Presented in part in abstract form at the 2nd joint meeting of the American Society for Bone and Mineral Research and the International Bone and Mineral Society, San Francisco, December 1998.
Morphometric X-Ray Absorptiometry and Morphometric Radiography of the Spine: A Comparison of Prevalent Vertebral Deformity Identification†
Article first published online: 18 FEB 2010
Copyright © 2000 ASBMR
Journal of Bone and Mineral Research
Volume 15, Issue 3, pages 564–574, March 2000
How to Cite
Rea, J. A., Chen, M. B., Li, J., Blake, G. M., Steiger, P., Genant, H. K. and Fogelman, I. (2000), Morphometric X-Ray Absorptiometry and Morphometric Radiography of the Spine: A Comparison of Prevalent Vertebral Deformity Identification. J Bone Miner Res, 15: 564–574. doi: 10.1359/jbmr.2000.15.3.564
- Issue published online: 18 FEB 2010
- Article first published online: 18 FEB 2010
- Manuscript Accepted: 9 NOV 1999
- Manuscript Revised: 3 NOV 1999
- Manuscript Received: 11 JUN 1999
- dual X-ray absorptiometry;
- morphometric X-ray absorptiometry;
- vertebral fractures
Prevalent vertebral deformities are associated with a substantially increased risk of subsequent vertebral and nonvertebral fractures. Knowledge of vertebral fracture status is an important component in the prediction of further fractures in patients with osteoporosis. This study reports a comparison of the quantitative identification of vertebral deformities on morphometric X-ray absorptiometry (MXA) scans and conventional radiographs (MRX) in 161 postmenopausal women (mean age ± SD, 64 ± 7.1 years) recruited from patients referred by their family doctor for bone density measurement (n = 119) and osteoporotic subjects with known vertebral deformities attending an osteoporosis clinic (n = 42). Each subject had MXA scans and MRXs of the thoracolumbar spine, to image the vertebrae from T4–L4, at a single visit. The scans and radiographs were analyzed by two trained observers using six points to quantify the shape of each vertebral body. From these points, three vertebral heights were measured: anterior, middle, and posterior. Vertebral deformities were identified using the algorithms proposed by Eastell and by McCloskey. Generally good to excellent agreement (per vertebra, κ = 0.87–0.93; per subject, κ = 0.81–0.91) was observed between the two algorithms used for quantitative vertebral deformity identification using MXA or MRX More moderate agreement (per vertebra, κ = 0.70–0.79; per subject, κ = 0.67–0.75) was seen when comparing the same algorithm between MXA and MRX Agreement between MXA and MRX for the McCloskey algorithm was better than for the Eastell algorithm, largely because of the lower number of false positives produced by the McCloskey methodology. Deformity identification by MXA was limited because of poor image quality, primarily in the upper thoracic spine. One in six MRX deformities were missed by MXA as they occurred in vertebrae not visualized sufficiently for analysis on the MXA scans. Deformity identification was poorer in the upper thoracic spine in analyzable vertebrae with a sensitivity of 50.0% for MXA in terms of MRX using the Eastell algorithm for the vertebral levels T4–T7, compared with 80.6% for L1–L4. MXA proved to be more effective at identifying moderate to severe MRX deformities producing a sensitivity of 22.0% for MXA in terms of identifying MRX grade 1 deformities using the Eastell algorithm, compared with 81.6% for grade 2 deformities. Although MXA image quality is inferior to that of conventional radiographs, MXA has distinct advantages such as a substantially reduced effective dose to the patient and acquisition of a single image of the spine. MXA is a potentially useful, relatively fast, low-radiation technique to identify prevalent vertebral deformities, particularly moderate to severe deformities in the middle/lower thoracic and lumbar spine, in conjunction with morphometric radiography in some patients.
Since its introduction in the 1960s morphometric radiography (MRX), the measurement of vertebral body height on conventional lateral radiographs of the thoracolumbar spine, has become an established tool in the identification of vertebral deformities.(1,2) Morphometric approaches to quantify vertebral body shape are not generally used to identify deformities in the individual osteoporotic patient, but rather in epidemiological studies and clinical trials of novel osteoporotic therapies, in which objectivity and reproducibility over long periods of time and a number of different observers is crucial.(3–6) Vertebral deformities are a classic hallmark of established osteoporosis, resulting in increased back pain and disability(7) and a substantially increased risk of subsequent vertebral and nonvertebral fracture.(7–9)
Numerous methods have been proposed to quantify vertebral body shape and to identify deformities.(10–16) Each may be used individually or in conjunction with various qualitative or semiquantitative (SQ) approaches to facilitate a differential diagnosis of deformity etiology.(17,18) Currently, there is no consensus agreement on a gold standard definition of vertebral deformity. Although conventional radiography frequently produces high-resolution images of the spine with little noise, this technique is adversely affected by inherent problems. These include magnification and distortion of the image associated with the use of a cone-beam and the exposure of the subject to a relatively large radiation dose.(19) Recently, an alternative method of acquiring the lateral images of the spine required for vertebral morphometry has been developed, termed morphometric X-ray absorptiometry (MXA), which utilizes dual X-ray absorptiometry (DXA) machines.(20–24) MXA scan image quality is inferior to that of the conventional radiograph; however, MXA has several advantages when compared with conventional radiography, which include a significantly lower radiation dose to the patient, acquisition of a single image of the whole spine, and straightforward supine patient positioning.(25)
This study reports a direct comparison of prevalent vertebral deformity identification using quantitative morphometric techniques on the MXA scans and conventional radiographic images acquired on a group of postmenopausal women, ranging from healthy individuals with “normal” bone density to subjects with severe osteoporosis and multiple vertebral deformities. The study was approved by the Guy's Hospital Research Ethics Committee and written informed consent was obtained from all participants.
