ARTHRITIS & RHEUMATOLOGYVol. 66, No. 7, July 2014, pp 1734–1744DOI 10.1002/art.38613© 2014, American College of Rheumatology

Effectiveness of Tumor Necrosis Factor � Blockers inEarly Axial Spondyloarthritis

Data From the DESIR Cohort

Anna Molto,1 Simon Paternotte,2 Pascal Claudepierre,3 Maxime Breban,4

and Maxime Dougados2

Objective. To estimate the frequency of use andeffectiveness in daily practice of tumor necrosis factor �(TNF�) blockers in a population with inflammatoryback pain suggestive of early axial spondyloarthritis(SpA).

Methods. The Devenir des SpondylarthropathiesIndifferenciees Recentes (DESIR) cohort is a prospec-tive, multicenter, observational cohort of 708 patientswith early (<3 years’ duration) inflammatory back painsuggestive of axial SpA. The percentage of patientsreceiving TNF� blockers over the first 2 years of fol-lowup was estimated by survival analysis. To evaluateeffectiveness, the primary outcome (40% improvement

in disease activity according to the Assessment of Spon-dyloArthritis international Society criteria [ASAS40])was compared in patients who received TNF� blockersversus control patients who received any other treat-ment (usual care). Controls were matched to the pa-tients based on a propensity score method.

Results. A total of 30.2% (95% confidence interval[95% CI] 26.7–33.7) patients received at least 1 TNF�blocker during the 24 months of followup. The percent-age of ASAS40 responders was 31.5% (62 of 197 pa-tients) in the group receiving TNF� blockers versus13.2% (26 of 197) in the control group (OR 2.99 [95% CI1.80–4.99], P � 0.0002). This effectiveness was morepronounced in the subgroup of patients with sacroiliitisidentified on magnetic resonance imaging, with 46% ofASAS40 responders receiving TNF� blockers versus15% of ASAS40 responders receiving usual care (OR4.99 [95% CI 2.17–11.51]).

Conclusion. Our study shows that TNF� blockersare frequently used in daily practice to treat patientswith early axial SpA. Our findings confirm the effective-ness of TNF� blockers as compared to any othertreatment, especially in the subgroup of patients withsacroiliitis on MRI.

The concept of spondyloarthritis (SpA) as agroup of interrelated diseases has been recognized sincethe early 1970s. The classification and recognition ofaxial SpA relied on the combination of consistent axialskeleton symptoms and unequivocal bilateral grade 2 orunilateral grade 3 radiographic evidence of sacroiliitisaccording to the modified New York criteria (1). Morerecently, the different criteria sets for SpA (e.g., theAmor [2], European Spondyloarthropathy Study Group[ESSG] [3], and ASsessment of SpondyloArthritis inter-

ClinicalTrials.gov identifier: NCT01648907.Supported by a grant from Pfizer Ltd. France. Dr. Molto’s

work was supported by a Master of Science grant from the FrenchSociety of Rheumatology. The Devenir des SpondylarthropathiesIndifferenciees Recentes (DESIR) cohort study is conducted as aProgramme Hospitalier de Recherche Clinique, which is sponsored byAssistance Publique Hopitaux de Paris and is supported by unre-stricted grants from the French Society of Rheumatology and PfizerLtd. France.

1Anna Molto, MD: Paris Descartes University, Cochin Hos-pital, AP-HP, INSERM U1153, PRES Sorbonne Paris Cite, Paris,France, and Universitat Autonoma de Barcelona, Barcelona, Spain;2Simon Paternotte, MSc, Maxime Dougados, MD: Paris DescartesUniversity, Cochin Hospital, AP-HP, INSERM U1153, and PRESSorbonne Paris-Cite, Paris, France; 3Pascal Claudepierre, MD: Uni-versite Paris Est Creteil, EA4393, and Henri-Mondor Hospital, AP-HP, Creteil, France; 4Maxime Breban, MD, PhD: University ofVersailles St.-Quentin-en-Yvelines and Ambroise Pare Hospital, AP-HP, Boulogne-Billancourt, France.

Dr. Claudepierre has received consulting fees, speaking fees,and/or honoraria from Pfizer, AbbVie, UCB, Roche, Merck, andBristol-Myers Squibb (less than $10,000 each).

