Identification of vertebral fractures: An update
- First Online:
- Cite this article as:
- Ferrar, L., Jiang, G., Adams, J. et al. Osteoporos Int (2005) 16: 717. doi:10.1007/s00198-005-1880-x
- 424 Views
Osteoporotic vertebral fracture is associated with increased morbidity and mortality. As a powerful predictor of future fracture risk, the identification of vertebral fracture helps target individuals who will benefit from anti-fracture therapy. The identification of vertebral fractures is problematic because (1) “normal” radiological appearances in the spine vary greatly both among and within individuals; (2) “normal” vertebrae may exhibit misleading radiological appearances due to radiographic projection error; and (3) “abnormal” appearances due to non-fracture deformities and normal variants are common, but can be difficult to differentiate from true vertebral fracture. Various methods of vertebral fracture definition have been proposed, but there is no agreed gold standard. Quantitative methods of vertebral fracture definition are objective and reproducible, but the major limitation of these methods is their inability to differentiate between vertebral deformity and vertebral fracture. The qualitative visual approach draws on the expertise of the reader, but it is a subjective method with poor interobserver agreement. Semiquantitative assessment of vertebral fracture is a standardized visual method, which is commonly applied in research studies as a surrogate gold standard. This method is more objective and reproducible than a purely qualitative approach, but can be difficult to apply. The established methods focus primarily on the identification of “reduced” or short vertebral height as an indication of vertebral fracture, but this is also a feature of some non-fracture deformities and normal variants. A modified visual approach known as algorithm-based qualitative assessment of vertebral fracture (ABQ) has recently been introduced, and this focuses on radiological evidence of change at the vertebral endplate as the primary indicator of fracture. Preliminary testing of the ABQ method has produced promising results, but the method requires further evaluation. Vertebral imaging by means of dual energy X-ray absorptiometry (DXA) scanner produces images of near-radiographic quality at a fraction of the radiation dose incurred by conventional radiography. There is growing interest in vertebral fracture assessment using this technique as a means of assessing a patient’s fracture risk. Given the increasing availability of new technology and the importance of accurate diagnosis of vertebral fracture, there is an urgent need for better awareness of and training in the definition of vertebral fracture. Methods of vertebral fracture definition should be validated by testing the association with clinical outcomes of vertebral fracture, in particular the prediction of incident fractures.
KeywordsAlgorithm-based qualitative assessment of vertebral fracture (ABQ)Dual energy X-ray absorptiometry (DXA)OsteoporoticRadiological evidenceVertebral fracture
Osteoporotic vertebral fractures are associated with increased morbidity [1–11] and mortality [12–16], but are often missed in clinical practice . Identification of vertebral fracture in osteoporosis is important, not least because there is evidence from several studies that it is a powerful independent predictor of future fracture risk at both the spine and the hip [18–21]. In a recent report from the MORE study, the presence of severe prevalent vertebral fracture was the strongest predictor of future vertebral and nonvertebral fractures . According to pooled estimates, the presence of a prevalent (existing or first) vertebral fracture confers a fourfold increased risk of further vertebral fractures in perimenopausal or postmenopausal women (with up to 19 times increased risk in other populations), and doubles the risk of hip fracture . Vertebral fracture is an important indicator of patients who are likely to benefit most from treatment to reduce fracture risk, and in clinical trials, the incidence of vertebral fracture is a major outcome measure . Accurate identification of vertebral fracture is clearly essential not only in clinical trials and epidemiological studies, but also for the clinical management of patients with osteoporosis.
Problems associated with the identification of prevalent vertebral fracture
The standard imaging approach for assessment of vertebral fracture is radiography of the thoracolumbar spine. At follow-up, the examination is often confined to lateral views only. The interpretation of the radiographic appearances is challenging and controversial, and there is no agreed “gold standard” for the definition of osteoporotic vertebral fracture. The first problem is that the identification of “abnormal” vertebral appearances from spinal radiographs requires a clear understanding of what is normal, and it is important that the definition of normality takes into account the wide range of intra-individual and inter-individual variation in vertebral body size and shape. Furthermore, normal vertebrae may exhibit misleading appearances on radiographs when the vertebral bodies are projected obliquely. This can be the result of patient positioning error, or may be due to the parallax effect caused by the divergent X-ray beam .
