The authors state that they have no conflicts of interest.
Comparison of Densitometric and Radiographic Vertebral Fracture Assessment Using the Algorithm-Based Qualitative (ABQ) Method in Postmenopausal Women at Low and High Risk of Fracture†
Article first published online: 10 SEP 2007
Copyright © 2008 ASBMR
Journal of Bone and Mineral Research
Volume 23, Issue 1, pages 103–111, January 2008
How to Cite
Ferrar, L., Jiang, G., Clowes, J. A., Peel, N. F. and Eastell, R. (2008), Comparison of Densitometric and Radiographic Vertebral Fracture Assessment Using the Algorithm-Based Qualitative (ABQ) Method in Postmenopausal Women at Low and High Risk of Fracture. J Bone Miner Res, 23: 103–111. doi: 10.1359/jbmr.070902
- Issue published online: 4 DEC 2009
- Article first published online: 10 SEP 2007
- Manuscript Accepted: 6 SEP 2007
- Manuscript Revised: 25 JUL 2007
- Manuscript Received: 13 APR 2007
- vertebral fracture;
- bone densitometry;
Using ABQ diagnosis, the sensitivity to detect VF of densitometric versus radiographic assessment in 755 postmenopausal women was 71-81% and specificity was 97%. Misdiagnosis was influenced by image quality and was more common for mild deformities.
Introduction: Using densitometric vertebral fracture assessment (VFA), prevalent fractures are identified when vertebral height appears reduced by ≥20%. However, this approach does not discriminate between osteoporotic vertebral fracture (VF) and nonosteoporotic deformity, which increases the false-positive rate. Algorithm-based qualitative diagnosis (ABQ) focuses on vertebral endplate fracture to exclude these deformities but has not been applied in VFA. We wished to determine whether densitometric image quality is adequate for ABQ assessment. Our aims were to (1) calculate agreement between VFA and radiography using ABQ to identify prevalent VF and (2) identify the primary reasons for any discordant diagnosis.
Methods: Radiographic and densitometric spine images for postmenopausal women at low risk (LR; n = 459) and high risk (HR; n = 298) of VF were assessed using ABQ. Agreement between imaging modalities for VF diagnosis was assessed by κ statistics using ABQ radiographic readings as the gold standard.
Results: The prevalence of VF was 11-29% (radiography) and 9-26% (VFA) in the LR and HR groups, respectively. Agreement between imaging modalities was good or very good (κ = 0.62-0.81 in the LR and HR populations). The sensitivity to detect women with VF by VFA was 71% and 84% in the LR and HR populations, respectively, and specificity was 97%. Fifty-two (77%) and 60 (61%) of vertebrae misclassified by VFA in the LR and HR populations were mild fractures and 37 (54%) and 62 (63%) were wedge fractures. One third of fractures missed by VFA were related to poor or unreadable image quality (n = 27 and 28 vertebrae in the LR and HR populations, respectively).
Conclusions: There was good agreement between VFA and radiography using ABQ to identify prevalent VF in women at LR or HR of osteoporotic VF. Vertebrae misclassified by VFA were primarily mild fractures or deformities, and two thirds of all fractures missed by VFA were related to poor or unreadable image quality.
Osteoporotic vertebral fracture (VF) is associated with increased morbidity and mortality.(1) Prevalent VF predicts future osteoporotic fractures independently of BMD. Patients with one or more VFs have a 4-fold risk of subsequent hip fractures and a 5-fold risk of further VF(2): once identified, these patients can be offered therapy to reduce their fracture risk. VFs are usually diagnosed from spine radiographs or digital images of the spine acquired by bone densitometry devices. The latter, known as densitometric vertebral fracture assessment (VFA), represents a safer alternative to radiography for the imaging of VF, because the radiation dose to the patient is substantially reduced in comparison with conventional radiography.(3)
A semiquantitative approach(4) is often applied to visual assessment of osteoporotic VF from spine radiographs. This method may also be applied for the assessment of VFA images and/or morphometric assessment may be performed using the scan analysis software. With these approaches, the definition of prevalent fracture is based on the appearance of “reduced” vertebral height, usually by ≥20%. This can be problematic, because of the presence of long-standing nonosteoporotic deformities that may also appear to have reduced vertebral height.(5) We have recently proposed a modified approach to visual diagnosis of osteoporotic vertebral fracture, known as the algorithm-based qualitative (ABQ) method.(6) The ABQ method was developed in an effort to reduce the false-positive rate caused by misclassification of nonosteoporotic deformities: it focuses on the appearance of the central vertebral endplates to identify prevalent fractures rather than on the appearance of “reduced” (short) height per se.
