Morphometric Dual-Energy X-Ray Absorptiometry of the Spine: Report of a Large Series and Correlation with Axial Bone Mineral Density

INTRODUCTION

OSTEOPOROTIC FRACTURES result in considerable mortality, morbidity, and high medical cost.^{1, 2} The vertebral body, proximal femur, and distal forearm are among the most commonly affected skeletal sites. In postmenopausal women, the lifetime risk of each of these fractures is approximately 15%.³ Although hip and wrist fractures can be clinically diagnosed and identified unambiguously on radiographs, there are no reliable criteria for identifying osteoporotic fractures in the spine. Vertebral fractures are identified by crush, biconcave, and wedge deformities on plain lateral radiographs of the vertebrae. However, there is considerable difficulty defining a “vertebral fracture,”^{4, 5} in part because only about 30% of clear vertebral deformities are clinically significant. In this regard, it is often difficult to discriminate between an osteoporotic fracture and normal variation in vertebral shape or a vertebral deformity or fracture that may have occurred a long time ago.

Clinical vertebral fractures are associated with significant morbidity.^6-9 However, moderate deformity of a single vertebra might not be a specific indicator of fracture; the involvement of two or more vertebrae, or the occurrence of one severe deformity, might be a better indicator of a symptomatic fracture.^9-11 Another aspect of the difficulty is technique variations in the lateral spine radiographs that are used for assessment, including variation in the angulation of the patient and under- or overexposure in different regions.¹² The procedure normally requires two lateral exposures (thoracic and lumbar regions) and results in effective radiation dosages of up to 2.1 mSv.¹³ In addition, there is a high degree of image distortion due to the cone-beam geometry of the imaging system and the difference in focal spot location on each film, typically centered over T8 and L3. The most critical problem, however, is variation in the qualitative assessment of deformity, even among experienced radiologists. In one study, radiologists reached agreement on “fracture” in only one-third of the cases.¹⁴ Differences among countries in the prevalence of vertebral fracture may reflect these uncertainties.¹⁵

Development of a consistent typology (biconcave, wedge, compression) can aid qualitative evaluation,¹⁶ but the major improvement over the past 15 years derives from measurement of the posterior, middle, and anterior heights of the vertebra on lateral spine radiographs, either directly or after digitization.^{13, 17-30} As an example, six markers manually located on bony landmarks are used for the height definitions. In this quantitative morphometric analysis, deformities may be detected by comparing vertebral height ratios (anterior/posterior or mid/posterior) derived from the measured vertebral heights with reference values in conjunction with a classification system,³⁰ or by direct comparison of the measured heights with predicted normal or reference values.³¹ When compared with reference values, changes in vertebral height ratios (deformities) are commonly taken as significant if they are substantially below normal at a given vertebral level, usually around 3 or 4 SD below. Less pronounced deformities are poorly associated with clinical manifestations; however, depending on the clinical presentation such deformities might still be generally considered fractures.¹⁰ Even with more uniform methods, the use of a standard criterion of abnormality, and vertebral height ratios, there are still substantial uncertainties in assessing subjects of different body size, and there remain inexplicable differences among populations in the prevalence of spinal deformity.^{15, 26, 32, 33}

The recent development of morphometric X-ray absorptiometry (MXA) allows a more uniform acquisition procedure, with little or no magnification, smaller effective dose to the patient, and a more consistent software-guided analysis.^34-36 In the present investigation, we have developed reference data for MXA in Brazilian women using an imaging densitometer. We also measured axial bone mineral density (BMD) in these women to examine its association with incipient vertebral deformity. It is well documented that vertebral deformities and fracture are associated with low axial BMD.^37-39 As a group, patients with spine fracture have BMD that is ∼3 SD below the level in young normal and ∼1 SD below the level in age-matched patients. In fact, there is a continuous gradient of risk between the frequency of fracture and BMD. However, we wanted to see if an association could be detected in normal women prior to the occurrence of significant anatomical alteration.

MATERIALS AND METHODS

Patients

Ambulatory, healthy white women (n = 605) from the São Paulo area volunteered for the study. They were recruited from the gynecologic outpatient clinic of our hospital where they were seen for routine checkup. They did not have any systemic diseases based on a clinical screen, nor were they taking any medication known to affect bone and mineral metabolism. These subjects were part of a larger prospective study that provided BMD reference data for white Brazilian women.⁴⁰ Their biographic and anthropometric data is presented in Table 1. The ages ranged between 22 and 97 years, but most women were in the range of 40–70 years of age.

Table Table 1.. Anthropomorphic Variables in the Brazilian Women by Age and the Z Score for Spine and Femur BMD

Morphometric X-ray absorptiometry

MXA was performed using an imaging densitometer (Expert XL; Lunar Corp., Madison, WI, U.S.A.). This instrument uses the dual-energy X-ray absorptiometry (DXA) method to obtain a lateral image of the spine (thoraco-lumbar segment, T4–L4). It has a fan-beam X-ray source, slit collimation, and a solid-state, linear-array detector.^{35, 36} The detector has submillimeter spatial resolution. The source and the detector are positioned at opposite ends of a rotatable C-arm. During the acquisition of the image, both source and detector are fixed on the lateral–lateral axis of the patient and moved parallel to the vertical axis of the spine. During the acquisition, the patients were positioned in the supine position, with the knees and hips flexioned in a 90° angle; the arms were flexed with both hands behind the head. Image acquisition required about 40 s with an absorbed radiation dose of ∼1.2 mGy.

