monitoring skeletal response to treatment: which site to measure in the femur?

IntroductionIn the past decade, osteoporotic fractures have

been recognized as one of the most serious problems

in public health. In white women aged 65–84 yr,90% of fractures of the hip and spine, 70% of fore-arm fractures, and 50% of fractures at other sites arecaused by osteoporosis (1). Compared with otherosteoporotic fractures, hip fractures incur the great-est morbidity and cost for health services (2). Aquarter of hip fracture patients die within a year oftheir fracture (3), and survivors frequently suffer sus-tained disability and loss of independence (4).

Monitoring Skeletal Response to Treatment:Which Site to Measure in the Femur?

Glen M. Blake, PHD, Nancy G. Preston, BA, Rajesh Patel, MSc,Ruth J. M. Herd, PHD, and Ignac Fogelman, MD

Department of Nuclear Medicine, Guy’s Hospital, London, UK

Abstract

In the past it was usual to interpret bone mineral density (BMD) scans of the femur using the femoral neck,trochanter, or Ward’s triangle sites. Recently, a report by the International Committee for Standards in BoneMeasurement recommended that the total hip should be the preferred site for the interpretation of femur BMD,and another study described a new central hip site that may offer improved precision. This article comparesthe longitudinal sensitivities of the different femur BMD sites for monitoring patient response to treatment.The study population was 152 postmenopausal women enrolled in a trial of a bisphosphonate therapy. Spineand hip BMD scans were performed at 0, 1, and 2 yr. The mean percentage change at 2 yr was calculated forsix sites in the hip, and the spine was also included for comparison. Treatment effect was defined as the dif-ference in the BMD change between the treated and placebo groups. Although the data analysis incorporateda term for a calibration change caused by a repair of the dual X-ray absorptiometry scanner, the effect of thisevent on the estimation of treatment effect was negligible. Longitudinal sensitivity was derived by dividing thetreatment effect by the root mean square error (RMSE) of the statistical model. Results (and standard errors)normalized to the ratio of treatment effect: RMSE for femoral neck BMD were as follows: femoral neck: 1.00;trochanter: 1.33 (0.38); intertrochanteric: 0.84 (0.41); total hip: 1.20 (0.38); Ward’s triangle: 1.03 (0.27); cen-tral hip: 1.09 (0.30); spine: 2.08 (0.45). At none of the femur sites was the change in BMD large enough toallow monitoring of response to treatment in individual patients. However, for studies involving the follow-upof a group of subjects, the longitudinal sensitivities of the different femur sites were equal within the statisti-cal errors of the study. In particular, total hip BMD appears to be as effective as femoral neck BMD for detect-ing response to treatment in the femur in the setting of a clinical trial or similar research study.

Key Words: Dual X-ray absorptiometry; femur bone mineral density measurements; longitudinal sensitivity.

Received 08/02/99; Revised 11/10/99; Accepted 12/20/99.Address correspondence to Dr. G. M. Blake, Department of

Nuclear Medicine, Guy’s Hospital, St Thomas St., London, SE19RT, UK. E-mail: [email protected]

149

Original Article

Journal of Clinical Densitometry, vol. 3, no. 2, 149–155, Summer 2000 © Copyright 2000 by Humana Press Inc. All rights of any nature whatsoever reserved. 0169-4194/99/2:149–155/$11.75

The increased recognition of the scale of morbid-ity and mortality attributable to osteoporosis has ledto a major effort by the pharmaceutical industry todevelop new therapeutic strategies for the preventionof fracture (5–7). In addition to these developments,there has been a rapid evolution of new radiologicaltechniques for the noninvasive assessment of skele-tal integrity (8,9). The technique most associatedwith the recent growth in bone densitometry is dualX-ray absorptiometry (DXA) (10). With its advan-tages of high precision, low radiation dose, and sta-ble calibration, DXA is well suited to meet the needfor scanning equipment to assist in the diagnosis ofosteoporosis and aid decisions about treatment.