The assessment of the vertebral endplates from spine radiographs can be complicated by geometric image distortion (the so-called parallax effect), because the endplates are projected obliquely, giving them an elliptical appearance.(7) An advantage of VFA is that this effect is virtually unseen because the X-ray beam is collimated to produce a narrow fan beam. However, although the spatial resolution for VFA has improved considerably over recent years, a disadvantage of VFA is that the spatial resolution remains slightly inferior to that of radiography. The ABQ method has not previously been applied to the assessment of VFA images. We therefore needed to determine whether the image resolution for VFA is sufficient to enable the detailed inspection of the vertebral endplates required to apply the ABQ method. The aims of this study were to (1) calculate agreement between VFA and radiography using the ABQ method to identify prevalent vertebral fractures and (2) determine the primary reasons for any discrepancies in diagnosis between VFA and radiography.
MATERIALS AND METHODS
We assessed spine radiographs and densitometric spine images for two groups of women, one at low and one at high risk of osteoporotic VF. The low-risk (LR) group was a population-based sample of postmenopausal women participating in the Sheffield arm of the Osteoporosis and Ultrasound Study (OPUS).(8) This is a prospective study based in five European centers. These are Aberdeen (Department of Medicine and Therapeutics, University of Aberdeen, Aberdeen, UK); Berlin (Center of Muscle and Bone Research, Charite-Universitätsmedizin Berlin, CBF Germany); Kiel (Medizinishce Physik, Klinik fur Diagnostische Radiologie, Univesitats-klinikum Schleswig-Holstein, Campus Kiel, Kiel, Germany); Paris (Center d'Evaluation des Maladies Osseuses, Service de Rhumatologie, Assistance-Publique, Hopital Cochin, Universite Rene Descartes, Paris, France); and Sheffield (Academic Unit of Bone Metabolism, University of Sheffield, Sheffield, UK). The study is coordinated by the Kiel center. Exclusion criteria for entry into the study were disorders that precluded valid quantitative ultrasound measurements (such as bilateral calcaneal fractures, hip prostheses, or disorders of the hand) or general inability to undergo the specified examinations.
For the LR population, this report is based on the analysis of data for women 55-79 yr of age (mean, 68 ± 7 [SD] yr) who attended OPUS baseline visits in Sheffield, UK, between 1999 and 2001. The women were recruited from local general practices in the Sheffield area. In the first contact, women were asked to complete a short questionnaire and state whether they might wish to participate in the study. Women who agreed to participate but failed to attend were offered a further appointment; those who did not attend the second appointment were excluded from the study. As recruitment progressed, response rates stratified by 5-yr age groups were monitored, and recruitment was adjusted to achieve a homogeneous distribution across age groups. Five hundred postmenopausal women completed baseline visits: this analysis was restricted to 459 women for whom both spine radiographs and densitometric spine images were available.
The high-risk (HR) group was made up of postmenopausal women attending the Metabolic Bone Center, Northern General Hospital, Sheffield, UK, for baseline assessments after a low-trauma fracture.(9) These women had recently a sustained low-trauma fracture (of the proximal femur, proximal humerus, or distal forearm), been diagnosed with prevalent VF, or had been receiving prednisolone therapy at a dose ≥5 mg daily (or equivalent) for >3 mo. The women with forearm or humeral fractures were consecutively recruited from the orthopedic fracture clinic, those with hip fractures were recruited from the orthopedic wards, and those with vertebral fractures were recruited from new referrals to the metabolic bone clinic. Women receiving prednisolone therapy were recruited from outpatient clinics. A total of 380 women completed baseline visits: this analysis was restricted to 298 women 55-80 yr of age (mean, 68 ± 7 yr) for whom both spine radiographs and densitometric spine images were available. The composition of this group was as follows: 185 women with forearm (n = 76), humeral (n = 64), or hip fracture (n = 43), 49 women referred to the Metabolic Bone Center with VF, and 64 women on prednisolone therapy.
All the research conducted in the two study populations was approved by the North Sheffield Local Research Ethics Committee, and written informed consent was obtained from each study participant.
