| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Full Paper |
1 Regional Medical Physics Department and3 Rheumatology Department, Sunderland Royal Hospital, Sunderland and 2 University of Sunderland, Sunderland, UK
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
PIXI uses a cone X-ray beam with dual energies (55 kVp and 80 kVp) with a scintillation screen and optically coupled charge-coupled device sensor and software for automated analysis of a circular region in the calcaneus. The reference ranges supplied by the manufacturer were used by the machine to provide T-scores. The threshold T-score values used for diagnosis of osteoporosis from axial measurements were considered to be inappropriate for use with peripheral measurements [8], so the possibility was pursued of categorizing the results as indicating low (for values of T>-1), moderate (T![{els]](/math/els.gif)
le;-1, >-1.6), and high (Tle;-1.6) fracture risk, in accordance with advice for Lunar equipment supplied by AuRa Scientific (Milton Ernest, UK). The potential for correspondence between these three categories, and the three used with axial densitometry, was also examined.
The study subjects were all patients who had been referred by general practitioners or hospital doctors for bone densitometry of hip and spine. Referral for investigation was based on the UK Department of Health's guidelines [9] and required the presence of at least one risk factor for osteoporosis. There was no other restriction on access to the service by all hospital doctors and general practitioners in the area served. After having the standard hip and spine scans, patients were invited to participate in the study. They were given both written and verbal explanations of the procedure, and written consent was obtained from those willing to participate. 100 subjects (6 male, 94 female; mean age 59.7 years, SD 10.8, range 41–83) then had bone densitometry of both heels carried out using the PIXI. All scans were analysed in accordance with the equipment manufacturer's procedures, using semi-automated routines. This study was approved by the Sunderland Local Research Ethics Committee.
Statistical methods
The use of bone mineral density values obtained for each heel as measures of the same quantity was examined using a Bland–Altman plot [8] of the differences between the heel measurements against their mean. Relationships between T-scores from the two densitometers were examined by linear regression. The kappa index was used to examine agreement between the classifications derived from them. To construct receiver operating characteristic (ROC) curves, the heel T-scores were input as a continuous variable into software that calculated maximum likelihood estimates [9] of statistical parameters of the ROC curves (ROCKIT 0.9B; C.E.Metz, University of Chicago, Chicago, IL). The "true" condition for each subject was the diagnostic classification from their axial densitometry score. The statistical test used to examine differences between the ROC curves for each heel was a z-score test of the areas under the curves.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
The relationship between the bone mineral density (BMD) values obtained for each heel was examined using a Bland–Altman plot [10] of the differences between the heel measurements against their mean (Figure 1
). The plot shows that differences were independent of the mean value. The overall mean BMD was 0.46 g cm-2, the mean of the signed differences was 0.01 g cm-2 (SD 0.05), and that of the unsigned differences was 0.03 g cm-2 (SD 0.04). With the exception of relatively large negative values in a few subjects, indicating that these had higher bone density in their left heel compared with the right, it could be concluded that there was no systematic difference between right and left sides. ROC analysis (see below) also supported this finding.
|
). Comparing heel scores with these axial scores, there was a closer relationship between the heel and the total hip T-scores (Figure 2a) than with the minimum axial T-score (Figure 2b). Results are shown for the left heel only, but those for the right heel and for the minimum score for either heel were similar.
|
|
), and also when the minimum T-score from either heel was used. For a diagnosis of osteoporosis (by axial T-score of -2.5 or less), the area under the ROC curve (Table 2) was larger, although we cannot show statistical significance, when diagnosis was based on total hip score (method (c)) than when all regions were used (method (a)). This suggests better agreement between heel and hip measurements than between heel and all regions, as also indicated by their closer correlation (Figure 2
). For a diagnosis of "osteoporosis or osteopenia" (by axial T-score of -1 or less), the areas under the ROC curves were generally smaller, suggesting a worse ability to discriminate between these categories.
|
|
Using these advised threshold values, the T-scores for the left heel placed 20 subjects in the high, 18 in the moderate and 62 in the low risk category. For the right heel the corresponding numbers were 24, 20 and 56. The right and left heel scores classified 82 of the subjects into the same fracture risk categories. The weighted kappa index was 0.80, indicating very substantial but not perfect agreement between classifications based on left and right heel measurements.
