SEARCH

SEARCH BY CITATION

Keywords:

  • absorptiometry;
  • deformity;
  • reference values;
  • vertebrae;
  • women

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

Our aim was to compare normal vertebral reference values for morphometric radiography (MRX) and morphometric X-ray absorptiometry (MXA) and to compare these methods for the identification of vertebral deformities. We calculated MXA reference values (Hologic QDR 4500 A) for 327 women (ages 22–88 years) randomly selected from local General Practice lists in Sheffield, U.K. MRX reference values were calculated from spinal radiographs for 123 of these subjects (ages 56–88 years). We used these reference values to identify deformities in the MRX and MXA reference populations and in 83 women with osteoporosis (ages 49–87 years). We observed differences in mean deformity of vertebral height ratios measured by MRX and MXA, especially for the mid-to-posterior ratio. We compared agreement between quantitative methods (MRX and MXA) and qualitative radiological assessment. Severity of deformity was defined by semiquantitative (SQ) assessment. Agreement was moderate for MRX (k = 0.59; 95% CI = 0.43–0.77) and for MXA (k = 0.47; 95% CI = 0.29–0.66) in the reference population. Agreement was good for MRX (k = 0.86; 95% CI = 0.82–0.89) and MXA (k = 0.71; 95% CI = 0.66–0.75) in the osteoporotic population. MRX and MXA correctly identified a greater proportion of moderate or severe deformities compared with mild deformities. Sensitivity, specificity, predictive values, and accuracy were slightly better for MRX than for MXA. Although MXA agrees well with qualitative radiological assessment, the large proportion of vertebrae excluded from analysis because of poor image quality limits the diagnostic value of the technique. Reference intervals should be technique specific.


INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

The identification of vertebral deformities is an important indicator in the initial diagnosis of osteoporosis and in the follow-up of patients receiving therapy. Methods currently available for detection of osteoporotic vertebral deformities include qualitative radiological assessment of spinal radiographs and quantitative vertebral morphometry using morphometric radiography (MRX) or morphometric X-ray absorptiometry (MXA).

Generally, the qualitative method has been regarded as the gold standard. This method has the advantage that the radiographs are assessed by radiologists who possess the expertise to differentiate between true osteoporotic deformities and changes in vertebral shape that may be associated with other conditions.(1) However, this is a subjective method, and although there have been attempts to develop a standardized approach to visual assessment, these methods may overestimate the prevalence of mild deformities.(2–4) Also, the radiation dose to the patient for spinal radiography is high.

Quantitative vertebral morphometry involves making measurements of vertebral dimensions, which are then matched against a normal reference interval. The measurements may be made on spinal radiographs (MRX) or on scan images acquired by fan beam densitometer (MXA). The accuracy of vertebral measurements in MRX is highly dependent on the quality and consistency of the radiographic technique used, and repeat exposures to correct poor quality radiographs are undesirable in view of the high radiation dose.(5) MXA is a relatively new but promising technique that may overcome some of the patient-positioning and exposure factor problems inherent in conventional radiography.(6) However, although the devices that are capable of MXA are readily available commercially, this technique has not yet been fully evaluated or validated for the detection of vertebral deformities.

Spatial resolution for MXA is inferior to that of conventional radiography, and image quality is adversely affected by obesity.(7) This may make vertebral dimensions more difficult to measure with accuracy, and more vertebrae may have to be excluded from morphometric assessment. The main attraction of MXA is that the radiation dose to the patient is considerably lower than for conventional radiography.(8)

Few researchers have published data from MXA studies, and most of the available data has been derived using an earlier Hologic device (QDR 2000, Hologic, Inc., Bedford, MA, U.S.A.). In a recent study, Rea et al. comprehensively evaluated the different scan modes available for MXA using the Hologic QDR 4500 A (Hologic, Inc., Bedford, MA, U.S.A.), namely, single-energy and dual-energy (array, fast lateral, and high definition) scan modes.(7) The single-energy scan takes only 12 s to acquire and offers a well-defined image, but soft tissue artifacts may obscure vertebral outlines. Dual-energy scans take between 6 minutes (array) and 12 minutes (fast lateral and high definition), and although they less affected by soft-tissue artifacts, these scan modes produce a “noisier” image. Side-by-side image analysis allows the operator to view both single-energy and dual-energy scans simultaneously to aid placement of vertebral markers. Rea et al. concluded that a combination of the single-energy and high-definition scan modes maximizes the number of vertebrae that can be visualized clearly for analysis. In this study, body mass index (BMI) (but not bone mineral density [BMD]) influenced image quality, but neither variable had any noticeable effect on precision. Reported short-term precision estimates for MXA range between 2% and 5% CV.(7,9,10)

There is evidence from studies performed using the Hologic QDR 2000 that MXA can detect vertebral deformities in osteoporotic patients, and there is good agreement between MXA and qualitative radiological assessment.(9–12) However, the numbers of subjects with vertebral deformities in these studies are relatively small. Also, the identification of deformities in the earlier studies relies on the manufacturer's reference range, which is based on MRX values.(10–12) In a larger study, Rea et al. suggest that reference ranges derived from MRX studies may not be applicable to MXA, in view of the observed differences between their MXA mean vertebral height ratios when compared with MRX values reported by other researchers.(13) Our own data for MXA in men supports this and also suggests that MXA reference ranges should be gender specific.(14) In our previous study, we witnessed a decline in mean height ratio with age in men measured by MXA. Rea et al. observed a similar decline for vertebral heights (but not for height ratios) in women measured by MXA.(13)

Apparent differences between MRX and MXA reference values could arguably be attributed to study design. Potential sources of variability may be minimized by comparing normal values for MRX and MXA in one group of individuals and by having one operator mark both radiographs and scan images.