Conventional spine radiographs and densitometric spine images of the thoraco-lumbar spine were obtained for all women. Spine radiographs were obtained in the lateral projection only (lateral decubitus position) following a standardized protocol. The densitometric images were acquired using a QDR 4500A densitometer (Hologic, Bedford, MA, USA), software version 9.03. A centerline scan of the thoraco-lumbar spine was first obtained (postero-anterior [PA] projection), followed by supine lateral projection. Lateral scans were acquired using the single-energy scan mode.
Identification of prevalent VFs
An experienced radiologist (GJ) assessed both spine radiographs and densitometric images using the ABQ method. The spine radiographs were digitized before ABQ assessment. The digitized radiographs were viewed using Adobe Photoshop for Windows version 5.0 (Adobe). The densitometric images were viewed on the same workstation using Hologic Discovery software (version 12.0). Morphometric measurements of vertebral height were not performed. The study radiologist evaluated vertebrae T4 through L4 for evidence of osteoporotic VF. The densitometric images were read first: the spine radiographs were read by the same radiologist after a time gap of at least 3 mo. The study radiologist was blinded to the results of the earlier readings of densitometric images.
Full details of the ABQ method have been previously published.(6) The ABQ diagnostic algorithm was applied for visual classification of each vertebra to one of three categories: these were (1) osteoporotic fracture, (2) nonosteoporotic short vertebral height (SVH), or (3) normal. The algorithm incorporates several criteria for identification of osteoporotic VF. The first of these is that there must be radiological evidence of depression of the central vertebral endplate (concave fracture), with or without fracture of the vertebral ring apophysis or vertebral body cortex (wedge or crush fracture). Further criteria for osteoporotic fracture are then examined. These have been described previously,(6) but in brief, the depression of the endplate should be generally concave in shape and should be located within the anterior and posterior vertebral ring; the depression should involve more or less the whole of the endplate within the ring and there should be no evidence to suggest traumatic fracture (such as bone fragments, narrowed disc space, or bridging osteophytes) or pathologic fracture (such as density variations and bone destruction); or deformity caused by metabolic disease other than osteoporosis (such as osteomalacia).
The severity of osteoporotic fracture identified by ABQ was determined by visual estimation of the apparent reduction in vertebral height as follows: ≤25%, grade 1 (mild); >25% to <40%, grade 2 (moderate); or >40%, grade 3 (severe fracture). This approach to grading the severity of fracture is similar to that used in the semiquantitative (SQ) method.(4) However, fracture is not identified by ABQ on the basis of apparent reduction in vertebral height unless there is also evidence of endplate depression. Furthermore, the ABQ and SQ methods differ in the grading of mild fractures (those with <20-25% apparent reduction in vertebral height). For the ABQ method, there is no minimum threshold for reduction in vertebral height (provided the criteria for depression of the endplate are satisfied), whereas using the SQ method, vertebral height must appear reduced by at least 20% to identify a fracture.
Vertebrae with apparent reduction in vertebral height (short height) but no radiological evidence of osteoporotic endplate fracture were categorized by ABQ as non-osteoporotic SVH due to other causes. This was assessed qualitatively; that is, SVH was identified when one or more heights were approximately ≥15% shorter than expected. The study radiologist took into account the variation commonly seen within and between vertebrae and the variation seen in different regions of the spine to classify a vertebra as SVH rather than osteoporotic fracture. For example, the middle vertebral heights are normally shorter than the anterior and posterior heights of lumbar vertebrae, and in thoracic vertebrae, the anterior heights are generally shorter than the middle and posterior heights.
Vertebrae that could not be reliably categorized as osteoporotic fracture, SVH, or normal were classified “uncertain.” The reasons for uncertainty were noted and could include, for example, atypical or ambiguous appearances or poor image quality. For osteoporotic fracture, the affected endplate (superior, inferior, or both) and type of fracture (concave, wedge, or crush) was recorded. Each vertebra classified as SVH was categorized as follows: (1) normal or developmental variation; (2) degenerative change or Scheuermann's disease (with or without degenerative change); (3) scoliosis or kyphosis; (4) nonosteoporotic (traumatic or pathologic) fracture, or (5) metabolic bone disease other than osteoporosis. The affected height(s) (anterior, middle, posterior, or all three) was also recorded for each vertebra identified with SVH. Minor endplate deformities such as small Schmorl's nodes or osteophytes were not classified as SVH, but were noted as supplementary information. Women with evidence of both osteoporotic vertebral fracture and nonosteoporotic SVH (at different vertebral levels) were classified to the VF group.