One of the most important applications of DXA hasbeen in trials of new treatments for osteoporosis(5,11–13). Follow-up scans are also widely used tomonitor the response to treatment of individualpatients. In longitudinal studies, it is important to max-imize the statistical significance of the measuredchanges by choosing measurement sites that combinethe advantages of a large response to treatment withhigh precision (14,15). For most purposes, the pos-teroanterior (PA) scan of the lumbar spine is ideal (16).However, because of the implications of hip fracturefor quality of life, morbidity, and mortality, longitudi-nal studies of femur bone mineral density (BMD) forthe purpose of research studies such as clinical trialsremain an important application of bone densitometry.

Although a number of regions of interest (ROIs)are available for the evaluation of DXA scans of thefemur (Fig. 1), until recently there was a widelyaccepted convention to make clinical decisionsbased on femoral neck BMD. However, two recentdevelopments suggest a need to reevaluate the opti-mum site for longitudinal studies in the hip. First, theInternational Committee for Standards in BoneMeasurements (ICSBM) recommended that the totalhip ROI should be the preferred site for the interpre-tation of femur BMD (17). Second, Takada et al.(18) described a new central ROI (Fig. 1) that com-bines improved precision with good response to thechanges associated with aging and osteoporosis.

To evaluate the total hip and central sites andcompare them with the femoral neck region, we havereviewed data from a recently published trial of abisphosphonate therapy (19) to determine the opti-mum ROI for longitudinal studies in the femur.

Materials and MethodsThe subjects studied were 152 postmenopausal

women all 1–10 yr after the onset of menopause whowere recruited into a double-blind, placebo-con-trolled trial of cyclical etidronate therapy (19).Subjects were randomly allocated to one of twotreatment groups: oral etidronate (400 mg/d) for 14d followed by 76 d of calcium supplements (500mg/d), or placebo for 14 d followed by 76 d of cal-cium. Each 90-d treatment cycle was repeated eighttimes for a total of 2 yr. The study was approved bythe Lewisham and North Southwark local ethicscommittee, and all participants gave written,informed consent.

150 Blake et al.

Journal of Clinical Densitometry Volume 3, 2000

Fig. 1. Sites for bone density evaluation in thefemur. The manufacturer’s routine scan analysis softwaregenerates BMD results for the femoral neck, trochanter,intertrochanteric, and Ward’s triangle regions. Total hipBMD is the area weighted mean BMD for the femoralneck, trochanter, and intertrochanteric sites. The centralROI was generated using the method described by Takadaet al. (18).

DXA StudiesPA spine and left hip scans were performed on a

QDR-2000 DXA system (Hologic, Bedford, MA) atbaseline, 1, and 2 yr. Scans were acquired using thearray mode and were analyzed according to the man-ufacturer’s standard protocols using the scan com-parison facility. The skeletal sites studied were thelumbar spine (L1–L4) and the five standard ROIsites in the femur (femoral neck, trochanter,intertrochanteric, total hip, and Ward’s triangle) (seeFig. 1). After publication of the study by Takada etal. (18) describing the central hip site, the femurscans were reanalyzed and the Ward’s triangle regionwas modified to create the central ROI as describedby Takada et al. (18).

Maintenance of DXA ScannerThe stability of the DXA system was monitored

by daily scanning of a Hologic spine phantom.During a preventive maintenance visit halfwaythrough the 2-yr scans, an oil leak was discoveredfrom the X-ray tube coolant caused by a perishedseal. The instrument was repaired and recalibratedby the manufacturer’s representative. The dailyscans of the phantom before and after this eventconfirmed that the system performed within themanufacturer’s specifications throughout thestudy.

