O

Dd

JBÉSa

b

c

d

e

f

g

h

i

j

k

l

m

AAA

KVMRAD

1

cd

1d

Joint Bone Spine 81 (2014) 169–174

Available online at

www.sciencedirect.com

riginal article

ifficulties in the diagnosis of vertebral fracture in men: Agreement betweenoctors

acques Fechtenbauma,∗, Karine Briota, Simon Paternottea, Maurice Audranb, Véronique Breuil c,ernard Cortetd, Franc oise Debiaise, Franck Grados f, Pascal Guggenbuhlg, Michel Larocheh,rick Legrandb, Éric Lespessailles i, Christian Marcelli j , Philippe Orcelk, Pawel Szulc l, Thierry Thomasm,ami Koltaa, Christian Rouxa, the bone section of the French Rheumatology Society

Paris Descartes University, Rheumatology department, Cochin Hospital, 27, rue du Faubourg-St-Jacques, 75014 Paris, FranceDepartment of rheumatology, LUNAM, CHU d’Angers, 49933 Angers cedex 09, FranceRheumatology, Hôpital l’Archet, 06202 Nice cedex, FranceDepartment of rheumatology, Hôpital Roger-Salengro, CHU de Lille, 59037 Lille cedex, FranceDepartment of rheumatology, CHU la Milétrie, 86021 Poitiers cedex, FranceDepartment of rheumatology, CHU Amiens Nord, 80054 Amiens cedex 1, FranceDepartment of rheumatology, Hôpital Sud, CHU de Rennes, 35203 Rennes cedex 2, FranceDepartment of rheumatology, Hôpital Purpan, 31059 Toulouse cedex 09, FranceDepartment of rheumatology, CHR la Source, 45067 Orléans cedex 2, FranceDepartment of rheumatology, CHU Côte de Nacre, 14033 Caen cedex, FranceDepartment of rheumatology, CHU Lariboisière, 75475 Paris cedex 10, FranceInserm, Hôpital de Lyon, 69437 Lyon cedex, FranceDepartment of rheumatology, Hôpital Nord, CHU de Saint-Étienne, 42055 Saint-Étienne, France

a r t i c l e i n f o

rticle history:ccepted 9 July 2013vailable online 23 January 2014

eywords:ertebral fractureale osteoporosis

adiographylgorithmiagnosis of vertebral fractures

a b s t r a c t

The agreement for vertebral fracture (VF) diagnosis in men, between doctors is poor.Objectives: To assess the agreement for VF diagnosis, in men, on standard radiographs, between experts,before and after consensual workshop and establishing an algorithm.Methods: The agreement between thirteen experimented rheumatologists has been calculated in thirtyosteoporotic men. Then, the group discussed in a workshop and 28 other radiograph sets of osteoporoticmen with follow-up radiographs and incident confirmed VF, have been reviewed. The experts identifiedand hierarchised 18 pathological features of vertebral deformation and established an algorithm of VFdiagnosis. Eleven experts have realized a second reading of the first set of radiographs. We compared theagreement between the 2 readings without and with the algorithm.Results: After consensus and the use of the algorithm the results are: number of fractured patients (with

at least 1 VF) according to the experts varies from 13 to 26 patients out of 30 (13 to 28 during the firstreading). The agreement between the experts at the patient level is 75% (70% at the first reading). Amongthe 390 vertebrae analyzed by the experts, the number of VF detected varies from 18 to 59 (18 to 98 at thefirst reading). The agreement between the experts at the vertebral level is 92% (89% at the first reading).The algorithm allows a good improvement of the agreement, especially for 8 of the 11 experts.Discrepancies for the VF diagnosis between experts exist. The algorithm improves the agreement.

© 2013 Société franc aise de rhumatologie. Published by Elsevier Masson SAS. All rights reserved.

. Introduction

Vertebral fractures (VF) have, in men and women, well-knownonsequences in term of mortality [1,2] and morbidity [3]. Theiriagnosis on standard radiographs is sometimes, difficult due to

∗ Corresponding author. Tel.: +33 1 58 41 25 84; fax: +33 1 58 41 13 70.E-mail address: jacques.fechtenbaum@cch.aphp.fr (J. Fechtenbaum).

