Introduction

Osteoporosis is the most common metabolic bone disease, and vertebral fractures are the most common type of osteoporotic fractures [1]. These fractures are associated with significant morbidity [2–7], mortality [8, 9], and high health-care costs. Given the ageing of populations, osteoporotic vertebral fractures are likely to become an increasingly important health issue. The costs of osteoporotic vertebral fractures were estimated to be €1.5 billion in Europe in 2010 [10] and US$ 1.1 billion in the US in 2005, and they are expected to have increased by more than 50% by 2025 [11].

The aetiology of osteoporotic vertebral fractures is believed to be multi-factorial, influenced by genetic and environmental factors [12, 13]. Osteoporosis is a disease characterised by low bone mass and micro-architectural deterioration of bone tissue, leading to enhanced bone fragility and a consequent increase in fracture risk [14]. Bone mineral density (BMD) and age are strongly predictive for most osteoporotic fractures, whereas other risk factors may vary according to fracture site [15]. Compared with non-vertebral fractures, which usually occur after a fall, it has been suggested that only ∼10–15% of osteoporotic vertebral fractures are preceded by a fall, with many resulting from low-grade trauma apparently derived from insignificant everyday activities [16].

Importantly, vertebral fractures are strong predictors of future osteoporotic fractures, including both non-vertebral and new vertebral fractures [17–20]. Vertebral fractures can be relatively asymptomatic in some cases, still, asymptomatic vertebral fractures remain strong predictors of subsequent risk of fractures and fracture-associated mortality [21]. The reason why radiological detection may be even more valuable. In clinical practice, therefore, prevalent osteoporotic vertebral fractures are considered as a strong indication for anti-osteoporotic treatment [22]. Yet, previous studies have shown that only one third of the patients with vertebral fractures come to clinical attention [23] and that vertebral fractures are commonly underreported in radiological practice [24–27]. The latter implies that applying standardised assessment methods of osteoporotic vertebral fractures might be beneficial to decrease reader subjectivity.

Currently, there is no gold standard for osteoporotic vertebral fracture diagnosis [28]. Several radiological scoring methods for osteoporotic vertebral fractures exist, each using different criteria for diagnosing and grading the fracture. Such grading definitions are currently under debate. This article will review different scoring methods for diagnosing osteoporotic vertebral fractures by discussing the benefits and limitations of the most commonly applied radiographic scoring methods. We will also discuss the role of alternative imaging techniques for assessing these fractures. In addition, this review will illustrate how prevalence of osteoporotic vertebral fractures is influenced by different scoring methods. Finally, application of scoring methods in research and patient care will be discussed.

Vertebral fracture assessment by radiography

Radiography is the standard imaging modality used for initially assessing vertebral fractures. Usually separate anteroposterior and lateral projections of the thoracic and lumbar spine are acquired, sometimes supplemented by additional views focused at the thoracolumbar junction. However, in the scientific research setting, occasionally only lateral radiographs are obtained. Radiographic capture is rapid, image quality is mostly high, and the radiation dose is relatively low, especially compared to CT. It is important that the spine is positioned parallel to the table to enable good assessment of vertebral endplates. Imperfect centering and collimation of the x-ray beam may, however, cause oblique projection and incorrect exposure, resulting in poor image quality [29]. Also, because the x-ray beam is conical, oblique projection is worst at the film areas furthest from the centre. This distortion may hinder correct appraisal of vertebral body shape and can in some situations wrongly suggest a biconcave shape [30]. Because of superimposition of the overlying shoulder girdle, the upper thoracic spinal region cannot be clearly visualised in many cases. The iliac wings of the pelvis can also exert a similar hindrance effect on images of the lower lumbar spine. In addition, superposition of the ribs and pulmonary vasculature on the thoracic vertebrae may occasionally confound the vertebral body margins on the image.