Having established that there is an abnormality, the next difficulty lies in deciding whether the appearances indicate true osteoporotic fracture [26–30]. The established methods of vertebral fracture definition rely mainly on the appearance of “reduced” or short vertebral height. This is problematic for the identification of prevalent fractures because true change in vertebral height can only be identified longitudinally. Vertebral height may appear to be reduced in comparison with the same vertebra or in comparison with other normally developed vertebrae. However, some adults have vertebrae that exhibit longstanding short anterior height in developmentally small vertebrae. This type of variation is present from adolescence, and may affect isolated vertebrae or several adjacent vertebrae. Other conditions such as Scheuermann’s disease are also associated with short vertebral height. It has been suggested that misclassification of non-fracture wedging may be avoided by limiting the identification of vertebral deformity to non-contiguous vertebrae. However, vertebral fractures can occur in adjacent vertebrae, but these are unlikely to have identical appearances. These vertebrae may be differentiated from normal variants or deformities due to other conditions by evidence of change in the vertebral endplate and lack of symmetry between adjacent vertebrae. In osteomalacia, for example, the pattern of endplate depression is very similar among adjacent vertebrae, whereas in osteoporosis, the appearances are unlikely to be symmetrical, even in the case of several adjacent fractures.
Non-fracture deformity is a collective term that describes a vertebra with abnormal variation in vertebral body height. This variation may be due to developmental abnormalities, with or without degenerative changes, or to pathological fracture. In addition to short vertebral height, non-fracture deformities of developmental origin may have deep or step-like endplates, whilst wedging and elongation of the vertebral bodies, irregularity or sclerosis of the endplates and Schmorl’s nodes are features associated with adolescent Scheuermann’s and/or degenerative disc disease. Non-fracture deformity is common and is arguably the most important source of false-positive identification of osteoporotic vertebral fractures.
Approaches to the identification of vertebral fracture
The ideal method of vertebral fracture definition is a standardized, reproducible approach that differentiates true osteoporotic fracture from normal variation or non-fracture deformity. Of the established approaches, there is no one method that satisfies all of these criteria. By applying the traditional qualitative approach to vertebral fracture definition, the expert reader may be able to eliminate erroneous sources of variation in vertebral shape, but because it is an entirely subjective method, qualitative assessment of vertebral fracture is limited by poor interobserver agreement and reproducibility [30–32].
Quantitative vertebral morphometry
Quantitative vertebral morphometry (QM) is both objective and reproducible. Several approaches have been proposed and tested [33–47], one of the most widely adopted methods being the classification of vertebral fracture according to Eastell and Melton . Using this method, any vertebral height ratio that is 3 or more standard deviations below the normal mean is defined as a vertebral “deformity.” This approach takes into account the normal variation in vertebral body shape across different vertebral levels and allows the classification of vertebral deformities according to type and severity. However, the use of population-based reference intervals may introduce greater error among individuals than the intra-individual variances across adjacent vertebrae, although the McCloskey approach  does attempt to overcome this problem.
The main weakness of QM in general is the lack of specificity for osteoporotic vertebral fracture. This is a problem associated particularly with methods based on continuous measures of vertebral deformity, such as the Minne algorithm , which has high sensitivity  but a weaker association with clinical outcomes of vertebral fracture compared with other morphometric approaches . A vertebral deformity identified quantitatively may represent true fracture, but may equally represent a non-fracture deformity. Because QM treats all changes in vertebral shape as vertebral deformities, this clearly increases the risk of false-positive identification of osteoporotic fracture. In the European Vertebral Osteoporosis Study (EVOS), 50% of moderate or severe vertebral deformities identified quantitatively were classified by radiologists as non-fracture deformities . Some researchers have attempted to address the problem of the high false-positive rate in QM by applying additional criteria for identification of vertebral deformity [46, 47]. Whilst these approaches may improve the specificity for detection of vertebral deformities, they may also increase the false-negative rate .