The aims of our study were to (1) compare mean vertebral dimensions measured by MRX and MXA in the same reference population and (2) test agreement between quantitative methods (MRX and MXA) and qualitative methods (visual radiological assessment) for the diagnosis of vertebral deformities.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

Subjects

We studied 327 women ages 22–88 years (mean 56 years ± 16 years). This group constituted women who originally had been selected randomly from local General Practice lists in Sheffield, U.K., to participate in three separate studies of bone density. These studies were as follows:

  • 1.
    A prospective population-based study of bone density in postmenopausal women. Details of the study cohort of 375 women have been published elsewhere.(15) A total of 242 women attended for follow-up bone density scans and spinal radiographs at 5 years, and of these, 123 women ages 56–88 (mean 67 ± 7 years) also consented to have MXA. These 123 women were similar in terms of mean age, height, weight, BMI, and BMD to the larger group of 242. The exclusion of the 119 women who did not have MXA was on the grounds of scoliosis, poor mobility, or unwillingness to undergo further examinations in addition to the requirements of the original study protocol.
  • 2.
    A feasibility study for recruitment into a study of the efficacy of a selective estrogen receptor modulator in the treatment of osteoporosis. Exclusion criteria were the use of any medication or existence of any disease or condition known to influence bone density; use of statins or diuretics; and history of neoplasia, mild stroke, deep vein thrombosis or psychiatric illness. Of 762 women approached, 142 responded. Sixty-two of these women were excluded with respect to the study exclusion criteria. Eighty women were then appointed for exploratory bone density scans. Seventy-three of these women also consented to have MXA, but 6 women were unable to be measured by MXA because of scoliosis, leaving a total of 67 women ages 62–82 (mean 71 ± 4 years).
  • 3.
    A cross-sectional study of bone density in young women. Exclusion criteria were use of medications or the existence of any disease or condition known to have a major influence on bone density and inability to give informed consent. A total of 219 women were approached, and of these 143 were enrolled in the study. Six women were unsuitable for MXA because of scoliosis, and the remaining 137 ages 22–55 (mean 40 ± 10 years) consented to have MXA. (We have no information regarding reasons for refusal to participate in any of these studies.)

We used two sample t-tests to compare mean vertebral dimensions measured by MXA for younger and older women by dividing the combined group of 327 into two groups of women <55 years (n = 101) and ≥56 years of age (n = 226). There were no significant differences in mean values for the two age groups, so all 327 women were included in the MXA reference population. Radiographs of the thoracolumbar spine (acquired on the same day as MXA scans) were available for the subset of 123 postmenopausal women ages 56–88 years (mean 67 years ± 7 years), who originally had been enrolled into study 1. This subset was used as our MRX reference population.

We also studied 83 women ages 49–87 (mean 70 ± 9 years), who had qualitative radiological evidence of osteoporotic vertebral fractures. These women had been referred to the metabolic bone unit at the Northern General Hospital Trust, Sheffield, U.K. Spinal radiographs were available for these women. All women from both the reference and the osteoporotic populations also had lumbar spine BMD (L1–L4) measured by dual-energy X-ray absorptiometry using the Hologic QDR 4500 A. All subjects gave informed consent and the study was approved by the North Sheffield Local Research Ethics Committee.

Vertebral morphometry

All subjects had MXA of vertebrae T4–L4, using the Hologic QDR 4500 A. Both single-energy and dual-energy (high definition) scans were acquired. However, in an earlier analysis, when comparing MXA reference values for single-energy and high-definition scans in the subset of 123 postmenopausal women from the reference population, we found that there were some significant differences between the two scan modes for mean height ratio and SD. For this reason we used side-by-side image analysis but marked only single-energy scans, using high-definition scan images as a visual aid to point placement.

Radiographs for the 123 women from the reference population and the 83 women from the clinic-based population were acquired after a standardized protocol. The women from the reference population had spinal radiographs and MXA on the same day. For the osteoporotic population, the majority of radiographs and MXA scans were acquired on the same day and no subject had a time gap between MXA and spinal radiography greater than 5 months. All radiographs and scan images were marked by one operator (L.F.) using the Melton method.(16) The posterior superior points were marked at the base of the uncinate process, and the inferior points were placed in vertical alignment with the superior points. Oblique projection of the vertebral bodies is unusual in MXA; however, if the subject is not positioned with the spine in perfect alignment with the long axis of the scan table or if the subject has a slight scoliosis, a double vertebral rim occasionally may be seen. In this eventuality, the midpoint for MXA was placed equidistant from the right and left vertebral rims.

Anterior, middle and posterior vertebral heights were measured on radiographs by transparent ruler to the nearest millimeter. Vertebral height ratios were calculated for ha/hp (wedge), hm/hp (end plate), hp/hp+1, and hp/h−1 (compression), where ha = anterior height, hm = midheight, hp = posterior height, hp+1 = posterior height of vertebra below the one under examination, and hp−1 = posterior height of vertebra above the one under examination.

Intraoperator variability for marking of radiographs and scan images

Radiographs and scan images for a subset of 32 postmenopausal women from the reference population ages 56–76 years (mean 65 years), who had been measured by both MRX and MXA, were marked twice by the same operator, once at the time the images were acquired and again after an interval of 1 year. Overall estimates for intraoperator variability (for vertebral height ratios) were 1.9% CV for MRX and 2.6% CV for MXA.

Normal values for vertebral dimensions

Normal values for MRX and MXA were derived from the two reference populations. We used an iterative trimming algorithm to remove outliers (height ratio values that fell more than 1.5 interquartile ranges beyond the interquartile range).(16) The mean height ratios and their SDs were calculated from the trimmed values for each vertebral level. Reference values originally were calculated separately for the 327 women in the reference population who had been divided into two groups by age. When no significant differences were found between the two groups, the un-trimmed data for all 327 women was pooled and the trimming process was repeated.