MATERIALS AND METHODS
Subjects were recruited from postmenopausal women referred by their general practitioner (GP) to the Osteoporosis Unit, Guy's Hospital, London, U.K. for osteopenia screening and from subjects attending the Metabolic Bone Clinic at Guy's. The clinic subjects were selected because they exhibited low bone density, with a T score less than or equal to −2 at the lumbar spine (lumbar spine bone mineral density [BMDLS]; posteroanterior [PA], L1–L4) and/or hip (total hip BMD [BMDTH]), and had at least one vertebral deformity previously diagnosed by radiologists at the patients' local hospitals. Subjects were recruited from these two sources to provide adequate numbers of normal vertebrae for reference data calculation (GP patients) and a relatively larger number of vertebral deformities for evaluation (clinic patients). Potential recruits were excluded from the study if they were obviously obese or when moderate to severe scoliosis was apparent on BMD scans or mentioned on previous referrals.
An MXA study of each subjects' spine was acquired using a QDR-4500A DXA machine (Hologic, Inc., Bedford, MA, U.S.A.). The procedure has been described in detail elsewhere.(23) Briefly, each subject was positioned supine with their legs raised and arms lifted above shoulder level. A PA centerline scan of the spine was then acquired to visualize spinal anatomy, identify the centerline of the vertebral column and set the start point for the subsequent lateral scan. Data from this centerline scan was used by the machine to maintain a constant distance between the spine and X-ray tube during the lateral scan, thus correcting for the magnification effect of the fan-beam. Two lateral scans of the vertebrae from L4 to T4 were then performed using the single energy (SE) and the dual-energy high-definition (HD) MXA scan modes. These scan modes have been identified previously as the optimal combination of those available on this DXA machine.(23) BMD scans of the lumbar spine and left hip also were acquired if they had not been obtained within the preceding 12 months.
On the same day anteroposterior (AP) and lateral lumbar and thoracic radiographs were acquired on each subject in a general X-ray room, using an over-couch tube and a Bucky technique, following a standardized protocol. For the lateral views subjects were positioned in the decubitus position lying on their left side with knees and hips flexed. Tube-to-film distance was set at 105 cm and films were centered at T7 and L3 for the thoracic and lumbar views, respectively. A breathing technique was used to blur the overlying ribs in the thoracic views. No extra positioning aids were used in subjects exhibiting any degree of scoliosis.
The MXA scans were analyzed by a trained operator (J.A.R.), ideally from L4 to T4, using the standard semiautomated analysis software supplied by the manufacturer.(23) Six points were used to describe the shape of each vertebral body visualized sufficiently for analysis on either the SE or HD scan images, which were viewed together on the analysis screen along with the PA centerline scan. From these points three vertebral heights were measured: anterior (Ha), middle (Hm), and posterior (Hp) for each vertebral body. If 5 or more of the 13 required vertebrae could not be visualized sufficiently for analysis, then the MXA scans for that subject were deemed unusable.
Vertebral levels were marked on the conventional radiographs after comparison with the appropriate MXA image to ensure consistency in vertebral level labeling. The radiographs were analyzed by a radiologist at the University of California, San Francisco (UCSF), CA, U.S.A. (M.B.C.), using a backlit X-Y digitizing table connected to a PC. The x,y coordinates for each point were recorded on an electronic grid, with a resolution of 0.1 mm, using the crosshairs of a cursor. Again, six points were used to describe the shape of each vertebral body following a standardized protocol that was as consistent as possible with the corresponding point placement protocol used to analyze the MXA scans.(26) Three vertebral heights (Ha, Hm, and Hp) were measured for each vertebral body using the coordinates stored in the PC.
Quantitative vertebral deformity assessment
Vertebral deformity assessment of both the MXA scans and the MRX was performed using two different algorithms:
Eastell algorithm. This method was based on the method originally proposed by Riggs, which was further developed by Eastell.(10,15) Four simple ratios of the measured vertebral body heights were calculated: the wedge ratio (Ha/Hp), the midwedge ratio (Hm/Hp), and the two crush ratios (I, Hpi/Hpi–1, and II, Hpi/Hpi+1, in which Hpi+1 and Hpi–1 refer to the Hp of the vertebrae immediately above and below the target vertebra). A vertebra was classified as deformed when at least one ratio fell below a deformity threshold, which was set at more than 3 SD below the normal reference mean for that ratio at that vertebral level. Each deformity was classified according to its severity—when a deformity was identified by one or more ratios greater than 3 but less than 4 SD below the reference mean, it was described as a grade 1 deformity, and when one or more ratios were 4 SD or more below the reference mean, then the deformity was classified as a grade 2 deformity. Deformities were categorized to type as wedge, midwedge, or crush, depending on the ratios that identified the vertebra as deformed.