Poor differentiation between true fracture and non-fracture deformity by QM principally affects the identification of mild deformities, and for some time it has been acknowledged that a large proportion of mild QM deformities may be unrelated to osteoporotic fracture [51, 52]. This is mainly due to lack of evidence in some studies of an association between mild deformities and low bone mineral density (BMD) [49, 50], but the evidence remains unclear. In one (cross-sectional) study for example, mild vertebral deformities (with height ratio reduction less than 3 SD below the mean)  in older women have been found to be associated with low BMD, and mild vertebral deformities identified by other methods have been shown to be predictive (although to a lesser degree than moderate to severe deformities) of future fracture . Severe, but not mild, isolated vertebral deformities have been linked with risk factors for secondary osteoporosis , and in some studies, clinical outcomes such as back pain and physical disability tend to be associated mainly with moderate to severe deformities [56–59].
Some researchers speculate that mild vertebral wedge deformities may represent “gradual” fractures, but the evidence for this remains unclear . In osteoporosis, loss of the horizontal trabeculae may lead to the development of microfractures on the vertical trabeculae of the subchondral area of the vertebral body, due to pressure from the nucleus pulposus . The accumulation of these microfractures may result in increased concavity, as the central endplate becomes depressed. It could be argued that anterior loading on vertebrae with an accumulation of microfractures could produce gradual wedge fractures in the same way. However, this process does not adequately explain gradual wedging, because for anterior vertebral height to be reduced, there must also be fracture of the anterior cortical bone. In a cross-sectional study, a trend was found towards greater anterior wedge angle in elderly women  and significantly greater wedge angle with age in elderly men, particularly in those with evidence of degenerative changes . This suggests that a degree of non-fracture wedging may occur with increasing age. The concept of gradual fracture remains controversial and is an area where there is a need for further research.
Several factors limit the comparison of quantitative data across different studies, making it difficult to determine the true prevalence of vertebral deformity in the population. Vertebral dimensions vary among populations from different countries [64–69] and between men and women [9, 70–72], and vertebral heights measured by densitometric methods differ from those measured from conventional radiographs [72–75]. The prevalence and incidence of vertebral deformity vary according to the morphometric method and cut-points used, [26, 48, 50, 76–81], and are also influenced by false-positive and false-negative identification of vertebral deformity arising from radiographic projection or marking error . The placement of middle height markers is particularly problematic when the divergent X-ray beam produces an oblique projection, giving the vertebral endplates an elliptical appearance . For identification of incident vertebral deformities, the precision error for vertebral height measurements is greater in patients with established osteoporosis [84–85], and may be further increased when there is progression in a previous fracture .
Semiquantitative identification of vertebral fracture
Visual methods of vertebral fracture definition have the advantage that the expert reader may rule out deformities that are unrelated to vertebral fracture. Genant and colleagues proposed a standardized visual approach to vertebral fracture definition known as the semiquantitative method (SQ) . This popular method has been used as a surrogate gold standard in several important studies of osteoporosis [88–90]. The approach is more objective and reproducible than qualitative assessment of vertebral fracture, with better interobserver agreement, and the grading system set out in the SQ method provides useful information for the study of the epidemiology of osteoporosis and for use in clinical trials.
Identification of a prevalent vertebral fracture in SQ is based on the appearance of apparent reduction in vertebral body height and the identification of radiological characteristics of fracture at the vertebral endplate. The evaluation of apparent change in vertebral dimensions is standardized, so that a vertebral fracture is identified if vertebral height appears to be shorter than expected by 20–25% or more, and fractures are graded according to severity and type. In the original publication , it is recommended that changes at the vertebral endplate and cortical margin, and lack of consistency with adjacent vertebrae, should also be taken into consideration, and more recently, a modified approach has been reported for better differentiation between non-fracture deformity and true vertebral fracture . However, there may be a need for greater clarification for the interpretation of the radiological characteristics of change at the vertebral endplate that may or may not signify fracture. It remains unclear, for instance, what the diagnosis should be if vertebral height appears to be 20% shorter than expected, but there is no evidence of change at the vertebral endplate, and vice versa. In the absence of clear guidelines, there is the risk that a normal vertebra with developmentally short height may be diagnosed as a vertebral fracture, on the basis of apparent reduction in vertebral height and lack of symmetry with adjacent vertebrae.