Statistical AnalysisThe effects of etidronate therapy were evaluated

by expressing the changes in BMD at 2 yr as the per-centage change from baseline. The treatment effectwas defined as the difference in percentage changebetween the etidronate and placebo groups. To inves-tigate a suspected change in in vivo calibration of theDXA scanner following the repair of the oil leak,multivariate regression analysis was performed withpercentage change in BMD from baseline to 2 yr asthe dependent variable and treatment group and scandate (before and after the repair) as independent vari-ables using Eq.1 (20):

% change in BMD = a + b × treatment code + c × QC code (1)

in which treatment code is 0 for the placebo groupand 1 for the etidronate group; is 0 for 2-yr scansperformed before the repair of the oil leak and 1

for scans performed afterward; a is the true changeat 2 yr in the placebo group; b is the true treatmenteffect; and c is the calibration shift followingrepair of the DXA scanner. Statistical analysis wasperformed using commercial software (Stata,College Station, TX). The coefficients b and cwere taken to be statistically significant if p < 0.05.To establish whether the six hip sites and the spineshowed statistically significant differences in treat-ment effect or calibration shift, the differences inthe baseline to 2-yr percentage changes in BMDbetween pairs of sites were analyzed using Eq. 1.The pairwise analysis allowed the correlationsbetween the BMD changes at different sites andavoided loss of statistical power in evaluating thedifferences among sites.

Evaluation of the longitudinal sensitivity of aDXA ROI requires knowledge of the precisionerrors of the BMD measurements as well as thetreatment effect, because a site with a large treat-ment effect might nevertheless prove unsuitablefor follow-up scans if the precision is poor. Thus,longitudinal sensitivity should be evaluated fromthe ratio of the treatment effect to the precisionerror. Depending on the application, there are twoways in which precision can be defined for thiscalculation: from the coefficient of variation (CV)of repeated scans (although this should reflect thelong-term rather than the short-term CV; [21]),and from the root mean square error (RMSE)obtained from the statistical model used to evalu-ate the clinical trial data. The latter differs from theCV in including the variations in clinical responseamong individual patients as well as the randomerrors arising in the scanning equipment. Theformer definition should be used when estimatinglongitudinal sensitivity for monitoring individualpatients, whereas the latter is more appropriate forresearch studies involving a group of patients.Because the latter context was the main objectiveof the present study, the longitudinal sensitivity ofthe six hip ROIs was compared by dividing the 2-yr treatment effect by the RMSE error obtainedfrom the multivariate regression model. Forcomparison purposes the spine was also included.The treatment effect: RMSE ratio for each femurROI and the spine were compared to the femoralneck site.

Which Site to Measure in the Femur? 151


ResultsOf the 152 subjects enrolled in the study, 135

(71 placebo, 64 etidronate) completed the trial at 2yr. A study on the prevention of postmenopausalbone loss by etidronate treatment was describedpreviously (19). The effect of the repair of theDXA system on its calibration and the significancefor instrumental quality control were alsodescribed previously (20). In the present study, weexamined the longitudinal sensitivity of the femurBMD data to determine the optimum ROI for fol-low-up scans in the hip. Of the 135 women com-pleting the 2-yr study, DXA data on the left femurwere available for 133. For one patient, the 2-yrfemur scan was lost because of the breakdown ofthe scanner, and data for another patient wereomitted because the right femur was scannedinstead of the left owing to an earlier hip replace-ment operation. Results for the central ROI werelost in another subject because the baseline scancould not be recovered from the archiving mediafor reanalysis. Of the 133 femur scans studied at 2yr, 66 (33 placebo, 33 etidronate) were completedbefore the repair of the oil leak and 67 (37 placebo,30 etidronate) afterward.

Baseline demographic data and BMD data forspine, femoral neck, trochanter, and Ward’s trianglewere not statistically significantly different betweenthe etidronate and placebo groups (19). Mean totalhip BMD at baseline was 0.796 g/cm2 for the

etidronate and 0.805 g/cm2 for the placebo group.Equivalent figures for the central ROI were 0.661and 0.673 g/cm2. Neither site was statistically signif-icantly different between the two groups.