McCloskey algorithm. This method was developed by McCloskey et al. in 1993.(16,27) Five ratios were calculated: the wedge (Ha/Hp) and midwedge (Hm/Hp) ratios, the same ratios as for the Eastell algorithm, plus three ratios using the predicted posterior height, Hpp (calculated using data from up to four vertebral bodies adjacent to the target vertebra)—Ha/Hpp, Hm/Hpp, and Hp/Hpp. A vertebral body was considered deformed when ≥2 of these ratios fell below a set deformity threshold, which as in the Eastell algorithm was the appropriate reference mean less 3 SD. Three types of deformity were identified by different combinations of ratios falling below the deformity threshold: an anterior wedge deformity when Ha/Hp and Ha/Hpp < (mean Ha/Hp − 3 SD); an end plate deformity when Hm/Hp and Hm/Hpp < (mean Hm/Hp − 3 SD); and a crush deformity when Hp/Hpp < (mean Hp/Hpp − 3 SD) and Ha/Hpp < (mean Ha/Hp − 3 SD). This was a slightly adjusted version of the original algorithm because it excluded the criteria proposed to define posterior wedge deformities.(28)
Reference data of vertebral dimensions
Reference data of normal vertebral dimensions for both algorithms was specific to the acquisition technique and was derived from within the study subjects. Mean, SD, and deformity thresholds, that is, mean less 3 and 4 SD for the appropriate ratios at each vertebral level, were calculated from the vertebral heights measured in a subset of 100 women (mean age, 63 ± 6.9 years; range 51–79 years) who participated in this study. These subjects exhibited no vertebral deformities (grade ≥ 1, a minimum reduction of approximately 20–25% in anterior, middle, and/or posterior vertebral height compared with adjacent vertebra[e]) according to SQ reading of their radiographs by an experienced radiologist (J.L.).(18) Subjects were not excluded from reference range calculations as a result of back pain, degenerative disease, or evidence of metabolic bone disease such as osteoporosis.
Agreement between the vertebral heights measured by MXA and MRX for subjects with and without vertebral deformities (according to the SQ result) was evaluated using both correlation and Bland-Altman plots.(29) Differences between the vertebral dimensions measured by MXA and MRX were described by calculating the mean difference of the vertebral heights and vertebral height ratios calculated for the Eastell algorithm measured by each of the techniques.
Agreement between MXA and MRX in the designation of individual vertebrae and subjects as “normal” or “deformed” was described using percentage agreement and κ-scores.(30) Comparisons were made using a binary system designating each vertebra or subjects as undeformed, normal for the purposes of this study, or deformed and did not take into account the different types or varying severity of the deformities identified. The number and location of the vertebrae visualized sufficiently for assessment on the MXA and conventional radiographic images from each subject were noted and each vertebral body was classified as normal or deformed by each vertebral deformity algorithm. Each subject was then given an overall classification according to his or her individual vertebra results. If all vertebrae in a single subject were classed as normal, then the individual was classified as normal. However, if one or more deformed vertebrae were identified then a subject was classified as deformed.
A κ-score is a single omnibus index that measures both positive and negative agreement that occurs above and beyond that expected by chance and may also be termed “chance-corrected proportional agreement”:
where po is the proportion of observations on which the vertebral deformity assessment methods or observers agree and pe is the proportion of units for which agreement is expected to occur by chance. Strength of agreement may be evaluated using arbitrary “benchmarks”, with moderate agreement requiring a κ > 0.4, substantial or good agreement requiring a κ > 0.6, and an almost perfect or excellent agreement requiring a κ > 0.8.(31)
Sensitivity and specificity of vertebral deformity identification by MXA was calculated in relation to the MRX results. Negative and positive predictive value (NPV and PPV) for MXA in respect to MRX was calculated to evaluate the proportion of false-positive and false-negative observations. When the Eastell and McCloskey algorithms were compared directly with each other using the same image, neither algorithm was treated as a gold standard for vertebral deformity identification. Therefore sensitivity and specificity, NPV, and PPV were replaced with indices of positive (Ppos) and negative (Pneg) agreement.(32,33)
One hundred and sixty-one postmenopausal women (mean age, 64 ± 7.1 years [SD]; range, 49–81 years) were recruited to take part in the study. One hundred and nineteen subjects were recruited from GP patients referred to the unit and 42 from patients attending the Metabolic Bone Clinic, who were known to have both low BMD (T score, mean ± SD; BMDLS, − 2.82 ± 0.90; and BMDTH, − 2.38 ± 1.01) and at least one vertebral deformity each. Single factor analysis of variance (ANOVA) found no statistically significant differences in age, height, weight, or body mass index (BMI, weight [kg]/height [m2]) between the subjects recruited from the two sources.
Comparison of vertebral dimensions measured by MXA and MRX
Excellent correlation (r = 0.96–0.97 for Ha, Hm, and Hp) was observed in the vertebral heights measured in normal subjects (Fig. 1A). The relationship between the vertebral heights measured by MXA and MRX in normal subjects also is illustrated by a Bland-Altman plot of anterior vertebral heights in Fig. 1C. The slope seen in the plots was explained by the cone-beam-induced magnification factor inherent in conventional radiography. Correlation also was good for subjects with vertebral deformities (r = 0.90–0.95 for Ha, Hm, and Hp) (Fig. 1B). A Bland-Altman plot for subjects with at least one vertebral deformity showed a larger number of outliers than was seen for normal subjects (Fig. 1D).
The relationship between the heights measured by MXA and by MRX varied according to vertebral level and the particular height in question (Fig. 2). Vertebral heights measured by MXA were on average 22.7% (6.5 mm), 24.1% (6.7 mm), and 24.8% (7.4 mm) smaller than the relevant MRX values for Ha, Hm, and Hp, respectively. The Hm and Hp heights measured by MXA tended to be smaller in relation to the equivalent heights measured by MRX than Ha heights, with an increase in this discrepancy moving down the vertebral column from T4 to L4. This led to a systematic difference, particularly in the wedge ratio and to a lesser extent the midwedge ratio, calculated for each vertebra using the vertebral heights measured by MXA and MRX (Fig. 3). Although the crush ratios showed generally less fluctuation between the MXA and MRX values, more variation was seen in the upper thoracic levels T4–T6.