We have recently calculated good interobserver agreement for ABQ assessment of spine radiographs for 203 elderly women referred for bone densitometry (κ = 0.74). The radiographs were read independently by an experienced radiologist (GJ) and a nonradiologist (LF) with experience in the definition of osteoporotic VF (unpublished data).
BMD was measured at the lumbar spine (vertebrae L1 through L4) and hip using DXA (Hologic QDR 4500A). For measurements made at the hip, we analyzed total hip BMD. Body height was measured in centimeters (using a wall-mounted stadiometer), and weight was measured in kilograms using a balance scale. Body mass index was calculated as weight in kilograms divided by height in square meters.
Comparisons between radiography and VFA
The ABQ readings of densitometric spine images (VFA) were compared with the ABQ readings of spine radiographs: the latter were considered the gold standard for this analysis. We compared the numbers of VFs and the numbers of women with prevalent VF identified in the LR and HR populations, respectively. We calculated percent values for sensitivity, specificity, and positive and negative predictive values (PPV and NPV) for VFA to identify VFs and women with VF using ABQ. For discordant diagnosis of VF identified by radiography and VFA, we calculated the frequency according to vertebral level, type and severity of fracture, and ABQ differential diagnosis of VF.
Two-sample t-tests were used to compare mean age, height, weight, body mass index, and BMD in women with and without VFs identified by ABQ. Agreement between ABQ readings of densitometric images and spine radiographs was assessed using κ statistics performed using MedCalc (B-9030; MedCalc, Mariakerke, Belgium). We calculated κ scores and 95% CIs for κ. κ scores were classified according to the system described by Altman.(10) κ scores were considered significantly different if there was no overlap in the 95% CI. Sensitivity and specificity for VFA were calculated as the percent women correctly identified with and without VF. PPV and NPV were calculated as the percent VFA-positive and VFA-negative women with and without true VF. Associations between misclassification of VF by VFA and poor image quality for VFA were analyzed by the χ2 test. The ABQ readings of spine radiographs were considered the gold standard for these analyses. For all analyses, p < 0.05 was considered statistically significant.
Identification of prevalent VF using ABQ readings of spine radiographs
The prevalence of VF according to ABQ reading of spine radiographs was 11% in the LR group (52 women were diagnosed with a total of 103 fractures) and 29% in the HR group (86 women were diagnosed with a total of 200 fractures; Fig. 1). In the LR group, 31 (60%) of the women with VF had one or more moderate or severe fractures: the corresponding figure for the HR group was 60 women (70%). The characteristics of women with and without VF identified from radiographs are shown in Table 1. Women with VF (in either study population) were older and had significantly lower BMD compared with those without fracture. In the HR group, the proportions of women with prevalent VF identified by ABQ were greater among those who had recently been identified clinically as having vertebral (n = 48/49; 98%) or hip fracture (n = 16/43; 37%) compared with those recently diagnosed with forearm (n = 8/76; 11%) or humerus fracture (n = 9/64; 14%). The corresponding figure for women receiving prednisolone therapy was 9/64 (14%).
Identification of prevalent VF using ABQ readings in VFA
The prevalence of VF according to VFA was 9% in the LR group (42 women were diagnosed with a total of 68 fractures) and 26% in the HR group (77 women were diagnosed with a total of 202 fractures). In the LR group, 22 (52%) of the women with VF had one or more moderate or severe fractures: the corresponding figure for the HR group was 59 women (77%). Women in either study population who had VF identified by VFA were significantly older (p < 0.05) and had significantly lower BMD (p < 0.001) compared with those without fracture (data not shown).
Agreement between radiography and VFA for identification of women with prevalent VF
Agreement between VFA and radiography for the identification of women with VF (assessed by κ statistics) was good in both study populations and was better in the HR group (Table 2). In the LR population only, the κ score was slightly higher for the identification of women with one or more moderate or severe fractures than for identification of those with only mild fractures, but these differences were not statistically significant. The κ scores for agreement between VFA and radiography were similar when calculated on a per-vertebra basis. When the agreement between radiography and VFA was tested after excluding vertebral levels that were not assessed by VFA (that is, those that could not be evaluated because of poor image quality), the κ scores were 0.76 (95% CI, 0.65, 0.87) and 0.88 (95% CI, 0.82, 0.95) for the identification of women with VF in the LR and HR groups, respectively.