Scoring methods

Measurements of vertebral shape

The first published standardised assessment methods use quantitative morphometry (QM), which entails direct measurement of vertebral body shape. With six-point morphometry, points are placed in the superior and inferior endplates at the anterior, middle and posterior aspects of the vertebral body. For example, two of the more recent and commonly applied QM scoring methods are those described by Eastell-Melton [31] and McCloskey-Kanis [32]. In these methods anterior, central, and posterior vertebral body heights are first measured on a lateral radiograph, and ratios between these heights are calculated. These ratios are then used to classify vertebral fractures, using cutoff values based on standard deviation reductions from normal-population means derived from epidemiological studies (Fig 1). Depending on which of the three heights are diminished, the Eastell-Melton [31] method distinguishes three types of fractures (i.e. wedge, biconcavity, or compression), and the McCloskey-Kanis method further classifies the wedge type into anterior and posterior [32]. Using ratios instead of absolute heights is preferable, as anatomical structures farther away from the film may be falsely magnified, depending on the distance of the x-ray tube from the subject. Also, vertebral height is partly associated with a person’s body height [31]. It is considered essential to appraise these ratios relative to population reference data, as it has been shown that the derived vertebral height ratios are normally (Gaussian) distributed [32]. In addition, several of these methods relate the values to adjacent vertebra within the same individual, as each vertebra has a different size [33]. Although QM measurements appear more objective and reproducible than visual methods, they are more laborious and time-consuming to acquire. This is an important consideration for large-scale epidemiological research as well as for its implementation in clinical practice.

Fig. 1
figure 1

Six-point quantitative morphometry (QM) and semi-quantitative (SQ) method. P, posterior; C, central; A, anterior. a Normal thoracic vertebrae. b Mild wedge deformity of T8 and c severe wedge deformities of T12 and L1. d Mild wedge deformity of T6, moderate wedge deformity of T7 and moderate biconcave deformities of T8, T9 and T10. e Crush deformity of L3 in an individual with confirmed history of spinal trauma; severe vertebral body fracture is seen with slight bulging of the posterior vertebral body margin. This fracture morphology is usually traumatic

Full size image

Semi-quantitative (SQ) method

Currently, the most widely used standardised grading method is the visual SQ method (i.e. according to Genant [34]). It is commonly applied as a surrogate gold standard in research [35]. Vertebral fractures are SQ graded by trained readers, who estimate the percentage of height and/or area reduction subjectively, without direct measurement. Vertebral deformities are graded according to shape and severity (Fig 1). The deformity’s shape is classified on the basis of anterior height loss (i.e. wedge), middle height loss (i.e. biconcave), or posterior and anterior height loss (i.e. crush). Severity of vertebral deformities is graded according to the extent of height and area loss, as mildly deformed, moderately deformed, and severely deformed (Table 1). Next, a spinal fracture index (SFI) can be calculated by summing the individual vertebral body grades. Genant et al. [34] have noted that height and area loss determined by morphometry alone fails to capture several other important characteristics of vertebral fracture, including endplate deformity, buckling of cortices, lack of parallelism of endplates and loss of vertical continuity of vertebral morphology.

Table 1 Semi-quantitative grading of severity of vertebral fractures according to Genant [34]
Full size table

Algorithm-based qualitative (ABQ) method

The more recent ABQ method by Jiang et al. diagnoses osteoporotic vertebral fractures on the basis of endplate depression, regardless of vertebral height reduction (Fig 2) [36]. The key assumption is that the endplate is always deformed in vertebral fractures, and therefore endplate depression has perfect specificity for vertebral fracture. The fracture occurs primarily at the centre of the endplate, and thus it follows that the endplate is centrally depressed in all types of vertebral fracture (i.e. concave, wedge, and crush). In addition, vertebral height may appear to be decreased as a result of oblique image projection, certain diseases, and anatomical variants that can mimic vertebral fractures. To deal with this misclassification, ABQ uses a flowchart to systematically rule out non-fracture deformities by examining certain radiological features. A skilled ABQ reader is needed to differentiate accurately between vertebral fractures and non-fracture deformities. If images are of poor quality, vertebral fractures with subtle endplate changes can easily be missed.

Fig. 2
figure 2

Algorithm-based qualitative (ABQ) method. a Superior endplate depression of T11. b Inferior endplate depression of L3. c Superior and inferior endplate depression of L3

Full size image

Non-radiographic imaging techniques

Vertebral fractures can also be detected and graded on radiological imaging investigations other than conventional radiography. Although developed for radiography, which remains the most commonly used technique for vertebral fracture assessment, the scoring methods described above can also be applied to other radiological techniques.