Vertebral deformity identification
MRX analyzed 99.5% (n = 2082) of the available 2093 vertebrae from T4 to L4 (Table 1). A smaller proportion, 94.4% (n = 1,975), of the available vertebrae could be visualized sufficiently for analysis on the MXA scans. The vast majority of vertebrae that could not be visualized sufficiently for point placement on the MXA scans were located in the upper thoracic spine (T4–T6) where image quality was poor. 71.4% of MXA scans were analyzed up to and including T4, compared with 95.0% up to T7. Nineteen hundred sixty-seven vertebrae could be analyzed on both the MXA and the MRX images and are subsequently referred to as “analyzable” vertebrae.
Prevalence of vertebral deformities varied according to vertebral level (Fig. 4). All methods showed a peak in deformity identification around the thoracolumbar junction (T12/L1) and to some extent in the midthoracic region (T7/T9). MXA methods identified a noticeably smaller number of deformities in the upper thoracic vertebrae (T4–T6) compared with MRX as a result of the number of unanalyzable vertebrae in this region on the MXA scans.
MXA versus MRX—Eastell algorithm
Per vertebra: MRX identified 87.9% of the 2093 vertebrae as normal, 11.6% as deformed, and 0.5% were not visualized sufficiently for analysis, whereas MXA identified 85.5, 8.8, and 5.6% of these same vertebrae as normal, deformed, or unanalyzable, respectively (Table 1). Indices of agreement between the two methods in designating each vertebra as normal or deformed are summarized in Table 2. The majority of deformities identified by MXAand by MRX were grade 2 deformities (MXA, 69.7%; MRX, 81.4%) and were most commonly wedge deformities (MXA, 41.1%; MRX, 47.5%). Of the vertebrae identified as deformed by both methods, 83.1% (118/142) were classified as the same deformity grade and 62.7% (89/142) were categorized as the same type of deformity. MXA was poorer at accurately identifying MRX deformities in the upper thoracic spine (T4–T7, sensitivity 50.0% and PPV 67.6%) when compared with the middle to lower thoracic (T8–T12, sensitivity 72.4% and PPV 79.7%) and lumbar spine (L1–L4, sensitivity 80.6% and PPV 80.6%).
MXA could not identify 15.7% (38/242) of deformities identified by MRX because they occurred in vertebrae not visualized sufficiently for analysis (Table 1). MRX deformities not identified by MXA in analyzable vertebrae were equally split into grade 1 and grade 2 deformities; however, MXA did not identify 78.0% (32/41) of MRX grade 1 deformities, compared with 18.4% (30/163) of the more abundant grade 2 deformities. The majority of MRX deformities not identified by MXA in analyzable vertebrae was end plate deformities (53.2%) and wedge deformities (40.3%), with four crush deformities missed. A significant number of false-positive results were produced by MXA (n = 41) when compared with MRX; these were spread over all three types of deformity.
Per subject: MRX categorized 40.4% of the 161 subjects as having one or more vertebral deformities and the rest as normal, whereas MXA identified 38.5% as having deformities, 60.3% as normal, and 1.2% had MXA scans that were deemed unanalyzable as a result of poor image quality (Table 1). MRX and MXA agreed that 84 subjects were normal and 50 subjects had one or more vertebral deformities, and indices of agreement between the two methods are summarized in Table 6. MXA identified 60.0% (15/25) of subjects found to have a single vertebral deformity by MRX as having one or more deformities and 92.5% (37/40) of subjects with two or more deformities according to MRX. All of the subjects classified as having one or more deformities by MXA but as normal by MRX exhibited a single deformity, which in the majority of cases (10/12) was a grade 1 deformity. Similarly, 10 of the 13 subjects categorized as having one or more deformities by MRX but as normal by MXA had a single deformity. In all cases there was at least one MRX deformity identified in the vertebrae deemed analyzable on the MXA scans.
MXA versus MRX—McCloskey algorithm
Per vertebra: MRX identified 88.9% of the 2093 vertebrae as normal, 10.6% as deformed, and 0.5% were not visualized sufficiently for analysis, whereas MXA identified 86.8, 7.6, and 5.6% of these same vertebrae as normal, deformed or unanalyzale, respectively (Table 3). Indices of agreement between the two methods in designating each vertebra as normal or deformed are summarized in Table 2. The majority of deformities identified by MXA and by MRX were wedge deformities (MXA, 37.7%; MRX 43.0%). Of the vertebrae identified as deformed by both methods, 57.2% (79/138) were classified as the same type of deformity. MXA was poorer at accurately identifying MRX deformities in the upper thoracic spine (T4–T7, sensitivity 63.4% and PPV 86.7%) when compared with the middle to lower thoracic (T8–T12, sensitivity 75.3% and PPV 85.9%) and lumbar spine (L1–L4, sensitivity 81.0% and PPV 91.1%).
MXA could not identify 16.3% (36/221) of deformities identified by MRX because they occurred in vertebrae not visualized sufficiently for analysis (Table 3). MRX deformities not identified by MXA in analyzable vertebrae were end plate deformities (57.4%). MRX crush deformities were most consistently identified by MXA in analyzable vertebrae (sensitivity 92.9%), followed by wedge (sensitivity 79.3%) and end plate (sensitivity 55.7%) deformities.