False-negative identification of VF by VFA occurred more often (5% of all women) in both study populations than false-positive results (2% of all women). In the LR population, 41% of women with radiographic vertebral fracture were either missed by VFA or were not analyzed because of unreadable image quality for VFA: in the HR population, the corresponding figure was 19% (Fig. 2). On a per-vertebra level, <1% of all vertebrae in women from the LR group were misclassified by VFA, and in the HR group, 2% of vertebrae were identified as false negative and 7% as false positive. If ABQ from spine radiographs is used as the gold standard, VFA had a sensitivity of 84% for the identification of women with VF in the HR group compared with 71% in the LR group (Table 3). The PPV in the LR group was better when the analysis was restricted to the identification of women with moderate or severe fractures (Table 3).
Analysis of discrepancies between VFA and radiography
Fifty-two (77%) and 60 (61%) of all vertebrae misclassified by VFA in the LR and HR populations, respectively, were reported as mild fractures (or deformities) (<25% apparent reduction in vertebral height), and 37 (54%) and 62 (63%) were classified as wedge fractures (or deformities) by radiography (VFA false positives) or VFA (false negatives). The vertebral levels most frequently misclassified by VFA were vertebrae T6 through T9 (46% of all false negatives, n = 36) and L1 (25% of all false positives, n = 5) in the HR population and vertebrae T4 (23% of all false negatives, n = 10) and T12 to L1 (48% of all false positives, n = 12) in the LR population. Eight of 43 vertebrae classified false negative and 3 of 11 classified false positive by VFA in the LR population were noted by the study radiologist to have apparent reduction in middle vertebral height of ∼15% or less. In the HR population, the corresponding proportions were 6 of 79 false negatives and 8 of 20 false positives. The largest single contributory factor to false-negative identification of vertebral fracture by VFA (in both populations) was poor or unreadable image quality: this accounted for approximately one third of all fractures missed by VFA (n = 27 and 28 false negatives in the LR and HR populations, respectively) (Fig. 3). Poor or unreadable image quality was significantly associated with false-negative identification of VF by VFA (p < 0.01, χ2 test), but not with false-positive identification of VF (p = 0.52 and 0.10 in the LR and HR groups, respectively). An example of false-negative identification of VF by VFA is shown in Fig. 4. Most of the vertebrae classified false positive by VFA in either study population (≥90%) were reported normal by radiographic diagnosis. High BMI (≥30 kg/m2) was observed in 19% and 26% of all women misclassified by VFA in the LR and HR populations, respectively, but was not significantly associated with misclassification of VF in either study population (χ2 test, p > 0.05).
These are the first published data on the application of the ABQ diagnostic method in densitometric VFA. As expected, the prevalence of VF identified by radiography was higher among women with known risk factors (prior history of low-trauma fracture or of prednisolone therapy) than in the population-based sample. We observed a similar pattern of VFs using the ABQ method to read the densitometric images, although the prevalence of VF was slightly lower in both the HR and LR study populations. The agreement between VFA and radiography for the identification of prevalent VF was good to very good (according to the Altman classification(10)) in the LR and HR populations, respectively. False-negative definition of VF by VFA (in either study population) was mainly related to difficulty in evaluating (or inability to evaluate) some vertebral levels because of poor (or unreadable) image quality. This, together with misclassification of traumatic fracture or nonosteoporotic SVH, was also one of the most common identifiable (but not statistically significant) reasons for false-positive definition of VF by VFA.
The International Society for Clinical Densitometry recently recommended(11) that, when evaluating densitometric vertebral scans, fractures should initially be assessed visually, followed by determination of fracture severity using either a SQ approach(4) or quantitative assessment using the VFA analysis software. The assessment of fracture severity is relatively straightforward. On the other hand, the differentiation between VF and nonosteoporotic deformity (which is much more common than fracture) requires radiological expertise and there has been a lack of clear criteria for the exclusion of these deformities. We developed the ABQ method in an attempt to address this problem. The key steps in using this method are to classify deformities that involve fracture of the endplate as osteoporotic fracture and to exclude deformities with short height but no fracture of the endplate as deformities due to other causes (e.g., developmental variation or degenerative change). Using this method to read conventional radiographs, we were able to show a closer association between ABQ fractures and low BMD compared with other traditional diagnostic approaches(6) and showed that non-osteoporotic SVH identified by ABQ is not associated with low BMD.(12) We wished to determine whether image quality for VFA was sufficient for close inspection of the vertebral endplates using the ABQ method.