We calculated critical values for the identification of vertebral deformities for both MRX and MXA as mean height ratio minus 3 SD (grade 1 deformity) and mean minus 4 SD (grade 2 deformity).(17) We also calculated critical values separately for MXA in the subset of 123 women (the MRX reference population), so that we could compare mean values for both methods in the same group of women.

Identification of vertebral deformities

For gold standard identification of vertebral deformities we used a consensus reading of qualitative visual assessments by an experienced radiologist (N.A.B.) and one of the authors (R.E.). An atlas of radiological variants was used as a guide to assessment.(18) A second radiologist (G.J.) used Genant's semiquantitative (SQ) method to grade the severity of deformities that had been identified by the consensus reading.(19)

Table Table 1.. Characteristics of the Study Populations
 MXA reference (n = 327)MRX reference (n = 123)Osteoporotic (n = 83)
  1. Results are expressed as means ± SD. MRX reference is a subset of MXA reference population who had both MRX and MXA. Means for reference populations were compared with means for osteoporotic population by two sample t-tests.

  2. Asterisks denote significant differences: *p < 0.05; **p < 0.001; ***p < 0.0001.

Age (years)56.4 ± 16.2***66.6 ± 7.3**69.7 ± 8.6
Height (m)1.60 ± 6.60***1.59 ± 0.06***1.55 ± 0.06
Weight (kg)67.8 ± 12.2***67.9 ± 11.5***61.4 ± 11.4
BMI   
 (kg/m2)26.2 ± 4.826.9 ± 4.4*25.4 ± 4.4
BMD   
 L1–L4   
 (g/cm2)0.971 ± 0.166***0.910 ± 0.169***0.727 ± 0.112

The critical values derived from the MRX and MXA reference populations were used to identify vertebral deformities in both the reference and the clinic-based populations. We tested interrater agreement between the quantitative methods (MRX and MXA) and qualitative method (consensus reading) on a per vertebra basis and calculated diagnostic values using the consensus reading as the gold standard for all analyses. We tested agreement for all vertebrae and repeated the test after excluding vertebrae that were not adequately visualized by MXA.

Statistical analysis

We used two sample t-tests to compare anthropometric data for the reference and osteoporotic populations. Paired t-tests were used to compare mean vertebral dimensions for MRX and MXA in the reference population. To test for significance, we used p < 0.05 with the Hochberg correction for multiple comparisons.(20) We tested agreement between methods by κ-statistics and calculated SE and 95% confidence intervals (CI) for κ. The κ-scores were classified according to Altman.(21) We also calculated sensitivity and specificity, positive and negative predictive values, and diagnostic accuracy for MRX and MXA. Statistical analysis was performed in Excel for Windows version 7.0 (Microsoft, Redmond, WA, U.S.A.), Statgraphics Plus version 2.0 (Manugistics, Rockville, MD, U.S.A.), and MedCalc version 4.15d (MedCalc Software, Mariakerke, Belgium).

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

Characteristics of the study populations

Table 1 shows the anthropometric data for the three study populations. The women with osteoporosis were older and had lower mean height, weight, and lumbar spine BMD compared with the two reference populations. BMI was also lower for these women than for women in the MRX reference population. Approximately 20% of women from the reference populations had BMI > 30, compared with 12% from the osteoporotic population.

Table Table 2.. Number of Vertebrae Excluded by MXA
 MXA reference (n = 327)MRX reference (n = 123)Osteoporotic (n = 83)
  1. Results are presented as number of vertebrae (%). MRX reference is a subset of MXA reference population.

T4186 (57)68 (55)51 (61)
T5131 (40)49 (40)38 (46)
T6101 (30)36 (29)28 (34)
T769 (21)20 (16)19 (23)
T839 (12)15 (12)14 (17)
T919 (6)4 (3)8 (10)
T1010 (3)4 (3)2 (2)
T116 (2)3 (2)1 (1)
T125 (2)3 (2)0
L13 (1)1 (1)0
L26 (2)3 (2)0
L314 (4)7 (6)1 (1)
L414 (4)5 (4)2 (2)
Total603 (14)218 (14)164 (15)
Median111

Marking of radiographs and scan images

For MXA in the reference population, 14% of all vertebrae was excluded from analysis because of poor image quality (Table 2). BMI did not influence the number of vertebrae excluded. Two vertebrae (<1%) could not be visualized adequately for MRX in this population.

In the osteoporotic population, 15% of all vertebrae were unanalyzable by MXA, and the number of vertebrae excluded was not influenced by BMI. For MRX, thoracic spine radiographs were unavailable for three patients, but we were able to measure vertebra T12 from the lumbar spine radiographs of these women. The lumbar spine radiograph was unavailable for one patient. Eleven vertebrae (1%) could not be marked by MRX. These were at the levels of T4, T5, and T6 (five, three, and three vertebrae, respectively).

Trimming

Trimming removed 4.3% and 3.3% of values from the upper and lower tails of the vertebral height ratio distribution for the MRX and MXA reference populations, respectively. After trimming, mean height ratio for MRX was increased by an average of 0.3, 0.06, and 0.07% for ha/hp, hm/hp, and hp/hp+1/–1, respectively. For MXA there was an average increase of 0.1% and 0.05% in means for ha/hp and hm/hp and mean hp/hp+1/–1 ratio decreased by 0.07%. SD was decreased by 0.7, 0.3, and 1.1% for MRX and by 0.7, 0.3, and 1.0% for MXA for ha/hp, hm/hp, and hp/hp+1/–1 ratios, respectively.

thumbnail image

Figure FIG. 1.. Trimmed mean height ratios and SD measured by MRX and MXA in the reference population. (A–C) Means and SDs by vertebral level measured by MRX and MXA for ha/hp, hm/hp, and hp/hp+1 height ratios, respectively. MRX values are represented by the closed circles connected by a broken line and MXA by open circles with a solid line. Asterisks indicate significant differences between methods when compared by paired t-tests (means), and F tests (SD) with the Hochberg correction for multiple comparisons (*p < 0.05; **p = < 0.01; ***p = < 0.001).