297-319X/$ – see front matter © 2013 Société franc aise de rhumatologie. Published by Eoi:10.1016/j.jbspin.2013.12.006

the high frequency of vertebral body deformation of different ori-gins. The frequency of the non-osteoporotic vertebral deformationsis higher in men than in women, due to traumatisms, disc andvertebral degeneration, Scheuermann disease. . . [4,5].

Different methods of VF diagnosis, on standard radiographs,

have been developed and are used in clinical studies: quantita-tive with vertebral height measurement and vertebral height ratioscalculation, semiquantitative evaluation with or without visualadjudication. These methods have been developed in an attempt to

lsevier Masson SAS. All rights reserved.

1 t Bone Spine 81 (2014) 169–174

rmm

dsiafot

2

2

pc

asbra

ha

2

idcgilad

vdoojftt1E(

2

2

tvrt

2

q

70 J. Fechtenbaum et al. / Join

educe the disagreement between physicians. However, the agree-ent between readers in women remains insufficient [6] and theethods used for VF diagnosis give different results in men [7].The main objective of our study is to assess the agreement for VF

iagnosis in osteoporotic men by rheumatologists. The goals of ourtudy are to identify and describe the radiological semiological find-ngs of fractured vertebrae in osteoporotic men by experts, to reach

consensus on the essential findings and to establish an algorithmor the definition of VF. The primary endpoint is the improvementf the agreement for the VF diagnosis in men between experts, afterhe use of the algorithm.

. Methods

.1. Patients

Patients’ radiographs were selected from the baseline data of arospective multi centre, international, double blind and placebo-ontrolled trial of an anti-osteoporotic drug.

Study population included 261 ambulatory Caucasian men oft least 65-years-old. All the patients were osteoporotic on den-itometric criteria. The bone mineral density (BMD) was measuredy Dual energy X-Ray Absorptiometry (DXA) devices. The T scoreanged between −2.5 and −4 standard deviations at lumbar spinend/or femoral neck and/or total hip.

A total of 30 radiographic sets among the study populationave been randomly selected form the different centres, taking intoccount the size of the centre in order to avoid a “centre’s effect”.

.2. Radiographs

Radiographs were performed with a standardized acquisitionn all centres. In addition, the “X-rays technical manual”, whichescribes the standard method of acquisition, was provided to eachentre [8]. Briefly, the patient positions to perform the spine radio-raphs are: for the antero-posterior views, the patient was lyingn the supine position and for the lateral views, the patient was inateral decubitus position, choosing the side that allows a betterlignment of the spine (left decubitus is preferable). The film-focalistance is 100 centimeters.

Each view was anonymized by the central reader. The T8 and T12ertebrae were labelled on each view in order to avoid any labellingiscrepancy. Five views were used: thoracic spine centered on T7n antero-posterior and lateral views, lumbar spine centered on L3n antero-posterior and lateral views, and a thoraco-lumbar lateralunction view centered on T12. The numerical views were trans-ormed from DICOM to JPEG, and plain films were transformedo JPEG. There was no deterioration of film quality due to theseransformations. The obtained files were copied on CD and sent to3 rheumatologists working in hospitals and experts in bone field.ach CD includes an excel file of 30 lines numbered from 1 to 3030 patients) and of 13 columns for the 13 vertebrae from T4 to L4.

.3. Assessment of vertebral fractures

.3.1. Evaluation by the expertsEach expert evaluated the presence or not of a VF, according

o his (her) experience, for each of the 30 patients and for eachertebra from T4 to L4 and reported the result on the excel file. Noecommendation was given to the experts and all of them knewhe different classification methods for VF assessment.