Before the development of ABQ, we had previously compared radiography and VFA using a qualitative diagnostic approach for the identification of VFs.(13) The results of this analysis indicate that, although some VFs were missed by VFA because of poor image quality, the agreement between VFA and radiography using the ABQ method was similar to that observed in our earlier study. Furthermore, the prevalence of VF in this analysis was much lower (11-29%) than in the previous study (94%). The observed levels of agreement between VFA and radiography and sensitivity and specificity for VFA in this analysis are also comparable to those previously obtained in similar studies in which other diagnostic approaches have been applied for the definition of VFs.(14-16)
Our results suggest that image quality for VFA is no more problematic for application of the ABQ approach than it is for other diagnostic approaches. The fracture shown in Fig. 4 (which was false negative by VFA) would probably also be diagnosed false negative using either SQ or quantitative definitions of fracture: the vertebral height does not appear reduced because the depressed endplate is virtually impossible to identify without the support of a radiograph. It is important also to note that good agreement between two imaging modalities (whatever definition of fracture is applied) is not necessarily an indicator of the accuracy of VF detection. Quantitative morphometry, for example, may be more objective and reproducible than qualitative approaches, but the accuracy of VF detection may be reduced if radiological differential classification of nonosteoporotic deformities is not performed.(17)
Similar proportions of women at LR and HR of VF were misclassified by VFA in this study, but at a per-vertebral level, more vertebrae were misclassified among women from the HR population. Analysis of discrepancies between VFA and radiography suggests that misclassification of VFs by VFA was influenced by the apparent severity of fracture, because a large proportion of all misclassified vertebrae were mild fractures (with <25% apparent reduction in vertebral height). Women were more frequently misclassified by VFA (if radiography is considered the gold standard) when they had one or more mild vertebral fracture only: 1% and 8% of women in the LR and HR populations, respectively, were identified with at least one moderate or severe fracture by radiography alone compared with 27% and 12% of women with mild vertebral fracture only (data not shown). The corresponding figures for women with VF identified by VFA alone were 2% and 1% of women (in the LR and HR populations) identified with at least one moderate or severe fracture and 24% and 7% of women with mild fracture only. However, the clinical significance and true nature of mild vertebral deformities (even when imaged by conventional radiography) is often called into question.(17)
We anticipated that vertebrae with apparent height reduction <20% might be difficult to identify with less than optimal image quality: these fractures were more commonly associated with false-positive than with false-negative identification of fracture by VFA (data not shown). The absence of a pattern to the vertebral-level distribution of misclassified fractures by VFA was surprising (because the upper thoracic vertebrae are often the least clearly imaged), but this might be partly because of the fact that the frequency of osteoporotic fracture in this region is generally relatively low. We observed several instances of suspected discrepancy between radiography and VFA in the segmentation of vertebral levels (data not shown). These were mostly seen in women who had more than one VF and therefore were likely to have influenced the numbers of fractures correctly identified by VFA rather than the numbers of women with fracture. This may explain why the proportions of false negatives were much greater when calculated per-vertebra rather than per-subject.
Poor or unreadable image quality was significantly associated with false-negative, but not false-positive, identification of VF by VFA. Fifteen women with prevalent VF in the LR group and 10 women in the HR group were not identified by VFA because the fractured vertebrae (identified by radiography) could not be reliably assessed. In comparison studies, vertebral levels that cannot be evaluated by VFA (because of poor image quality) are often excluded from statistical analysis. When we applied this approach, the agreement between VFA and radiography for the identification of women with vertebral fracture was improved considerably in both study populations. In clinical practice, radiographs could be obtained for those patients who cannot be fully evaluated by VFA: however, in our study population, this would mean performing radiography in a large proportion of women without radiographic evidence of VF.