Download figure to PowerPoint

Comparison of mean height ratios for MRX and MXA in the reference population

Mean ha/hp height ratios in the 123 women measured by both quantitative methods were predominantly higher for MXA than for MRX, but after correction for multiple comparisons, none of the differences reached significance (Fig. 1). Mean hm/hp ratio was lower for MXAthan for MRX at all vertebral levels, with significant differences at vertebrae T6, T8, T10, and T11 (Fig. 1). There was no clear pattern of differences for compression deformity.

Comparison of SD for MRX and MXA in the reference population

SD of mean height ratios in the reference population was relatively low for both MRX and MXA in comparison to SD reported for other studies and was approximately 2–4% of the mean height ratio. We observed differences (some significant) in SD between the two quantitative methods (Fig. 1). Although the difference between SD values for mean height ratios at any given vertebral level was no more than 0.02, the shape of the data plots for SD by vertebral level was markedly different for the two quantitative techniques.

Critical values

Tables 3–6 give the reference values for MRX and MXA, which were used for identification of vertebral deformities. MXA critical values derived from the reference population (n = 327) were predominantly lower in the thoracic spine for hm/hp and hp/hp+1/–1 ratios compared with MRX. This mainly was caused by greater means for MXA hm/hp ratios and greater SDs for MXA hp/hp+1/–1 ratios. In the lumbar spine, MXA critical values were lower for ha/hp ratios (because of greater SD) and higher for hm/hp ratios (because of smaller SD) in comparison with MRX.

Identification of vertebral deformities in the reference population

Table 7 shows the prevalence of deformity identified by the consensus reading and by the quantitative methods in the reference and osteoporotic populations. The prevalence of vertebral deformities identified by the consensus reading in the reference population was <2% (26 deformities out of a total of 1599 vertebrae). Table 8 shows the inter-rater agreement per vertebra between the qualitative consensus reading and the quantitative methods (MRX and MXA) in the subset of 123 women from the MXA reference population who were measured by both methods and in the 83 women with osteoporosis. In the reference population, agreement was only moderate for the quantitative methods but was better for MRX than for MXA. The 95% CIs for k in the reference population were wide (possibly because of the low prevalence of vertebral deformity).

Table Table 3.. Vertebral Reference Values for ha/hp Height Ratios Measured by MRX and MXA
 MXA (n = 327)MRX (n = 123)
 MeanSDCritical valueMeanSDCritical value
  1. Values for MRX were derived from a subset of the MXA reference population who were measured by both methods. Mean is the trimmed mean ha/hp height ratio; SD is the trimmed SD of ha/hp ratio; critical value is the mean ha/hp height ratio minus 3 SD.

T40.9550.0410.8340.9550.0400.835
T50.9470.0450.8100.9450.0490.798
T60.9270.0490.7810.9200.0420.794
T70.9280.0490.7820.9200.0430.791
T80.9430.0430.8040.9230.0480.779
T90.9770.0520.8210.9490.0360.841
T100.9840.0470.8440.9770.0300.887
T110.9550.0440.8220.9480.0440.816
T120.9690.0430.8390.9590.0420.833
L10.9900.0390.8720.9910.0290.904
L21.0300.0450.8941.0340.0470.893
L31.0530.0490.9081.0530.0460.915
L41.0780.0530.9181.0680.0450.933
Table Table 4.. Vertebral Reference Values for hm/hp Height Ratios Measured by MRX and MXA
 MXA reference (n = 327)MRX reference (n = 123)
 MeanSDCritical valueMeanSDCritical value
  1. Values for MRX were derived from a subset of the MXA reference population who were measured by both methods. Mean is the trimmed mean hm/hp height ratio; SD is the trimmed SD of hm/hp ratio; critical value is the mean hm/hp height ratio minus 3 SD.

T40.9540.0350.8490.9660.0330.866
T50.9560.0340.8550.9610.0300.872
T60.9460.0340.8430.9560.0350.852
T70.9430.0330.8430.9550.0340.853
T80.9460.0320.8500.9570.0290.870
T90.9600.0360.8510.9610.0260.884
T100.9590.0330.8590.9660.0320.871
T110.9420.0280.8590.9520.0290.864
T120.9570.0270.8760.9540.0340.851
L10.9620.0250.8880.9630.0340.862
L20.9700.0290.8840.9720.0360.864
L30.9770.0270.8960.9810.0350.877
L41.0070.0320.9101.0110.0390.895
Table Table 5.. Vertebral Reference Values for hp/hp+1 Height Ratios Measured by MRX and MXA
 MXA reference (n = 327)MRX reference (n = 123)
 MeanSDCritical valueMeanSDCritical value
  1. Values for MRX were derived from a subset of the MXA reference population who were measured by both methods. Mean is the trimmed mean hp/hp+1 height ratio; SD is the trimmed SD of hp/hp+1 ratio; critical value is the mean hp/hp+1 height ratio minus 3 SD.