Dual-energy X-ray absorptiometry (DXA) at the lumbar spine and hip to measure BMD is a routine investigation in osteoporosis, because BMD constitutes one of the strongest predictors of future fracture [37, 38]. Several studies have shown that the risk of incident vertebral fractures doubles for each SD reduction of lumbar spine BMD [37, 39]. Note, however, that many fractures occur when BMD is in the osteopenic or normal range of values [37]. In addition to artefacts (i.e. osteophytes, calcifications), BMD measurements of the lumbar spine may be falsely elevated in the presence of vertebral fractures because impacted fracture or fracture healing result in higher areal BMD [40]. The World Health Organisation’s FRAX® tool can be used to calculate the 10-year fracture risk for individual patients, using validated risk factors (with or without femoral neck DXA BMD) [41]. The clinical risk factors used in the calculation include age, gender, height, weight, previous low trauma fracture (including vertebral fractures), parental hip fracture, oral glucocorticoid therapy, rheumatoid arthritis, current smoking, alcohol consumption of more than three units per day, and secondary causes of osteoporosis. In recent years, the use of densitometers has extended beyond BMD assessment to identify vertebral fractures from DXA images. The so-called lateral densitometric vertebral fracture assessment (VFA) is gaining popularity because of the considerable improvement in image resolution, and is currently offering complementary and independent information about fracture risk (Fig 3a) [42]. The implementation of fan-beam technology in the DXA devices has allowed capturing the whole spine in one image, with virtually no divergent beam issues due to parallax effect. Also, VFA has a low radiation dose, making it very suitable for screening in the clinical setting. For those more recently introduced DXA devices with a rotating C-arm, the lateral examination can even be done without moving the patient from the supine position used for the BMD measurements. In addition, the rotating C-arm may enable three-dimensional DXA scans, allowing the direct measurement of geometric parameters of the vertebrae [43]. It has been demonstrated that image quality can differ greatly between types of densitometers [44]. Still, radiographs have superior spatial resolution, which facilitates identification of more subtle abnormalities [45].

Fig. 3
figure 3

Non-radiographic imaging modalities. a Lateral VFA shows a biconcave deformity of T12. b Three-dimensional visualisation of the thoracolumbar spine with CT. c Midline sagittal CT reformation shows an osteoporotic vertebral fracture of L1 (arrow), in addition to degenerative changes and endplate irregularities at multiple levels. d MRI: Sagittal short tau inversion recovery (STIR) sequence shows endplate deformity, height reduction, and bone marrow oedema at the T11, L3, and L4 levels (arrows), indicating recent osteoporotic vertebral fractures

Full size image

Unlike two-dimensional radiography, computed tomography (CT) and magnetic resonance imaging (MRI) offer three-dimensional visualisation of the vertebra (Fig 3b). In addition, CT and MRI can differentiate between old and recent vertebral fractures by assessing the integrity and shape of the cortical margins (Fig 3c). MRI does not use ionising radiation and can demonstrate bone marrow oedema, which distinguishes recent from old fractures (Fig 3d). The images produced by CT have a much higher spatial resolution than those of MRI and DXA. It has been shown that sagittal reformations need to be used to demonstrate vertebral fractures on CT [21]. Despite the introduction of several dose reduction techniques, the ionising radiation exposure of CT is still substantial, which is a major disadvantage of the imaging technique, especially in the research setting [46]. CT scout images may also be used for assessing vertebral fractures [47]. Novel quantitative and high-resolution CT techniques are being developed to enable separate analysis of trabecular and cortical bone compartments [48, 49]. High-resolution MRI can be used to assess bone trabeculation in the extremities, but this application of MRI at the spine is more challenging [50]. Drawbacks of MRI are, however, the long imaging time and high costs. Hence, MRI is usually used for other conditions that specifically require MRI, such as spinal cord compression and paraspinal soft tissue abnormalities [51, 52]. If a malignant aetiology for vertebral fracture is suspected, then MRI or CT have advantages over conventional radiography [53, 54].

Image analysis software

Automated image analysis software packages (e.g. SpineAnalyzer®, Optasia Medical Ltd, Cheadle, UK [55]) have been developed to facilitate efficient and standardised vertebral fracture scoring of large data sets [56, 57]. The software can handle lateral spine radiographs, VFA or CT scout films. So far, software packages have been dedicated to recording QM and SQ.