Per subject: MRX categorized 37.3% of the 161 subjects as having one or more vertebral deformities and the rest as normal, whereas MXA identified 32.3% as having deformities, 66.5% as normal, and 1.2% had MXA scans that were deemed unanalyzable as a result of poor image quality (Table 3). MRX and MXA agreed that 95 subjects were normal and 46 subjects had one or more vertebral deformities, and indices of agreement between the two methods are summarized in Table 6. MXA identified 63.6% (14/22) of subjects found to have a single vertebral deformity by MRX as having one or more deformities and 89.5% (34/38) of subjects with two or more deformities according to MRX. All of the subjects classified as having one or more deformities by MXA but as normal by MRX exhibited a single wedge deformity. Twelve subjects, 75% of which had a single deformity identified by MRX, were classified as having one or more deformities by MRX but as normal by MXA. In two of these cases all the deformities identified by MRX occurred in vertebrae deemed analyzable on the MXA scans.
|Per vertebra||Per subject|
|Vertebral deformity assessment method|
|MXA vs. MRX||Eastell vs. McCloskey|
|Kappa 95% CI||0.65–0.76||0.74–0.84||0.90–0.95||0.83–0.91|
|Sensitivity % (cfc)||69.6 (66.5)||74.6 (72.4)||—||—|
|Sensitivity uta % (cfc)||58.7 (54.7)||62.4 (59.4)||—||—|
|Specificity % (cfc)||97.7 (75.0)||98.9 (86.6)||—||—|
|Specificity uta % (cfc)||93.6 (55.7)||94.7 (60.2)||—||—|
Eastell algorithm versus McCloskey algorithm
MRX: The Eastell algorithm identified 242 vertebrae as deformed, whereas the McCloskey identified 220 deformities in the 2082 analyzable vertebrae (Table 4). The two methods agreed that 216 vertebrae were deformed and indices of agreement between the two methods in designating each vertebra as normal or deformed are summarized in Table 2. Twenty-six deformities (25 grade 1 and 1 grade 2 deformity) identified by Eastell were classified as normal by McCloskey and four deformities (all crush deformities) identified by McCloskey were classified as normal by Eastell. Little difference was found in the level of agreement between the two algorithms when analysis agreement was evaluated separately in the upper thoracic, middle to lower thoracic, and lumbar spine. A slightly lower Ppos was observed for the upper thoracic region of the spine (T4–T7) where Ppos = 89.1% compared with the middle/lower thoracic (T8–T12) in which Ppos = 96.8%.
Eastell identified 40.4% of the 159 subjects as having one or more vertebral deformities, whereas McCloskey identified 37.3% of subjects with deformities (Table 4). The two algorithms agreed that 59 subjects had one or more deformities and 95 were normal, and indices of agreement between the two methods in designating each subject as normal or deformed are summarized in Table 6. Six subjects identified as having one or more deformities by Eastell were classified as normal by McCloskey (five had a single grade 1 deformity identified by Eastell) and one subject classified as having one or more deformities by McCloskey (a single crush deformity) was classified as normal by Eastell.
|Per vertebra||Per subject|
|Per vertebra||Per subject|
|Per vertebra||Per subject|
MXA: The Eastell algorithm identified 185 vertebrae as deformed, whereas the McCloskey algorithm identified 158 deformities in the 1975 analyzable vertebrae (Table 5). The two algorithms agreed that 151 vertebrae were deformed and indices of agreement between the two methods in designating each vertebra as normal or deformed are summarized in Table 2. Thirty-four deformities (33 grade 1 and 1 grade 2 deformity) identified by Eastell were classified as normal by McCloskey and 7 deformities (all crush deformities) identified by McCloskey were classified as normal by Eastell. Little difference was found in the level of agreement if analysis was evaluated in different regions of the spine—upper thoracic, middle/lower thoracic, and lumbar.
Eastell identified 39.0% of the 159 subjects with analyzable MXA scans as having one or more vertebral deformities, whereas McCloskey identified 32.7% of subjects with deformities (Table 5). The two algorithms agreed that 50 subjects had one or more deformities and 95 were normal, and indices of agreement between the two methods in designating each subject as normal or deformed are summarized in Table 6. Twelve subjects identified as having one or more deformities by Eastell were classified as normal by McCloskey (each had a single grade 1 deformity identified by Eastell) and two subjects classified as having one or more deformities by McCloskey (each had a single crush deformity) were classified as normal by Eastell.
|Vertebral deformity assessment method|
|MXA vs. MRX||Eastell vs. McCloskey|
|Kappa 95% CI||0.55–0.79||0.64–0.86||0.84–0.97||0.71–0.90|
|Sensitivity (cfc)||79.4 (66.2)||79.3 (69.3)||—||—|
|Sensitivity (cfc)||87.5 (67.9)||94.1 (81.8)||—||—|
Comparison of MXA and MRX reference data
The MXA and MRX reference values for normal vertebral dimensions were calculated using the vertebrae of 100 study participants who were designated as normal by SQ reading of the conventional radiographs (J.L.).(18) The wedge and midwedge ratios showed a slight difference in the reference mean calculated with the MXA value generally higher than the corresponding MRX value (Fig. 5A). The reference SD was approximately comparable up to and including T9 between the two reference ranges; however, the MXA SD tended to be somewhat higher in the thoracic region (T5–T8) compared with the relevant MRX value (Fig. 5B), where the wedge ratio SD was on average 26.0% larger than the corresponding value for MRX. The deformity cut-off calculated from the reference values showed MXA produced a slightly higher threshold than MRX from T9–L4 and more comparable values from T4–T8 (Fig. 5C).
The crush ratios showed very consistent reference mean values for both MXA and MRX (Fig. 5D). However, the reference SD calculated from MXA was consistently larger than the corresponding value for MRX, particularly in the thoracic region (Fig. 5E), with the MXA SD for the wedge ratio on average 35.3% larger than the appropriate MRX value. As a result of this larger SD the MXA data produced a slightly lower deformity cut-off compared with MRX data (Fig. 5F).