T40.9700.0400.8510.9730.0370.861
T50.9740.0410.8510.9790.0340.876
T60.9770.0420.8520.9780.0320.882
T70.9820.0390.8640.9790.0300.889
T80.9760.0400.8550.9760.0320.878
T90.9390.0400.8190.9430.0330.844
T100.9270.0340.8240.9240.0260.847
T110.9350.0330.8350.9330.0350.829
T120.9520.0300.8610.9650.0430.836
L10.9820.0300.8910.9860.0280.901
L20.9960.0320.9010.9830.0240.910
L31.0220.0410.8981.0160.0470.874
Table Table 6.. Vertebral Reference Values for hp/hp−1 Height Ratios Measured by MRX and MXA
 MXA reference (n = 327)MRX reference (n = 123)
 MeanSDCritical valueMeanSDCritical value
  1. Values for MRX were derived from a subset of the MXA reference population who were measured by both methods. Mean is the trimmed mean hp/hp−1 height ratio; SD is the trimmed SD of hp/hp−1 ratio; critical value is the mean hp/hp−1 height ratio minus 3 SD.

T51.0310.0420.9061.02950.03910.9121
T61.0270.0430.8971.02230.03590.9146
T71.0220.0470.8791.02320.03340.9231
T81.0180.0410.8951.02210.03130.9281
T91.0250.0410.9011.02580.03440.9225
T101.0640.0430.9351.06170.03680.9514
T111.0790.0390.9601.08280.03020.9922
T121.0710.0380.9551.07330.04000.9534
L11.0510.0340.9491.03720.04630.8983
L21.0190.0320.9251.01530.02890.9287
L31.0050.0320.9091.01850.02530.9427
L40.9780.0400.8590.98330.04310.8539
Table Table 7.. Prevalence of Deformity
 Reference (n = 123)Osteoporotic (n = 83)
  1. Results are presented as number of vertebral deformities detected by consensus reading (qualitative) with severity of deformity defined by SQ grade, and the quantitative methods (MRX and MXA reference mean height ratios minus 3 and 4 SD) in the reference and osteoporotic populations.

SQ grade 118155
SQ grade 27117
SQ grade 3173
MRX mean − 3 SD18122
MRX mean − 4 SD9185
MXA mean − 3 SD18103
MXA mean − 4 SD14175

Sensitivity for MRX and MXA in the reference population, when tested against the consensus reading, was <60% for both methods (Table 9). Specificity and negative predictive values were high and were similar for both quantitative methods, but all other diagnostic values were slightly better for MRX (Table 10).

Both MRX and MXA correctly detected a relatively high percentage of moderate and severe deformities (SQ grades 2 and 3) in comparison with mild (SQ grade 1) deformities. We tested inter-rater agreement between the consensus reading and MRX and MXA by severity of deformity. For this analysis we used mean MRX and MXA height ratios minus 3 SD to define a mild deformity and mean ratio minus 4 SD to define a moderate to severe deformity. The κ-scores for the identification of moderate to severe deformities were considerably higher for both MRX and MXA in comparison with those yielded for mild or all deformities (Table 11). Both MRX and MXA were less effective in detecting end plate in comparison with wedge or compression deformities. MXA and MRX correctly identified 68% and 63%, respectively, of wedge, and 17% and 50% of end plate deformities detected by the consensus reading. A single compression deformity identified by the consensus reading in the reference population also was detected by MRX but was missed by MXA.

Identification of vertebral deformities in the osteoporotic population

The prevalence of vertebral deformity identified by consensus reading in the osteoporotic population was 32% (345 deformities out of 1064 vertebrae; Table 7). Agreement with the consensus reading was good for both MRX and MXA (Table 8). When vertebrae that could not be analyzed by MXA were excluded from the calculation of k, agreement for both quantitative methods was very good. The 95% CIs for k were relatively narrow compared with those for the reference population. Sensitivity (Table 9) and diagnostic values (Table 10) also were better for MRX and MXA in this population, and values were better for MRX compared with MXA.

MRX and MXA also correctly detected a greater proportion of moderate or severe deformities in comparison with mild deformities in the women with osteoporosis. The proportion of moderate and severe deformities correctly identified was similar for both quantitative methods, but MRX was superior to MXA for the identification of mild deformities. Inter-rater agreement between quantitative methods and the consensus reading for the identification of moderate to severe deformities was excellent for both MRX and MXA (Table 11). Both quantitative methods missed a greater proportion of end plate compared with wedge or compression deformities. The percentages of deformities correctly identified by type were 93% and 85% of wedge, 73% and 72% of end plate, and 88% and 92% of compression deformities for MRX and MXA, respectively.

Influence of excluded vertebrae and BMI on the identification ofvertebral deformities by MXA

Three vertebral deformities were missed by MXA in the reference population because of vertebrae excluded from analysis. BMI did not influence the percentage of false positives (approximately 1% of all vertebrae) or the percentage of false negatives (<1%) identified by MXA in this population, and the results were similar for MRX.

In women with osteoporosis, 40 vertebral deformities were missed by MXA because of the exclusion of inadequately visualized vertebrae. MRX correctly identified 38 of these 40 deformities. High BMI slightly increased the false-positive rate for MXA (9% for women with BMI ≥ 30 compared with 2% for women with BMI < 30) but had no effect on the false-negative rate (6%). For MRX the false-positive rate was 2.2% and 0.9% for BMI ≥ 30 and BMI < 30, respectively, and the false-negative rate was lower in the heavier women (1% compared with 6% in the lighter women)

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

The normal means for ha/hp ratios in our study were slightly (but not significantly) higher for MXA compared with MRX, but for hm/hp ratios the means for MRX were consistently lower. The differences in mean hm/hp ratios may be related to the different projection effects for the two imaging modalities. Obliquity of the vertebral bodies (which influences the measurement of midheight) is rarely seen on MXA images, whereas geometric distortion and patient-positioning problems mean that this is a commonly seen feature on conventional radiographs. We also observed some significant differences in SD between MRX and MXA. SD was greater in the thoracic spine for MXA, and this could be related to variability in point placement caused by poor image quality for MXA. However, for the hm/hp ratio in the lumbar spine, SD was greater for MRX, and we suggest this may be caused by the oblique projection of lumbar vertebrae on conventional radiographs.