.3.2. Centralised readingA centralized reading was done, using 5 classifications: 3 semi-

uantitatives, 1 qualitative and 1 quantitative. The first method is

Fig. 1. a: vertebral fracture assessment by the semiquantitative method of Genant;b: vertebral fracture assessment by Algorithm-Based Qualitative (ABQ).

the most commonly used, called SQ for semiquantitative assess-ment described in 1993 in women by Genant et al. [9] (Fig. 1a).The first question to begin the assessment is: “is there a heightreduction of the vertebral body?” by visual estimation. The heightreduction could affect the anterior and/or the middle and/or theposterior part of the vertebral body. The severity of the reductionis appreciated, by visual estimation, reduction of 20 to 25% for thegrade 1 or mild; 25 to 40% of reduction for the grade 2 or moder-ate and more than 40% for the grade 3 or severe VF. The secondmethod is called SQ adjudicated for a semiquantitative assessmentthen adjudication by expert opinion also described by Genant, inorder to eliminate, among the vertebral deformities detected by SQ,what is not an osteoporotic VF. With careful attention paid on thenon-fractured deformities as: poor quality of the radiograph withan excessive obliquity of the X-ray beam, Scheuermann’s diseaseand Schmorl’s nodes, degenerative changes due to osteoarthritis.We take into consideration the normal variants: variation in thevertebral body size and shape. We, also distinguish the differencesbetween thoracic vertebrae and lumbar vertebrae. Thoracic verte-brae are often wedge and lumbar vertebrae could be biconcave(Fig. 2a and 2b). The method 3 is based on the SQ assessment ofGenant where the author stated that the grade 1 does not existat the thoracic level in men [10]. The method 4 is called ABQ forAlgorithm-Based Qualitative assessment. This method has beenextensively described in women [11] and in men [12] (Fig. 1b). Tosummarize, the first question to begin the assessment, is: “is therea depression of the endplate?” This method eliminates the normalvariants and the short vertebral heights (Fig. 1b). The fifth method is

quantitative and uses a software developed by Optasia Medical, UK[13,14]. This vertebral morphometric software permit after man-ual vertebral labelling to rebuild automatically the vertebral bodyfrom many points sticked by the software on the vertebral body

J. Fechtenbaum et al. / Joint Bone Spine 81 (2014) 169–174 171

Fig. 2. a: thoracic vertebrae: wedge and non-fractured; b: lumbar vertebrae: bicon-cave and non-fractured; c: Good concordance for the diagnosis of vertebral fracture(df

[tbber

2

wetboetftv

2

gdb

2

fkoW

Fig. 3. a: comparison between experts and other reading methods at the patient

13 experts out of 13 diagnosed a vertebral fracture); d: poor concordance for theiagnosis of vertebral fracture at L1 (7 experts out of 13 diagnosed a vertebralracture).

15]. The radiographs in DICOM format are automatically read byhe software, the other radiographs are transformed to TIFF formatefore assessment. As every automatic software of the vertebralody edge detection, a manual correction is possible and often nec-ssary [16,17]. The centralized assessment has been realized by oneeader (KB) for ABQ and another one (JF) for SQ and morphometry.

.3.3. Consensus workshopEleven of the 13 experts have participated to the consensus

orkshop. During the 5 hours of the workshop, the experts havevaluated 28 different radiographic sets of patients included inhe same study. They presented a VF according to all the expertsecause it was an incident one (that was not present on a previ-us radiograph 1 or 2 years earlier). From these incidents VF, thexperts established the semiological radiological features they usedo obtain the VF diagnosis, creating a verbatim of 18 characteristiceatures. Then a Delphi method was used to hierarchise and selecthe 5 most discriminating signs. The decision-making algorithm ofertebral deformations has thus been created.

.3.4. Second assessment by the expertsEleven out of the 13 experts have read again the first 30 radio-

raph sets, used for the reproducibility. This second reading wasone after the workshop with the same design as the first reading,ut with the help of the algorithm.

.4. Statistical analysis

The concordance was evaluated by K statistics. The PABAK score

or prevalence-adjusted bias-adjusted kappa was preferred to the

score because it is able to correct the bias due to a poor frequencyf an event. The VF probability is less than that of a normal vertebra.e have calculated:

level; b: comparison between experts and other reading methods at the vertebrallevel.

• the concordance between experts before consensus workshop;• the concordance between experts and central reading.