The same operator analyzed all scans using the standard software supplied by the manufacturer. Six edge points were placed on the border of the vertebral body using the standard procedure. The point placement protocol used is described in details elsewhere.^{19, 41} The anterior (H_A), middle (H_M), and posterior (H_P) heights of each vertebral body were measured using the analysis software, and average height (H_AVG) was calculated. The wedge (H_A/H_P) and midwedge (H_M/H_P) ratios of these vertebral heights were then calculated at each vertebral level for each individual. Adjustments of the brightness and contrast were made in order to improve visualization of vertebras at different spine levels. This is important during analysis because soft tissue attenuation is different along the spine. The reported precision of vertebral heights in repeat scans in vivo is –1 mm (3–4%), while the precision for ratios is –0.05 mm (5%).^41-44

The observed vertebral heights were compared with those expected based on a large (n = ∼3900) database of white women compiled from published studies.^{20-22, 24, 25, 29, 31} For determining expected normal vertebral heights, we have used a previously reported approach,³¹ which essentially relies on the nearly constant relationship between vertebral heights at each level. However, to calculate a correction factor, L2–L4 vertebrae were used rather than T4, which is small and often difficult to visualize and measure by MXA. For example, if an individual's L2–L4 H_AVG is 94% of that of the reference height, then the expected heights on the other vertebrae will also be 94% of the reference values. Alternatively, at each vertebral level the expected heights will be 94% of the actual reference heights. These expected heights are used to determine expected H_A/H_P and H_M/H_P ratios. This relationship allows us to ascertain a “Z score” for heights or ratios, which is defined by:

The SD in vertebral heights at each vertebral level is ∼8%, while the SD for the H_A/H_P and H_M/H_P is ∼6.5%. Based on the measured heights we can also ascertain a “T score,” which is defined by:

This normalization approach reportedly equalizes vertebral height data from white subjects of widely divergent stature and had even been shown to adequately normalize results for Asian women. Given the large number of individuals involved in the present investigation, we could examine the adequacy of the normalization procedure for vertebral heights in our sample of Brazilian women. It is important to note that the ratios (H_A/H_P and H_M/H_P) are not normalized and that we did not perform an iterative trimming procedure used in other reports.^{21, 26}

RESULTS

Morphometric reference data

The heights (156 ± 6.6 cm) and body weight (64.9 ± 11.2 kg) were typical of Brazilian women; the average body mass index (BMI) was 26.7 (Table 1). The age changes of spine and femoral neck BMD were entirely normal compared with reference values. The Z score for spine BMD was –0.5 in women under 50 years, but the corresponding value for femoral neck BMD was 0.1. Only 20.6% of women over age 50 years had a T score for spine BMD lower than –2.5 SD; for the femoral neck, this percentage was even lower at 6.9%.

Table 2 displays anterior, middle, posterior, and average vertebral heights. Figures FIG. 1., FIG. 2. show the heights and ratios profiles according to the vertebral level. The H_A and H_P averaged 5% (−0.56 T score) lower than values expected from reference data, but H_M was almost 10% lower (T = −1.2) (Table 3). The H_AVG was about 2 mm (or 6%) lower than reference values (T = −0.76). The postnormalization Z score was close to zero for all heights and for H_A/H_P (Table 4). In contrast, the H_M/H_P averaged about 0.04 lower than the reference values, or about −0.72 SD. The pattern of heights and ratios along the spine was close to that expected, with the exception that L4 had lower heights and ratios than expected.

Table Table 2.. Values for Anterior, Middle, Posterior, and Average Heights and Ratios

Table Table 3.. T Scores for Vertebral Heights

Table Table 4.. Z Scores for Vertebral Heights and Ratios

We examined the SD of the Z scores for heights at individual vertebral levels. Theoretically the Z score should be zero and the SD of the Z score should be about 1 (for the normalized vertebra T4–L1) if the normalization procedure is adequate. We observed an average SD of 0.99, 0.94, and 0.83 for H_A, H_M, and H_P, respectively (Tables 4 and 5).

Table Table 5.. Average T Scores and Z Scores for MXA Variables

Vertebral heights and BMD

The linear regression of H_A/H_P, and H_M/H_P for individual vertebral levels with spine BMD and with femoral neck BMD showed low correlations (−0.1) (data not shown). We then calculated the average Z score for the ratios for each individual (based on all observed vertebrae in each subject) and correlated these with BMD. The correlation coefficient between H_A/H_P and spine BMD was 0.05 (NS), while that for femur neck BMD was 0.09 (p = 0.04). The corresponding correlations for H_M/H_P were 0.21 and 0.25; while the correlations were low, both were highly significant (p < 0.001).