Multivariate regression analysis (Eq. 1) confirmeda statistically significant calibration shift for five ofthe six femur ROIs (20). The calibration shift(and standard error [SE]) was +1.67% (0.68%)(p = 0.015) for femoral neck BMD, +2.14% (0.51%)(p < 0.0001) for total hip BMD, and +2.17% (0.61%)(p = 0.0006) for trochanter BMD. Only the result forWard’s triangle (+1.48% [1.16%]) was not statisti-cally significant. The pairwise comparison of sitesshowed that there was no significant difference incalibration shift between any femur ROI. Althoughthe calibration shift for PA spine BMD (–1.05%[0.62%]) was not significantly different from zero, itwas significantly different from the shift for everyfemur site (20).

The treatment effect at 2 yr, defined as the dif-ference in the mean percentage changes betweenthe etidronate and placebo arms of the study, wasstatistically significantly different from zero forthe spine and all six sites in the femur (Table 1).When the analysis was repeated dropping the cali-bration shift term in Eq. 1, the difference in thetreatment effect between the two analyses was neg-ligible (Fig. 2). Pairwise analysis showed that theonly statistically significant difference in treatmenteffect between the femur ROIs was that between

152 Blake et al.


Table 1Treatment Effect and RMSE after 2 yr

Treatment effect Precision

Site of Etidronate-placebo Ratio of femoral Treatment effect: RMSEmeasurement % change (SEM) neck (and SEM) RMSE (SE) (%) ratio to femoral neck

Femoral neck 2.00 (0.68); p = 0.004 1.00 3.90 (0.24) 1.00Trochanter 2.41 (0.62); p < 0.001 1.21 (0.34); NS 3.54 (0.22) 1.33 (0.38); NSIntertrochanteric 1.46 (0.59); p = 0.014 0.73 (0.36); NS 3.37 (0.21) 0.84 (0.41); NSTotal hip 1.81 (0.51); p = 0.001 0.91 (0.29); NS 2.95 (0.18) 1.20 (0.38); NSWard’s triangle 3.54 (1.17); p = 0.003 1.77 (0.47); NS 6.71 (0.41) 1.03 (0.27); NSCentral hip 2.20 (0.69); p = 0.002 1.10 (0.31); NS 3.93 (0.24) 1.09 (0.30); NSSpine (L1–L4) 3.85 (0.63); p < 0.001 1.93 (0.41); p = 0.032 3.59 (0.22) 2.08 (0.45); p < 0.02

the Ward’s triangle and intertrochanteric sites(p = 0.039). Pairwise analysis between the spineand each femur ROI showed a statistically signifi-cantly larger treatment effect in the spine thanthe femoral neck (p = 0.032), intertrochanteric(p = 0.003), and total hip (p = 0.004) ROIs. Thetreatment effect for Ward’s triangle was not statis-tically significantly different from the spine, nei-ther were the trochanter (p = 0.064) nor central hipsites (p = 0.051).

The results for longitudinal sensitivity calculatedfrom the ratio of the 2-yr treatment effect dividedby the RMSE showed no significant differencebetween any of the femur sites (Fig. 3). Althoughthe study confirmed that the spine is the optimumsite for longitudinal studies, the spine ROI was notsignificantly different from the total hip ortrochanter (Table 1). However, the spine ratio wassignificantly larger than the femoral neck (p <0.02), intertrochanteric (p < 0.01), Ward’s triangle(p < 0.01), and central (p < 0.02) sites.

DiscussionDXA scanning provides a precise and sensitive

method of measuring BMD changes at selected sitesin the skeleton. Because of these advantages, one ofthe most important applications of DXA has been inclinical trials of new treatments for osteoporosis(5,19). The stable calibration of DXA equipment hasbeen a major reason for its successful use in such tri-als (11). The present study was therefore unusual infinding evidence of changes in in vivo calibrationthat could not be predicted or explained by the usualquality control procedures. Discussions with themanufacturer suggest that the calibration change wascaused by a misalignment of the collimator slit andthe detector array when the system was repaired. Adetailed explanation was presented elsewhere (20).