Instead of having to manually define vertebral contours and height, users only need to place one point in the centre of each vertebra to define the vertebral level. Next, vertebral contours are identified by the software using automated segmentation techniques, and vertebral height is measured (Fig 4a). A table with percentage height loss and presence or absence of deformity per vertebral level based on QM and SQ is generated (Fig 4b), after which the data can be exported to a database. The data that can be saved include the exact coordinates of the endplates anteriorly, centrally, and posteriorly. This information can be valuable in the research setting, where analysis of crude vertebral heights could be meaningful to explore optimisation of current vertebral fracture definitions. Relatively inexperienced users are deemed to be capable of using the software after a brief training [58]. These software packages can be further improved by incorporating population reference data for QM and by reducing the need for manual adjustment of vertebral contour definition, a procedure that is still required routinely. Automated VFA packages are nowadays integrated in DXA equipment.

Fig. 4
figure 4

Image analysis software (examinations with SpineAnalyzer®). a Automated contour detection. b Automated analysis of shape abnormality

Full size image

Differential diagnosis

There are a number of differential diagnoses that have to be considered in individuals with vertebral deformities [59]. In the 1960s, Hurxthal described several criteria for vertebral measurements [30]. Basically, all artefacts that can interfere with vertebral height measurement should be considered by the reader. Hook-shaped protuberances at the posterosuperior (called uncinate process by some) and posteroinferior borders of the vertebrae, any Schmorl’s nodes, and osteophytes should be excluded from vertebral height measurement. Six-point morphometry alone is unable to distinguish fractures and vertebral deformities due to other causes. In the description of the SQ method [34], several conditions that can mimic vertebral fracture such as scoliosis and vertebral body remodelling due to degenerative disc disease are listed. Moreover, the ABQ method introduced a very comprehensive decision-making algorithm, which provides a guideline for systematically assessing various non-fracture deformities [35].

Normal anatomical variation in the shape of individual vertebrae and of the spinal column as a whole should be taken into account. Viewed laterally, the spine has a natural curvature. Vertebrae in the mid-thoracic region are more wedge-shaped, causing a mild kyphosis. Lumbar vertebrae tend to be biconcave rather than wedge-shaped, and this gives rise to a normal lordotic curve, because of the relatively shorter posterior height. In addition, some adults have vertebrae that have longstanding short anterior height in developmentally small thoracic vertebrae [35]. Therefore, the normal spine shape must be known if SQ and ABQ readers are to avoid false-positive fracture diagnosis. Some QM methods that compare ratios to population reference data may classify short vertebral height correctly as non-fracture. In addition, the anterior vertebral wedge angle has been shown to increase concurrently with age-related degenerative change. Degenerative signs include degenerative disc disease, osteophytes, and endplate irregularities (Fig 5a). The ABQ method incorporates additional differential diagnoses including previous (e.g. during childhood) fractures, metabolic diseases (e.g. osteomalacia), and developmental anomalies, including anterior step deformity (depressions in the anterior portion of the vertebral endplate) in thoracic vertebrae, balloon disc, or cupid’s bow with deep inferior endplates in the lumbar vertebrae [35].

Fig. 5
figure 5

Conditions that mimic vertebral fractures. a Degenerative changes. Very mild anterior vertebral wedging of two mid-thoracic vertebral bodies is seen along with mild spondylotic changes at the anterior vertebral margins. Note that the endplates are intact and only show mild degenerative irregularities. b Scheuermann’s disease. In addition to marked endplate irregularity, mild anterior wedging of multiple midthoracic vertebrae is seen, resulting in increased thoracic kyphosis

Full size image

A frequent condition that resembles vertebral fractures is Scheuermann’s disease. Radiographic criteria of Scheuermann’s disease are a thoracic kyphosis greater than 45° and at least three adjacent wedge-shaped vertebral bodies of 5° or more [60, 61]. Vertebral wedging is frequently associated with endplate irregularity and Schmorl’s nodes. Elongated vertebrae and disc space narrowing can also be found in Scheuermann’s disease (Fig 5b). This vertebral wedging may be mistaken for mild vertebral fractures by QM or SQ, and Schmorl’s nodes may mimic endplate depression. Occasionally, but most importantly, osteoporotic vertebral fractures need to be distinguished from those resulting from malignant aetiologies, such as metastases (most commonly of primary breast, kidney, prostate, or lung neoplasms), multiple myeloma, or primary bone tumours [53]. In the majority of osteoporotic vertebral fracture cases, posterior margins of vertebrae maintain a straight or concave shape, whereas in malignant aetiologies the posterior margin is often convex.