An alternative approach is to examine the association in grouped data between vertebral heights and BMD. The T score for H_AVG was below −1 in 30% of subjects. The average BMD values in these subjects was 10% lower than BMD values in subjects with a T score for H_AVG > 0. In other words, bigger subjects had higher BMD values. However, there was no clear association when the H_AVG was expressed as a Z score (Table 6). There was only a 2–3% difference of BMD between the quartile with the lowest Z scores and those with the highest Z score for H_AVG (Fig. 3A). The chi-squared tests showed no significant trend. In contrast, there was a strong association between femur BMD and Z scores for H_M/H_P. The chi-squared tests were significant (p < 0.05 and p < 0.01) for femoral neck and Ward's triangle, respectively, but not for trochanteric BMD. The quartile with the highest Z score for H_M/H_P had axial BMD levels 8–15% above the lowest quartile (Fig. 3C). There was only a slight tendency for H_A/H_P to increase with BMD (Fig. 3B); the chi-square tests for trend were not significant.

Table Table 6.. Distribution of Axial BMD by Quartiles of MXA Variables (n = 427); the Chi-Squared, Test for Trend Was Used

DISCUSSION

Ascertainment of vertebral fracture is important, both clinically and epidemiologically, because it allows the assessment of bone fragility independent of BMD and the evaluation of treatment efficacy. Quantitative methods provide greater objectivity for assessment of vertebral fracture than the qualitative methods. The images obtained in the present investigation were of sufficient resolution to allow satisfactory identification of anatomical details, even though there was less spatial resolution than conventional radiography. In our hands, this limitation was evident in the upper thoracic vertebrae (approximately between T4 and T6), where the superposition of soft tissue, particularly the lung and the branchial tree, as well as ribs, hindered the visualization of the vertebral borders as well as the intervertebral spaces. As a result, only ∼58% of T4 and ∼80% of T5 were measurable; 95% of T6 vertebrae could be measured. These figures are comparable to a recent report where MXA was also used.⁴⁷ It is important to note, however, that image quality is dependent on the BMI and bone density of the subjects. Therefore, identification of anatomical details is expected to be further impaired in obese and/or osteoporotic patients.

The analysis of the data indicated a decrease of only 0.2 mm in vertebral heights in older women, that was due entirely to their 2 cm smaller stature. The decrease was small because there was only a 14-year difference between the average age of the younger and older women. Rea et al.⁴⁴ reported a larger decrease (0.5 mm) in vertebral heights of older women using MXA, but like us, they found no influence of age on vertebral ratios. Diacinti et al.⁴⁸ found a large decrease in vertebral heights and ratios with age in Italian women using radiographic morphometry. Sone et al.⁴⁹ reported a decrease in both H_A/H_P and H_M/H_P in older Japanese women using radiographic morphometry. Apparently the aging effects seen in radiographic morphometry may be less evident using MXA.

Vertebral heights in Brazilian women averaged 8% lower than the reference population consisting of American and European women, but this difference was adequately compensated for in the normalization procedure used in the analysis software. The Z score for normalized heights had a mean close to 0 and an SD close to 1, which conforms to theoretical expectations. Our normalization was based on the entire L2–L4 sequence, rather than on T4 as proposed by others.³¹ T4 is difficult to measure, even when it can be visualized, and hence is not useful.²⁷ L4 can also be used to normalize height results. However, our experience shows that L4 is sometimes difficult to measure, and is often highly deviant, as noted by Blake et al.³⁴ The L2–L4 sequence may be superior to L4 alone.

Our results for H_A/H_P were very close to those expected, but the actual ratio differs from those observed using radiographic morphometry and MXA with other devices. The reported H_A/H_P and H_M/H_P for Hologic (Waltham, MA, U.S.A.) MXA results on white European women (n = 1019) were about 0.02 and 0.07 higher than our values⁴⁴; values for radiographic morphometry on H_A/H_P are slightly lower than ours, and H_M/H_P values are similar.^{9, 29} The SDs for heights and ratios were close to expected, and similar to those observed in other studies using both MXA and radiographic morphometry.

There was no significant association of vertebral heights with axial BMD. Larger individuals had slightly higher BMD values and larger vertebrae, but this association disappeared when the vertebral heights were normalized (see Table 6). We observed no trend of H_A/H_P with BMD in these healthy women (with normal BMD and without significant deformity). In contrast, there was a modest association between H_M/H_P and BMD on an individual basis (r = −0.2), and a significant trend on a group basis. The approximate 1 SD difference between the lowest and highest quartile for H_M/H_P was associated with an 8–15% (0.5–1.0 SD) difference of BMD. This association of low BMD and low H_M/H_P needs to be explored in greater detail in normal postmenopausal women. Several groups have shown that axial BMD correlates better with H_M/H_P than with H_A/H_P if abnormal cases are included in the postmenopausal sample.^{11, 49, 50}

In conclusion, our results indicate that MXA assessment of vertebral deformity can be considered together with axial BMD in patient evaluation and may provide better prediction of future clinical fracture than either variable alone.^{51, 52}