The primary objective of the present study was toevaluate the longitudinal sensitivities of the total hipand central ROIs and to compare them with the con-ventional femur sites and the spine. To achieve thisobjective, it was first necessary to evaluate theeffects of the calibration shift on the DXA measure-ments. The analysis showed that the effects of thecalibration shift on the estimation of treatment effect



Fig. 2. Plot of the treatment effect defined as the dif-ference in the mean percentage changes in BMD frombaseline to 2 yr measured in the etidronate- and placebo-treated groups. Results are shown for multivariate regres-sion analyses (Eq. 1) that both include and drop thecalibration shift term. The error bars are ±1 SE. The ran-dom statistical errors owing to the finite number of sub-jects enrolled in the study are much larger than thecorrection owing to the calibration shift.

Fig. 3. Longitudinal sensitivity of the spine and dif-ferent BMD sites in the femur to monitoring response totreatment calculated by dividing the treatment effect bythe RMSE of the multivariate regression analysis. Theresults show this ratio normalized to femoral neck BMD.The error bars are ±1 SE.

was negligible (Fig. 2). Our earlier report (20)showed that this latter finding was explained by thethorough randomization of subjects in the placeboand etidronate arms of the trial. Instead, the principallimitation to the determination of the treatmenteffect was the statistical errors owing to the finitenumber of subjects enrolled in the study (Fig. 2).

The femur ROI showing the largest treatmenteffect was Ward’s triangle (Fig. 2). However, this sitealso showed the largest RMSE (Table 1), and, as aresult, the utility of this site for follow-up scans waslimited. The central ROI described by Takada et al.(18) showed a treatment effect and RMSE compara-ble to that of the femoral neck region. The total hipROI recently recommended for femur evaluations bythe ICSBM committee (17) showed a slightlysmaller treatment effect. However, the smallerRMSE meant that when evaluated by the ratio oftreatment effect over RMSE the total hip performedwell (Fig. 3). The optimum femur site was thetrochanter. However, none of the femur sites werestatistically significantly different in their longitudi-nal sensitivity.

The results of the present study can be comparedwith the data reported by Faulkner et al. (15) for 2-yr treatment of early postmenopausal women with 5mg of alendronate daily. In their study, the treatmenteffect was as follows: PA spine BMD, 5.24 vs 3.85%in the present study; femoral neck BMD, 2.84 vs2.00%; trochanter 3.88 vs 2.41%; and total hip, 3.27vs 1.81%. After allowing for the larger treatmenteffect of alendronate, the comparisons among siteswere similar to those of the present study.

As demonstrated previously by Eastell (16), eventreatment with a potent bisphosphonate such as alen-dronate does not bring about large enough changesin femur BMD to allow monitoring of the responseof individual patients using DXA scanning of thehip. Only for lumbar spine BMD is the response toalendronate sufficient to allow the measurement ofstatistically significant changes in individual sub-jects. It was therefore not surprising that for thestudy reported here, using a less potent therapy, thechanges in femur BMD were too small to permitmonitoring of individual patients. Indeed, even thechanges in lumbar spine BMD were marginal forthis purpose (Table 1).

Given this limitation, the principal aim of the pre-sent study was to evaluate the optimum femur ROIfor research studies involving follow-up of a groupof subjects. Within the statistical limitations of thestudy, because of the finite number of subjectsenrolled and the relatively weak treatment effect ofetidronate, the longitudinal sensitivities of the differ-ent femur ROIs were found to be equal. In particular,the total hip BMD recommended by the ICSBMcommittee for the diagnosis of osteoporosis usingfemur DXA (17) appeared to be as effective asfemoral neck BMD for detecting response to treat-ment in the femur in the setting of a clinical trial orsimilar research study.

References1. Melton LJ, Thamer M, Ray NF, et al. 1997 Fractures attrib-

utable to osteoporosis: report from the NationalOsteoporosis Foundation. J Bone Miner Res 12:16–23.

2. Ray NF, Chan JK, Thamer M, Melton LJ. 1997 Medicalexpenditures for the treatment of osteoporotic fractures inthe United States in 1995: report from the NationalOsteoporosis Foundation. J Bone Miner Res 12:24–35.