Chappard et al. reported a similar pattern of differences for normal mean ha/hp and hm/hp ratios measured by MRX and MXA to those observed in our reference population.(9) Conversely, Rea et al. reported lower mean ha/hp ratios and higher mean hm/hp ratios for MXA.(13) However, this study compared MXA with MRX values generated from other studies, and the authors suggest that the differences they observed may be explained by interobserver variability. The differences in our study cannot by explained in this way, because all images and radiographs were marked by the same operator (although this approach could potentially introduce an element of bias, in that the marking of radiographs and scan images was not blinded).

Table Table 8.. Agreement Between Methods for the Identification of Vertebral Deformities
 Reference populationOsteoporotic population
 All vertebrae (n = 1599)Vertebrae analyzed by MXA (n = 1381)All vertebrae (n = 1064)Vertebrae analyzed by MXA (n = 915)
  1. Inter-rater agreement was tested in the subset of women from the MXA reference population who were measured by both methods (n = 123) and in the osteoporotic population (n = 83). Results are presented as κ-scores (SE, 95% CI) for agreement between MRX and MXA (quantitative) and the consensus reading (qualitative) based on individual vertebrae. Agreement was calculated separately for all vertebrae and for vertebrae that could be analyzed by both the consensus reading and MXA.

MRX0.60 (0.09; 0.43–0.77) 0.86 (0.02; 0.73–0.82) 
MXA0.47 (0.10; 0.29–0.66)0.49 (0.10; 0.30–0.68)0.71 (0.02; 0.66–0.75)0.78 (0.02; 0.73–0.82)
Table Table 9.. Sensitivity and Specificity of MRX and MXA for the Identification of Vertebral Deformities
 Reference population (n = 123)Osteoporotic population (n = 83)
 SensitivitySpecificityNumber of vertebraeSensitivitySpecificityNumber of vertebrae
  1. Results are presented as percent sensitivity and specificity calculated for all vertebrae, (percent sensitivity and specificity after exclusion of vertebrae that could not be analyzed by MXA). For number of vertebrae, values in parentheses are the number of vertebrae that were analyzed by both the consensus reading and MXA. The consensus reading was used as the gold standard for all analyses.

MRX6299159985981064
MXA54 (58)99 (97)1599 (1381)72 (82)96 (95)1064 (915)
Table Table 10.. Predictive Values and Accuracy of MRX and MXA for the Identification of Vertebral Deformities
 Reference population (n = 123)Osteoporotic population (n = 83)
 PPVNPVAccuracyNumber of vertebraePPVNPVAccuracyNumber of vertebrae
  1. Results are presented as percent values calculated for all vertebrae (percent values after exclusion of vertebrae that could not be analyzed by MXA). For number of vertebrae, values in parentheses are the number of vertebrae that were analyzed by both the consensus reading and MXA. The consensus reading was used as the gold standard for all analyses.

  2. PPV, positive predictive value; NPV, negative predictive value.

MRX59999915999593941064
MXA44 (44)94 (99)98 (98)1599 (1381)89 (89)88 (91)88 (90)1064 (915)

Reference interval means for MXA height ratios in our study are very similar to those reported by Rea et al., and the shape of the means plots for the two studies are almost identical.(13) However, mean values for the Chappard study appear to be somewhat lower.(9) This probably is because of the different approaches used for defining reference intervals. In the Chappard study, normal values were derived by excluding subjects with no qualitative radiological evidence of vertebral deformities, rather than by statistical trimming. This may mean that vertebrae with deformities resulting from developmental abnormalities or degenerative disease would not be excluded as osteoporotic fractures and subsequently this may lead to lower mean height ratios.

The close similarity in mean MXA values between our reference data and that given by Rea et al. is perhaps surprising in view of the differences in study populations and methods used for vertebral point placement. However, SD of MXA mean height ratios measured by Rea et al. is consistently higher for all height ratios compared with the SD in our reference population. This may be explained by sampling differences for the two studies. Rea et al. derived reference values from three discrete study populations, measured at separate centers, using a combination of different generation Hologic devices for MXA measurements. Because our subjects were all measured locally using the same device, there may have been less variability in our data. Statistical trimming of mean height ratios usually lowers the SD (particularly for small sample sizes).(22) We used the same trimming method as Rea et al. but had a smaller number of subjects in our reference population. Altman recommends at least 200 observations for calculation of a reference interval, and Black reports that accuracy for SD of mean height ratios may become limited for sample sizes of <500.(21,22) However, the analysis described by Black in this context was performed using the Gaussian trimming method (which tends to overestimate SD), and Altman shows that when the width of 95% CIs are plotted against sample size, the curve begins to plateau at a sample size of 150 and decreases only minimally between sample sizes of 200 and 500. Also, the decrease in SD in our reference population after trimming was relatively small (between 0.06% and 1.10% of the mean height ratio). We therefore believe that our MXA reference interval based on a sample size of 327 is fairly robust but concede that comparisons of MXA with MRX values may be compromised by the smaller number of subjects (123) in the MRX reference population.