In order to calculate the concordance between each expertassessment and each of the five central reading methods, we estab-lished a “gold standard”. We defined empirically this reference asthe diagnosis according to at least 10 experts out of the 13. We havecompared this gold standard to the central readings.

The concordance between experts after consensus workshop,using the algorithm.

3. Results

3.1. First assessment by experts

The number of patients considered as fractured, that is pre-senting at least one fractured vertebra on the analyzed vertebraefrom T4 to L4, according to the expert’s opinion is reported onFig. 3a. The number of fractured patients ranges from 13 to 28patients among 30. The median number of fractured patients is20. The concordance percentages between experts at the patient’slevel ranges from 50% to 100% with a mean of 70.3%. The PABAKscores range from 0 to 1 with a mean of 0.40.

Then, we have calculated for each expert, the number of frac-tured vertebrae among the 390 analyzed vertebra (Fig. 3b). Thisnumber ranges from 18 to 98 vertebrae. Fig. 2c and 2d illustratetwo examples. The median number of fractured vertebrae is 36.The mean concordance percentage between experts at the verte-

bral level is 89% and ranges from 79% to 97%. The PABAK scoresrange from 0.57 to 0.93 with a mean of 0.78. The mean number ofdiscrepancies between experts on the diagnosis, at the vertebrallevel is 43 out of the 390 analyzed vertebrae and ranges from 13

172 J. Fechtenbaum et al. / Joint Bone Spine 81 (2014) 169–174

Table 1Percentage of concordance between experts according to the vertebral level before(reading 1) and after the use of algorithm (reading 2).

Reading 1 (%) Reading 2 (%)

T4 93.2 86.2T5 90.9 90.0T6 86.6 93.1T7 82.2 92.2T8 82.4 89.8T9 86.0 94.9T10 91.7 94.2T11 85.4 94.1T12 84.2 86.1L1 87.0 91.2L2 90.0 90.2L3 95.9 96.8

fr

3c

ovo346

3

siapooi

3a

oa7i0aai0atcwtiv[

Fig. 4. a: verbatim of the semiological radiological features of vertebral fracture inmen established by the experts; b: algorithm of the decision of vertebral deformationin osteoporotic men defined by the experts.

L4 96.1 95.8

or the 2 most concordant readers to 84 for the 2 less concordanteaders.

.2. Concordance between the experts’ reading and theentralized reading

We found only 15 vertebrae for which at least 10 readers outf 13 agreed with the vertebral fracture diagnosis. Among these 15ertebrae, 12 were detected by method 1: SQ, that is a concordancef 80%; 15 for method 2: SQ adjudicated, that is 100%; 11 for method: without grade 1 at the thoracic level, that is 73.3%; 11 by method: ABQ, that is 73.3% and 10 for method 5: morphometry, that is7%.

.3. Consensus workshop

Fig. 4a describes the 18 semiological radiological criteria con-idered pertinent by the experts. The vertebral fracture diagnosiss on Fig. 4b. First the radiograph quality has to be evaluated that is:ll the vertebral body contours are eligible, the vertebral body end-lates are well superimposed, correct exposition. . . The secondarysteolysis are excluded. The following steps search for the presencef a vertebral deformation, and then confirm that this deformations an osteoporotic vertebral fracture.

.4. Second experts’ assessment after the workshop and with thelgorithm

Fig. 5a and 5b show the results for each reader with and with-ut the algorithm. The concordance percentages between expertst the patient’s level range from 53.3% to 93.3% with a mean of5.03% (from 50 to 100% and mean 70.3% during the first read-

ng). The PABAK scores range from 0.10 to 0.87 with a mean of.49 (respectively 0; 1; 0.4 at the first reading). The mean concord-nce percentage between experts at the vertebral level is 91.84%nd ranges from 83.1 to 97.9% for 89% [79%–97%] at the first read-ng. The PABAK scores range from 0.66 to 0.96 with a mean 0.84;.57 to 0.93 with a mean 0.78 at the first reading. The concord-nce improvement between the 2 readings is principally based onhe thoracic vertebrae T6 toT10 where are the most important dis-repancies as shown in Table 1. When we re-analyze the resultsithout the 3 readers remaining discordants (readers 1, 6 and 7)

he concordance percentage, for the 8 readers, at the patient’s level

s 81% [70–93,3] with a PABAK score of 0.6 [0.3–0.87] and at theertebral level 94.3% [88.2–97.9] of agreement and a PABAK of 0.890.76–0.96].