The proportion of mild vertebral deformities identified by the SQ method tends to be greater than for other approaches. In one report, for example, the prevalence of vertebral deformity according to SQ is similar to that of QM using a criterion of mean height ratio minus 2 SD , and in the Study of Osteoporotic Fractures (SOF), almost four times as many mild vertebral deformities were identified by SQ than by various different quantitative approaches . The associations with clinical outcomes of vertebral fracture for mild vertebral deformities, however, were similar for all diagnostic methods applied in this study, and in another study, vertebral fractures identified by the SQ method have been shown to be predictive of both vertebral and nonvertebral fractures . In another study, mild vertebral fractures identified by SQ in men were not found to be correlated with BMD measurements .
Algorithm-based qualitative identification of vertebral fracture
Recently, a modified visual approach known as algorithm-based qualitative assessment of vertebral fracture (ABQ) has been proposed for the identification of prevalent vertebral fractures . The ABQ method differs from the SQ method in two ways. Firstly, the focus for identification of prevalent vertebral fracture using ABQ is on depression of the central endplate. This means that a concave fracture may be identified with apparent height reduction of less than 20%. Secondly, the ABQ method introduces the concept of differential diagnosis of short vertebral height. The method was developed after observing the characteristics of incident vertebral fractures and noting that new vertebral fractures always demonstrated radiological evidence of change at the vertebral endplate. This change usually exhibits subtle differences from those associated with non-fracture deformities, and it is the classification of these changes that forms the basis of the ABQ approach.
The algorithm shown in Fig. 4 serves only as a basic guideline. Observers need to be fully trained in the application of the method, and the algorithm should be applied with recourse to reference notes on differential diagnoses and an atlas of radiographic appearances. The method has been evaluated alongside other established methods in a preliminary analysis . Fewer fractures were identified using the ABQ method compared with other approaches, and most of the sources of disagreement were related to fractures identified by other methods that were identified by ABQ as non-fracture deformities or normal variation in vertebral shape.
Because there is no agreed gold standard, it is difficult to be certain of the validity of mild fractures, although age-adjusted BMD was lower in the women with ABQ fractures than in those with fractures identified by other methods, and was lower in women with vertebral fractures identified by all methods than in those with discrepant fractures. In addition, the distribution of vertebral fractures identified by ABQ closely matches the expected distribution of incident fractures. However, further studies are indicated to validate the ABQ approach in other study populations and to evaluate interobserver agreement for the method.
Identification of incident vertebral fractures
The identification of incident vertebral deformities can be problematic when radiographic technique is inconsistent, regardless of the technique used. Using quantitative approaches, for example, incident deformities are often defined as a 20% reduction in vertebral height , but this approach is complicated by differences in image magnification between visits. An approach has been developed in which serial quantitative measurements are corrected for differences in magnification between visits . This method corrects for the average difference in magnification for the whole study population. An alternative approach proposed by Jiang et al.  corrects the magnification differences for each individual, and separately for the thoracic and lumbar spine radiographs. This approach may improve the precision for identification of incident deformities, but adds to the time required to perform the morphometric analysis. Poor reproducibility for patient positioning and the use of different film or X-ray focal spot size can lead to variation in the magnification and obliquity of the vertebral bodies between serial radiographs, leading to marking error in quantitative morphometry and so-called reversal of deformity or un-fracturing at follow-up [51–52, 77, 107], particularly when the point-prevalence method is used. In addition, the reproducibility of morphometric measurements is likely to be worse in elderly patients or those with scoliosis or advanced degenerative disease, which make ideal radiographic positioning difficult.