Table Table 11.. Inter-rater Agreement for Identification of Deformities by Severity
 Reference population (n = 1599 vertebrae)Osteoporotic population (n = 1064 vertebrae)
 SQ grade 1SQ grades 2 and 3SQ grade 1SQ grades 2 and 3
  1. Results are presented as κ-scores (SE, 95% CI) for agreement between MRX and MXA (quantitative) and consensus reading (qualitative) for the identification of deformities in the subset of women from the reference population who were measured by both methods (n = 123) and in the osteoporotic population (n = 83). Agreement was calculated for all vertebrae. The severity of deformities identified by the consensus reading was defined by SQ grade.

MRX0.44 (0.13; 0.19–0.68)0.94 (0.06; 0.83–90)0.79 (0.03; 0.73–0.85)0.95 (0.01; 0.93–0.98)
MXA0.38 (0.13; 0.12–0.64)0.63 (0.13; 0.38–0.89)0.65 (0.04; 0.58–0.72)0.84 (0.02; 0.80–0.88)

The differences between values in our reference population (mainly in SD, but also in some means) for MRX and MXA led to a tendency for lower critical values for MXA. The use of MRX critical values for detection of vertebral deformities in patients measured by MXA therefore may lead to an increased prevalence of false positives. The differences we observed cannot be attributed to interobserver variability, geographical population differences, or systematic variability between different diagnostic centers. This suggests that there are true differences between vertebral measurements made by MRX and MXA, and our data support the conclusions of other researchers who recommend the use of MXA-specific reference ranges (9,13).

We saw no evidence of the age-related decline in vertebral height ratios that we witnessed in our study of MXA in men.(14) The observed changes in men may have been related to degenerative change and modeling of the vertebral end plates, and these features may be seen less commonly in women.

Our κ-scores for agreement between MXA and the consensus reading in the osteoporotic population are similar to those reported by Chappard et al.(9) The lower κ-scores for MXA were influenced by two factors. First, the false-positive rate was higher for MXA than for MRX. This probably was because of the relatively poor resolution of MXA images compared with conventional radiographs. Second, (also because of differences in image quality), there were fewer vertebrae available for analysis by MXA compared with MRX, and MRX was able to identify correctly 95% of the deformed vertebrae that had been excluded by MXA.

The κ-scores used alone give limited information about the diagnostic effectiveness of a test. Although agreement was better in the osteoporotic population compared with the reference population, for example, accuracy was better in the reference compared with the osteoporotic population for both MRX and MXA. Sensitivity and specificity are not affected by prevalence, but predictive values are, and this probably explains the poor positive predictive values we obtained for both MRX and MXA in the reference population (which had a low prevalence of deformities).

Although MRX and MXA were less effective at detecting the proportion of true deformities (defined by the consensus reading) in the reference population, the negative predictive values for detection of deformities were good for both MRX and MXA in both the reference and the osteoporotic populations. Rea et al. calculated similar negative predictive values for qualitative visual assessment of MXA scans, and these results suggest that MXA may have the potential for use as a prescreening tool.(23) This would mean that only MXA-positive patients need undergo conventional radiography, saving time and minimizing financial costs and radiation dose to the patients. However, based on our data, only 1357 out of 1573 vertebrae (86%) in the reference population and 523 out of 719 vertebrae (73%) in the osteoporotic population would be classified correctly as normal. These figures are influenced by the proportion of vertebrae excluded from analysis by MXA, because no assumption can be made as to whether MXA would have classified these vertebrae as normal.

The large proportion of vertebrae excluded from analysis by MXA because of poor image quality is probably the main disadvantage of MXA in its current form. However, in view of the relatively low radiation dose to the patient in comparison with conventional radiography, and the excellent agreement shown between MXA and the consensus reading for the identification of moderate and severe deformities, the following approach may be worthy of consideration: (1) prescreen all patients by MXA, (2) classify patients as normal if all vertebrae are visualized adequately and classified as normal by MXA, (3) acquire conventional radiographs if all vertebrae are not visualized adequately, and (4) if one or more deformities are detected by MXA, acquire conventional radiographs to check for further prevalent deformities and to identify the nature of the deformity.

The comparison between MRX and MXA for the identification of vertebral deformities in our study, as in others, is possibly subject to bias, in that MRX has the advantage of being compared with the gold standard technique, which makes an assessment from the same image, that is, spinal radiographs. Visual assessment of MXA scans by an experienced operator has been compared with a qualitative radiological assessment of spinal radiographs.(23) The κ-scores for agreement were similar to those observed for agreement between quantitative MXA and the gold standard, and this work suggests that MXA images could potentially be used by experienced radiologists for both qualitative and quantitative assessments. A consensus reading of MXA scans comprising a quantitative analysis and a qualitative visual assessment by an experienced radiologist could be used in conjunction with the MXA prescreen approach. MXA scans with positive results from quantitative analysis could be checked by the radiologist to rule out differential causes of vertebral deformity other than osteoporotic fracture. In cases of uncertainty about point placement in quantitative analysis because of poor image resolution, the experienced radiologist may be more adept at discerning vertebral outlines, and this may lead to fewer vertebrae being excluded from analysis.