An approximately constant magnification factor between the MXA and MRX vertebral height measurements was expected across the vertebral levels. However, in this study an inconsistent relationship was observed with the three vertebral heights exhibiting slightly different magnification factors, which fluctuated at different vertebral levels and resulted in systematic differences, particularly in the wedge and midwedge ratios, calculated from the MXA and MRX data. These differences may result from a combination of differences in point placement by the two observers who performed the MXA and MRX quantitative analysis, precision errors, and varying magnification factors in MRX. Magnification of the vertebral body image on conventional radiographs vary slightly from the center to the edge of a film as a result of the cone-beam geometry employed in conventional radiography and because of the variable thickness of subcutaneous fat changing vertebra-film distances between subjects. However, these factors would not account for the differences observed in this study because any effect should be approximately consistent across the three vertebral heights measured on any single vertebra. Therefore differences in point placement protocols and image resolution between MXA and MRX must account for this phenomenon. This would confirm previous speculation that reference data of normal vertebral dimensions, used in many quantitative methods to identify vertebral deformities, should be specific to the method of image acquisition.(24)
Generally, good to excellent agreement was observed between the two algorithms used for quantitative vertebral deformity identification using MXA or MRX on a per vertebra and per subject basis as expected because the algorithms compared were based on the same vertebral height measurements from a single image. More moderate agreement was seen when comparing the same algorithm between MXA and MRX, partly as the result of differences in normal reference data calculated for MXA and MRX. Agreement between MXA and MRX for the McCloskey algorithm was better than for the Eastell algorithm, largely as a result of the lower number of false positives produced by the McCloskey methodology, with a PPV for MXA of 87.9% in relation to MRX, compared with the Eastell algorithm, which produced a PPV of 77.6% for MXA in terms of MRX on a per vertebra basis.
Deformity identification by MXA was limited as a result of poor image quality, primarily in the upper thoracic spine, with one in six MRX deformities missed by MXA because they occurred in vertebrae that were not visualized sufficiently for analysis on the MXA scans. Deformity identification was poorer in the upper thoracic spine even when the vertebrae were deemed analyzable on the MXA scans with a sensitivity of 50.0% for MXA in terms of MRX for the vertebral levels T4–T7 using the Eastell algorithm and 80.6% for L1–L4. MXA proved more effective at identifying moderate to severe MRX deformities. This was clearly shown using the Eastell algorithm with MXA producing a sensitivity of 22.0% for identifying MRX grade 1 deformities, compared with a sensitivity of 81.6% for identifying grade 2 MRX deformities. Two MXA scans were totally unanalysable because of poor image quality, which resulted from the a combination of subjects' obesity, low bone density, and the presence of multiple vertebral deformities.
Although the feasibility of using MXA to identify prevalent vertebral deformities compared with conventional radiographic methods has been investigated in a number of studies, the results have frequently only been published as abstracts.(20,21,34–44) Although the majority of these studies have concluded that MXA is potentially a useful, low-radiation method for identifying prevalent vertebral deformities, direct comparison between these studies is often difficult for a number of reasons. Studies have used a number of different models of DXA machine, which produce MXA scans of varying image quality and offer different analysis software. The method of quantitative vertebral deformity identification is frequently insufficiently described and the methods described range from a 25% reduction in Ha, Hm, or Hp to the use of the more established vertebral deformity algorithms.(34–36,39–41,43)
Although the vertebral levels from T4 to L4 have been used more frequently, different vertebral levels have been assessed in many studies, such as the addition of L5 or the exclusion of one to four upper thoracic vertebral levels.(36,38,43) MXA image quality has been shown to be dependent, particularly in the upper thoracic spine, on the physical characteristics of the subjects and the specific MXA lateral scan mode performed.(23) Frequently, these factors are not described; therefore the problem of poor MXA image quality in the upper thoracic spine, which has been noted in a number of studies, cannot be evaluated properly.(34,39,40,42,44) Subjects used in these studies have varied from a mixture of postmenopausal women with and without vertebral deformities to osteoporotic individuals with previously established vertebral deformities.(34,35,37) However, in all cases the number of subjects with vertebral deformities has tended to be low, ranging from 13 to 40 subjects.(34,39)
Perhaps most importantly, none of the studies that compare vertebral deformity identification using MXA and MRX have characterized the distribution, type, or severity of deformities identified and in some cases even the total number of deformities identified was not given.(36,40,43) It seems obvious that a major factor in the large variation in vertebral deformity identification agreement between MXA and MRX in the various studies results from differences in the number, type, and severity of deformities identified in each study. The higher the prevalence of deformities in any study the higher the calculated κ-score is likely to be and the larger the proportion of gross moderate to severe deformities identified the more agreement is likely to be found.
A range of κ-scores have been produced when comparing vertebral deformity identification by various quantitative approaches using MXA and MRX on a per vertebra basis. These range from κ = 0.32 to κ = 0.71 depending on the vertebral levels evaluated.(21) In general the κ-scores obtained in other studies have been below those found in this study. One study comparing prevalent vertebral deformity identification using MXA scans and conventional radiographs was published recently by Chappard et al. using a group of 98 women, 31.6% of whom exhibited vertebral deformities.(21) Although the MXA scans were acquired on an earlier model of DXA machine and the conventional radiographs were performed without the use of a standardized acquisition, the Chappard study is surprisingly similar to the results found in this study. Comparison of vertebral deformity identification by quantitative analysis (using a method closer to the Eastell rather than the McCloskey algorithms used in this study) of the MXA scans and the conventional radiographs produced a κ-score of 0.67, which compares well with κ = 0.70 when MXA and MRX were compared using the Eastell algorithm in analyzable vertebrae in this study. Chappard also identified a sharp decrease in vertebral deformity identification in the upper thoracic spine (T4–T7) by the MXA methods.