The potential for clinical use of the fracture risk categories produced from the heel measurements was assessed by comparing the results for each risk category with the "true" diagnosis obtained from axial measurements (the reference method). For this purpose the associations between the high risk category and osteoporosis, moderate risk and osteopenia, and low risk and normal bone density, were examined (Table 3). Comparisons were made separately for each heel, and with each approach to the diagnosis of osteoporosis. Result categories were then reduced to two, i.e. denoting condition present or absent, as follows. The sensitivity and specificity were calculated first for identification of "high risk" with osteoporosis, and then for "high or moderate risk" with osteopenia or osteoporosis (Table 3). Generally, specificity was relatively high but sensitivity was relatively low.
|
). Results were similar for each heel.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
For our whole group data, the means of measurements on either heel were not significantly different. However, for some individuals there were relatively large differences in BMD; the mean difference was 6.5% (SD 8.7%). These differences are similar to those reported from bilateral ultrasound measurements [13]. It is therefore questionable whether heel measurements should be used interchangeably as predictors of fracture risks for individuals; in only 82% of our subjects was there agreement on the resulting diagnostic category. It may therefore be prudent to measure both heels. This contrasts with hip bone density, where side-to-side differences are generally smaller, although not negligible [14].
The proportion of people diagnosed as having osteoporosis will depend upon the assessment methods used and the population studied [15]. An alternative method for determining the threshold T-score for heel BMD using similar equipment [16] gave a value of -1.3 and, for a study population similar to ours, higher sensitivity, lower specificity and lower false negative rate, as might be expected. A study of heel BMD using single X-ray absorptiometry [17] using a threshold T-value of -1.2 and a similarly aged population gave still higher sensitivity and lower specificity. Thus there is no optimum threshold level, but a choice can be made of the desired sensitivity or specificity of the test, based upon the relative values assigned to them.
The prime clinical use of bone densitometry is as an aid to making therapeutic decisions [9, 18, 19]. A possible use of heel bone densitometry in this context is as a more readily accessible first-line investigation, instead of hip and spine scans. We can assess its potential role by applying our figures (from Table 3, means of values for both heels) for sensitivity (71%) and specificity (87%) for diagnosis of osteoporosis, using total hip measurements as the standard, to three groups with differing prior likelihood of disease. These examples are intended to provide an illustration of the changes in predictive value of scans with changing disease prevalence. The additional effects of the different sensitivity and specificity values corresponding to a threshold T-score of -1.8 rather than -1.6 on the following analyses are given in parentheses. While there are consequential small changes in the numbers of false negative and false positive cases, we consider that the conclusions described below are not changed by the use of a slightly different threshold value.
First, we considered a population with a low prevalence of osteoporosis, such as peri-menopausal women without additional risk factors. Such women would not have bone densitometry performed under existing guidelines in the UK, but might be screened if USA guidance were being used [20]. Taking a prevalence of 4% [15] as an estimate for this group, out of 100 women investigated, 15 (10) would be judged osteoporotic on heel measurements, but since we expect 12 (8) to be falsely positive, it would be reasonable to proceed to axial densitometry for all of these in order to provide convincing evidence for prophylactic treatment and lifestyle alteration, and a basis for any subsequent monitoring. There would be only one (two) false negative.
In contrast, an elderly population (women aged over 80 years) would have a much higher prevalence of osteoporosis; Kanis and Gluer [3] quote 47.5%. In such a group, of 100 women scanned we would expect to find 41 (31) in the osteoporosis group, of which 7 (4) would be falsely positive. Since the appropriate treatment in this age group is with calcium and vitamin D, which is relatively inexpensive, it would be reasonable to proceed to this treatment without performing axial densitometry. Of more concern would be the 14 (20) falsely negative results. For this reason this approach seems appropriate only where there is no alternative, for example for the residents of nursing or care homes, or other elderly people who may be unwilling to travel for an investigation.