To conclude, currently, we do not recommend that MXA is an adequate substitute for conventional radiography, but with technique-specific reference ranges and refinements in technology to improve the image resolution of scan images, MXA shows promise as a low-dose method of diagnosing prevalent vertebral deformities. A consensus of both quantitative and qualitative assessments of MXA images could potentially maximize the use of this technique, and further research involving both radiologists and trained operators is indicated to explore the feasibility of this approach.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES

This work was funded in part by the National Osteoporosis Society, U.K., and by an Arthritis Research Campaign Programme Grant. We also wish to give thanks to Diana Greenfield, Deborah Swindell, and Yvette Henry for recruiting subjects and to the staff of the Osteoporosis Centre, Northern General Hospital, Sheffield, U.K., in particular the radiographers, for acquisition of MXA scans.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. MATERIALS AND METHODS
  5. RESULTS
  6. DISCUSSION
  7. Acknowledgements
  8. REFERENCES
  • 1
    Tan TCF, Sartoris DJ, Resnick D 1995 Differential diagnosis of osteoporotic vertebral fractures: Pathology and radiology. In: GenantHK, JergasM, Van KuijkC (eds.) Vertebral Fracture in Osteoporosis. Radiology Research and Education Foundation, San Francisco, CA, U.S.A., pp. 7194
  • 2
    Harrison JE, Patt N, Müller C, Bailey TA, Budden FH, Josse RG, Murray TM, Sturtridge WC, Strauss A, Goodwin S 1990 Bone mineral mass associated with postmenopausal vertebral deformities. Bone Miner 10: 243251.
  • 3
    Kleerekoper M, Nelson DA 1992 Vertebral fracture or vertebral deformity? Calcif Tissue Int 50: 56.
  • 4
    Spector TD, McCloskey EV, Doyle DV, Kanis JA 1993 Prevalence of vertebral fracture in women and the relationship with bone density and symptoms: The Chingford study. J Bone Miner Res 8: 817822.
  • 5
    Banks LM, Van Kuijk C, Genant HK 1995 Radiographic technique for assessing osteoporotic vertebral deformity. In: GenantHK, JergasM, Van KuijkC (eds.) Vertebra Fracture in Osteoporosis. Radiology Research and Education Foundation, San Francisco, CA, U.S.A., pp. 131147
  • 6
    Jergas M, Lang F, Fuerst T 1995 Morphometric x-ray absorptiometry. In: GenantHK, JergasM, Van KuijkC (eds.) Vertebral Fracture in Osteoporosis. Radiology Research and Education Foundation, San Francisco, CA, U.S.A., pp. 331348
  • 7
    Rea JA, Steiger P, Blake GM, Fogelman I 1998 Optimizing data acquisition and analysis of morphometric x-ray absorptiometry. Osteoporos Int 8: 177183.
  • 8
    Njeh CF, Fuerst T, Hans D, Blake GM, Genant HK 1999 Radiation exposure in bone mineral density assessment. Appl Radiat Isot 50: 215236.
  • 9
    Chappard C, Kolta S, Fechtenbaum J, Dougados M, Roux C 1998 Clinical evaluation of spine morphometric x-ray absorptiometry. Br J Rheumatol 37: 496501.
  • 10
    Steiger P, Cummings SR, Genant HK, Weiss H 1994 Morphometric x-ray absorptiometry of the spine: Correlation in vivo with morphometric radiography. Osteoporos Int 4: 238244.
  • 11
    Steiger P, Slosman D, Sebert JL, De Vernejoul MC, Grados F, Muller CH, Birman P, Kelly T, Tsouderos Y 1995 Morphometric x-ray absorptiometry (MXA) and morphometric radiography (MRX) in osteoporotic subjects: A comparative study. J Bone Miner Res 10: S369 (abstract).
  • 12
    Coombes G, McCloskey E, Bernard J, de Takats D, Reaney L, Kanis J 1996 Morphometric x-ray absorptiometry (MXA) and radiographic vertebral morphometry in established postmenopausal osteoporosis. In: RingEFJ, ElvinsDM, BhallaAK (eds.) Current Research in Osteoporosis and Bone Mineral Measurement IV 1996. British Institute of Radiology, London, U.K., pp. 96
  • 13
    Rea JA, Steiger P, Blake GM, Potts E, Smith IG, Fogelman I 1998 Morphometric x-ray absorptiometry: reference data for vertebral dimensions. J Bone Miner Res 3: 46474.
  • 14
    Ferrar L, Eastell R 1998 Identification of vertebral deformities in men: Comparison of morphometric radiography and morphometric x-ray absorptiometry. Osteoporos Int (in press)
  • 15
    Peel NFA, Barrington NA, Blumsohn A, Colwell A, Hannon R, Eastell R 1995 Bone mineral density and bone turnover in spinal osteoarthrosis. Ann Rheum Dis 54: 867871.
  • 16
    Melton LJ III, Lane AW, Cooper C, Eastell R, O'Fallon WM, Riggs BL 1993 Prevalence and incidence of vertebral deformities. Osteoporos Int 3: 113119.
  • 17
    Eastell R, Cedel SL, Wahner HW, Riggs BL, Melton LJ III 1991 Classification of vertebral fractures. J Bone Miner Res 6: 207215.
  • 18
    Jergas M, Genant HK, Hackl F and Van Kuijk C 1995 Atlas of vertebral deformity and fracture. In: GenantHK, JergasM, Van KuijkC (eds.) Vertebral Fracture in Osteoporosis. Radiology Research and Education Foundation, San Francisco, CA, U.S.A., pp. 371450
  • 19
    Genant HK, Wu CY, Van Kuijk C, Nevitt C 1993 Vertebral fracture assessment using a semiquantitative technique. J Bone Miner Res 8: 11371148.
  • 20
    Hochberg Y 1988 A sharper Bonferroni procedure for multiple tests of significance. Biometrika 4: 800802.
  • 21
    Altman DG 1991 Practical Statistics for Medical Research, Chapman and Hall, London, U.K.
  • 22
    Black D, Cummings SR, Stone K, Hudes E, Palermo L, Steiger P 1991 A new approach to defining normal vertebral dimensions. J Bone Miner Res 6: 883892.
  • 23
    Rea JA, Chen M, Li J, Blake GM, Steiger P, Genant HK, Fogelman I 1998 Visual vertebral deformity assessment byx-ray absorptiometry: A highly predictive method to rule out vertebral fractures. Bone 23: S160 (abstract).