We tested VFA against conventional radiography. This is the current standard for identification of VF but is subject to geometric image distortion of the vertebral endplates. This distortion is virtually eliminated in VFA because the beam is emitted through a narrow slit aperture. We considered the possibility that, in some cases of discordance between imaging modalities, the true diagnosis could have been made by VFA and evaluated this by comparing mean BMD T-scores (young normal-adjusted SD units) in those with true and false diagnosis of VF by VFA. The mean T-scores for both lumbar spine and total hip BMD were significantly lower (p < 0.01) in women (from either population) who were classified false negative for VF by VFA than in women classified true negative (data not shown). This implies that most of fractures in women identified by radiography alone were true fractures, because they were associated with lower BMD. However, it is important to note that a substantial proportion of these were classified negative by VFA because the relevant vertebral levels were not evaluated (they were reported unreadable because of poor image quality). We could have excluded these, but this would have left too few values for comparison of mean values in women evaluated by both radiography and VFA. In general, women with VF identified by VFA only (false positives) had similar mean BMD T-scores to those who were identified by both VFA and radiography. The exception was for total hip BMD: the mean T-score was significantly higher (p < 0.01) in those identified with fracture by VFA alone (data not shown).
Our study methodology had several strengths. A single expert reader assessed both densitometric and radiographic images using the same diagnostic approach (the ABQ method): this meant that we could rule out discrepancies between imaging modalities related to differences between readers or different definitions of fracture. We studied two discrete groups of women: this enabled us to test ABQ assessment of VFA scans in populations with a different prevalence of fracture. There were some limitations to our analysis. First, there is no consensus on the optimal approach to the identification of prevalent VF and, as with any method, we cannot be certain of the accuracy of ABQ to detect true osteoporotic fractures. However, the main objective of this analysis was to determine the feasibility of applying ABQ for the assessment of densitometric vertebral images. Second, the VFA scans in our study populations were acquired using the single-energy mode only: an additional dual-energy scan could have provided useful information when one or more vertebrae were not clearly imaged on the single-energy scan. Also, we used an earlier version of Hologic acquisition software to obtain densitometric images. There have since been improvements in image quality with the introduction of high-resolution imaging. However, the detector resolution of the more recent Hologic Discovery system is almost the same as for the earlier Hologic devices, so our results should be applicable. The women reported here who are currently attending 6-year follow-up visits are undergoing repeat VFA using a Discovery system. We examined a small number of these scans (n = 24) acquired in early follow-up visits to evaluate the likely impact of high-resolution imaging. The study radiologist (GJ) used the ABQ method (as described in this report) to assess the scans, blinded to the results of the baseline assessments. Women who had been identified with VF at baseline were also identified at follow-up, and although fewer vertebrae were considered completely unreadable on follow-up (5%) compared with baseline scans (9%), suboptimal image quality reported at baseline was also reported for the same women and at the same vertebral levels on the high-resolution scans acquired at follow-up. Analysis of the full dataset on completion of the follow-up studies will enable us to evaluate this more accurately. Finally we evaluated VFA using a Hologic device: image quality may vary for vertebral scans acquired using other manufacturers' systems.
We conclude that, among postmenopausal women with LR or HR of osteoporotic vertebral fracture, (1) there was good agreement between VFA and conventional radiography for ABQ assessment of prevalent VF: the level of agreement was comparable to that obtained in previous studies using established diagnostic approaches, and (2) vertebrae that were misclassified by VFA were largely mild wedge fractures, and women with mild fracture only were much more frequently misclassified than women with at least one moderate or severe fracture: poor or unreadable image quality reported by the study radiologist accounted for approximately one third of all fractures missed by VFA.
LF was funded by the Medical Research Council, UK. The authors thank Margaret Paggiosi and Anne Stapleton for performing VFA and BMD measurements and Debbie Swindell for recruiting study participants.
- 71995 Radiographic technique for assessing osteoporotic vertebral deformity. GenantH, JergasM, van KuikC (eds.) Vertebral fracture in osteoporosis. Radiology Research and Education Foundation, San Francisco, CA, USA, 131–147., ,
- 82004 Association of five quantitative ultrasound devices and bone densitometry with osteoporotic vertebral fractures in a population-based sample: The OPUS Study. J Bone Miner Res 19: 782–793., , , , , , , , , , , ,Direct Link:
- 101991 Practical Statistics for Medical Research. Chapman and Hall, London, UK.