Traumatic fractures should also be distinguished from the typically low-grade trauma osteoporotic fractures. Posterior height loss was regarded as posterior wedge in the McCloskey-Kanis method [2] and as crush deformity in SQ [4]. However, fractures involving the posterior vertebral part are typically attributable to malignancy or high-energy trauma [62] rather than to low trauma, which is most common in osteoporosis.

Inter- and intraobserver agreement of scoring methods

Inter- and intraobserver agreement seems to vary considerably within and between scoring methods. However, agreement is about precision of a study and may not necessarily relate to its validity. Nevertheless, there are several aspects that need to be considered when comparing methods. Point-placement in SQ and QM is said to be somewhat subjective, and hence inclined to influence fracture discrimination. This is particularly pertinent in the presence of borderline deformities. Also, reproducibility of SQ and ABQ scoring may to some extent depend on the reader’s training and experience [34, 44, 63].

Kim et al. have evaluated intra- and inter-reader agreement of a semi-automated quantitative morphometry software algorithm on lateral CT scout views [58]. They found intraclass correlation coefficients of 0.96 to 0.98 for vertebral heights, while kappa statistics were 0.59 to 0.69 for intra-reader and 0.67 for inter-reader agreement. Agreement for vertebral fracture classification was worse than agreement for height measurements. This was explained by the small variation of height measurement around fracture classification thresholds. Such clinically insignificant variation in height measurement can actually lead to two different fracture classifications in a considerable number of cases. Furthermore, kappa scores did not improve much even when the fracture definition was changed to include only moderate and severe fractures (i.e. deformity ≥25%).

Several publications have evaluated the inter- and intraobserver agreement of SQ alone and compared with QM methods [34, 36, 63–65]. Kappa statistics reported for SQ interobserver agreement ranged from 0.51 to 0.80 and from 0.76 to 0.93 for intraobserver agreement, respectively. The kappas for agreement between SQ and several QM methods have been reported to be lower, ranging from 0.23 to 0.59, with some improvement when fracture definition included only moderate and severe fractures (i.e. deformity ≥25%). Obviously, the agreement between different QM methods will depend on the fracture threshold chosen. Recently, semi-automated QM reading using Genant’s criteria by a non-radiologist was compared with conventional SQ grading performed by experienced radiologists, finding a kappa for agreement of 0.78 [66].

Ferrar et al. have examined interobserver agreement for ABQ diagnosis of prevalent vertebral fracture in approximately 200 elderly women, finding kappa statistics of 0.74 for interreader agreement [63]. In general, the ABQ method has displayed low to moderate concordance with other methods. Jiang et al. found kappa statistics between 0.39 and 0.64 comparing ABQ with the QM methods developed by Eastell-Melton and McCloskey [36]. Also, ABQ has been compared with SQ observing kappa statistics of 0.30 to 0.58 [67].

Influence of scoring methods on vertebral fracture prevalence and incidence

All methods assess osteoporotic vertebral fractures with different criteria, which results in different estimates of the prevalence of the disease [36, 68]. For example, QM and SQ would not diagnose vertebral fractures in the case of endplate depression without reduced vertebral height (Fig 6a). Conversely, ABQ would not diagnose a QM-based vertebral fracture with reduced height but intact endplates (Fig 6b). In general, SQ would yield a higher number of fractures than when applying QM, asserting that SQ would be more sensitive particularly for the detection of mild deformities [68]. However, Melton et al. have demonstrated that depending on the morphometric definition used, the prevalence of vertebral fractures ranged from 3 to 90% in their study [69]. Of all the methods, the ABQ reading results in the lowest estimations of vertebral fracture prevalence. The question remains if the higher estimates from other methods are actually due to false-positive classification of non-fracture deformities [36].