J. Fechtenbaum et al. / Joint Bone

Fb

4

ffiiiotdddcft

oGdsirtfdemnavditT

different readers, imaging modalities, and diagnosis approaches. J Bone Miner

ig. 5. a: variation of the readings without and with algorithm, at the patient level;: variation of the readings without and with algorithm, at the vertebral level.

. Discussion

This is the first study describing the diagnosis of vertebralracture in osteoporotic men, using experience of experts in thiseld. We have noticed important discrepancies between experts

n vertebral fracture diagnosis [5]. Vertebral fracture assessments sometimes difficult in clinical practice. Vertebrae present a lotf different shapes. These aspects could be modified by consti-utional or acquired deformations. Traumatisms, including whileoing sports, lead to vertebral deformations that are most often notocumented during life. The aging of the spine and osteoarthriticegenerative disc disease lead also to vertebral body shape modifi-ations. All these vertebral deformations have to be differentiatedrom vertebral fractures. Vertebral deformations due to infection,umor and metabolic diseases are more easily eliminated.

Diagnostic methods of VF in women and men have been devel-ped in order to improve the diagnostic concordance of VF. Theenant semiquantitative classification has improved the repro-ucibility and lead to a homogeneity in VF diagnosis in clinicaltudies; it gives good results especially with a centralised read-ng. Some difficulties remain to classify the vertebrae which heighteduction does not reach 20%, while the fracture is certain and forhe vertebrae which height reduction is more than 20% but is notractured. The qualitative method ABQ is based on the endplateepression and permits to eliminate some diagnosis discrepancies,specially at the thoracic level. The thoracic vertebrae are often nor-ally wedged and the VF diagnosis is very difficult at this level. The

ecessity of the vertebral endplate depression for the VF diagnosist the thoracic level permits the elimination the non-osteoporoticertebral deformations. However, this depression is much moreifficult to appreciate at the lumbar spine where the vertebra

s spontaneously biconcave. Ferrar and al. [18] have introducedhe notion of short vertebral middle height at the lumbar level.he vertebral morphometry permits to obtain a vertebral height

Spine 81 (2014) 169–174 173

measurement with positioning of points along the vertebral bodyand to reconstitute the vertebral anatomy. However, vertebral bodyshape automatic detection is often false; the manual correction ofpoint position is in that case mandatory. A discrepancy betweenthe different readers could be introduced. Even if the vertebralbody shape analysis is correct, vertebral morphometry detects allthe deformations without any further information on whether ofosteoporotic origin or not. These methods assess differently the VFand there are discrepancies between them in most publications:kappa score about 0.5 [7,11] and in our study too. Therefore, wehave looked for a vertebral fracture classification in men, applica-ble in clinical practice and we have established an algorithm, whichimproves the diagnosis agreement. We have chosen an algorithm-based on the experience of experts who have integrated all the VFreading methods. We have noticed diagnostic discrepancies duringthe first reading. The consensus workshop succeeded to establishthe radiological semiological features of VF and to hierarchise thesefeatures in an algorithm of decision. We noticed an improvementof the concordance between experts during the second reading,unfortunately without a complete consensus. However, the con-cordance has been well improved between 8 of the 11 readers. Inclinical practice, standard radiographs can be completed by otherdiagnostic methods.

The strong points of our study are the original method withexperimented readers in this field. The VF assessment has beenrealized in clinical conditions. The radiographic acquisition wasstandardized. The choice of osteoporotic patients was justified bythe need of VF but this point could limit the external validity ofour results. Our study limitations are the views number, limitedto 5, and the impossibility to access to other exams, which couldbe useful in clinical practice. We did not take into account the VFseverity. We are also unable to differentiate the workshop’s and thealgorithm’s effect on the second reading variations. Discrepanciesbetween readers for the VF diagnosis in osteoporotic men exist. Theconcordance could be improved by a decisional algorithm.