The third category we considered had an intermediate prevalence of osteoporosis (14%), as for the subjects of our study (Table 1), was predominantly female and approximately 60 years old. For this group, of 100 people scanned, there would be 21 (15) in the osteoporosis group, of which 10 (8) would be true positives. It would be appropriate to perform axial densitometry on all those assessed as positive, since the appropriate treatment for this age group is likely to include bisphosphonates, a relatively expensive therapy, and where it is likely that subsequent monitoring would be considered desirable. Problems are likely to arise with those assessed as negative: since it is known that they would include four (six) false negatives, it would be inappropriate to provide all with reassurance that they do not have osteoporosis. From the patient's point of view this would not be satisfactory, so some of this group are therefore likely to request axial densitometry, which would negate any economies or advantages obtained by performing the peripheral measurements. This use of hip and spine densitometry as a second line investigation could have the adverse effect that the inevitable discordance between some of the results obtained using the two techniques is likely to reduce confidence in bone densitometry among both patients and clinicians and thus reduce its value for diagnosing and treating osteoporosis. Compliance with treatment is a common medical problem and therefore important for osteoporosis where the treatment is long-term. The addition of controversy into the area of the best available objective assessment in clinical use would be unfortunate.
Thus we may conclude that peripheral densitometry can be useful for assessing fracture risk in populations with either a very low or a very high prevalence of osteoporosis, particularly where they would probably not otherwise have any densitometry performed. For those with an intermediate prevalence, and in particular where treatment with bisphosphonates is likely to be appropriate, axial densitometry is the method of choice.
Peripheral measurements are not suitable for monitoring treatment, either because they have been shown not to reflect response at other sites, as with the forearm [21], or because of lack of evidence, as with the heel. Therefore those found to have low bone density would also need axial scans in order to set a baseline for monitoring treatment [9], although there is some controversy about the value of monitoring by repeated bone density measurements [22–24]. If monitoring is considered appropriate, such an approach would result in a large proportion of the patients involved ultimately having axial densitometry. Considering another approach to the subject, Baran et al [25] have recommended considering treatment for those with a peripheral T-score less than -2, to perform hip and spine densitometry for T-scores of -2 to 1, and to reassure those with a T-score above 1. For our series, this advice would mean considering treatment in 14 or 16 patients (depending which heel were measured), reassuring 15 or 12 patients, and performing hip and spine scans on the remaining 71 or 72 patients. With this approach the majority of patients could have both peripheral and axial scans.
While peripheral measurements may be better than the use of a clinical assessment alone since risk factors are poorer predictors of bone density [26], it seems prudent not to assume that they can diagnose osteoporosis [27] and to restrict their use to settings where there is no alternative, such as the high and low prevalence groups we have discussed. In an urban population such as ours, where currently the majority of demand is for scans of the intermediate prevalence group, the appropriate provision should be (and this is our aim) access to a hip and spine densitometry service, within a short travel distance and with no restrictions other than the medical decision that standard referral guidelines are met. It should then be possible to avoid the confusion and waste of resources that can be the consequence when the same population is exposed to two different assessment methods that are not interchangeable, for the same purpose.
|
Acknowledgments |
|---|
Received for publication October 23, 2001. Revision received September 9, 2002. Accepted for publication September 19, 2002.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  E McCauley, A Mackie, D Elliott, and A Chuck Heel bone densitometry: device specific thresholds for the assessment of osteoporosis. Br. J. Radiol., June 1, 2006; 79(942): 464 - 467. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J.L. Newton, D.E.J. Jones, K. Wilton, J. Pairman, S.W. Parry, and R.M. Francis Calcaneal bone mineral density in older patients who have fallen. QJM, April 1, 2006; 99(4): 231 - 236. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. J. Aspray, P. Stevenson, S. E. Abdy, D. J. Rawlings, T. Holland, and R. M. Francis Low bone mineral density measurements in care home residents--a treatable cause of fractures Age Ageing, January 1, 2006; 35(1): 37 - 41. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R Kullenberg Evaluation of a device is not an evaluation of a measurement site Br. J. Radiol., October 1, 2003; 76(910): 755 - 755. [Full Text] [PDF]
|
|
|
|
 |
|
 Robin Goodfellow (42-6) Rheumatology, June 1, 2003; 42(6): 813 - 813. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| BJR | DMFR | IMAGING | ALL BIR JOURNALS |
, female; •, male.
, right heel) are categorized operating points from the original continuous data. The areas under the fitted curves for each heel (left 0.89, right 0.88) were not significantly different (p=0.6).