Fig. 6
figure 6

Discrepancy between scoring methods. a Vertebral fractures diagnosed with ABQ L2 and L3 based on endplate depression, classified as normal with QM because of normal vertebral height. b Vertebral fracture of T12 diagnosed with QM, based on height reduction, classified as normal with ABQ because of intact endplates

Full size image

Research implications

Misclassification of vertebral fractures may result in non-differential information bias, leading to dilution of observed effects. As a consequence, true associations in studies may go undetected. This ascertainment bias can occur both in case control studies (including clinical trials investigating drug effects) and in observational studies. Therefore, scoring methods should procure the optimal classification of true vertebral fractures.

Large-scale application of standardised scoring can be difficult, with purely morphometric approaches being laborious, while the other methods will require thorough training of observers. In very large studies, especially population-based studies with an expected low prevalence of vertebral fracture cases, a technician triage system may reduce the work burden of scoring thousands of radiographs [65, 68, 70]. First, trained research technicians can triage radiographs as definite vertebral fracture, uncertain fracture, or definite normal. Finally, an expert reader may review the difficult cases and confirm vertebral fractures. Also, a stepwise evaluation process combining morphometry and qualitative assessment represents a possible procedure to achieve a final diagnosis of vertebral osteoporosis [71].

To date, there have been few large-scale comparisons of vertebral fracture assessment methods. We are currently applying both ABQ and software-assisted QM methods to radiographs from the Rotterdam Study (all image examples included in the present article originate from this study). This study is a prospective population-based cohort that has been studying disease and disability in more than 15,000 individuals aged 45 and over since 1990 [72]. Within the on-going research program, radiographs of approximately 11,000 participants are available, with a follow-up duration of maximally 15 years. An aim of the study is to compare the methods applied for identifying vertebral fractures. In addition, data on numerous outcomes and risk factors are available, including a comprehensive assessment of clinical fractures, BMD, and genetic determinations.

Clinical implications

It is estimated that only about one third of all vertebral fractures come to clinical attention [23]. However, assessment of vertebral fracture status, in addition to BMD, provides practical and relevant clinical information to aid the prediction of subsequent fracture risk [73]. Symptomatic and non-symptomatic vertebral fractures are both associated with decreased quality of life [2–7] and increased mortality risk [8, 9]. In the case of vertebral fracture, pharmacologic therapy is considered necessary to prevent the occurrence of future osteoporotic fractures [74]. However, as all interventions have costs and potential side effects, correct assessment of vertebral fractures is of utmost importance. Over- and underdiagnosis can have major consequences, particularly at the population level. Misdiagnosis of osteoporotic vertebral fractures will result in under- or overtreatment of patients and subsequently unnecessary costs, increased morbidity, and higher mortality.

Current definitions used by vertebral fracture scoring methods seem to be based on arbitrary cutoffs. At most, some QM methods have been established by deriving standard deviations from measurements in a sample of healthy individuals, but variation from the mean is not necessarily abnormal. The classifications show association with osteoporosis-related outcomes such as BMD and the risk of future non-vertebral and new vertebral fractures [17, 36, 65, 70, 75]. Yet, from a more clinically oriented perspective the definition of vertebral fractures should be based on cutoffs that were defined based on their ability to predict relevant outcomes, such as future osteoporotic fractures. This will require the optimal combination of true- and false-positive ratios that yield the greatest expected utility for the patient at acceptable costs to society. For optimal appraisal of future osteoporotic fracture risk it might prove necessary to refine currently available vertebral fracture scoring after comprehensive comparative studies and integrate more quantitative information that can be derived from imaging, for example three-dimensional reconstruction of vertebral shape, BMD, and measurements able to appreciate the integrity of the endplates and microarchitecture.

In conclusion, standardised and accurate scoring methods for osteoporotic vertebral fractures are desirable. There are several radiological scoring methods for osteoporotic vertebral fractures, which can be characterised as quantitative, qualitative, or semi-quantitative. Also, these standardised scoring methods can be implemented for different imaging modalities. The scoring methods each use different definitions for the diagnosis of vertebral fracture and the classification of severity. Such differences have implications for patient care and scientific research. Accurate diagnosis of vertebral fractures and differentiation from non-fracture deformities is an important aspect that depends on the expertise of the reader. Future evaluation of the concordance between methods will allow establishing their benefits and limitations, and most importantly, optimise their effectiveness for application in clinical and research scenarios.