Disclosure of interest

The authors declare that they have no conflicts of interest con-cerning this article.

Acknowledgments

To the Optasia Medical Ltd society, UK, for the provision of themorphometric software Spine Analyzer®.

References

[1] Khosla S, Amin S, Orwoll E. Osteoporosis in men. Endocr Rev 2008;29:441–64.[2] Bliuc D, Nguyen ND, Milch VE, et al. Mortality risk associated with low-

trauma osteoporotic fracture and subsequent fracture in men and women.JAMA 2009;30:1513–21.

[3] Fechtenbaum J, Cropet C, Kolta S, et al. The severity of vertebral fractures andhealth-related quality of life in osteoporotic postmenopausal women. Osteo-poros Int 2005;16:2175–9.

[4] O’Neill TW, Felsenberg D, Varlow J, et al. The prevalence of vertebral deformityin European men and women: the European Vertebral Osteoporosis Study. JBone Miner Res 1996;11:1010–8.

[5] Silman AJ, O’Neill TW, Cooper C, et al. Influence of physical activity on ver-tebral deformity in men and women: results from the European VertebralOsteoporosis Study. J Bone Miner Res 1997;12:813–9.

[6] Fechtenbaum J, Cropet C, Kolta S, et al. Reporting of vertebral fractures on spineX-rays. Osteoporos Int 2005;16:1823–6.

[7] Ferrar L, Jiang G, Schousboe JT, et al. Algorithm-based qualitative and semi-quantitaive identification of prevalent vertebral fracture: agreement between

Res 2008;23:417–24.[8] Kaufman JM, Audran M, Bianchi G, et al. Efficacy and safety of strontium

ranelate in the treatment of osteoporosis in men. J Clin Endocrinol Metab2013;98:552–601.

1 t Bone

[

[

[

[

[

[

[

74 J. Fechtenbaum et al. / Join

[9] Genant HK, Wu CY, Van Kujik C, et al. Vertebral fracture assess-ment using a semiquantitaive technique. J Bone Miner Res 1993;8:1137–48.

10] Szulc P, Munoz F, Marchand F, et al. Semiquantitive evaluation of prevalent ver-tebral deformities in men and their relationship with osteoporosis: the MINOSStudy. Osteoporos Int 2001;12:302–10.

11] Jiang G, Eastell R, Barrington NA, et al. Comparison of methods for the visualidentification of prevalent vertebral fracture in osteoporosis. Osteoporos Int2004;15:887–96.

12] Ferrar L, Jiang G, Cawthon PM, et al. Identification of vertebral fracture andnon-osteoporotic short vertebral height in men: the MrOS Study. J Bone MinerRes 2007;22:1434–41.

13] Kim YM, Demissie S, Eisenberg R, et al. Intra- and inter-reader reliabilityof semi-automated quantitative morphometry measurements and vertebral

[

[

Spine 81 (2014) 169–174

fracture assessment using lateral scout views from computed tomography.Osteoporos Int 2011;22:2677–88.

14] Steiger P. Use of imaging biomarkers for regulatory studies. J Bone Joint SurgAm 2009;91:132–6.

15] Smith-Bindman R, Cummings SR, Steiger P, et al. A comparison of morphomet-ric definitions of vertebral fracture. J Bone Miner Res 1991;6:25–34.

16] Roux C, Fechtenbaum J, Kolta S, et al. Prospective assessment of thoracickyphosis in postmenopausal women with osteoporosis. J Bone Miner Res2010;25:362–8.

17] Briot K, Kolta S, Fechtenbaum J, et al. Increase in vertebral body size in post-menopausal women with osteoporosis. Bone 2010;47:229–34.

18] Ferrar L, Roux C, Felsenberg D, et al. Association between incident and baselinevertebral fractures in European women: vertebral fracture assessment in theOsteoporosis and Ultrasound Study (OPUS). Osteoporos Int 2012;23:59–65.