Several limitations of this study should be acknowledged. Currently, no true gold standard for identifying vertebral deformities exists to which the various qualitative, SQ, and quantitative methods proposed to evaluate vertebral deformities on conventional radiographs or MXA scans can be compared. Although several comprehensive studies have compared various definitions and/or different deformity thresholds for identifying vertebral deformities and tested their correlation with clinical symptoms commonly associated with vertebral deformities, such as back pain, loss of height, and physical disability, no single method stands out as better than many others.(26,28) In this study we have nominally implemented the algorithms proposed by Eastell and McCloskey.(15,16) It should be recognized that different algorithms may produce a different relationship between MXA and MRX than has been characterized in this study.
Many different quantitative morphometric methods have been developed, especially over the past decade (e.g., see Refs. 10 and 12–17). They have the potential to make the quantification of any changes in vertebral dimensions over time straightforward, objective, and easily describable, usually with good reproducibility. However, the limitations of any quantitative morphometric technique, whether it is based on MXA scans or conventional radiographs, should be recognized. Any quantitative method requires the specification of a deformity threshold. However, low specificity tends to result in cut-offs giving high sensitivity and low sensitivity tends to result in cut-offs giving high specificity. Morphometric methods can only identify deformities and unlike the qualitative or SQ view of an experienced radiologist, cannot distinguish between deformities that are true osteoporotic fractures or deformities that are the result of other factors such as degenerative change.(18) Point placement in vertebral morphometry is a subjective decision that will vary between different observers however well defined the point placement protocol and will always be dependent on the level of experience and training of the observers involved. Although repeat analysis precision of vertebral height measurement using MXAand MRX has previously been shown to be good to excellent and well below the usual deformity threshold.(45)
Study subjects were prescreened before participation in this study to exclude individuals with moderate to severe scoliosis. It should be remembered that this would have removed a proportion of subjects, sometimes with vertebral deformities, who perhaps could have been evaluated by conventional radiographic methods. Conventional radiography has more flexibility in adjusting image acquisition parameters and may produce analyzable images when MXA scans of sufficient image quality cannot be acquired.
Before commencing the quantitative MXA and MRX analysis required in this study the observers concerned met to define a common protocol for point placement on each vertebral body. This was straightforward for anterior and posterior points but more problematic for midvertebral points. In MRX midpoints were conventionally placed in the center of the end plate ellipse on all vertebrae. Because of the scanning geometry of MXA scans, elliptical end plates were not usually seen, except when some lateral curvature of the spine is present and even then because of the inferior image quality of MXA scans they were frequently ill defined. Midpoints were placed directly on what was seen as the end plate of normal vertebrae and were positioned to describe the minimum distance between the superior and inferior end plates on deformed vertebrae. Therefore, it is likely that MXA produced midwedge ratios in deformed vertebrae that were smaller than MRX. This was unlikely to affect the results of this study except in the case of borderline or very mild deformities in which a small change in point placement may redefine a vertebra as normal or as deformed. In more severe deformities it was unlikely to have any impact on deformity identification. It should be recognized that differences between MXA and MRX in this study also may result from differences in the subjective view of each vertebra by the two different observers who performed the MXA and MRX analyses.
This study has only investigated the identification of prevalent vertebral deformities and not incident vertebral deformities. Further work is required to evaluate the utility of MXA scans compared with conventional radiographs in this instance.
In this study the utility of morphometric assessment of prevalent vertebral deformities using MXA scans and conventional radiographs have been compared in a diverse sample of postmenopausal women. Despite generally good agreement in analyzable vertebrae, particularly for moderate and severe deformities, the inherent problem of the image quality of MXA scans depletes the number of vertebrae visualized sufficiently for analysis and therefore the number of deformities identified, particularly in the upper thoracic spine. Although MXA image quality is clearly inferior to that of conventional radiographs, MXA has distinct advantages such as a substantially reduced effective dose to the patient, acquisition of a single image of the spine, and the removal of the nonlinear cone-beam–induced magnification inherent in conventional radiography.(19) These advantages make MXA a potentially useful, relatively fast, low-radiation technique to identify prevalent vertebral deformities, in conjunction with morphometric radiography in some patients. It will depend on the requirements of any particular study or clinician whether an MXA scan has the capability to fulfil the role required.
This study was supported financially by Hologic, Inc. (Bedford, MA, U.S.A.).
- 11960 The radiological diagnosis of osteoporosis: A new approach. Clin Radiol 11: 166–174.,
- 21968 Measurement of anterior vertebral compressions and biconcave vertebrae. AJR Am J Roentgenol 103: 635–644.
- 3the European Vertebral Osteoporosis Study Group 1996 The prevalence of vertebral deformity in European men and women: The European Osteoporosis Study. J Bone Miner Res 11: 1010–1018., , , , , , and
- 4National Osteoporosis Foundation Working Group on Vertebral Fractures 1995 Assessing vertebral fractures. J Bone Miner Res 10: 518–523.