In most studies and clinical trials, a visual assessment of vertebral fracture is also performed in addition to the quantitative analysis, and the visual analysis is generally considered to be the surrogate gold standard. The SQ approach  to the identification of incident fractures requires a vertebra to be classified one or more grades higher than at the previous diagnosis. This approach captures both incident fractures occurring in previously normal vertebrae and those due to worsening of existing fractures. It represents a reliable approach that is easy to apply. However, as with any method, the original or previous diagnosis of vertebral fracture may have to be revised, if review of serial radiographs demonstrates that the appearances at the previous examination were related to radiographic error. The SQ grading system may also be useful for predicting the severity of incident vertebral fractures. In a recent study, the determinants of various estimates of the reduction in vertebral size for incident fractures were examined , and an association was found between moderate or severe prevalent vertebral fractures identified by SQ, and the cumulative reduction in vertebral size for subjects with more than one incident fracture. This analysis was based on computer-simulated SQ grading of prevalent fractures that had been previously confirmed by qualitative radiological assessment.
For identification of a new incident fracture, the ABQ method requires evidence of change from normal appearances on the previous examination to the appearances of prevalent fracture according to the algorithmic criteria (Fig. 1). For worsening of a pre-existing fracture, there must be either (1) new fracture at the opposite endplate (in a vertebra with prevalent fracture at one endplate only) or (2) further reduction in vertebral height at the same endplate. This reduction should be approximately 4 mm or more, after making allowance for any magnification differences between visits.
The visual identification of incident vertebral fractures is generally less problematic than the identification of prevalent fractures because non-fracture deformities or normal variants either take a long time to develop or have been present since early adulthood. Acquisition of spinal radiographs, radionuclide bone scans  and measurements of stature in between scheduled study visits may be useful for validation of incident fractures in patients who report new back pain in clinical trials, or for monitoring of patients with osteoporosis. However, back pain, reduction in stature and increased uptake of tracer on radionuclide imaging are all relatively nonspecific assessments, so these may not be entirely conclusive.
New imaging technologies and the need for training in the identification of vertebral fracture
There is a general need for the development of better training materials for the identification of vertebral fracture, including a detailed radiological atlas to demonstrate the characteristics of all types of vertebral fracture and non-fracture deformity. The only comprehensive text that has focused exclusively on the identification of vertebral fracture in osteoporosis has not been updated since its publication almost a decade ago . A training package on the identification of osteoporotic vertebral fracture has recently been produced by the International Osteoporosis Foundation (IOF) and the European Society of Musculoskeletal Radiology. Based on the SQ method, this is available on compact disc . Alongside the recently published IOF resource document , this represents a good start in highlighting training needs.
The development of training materials is particularly important at this time, due to the evolution of alternative techniques for the imaging and identification of vertebral fractures. These include the acquisition of diagnostic vertebral images by dual-energy X-ray absorptiometry (DXA) scanners, and the automated detection of vertebral fracture by computer modeling of vertebral appearance and shape . Quantitative assessment of vertebral deformities from scan images acquired by X-ray absorptiometry has been shown to have moderate-to-good agreement with radiological assessment of spinal radiographs [72–75, 114–115], but the approach has been limited by problems associated with the quantitative methods in general, and by the relatively poor spatial resolution.
With recent improvements in image quality for X-ray absorptiometry, attention has focused on visual assessment of the scan images. This approach has been tested in preliminary studies with promising results [116–120], and because the radiation dose to the patient is minimal compared with conventional radiography , this may be useful for routine screening of patients referred for bone density assessments, or for monitoring of patients on therapy. Many applications also incorporate the option for quantitative measurements to supplement visual assessments. These may be used to confirm or evaluate the severity of fractures identified visually. Further evaluation of the visual approaches to the identification of vertebral fracture using X-ray absorptiometry is an important priority, given the increasing availability and competitive pricing of devices that offer the capability for vertebral imaging, particularly in the USA.
In general, more needs to be done to reach a full understanding of the characteristics of vertebral fracture, particularly in relation to the identification of mild fractures. Future research might include studies based on alternative imaging technologies such as CT and MRI. Key areas for research may include the determination of the true nature of vertebral deformities that exhibit wedging without obvious depression of the endplate, the concept of gradual fracture and the influence of spinal location and biomechanics. In the absence of a gold standard, any method used for the definition of vertebral fracture needs to be validated by testing the association with clinical outcomes of osteoporosis and vertebral fracture, such as the prediction of new vertebral and nonvertebral fractures, back pain, loss of body height, bone mineral density and distribution of incident vertebral fracture. This type of approach is required for the testing of modified approaches such as ABQ.