3. Jacobsen SJ, Goldberg J, Miles TP, Brody JA, Stiers W,Rimm AA. 1992 Race and sex differences in mortality fol-lowing fracture of the hip. Am J Public Health82:1147–1150.

4. Koval KJ, Skovron ML, Ahanonoff GB, Meadows SE,Zuckerman JD. 1995 Ambulatory ability after hip fracture:a prospective study in geriatric patients. Clin Orthop310:150–159.

5. Liberman UA, Weiss SR, Bröll J, et al. 1995 Effect of oralalendronate on bone mineral density and the incidence offractures in postmenopausal osteoporosis. N Engl J Med333:1437–1443.

6. Black DM, Cummings SR, Karpf DB, et al. 1996Randomised trial of the effect of alendronate on risk of frac-ture in women with existing vertebral fractures. Lancet348:1535–1541.

7. Delmas PD, Bjarnason NH, Mitlak BH, et al. 1997 Effectsof raloxifine on bone mineral density, serum cholesterolconcentrations, and uterine endometrium in postmenopausalwomen. N Engl J Med 337:1641–1647.

8. Genant HK, Engelke K, Fuerst T, et al. 1996 Noninvasiveassessment of bone mineral and structure: state of the art. JBone Miner Res 11:707–730.

9. Grampp S, Genant HK, Mathur A, et al. 1997 Comparisonsof non-invasive bone mineral measurements in assessingage-related loss, fracture discrimination and diagnostic clas-sification. J Bone Miner Res 12:697–711.

10. Blake GM, Wahner HW, Fogelman I. 1999. The evaluationof osteoporosis: dual energy x-ray absorptiometry and ultra-sound in clinical practice, 2nd ed. London: Martin Dunitz.

154 Blake et al.


11. Orwoll ES, Oviatt SK. 1991 Longitudinal precision of dual-energy X-ray absorptiometry in a multicenter study. J BoneMiner Res 6:191–197.

12. Glüer C-C, Faulkner KG, Estilo MJ, Engelke K, Rosin J,Genant HK. 1993 Quality assurance for bone densitometryresearch studies: concept and impact. Osteoporos Int3:227–235.

13. Faulkner KG, McClung MR. 1995 Quality control of DXAinstruments in multicenter trials. Osteoporos Int 5:218–227.

14. Blake GM, Herd RJM, Fogelman I. 1996 A longitudinalstudy of supine lateral DXA of the lumbar spine: a compar-ison with posteroanterior spine, hip and total-body DXA.Osteoporos Int 6:462–470.

15. Faulkner KG, McClung MR, Ravn P, et al. 1996 Monitoringskeletal response to therapy in early postmenopausalwomen: which bone to measure? J Bone Miner Res11(Suppl. 1):S96.

16. Eastell R. 1998 Treatment of postmenopausal osteoporosis.N Engl J Med 338:736–746.

17. Hanson J. 1997 Letter to the editor: standardization offemur BMD. J Bone Miner Res 12:1316, 1317.

18. Takada M, Grampp S, Ouyang X, Engelke K, Genant S.1997 A new trabecular region of interest for femoral dual X-ray absorptiometry: short-term precision, age related boneloss and fracture discrimination compared with currentfemoral regions of interest. J Bone Miner Res 12:832–838.

19. Herd RJM, Balena R, Blake GM, Ryan PJ, Fogelman I.1997 The prevention of early postmenopausal bone loss bycyclical etidronate therapy: a 2-year double-blind placebo-controlled study. Am J Med 103:92–99.

20. Blake GM, Preston NG, Patel R, Herd RJM, Fogelman I.1999 An unexpected change in DXA calibration notdetected by routine quality control checks. Osteoporos Int9:115–120.

21. Glüer C-C, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK.1995 Accurate assessment of precision errors: how to mea-sure the reproducibility of bone densitometry techniques.Osteoporos Int 5:262–270.



monitoring skeletal response to treatment: which site to measure in the femur?

Documents