁰,³⁵,³⁶. The anatomical distribution of mean differences in cortical sBMD between subjects included in the thoracic spine fracture subgroup and their respective subjects in the control subgroup shows more pronounced differences in the end plates³³-³⁶. The results are consistent with biomechanical studies showing the thickness and density of the end plate, and the density of the adjacent trabecular bone as good predictors of local stiffness and strength.

The most significant limitation of the present study is the small number of subjects included. The main difficulties in including subjects in the experimental group were to find patients with DXA images from before the incident fracture, since most patients go to the doctor's office after the fracture event, and to ensure that the subjects did not present prevalent osteoporotic fractures in any bone at the time of the initial baseline visit. Furthermore, our study is monocentric, only includes postmenopausal Caucasian females and not all of them have the same fractured vertebra. Therefore, the results can only be extrapolated to populations with similar characteristics. Besides, due to the design of our study (retrospective and casecontrol), we cannot directly imply a causal association between the reduction of 3D measurements derived from DXA and osteoporotic fracture. Another limitation is that the participants included in this study did not undergo a QCT examination. Therefore, we were unable to make a direct comparison between the results obtained using 3D measurements derived from DXA and measurements derived from QCT. Nor was a comparison made of 3D measures derived from DXA and other methods used in clinical practice to predict fracture risk (such as the Trabecular Bone Score –TBS– or the FRAX^® tool). Furthermore, the presence/ absence of vertebral fracture was confirmed by anteroposterior DXA scans and vertebral morphometry (VFA, Vertebral Fracture Assessment). It would have been interesting to include other imaging modalities such as QCT or X-ray to further assess vertebral fractures.

CONCLUSIONS

This case-control study showed the association between 3D measurements derived from lumbar DXA and incident osteoporotic fractures in the dorsal vertebrae. The individuals in the experimental group showed lower values of vBMD measured in different vertebral regions and compartments compared to the values measured in the group of control subjects. Trabecular vBMD in the vertebral body was the measure that best discriminated between the experimental and control groups. Methods based on 3D modelling based on DXA could be a valuable option to complement standard 2D measurements derived from DXA in the management of osteoporosis. Similar studies involving larger cohorts will be conducted in future researches to determine whether 3D measurements derived from lumbar DXA could improve the prediction of fracture risk in clinical practice. Case-control studies will also be carried out with subjects presenting exclusively osteopenia according to the aBMD criteria.