- 5Food and Drug Administration 1994 Guidelines for Preclinical and Clinical Evaluation of Agents Used in the Prevention or Treatment of Postmenopausal Osteoporosis. Division of Metabolism and Endocrine Drug Products, Food and Drug Administration, Washington
- 61995 Recommendations for the registration of new chemical entities used in the prevention and treatment of osteoporosis. Calcif Tissue Int 57: 247–250.(on behalf of the Group for the Respect of Ethics and Excellence in Science [GREES])
- 71998 The association of radiographically detected vertebral fractures with back pain and function: A prospective study. Ann Intern Med 128: 793–800., , , , , , , ,
- 81991 Pre-existing fractures and bone mass predict vertebral fracture incidence in women. Ann Intern Med 114: 919–923., , ,
- 9for the Study of Osteoporotic Fractures Research Group) 1999 Prevalent vertebral deformities predict hip fractures and new vertebral deformities but not wrist fractures. J Bone Miner Res 14: 821–828., , , , , (
- 101982 Effect of the fluoride/calcium regimen on vertebral fracture occurrence in postmenopausal osteoporosis. N Engl J Med 306: 446–450., , , ,
- 111988 Vertebral morphometry in diagnosis of spinal fractures. Bone Miner 5: 59–67.,
- 121988 A newly developed spine deformity index (SDI) to quantitate vertebral crush fractures in patients with osteoporosis. Bone Miner 3: 335–349., , , , , , , ,
- 131989 Epidemiology of vertebral fractures in women. Am J Epidemiol 129: 1000–1011., , , , ,
- 141990 Assessment of osteoporotic spine deformity. Skeletal Radiol 19: 91–97., , ,
- 151991 Classification of vertebral fractures. J Bone Miner Res 6: 207–215., , , ,
- 161993 The assessment of vertebral deformity: A method for use in population studies and clinical trials. Osteoporos Int 3: 138–147., , , , , ,
- 171984 Validity in diagnosing osteoporosis. Eur J Radiol 4: 1–3., , ,
- 181993 Vertebral fracture assessment using a semiquantitative technique. J Bone Miner Res 8: 1137–1148., , ,
- 191995 Patient dose in morphometric X-ray absorptiometry. Osteoporos Int 5: 281–282 (letter).,
- 20Study of Osteoporotic Fractures Research Group 1994 Morphometric X-ray absorptiometry of the spine: Correlation in vivo with morphometric radiography. Osteoporos Int 4: 238–244., , , ,
- 211998 Clinical evaluation of spine morphometric X-ray absorptiometry. Br J Rheumatol 37: 496–501., , , ,
- 221998 Comparison of the precision of two vertebral morphometry programs for the Lunar Expert-XL imaging densitometer. Br J Radiol 71: 388–398., , , ,
- 231998 Optimizing data acquisition and analysis of morphometric X-ray absorptiometry. Osteoporos Int 8: 177–183., , ,
- 241998 Morphometric X-ray absorptiometry: reference data for vertebral dimensions. J Bone Miner Res 13: 464–474., , , , ,
- 251997 Vertebral morphometry studies using dual energy X-ray absorptiometry. Semin Nucl Med 27: 276–290., ,
- 261991 A comparison of morphometric definitions of vertebral fracture. J Bone Miner Res 6: 25–34., , ,
- 271994 Assessing vertebral deformities. Osteoporos Int 4: 117–119.
- 28for the Study of Osteoporotic Fractures Research Group 1995 Comparison of methods for defining prevalent vertebral deformities: The study of osteoporotic fractures. J Bone Miner Res 10: 890–902., , , , , , ,
- 291986 Statistical methods for assessing agreement between two methods of clinical measurement. Lancet i: 307–310.,
- 301960 A coefficient of agreement for nominal scales. Educ Psych Meas p. 37–46
- 311977 The measurement of observer agreement for categorical data. Biometrics 33: 159–174.,
- 321990 High agreement but low kappa: I. The problems of two paradoxes. J Clin Epidemiol 43: 543–549.,
- 331990 High agreement but low kappa: II. Resolving the paradoxes. J Clin Epidemiol 43: 551–558.,
- 341995 Morphometric x-ray absorptiometry (MXA) and morphometric radiography (MRX) in osteoporotic subjects: A comparative study. J Bone Miner Res 10(Suppl 1): S369 (abstract)., , , , , , ,
- 351995 A comparison of radiographic vertebral morphometry and MXA in established vertebral osteoporosis. J Bone Miner Res 10(Suppl 1): S473 (abstract)., , , , , ,
- 361995 Expert-XL: Clinical evaluation of a new morphometric technique on 12 patients with vertebral fracture. Osteoporos Int 6: 79 (abstract)., , , , ,
- 371996 Morphometric X-ray absorptiometry (MXA), comparison with quantitative radiographic measurement (QRM). Osteoporos Int 6(Suppl 1): S191 (abstract)., , ,
- 381996 Vertebral deformities: evaluation with morphometric absorptiometry and conventional radiographs in normal and osteoporotic women. Osteoporos Int 6(Suppl 1): S207 (abstract)., , , ,
- 391996 Evaluation of morphometric X-ray absorptiometry. Osteoporos Int 6(Suppl 1): S212 (abstract)., , , ,
- 401996 Role of MXA in screening osteoporosis patients for vertebral fracture. Osteoporos Int 6(Suppl 1): S155 (abstract)., , ,
- 411997 Comparison of DXA morphometry (MXA) and conventional radiography in identifying spinal abnormalities, including vertebral fractures. Osteoporos Int 7: 265 (abstract)., , , , ,
- 421997 The use of automatic vertebral morphometry to screen women with vertebral fractures. J Bone Miner Res 12(Suppl 1): S264 (abstract)., , ,
- 431997 Vertebral morphometry by morphometric radiography (MR) and morphometric X-ray absorptiometry (MXA) using the same reference population. J Bone Miner Res 12: 1532 (abstract)., , ,
- 441997 A preliminary evaluation of the Lunar Expert-XL for bone densitometry and vertebral morphometry. J Bone Miner Res 12: 136–143., , , , , , , ,
- 451999 Morphometric X-ray absorptiometry and morphometric radiography of the spine: A comparison of analysis precision in normal and osteoporotic subjects. Osteoporos Int 9: 536–544., , , , , , , , ,