Address correspondence to Anna Molto, MD, Paris Des-cartes University, Faculty of Medicine, AP-HP, Cochin Hospital,Rheumatology B, 27 Rue du Faubourg St. Jacques, 75014 Paris,France. E-mail: anna.molto@cch.aphp.fr.

Submitted for publication July 26, 2013; accepted in revisedform February 25, 2014.

1734

national Society [ASAS] [4] criteria) have made itpossible to recognize axial SpA in the absence of chronicradiographic changes of the sacroiliac (SI) joints.

Tumor necrosis factor � (TNF�) blockers haveproven to be an effective treatment of SpA, with a largeamount of research evidence (5). Many randomizedcontrolled trials (RCTs) (6–8) have provided data con-firming the efficacy of TNF� blockers in patients withaxial SpA meeting the modified New York criteria (e.g.,radiographic axial SpA or ankylosing spondylitis [AS]patients), but only limited data are yet available forpatients lacking definite radiographic sacroiliitis (9–11).In these and other RCTs, age (12), disease duration(12), and baseline characteristics of disease activity, suchas an elevated C-reactive protein (CRP) level (6,12,13),elevated erythrocyte sedimentation rate (ESR) (6,12),presence of enthesitis (14), scores on the Bath Ankylos-ing Spondylitis Disease Activity Index (BASDAI) (8)and Bath Ankylosing Spondylitis Functional Index(BASFI) (8), HLA–B27 positivity (6), and lack of pre-vious exposure to TNF� blockers (15), have been iden-tified as factors associated with a better response toTNF� blockers.

Only very limited data have been published re-garding the therapeutic effect of TNF� blockers inpatients with axial SpA in clinical practice (16). Al-though RCTs are extremely useful in generating thera-peutic response data, the inclusion criteria are very strictand are not always applicable in daily clinical practice.Therefore, studies aiming to evaluate the therapeuticresponse of TNF� blockers in daily practice conditionsare required.

Devenir des Spondylarthropathies Indifferen-ciees Recentes (DESIR) is a prospective longitudinalcohort study involving 25 rheumatology centers inFrance and 708 patients with SpA. The aim of DESIR isthe comprehensive analysis of the nature and outcomeof axial SpA from early symptom onset (17). Evaluatingthe natural history of the disease in a cohort of patientspresenting with recent onset (�3 years) of inflammatoryback pain suggestive of axial SpA is not an unconven-tional study. However, the hallmark of this cohort is thatthe patients can receive any treatment according to theirrheumatologist’s preference (18), and therefore, such acohort presents a unique opportunity for evaluation ofcurrent practice and the effectiveness of TNF� blockersin daily practice.

The main objective of this study was to evaluateTNF� blockers in clinical practice in a population withearly axial SpA by estimating their frequency of use,

evaluating their effectiveness, and exploring the interac-tion of baseline variables with this effectiveness.

PATIENTS AND METHODS

Patients. A total of 708 patients with early inflamma-tory back pain were included in the DESIR cohort, withscheduled visits every 6 months. Ten-year followup is currentlyongoing; the analyses in the present study include the first 2years of followup.

Patients between the ages of 18 and 50 years withinflammatory back pain involving the thoracic spine, lumbarspine, or buttocks area for �3 months but �3 years and withsymptoms suggestive of SpA according to the assessment bytheir rheumatologist (score �5 on a Numerical Rating Scale of0–10, where 0 � not suggestive of axial SpA and 10 � verysuggestive of axial SpA) were included in the study. Patientswere required to fulfill the Calin et al criteria (19) or the Berlincriteria (20) for inflammatory back pain. Patients with adefinitive diagnosis of nonaxial SpA back pain, any conditionthat could affect the validity of the informed consent and/orprevent the patient from achieving optimal compliance (e.g.,alcoholism, psychiatric disorder), or a history of previous useof a TNF� blocker were excluded. All patients (e.g., thosefulfilling the ESSG, Amor, or ASAS classification criteria foraxial SpA, as well as those not fulfilling any criteria set) wereincluded in our analysis, except for those who were receiving abiologic agent other than a TNF� blocker as the first biologictreatment. The data set used for this analysis was locked onDecember 3, 2012.

Definition of visits. Patients could be started on treat-ment with a TNF� blocker at any time during the followup, attheir rheumatologist’s discretion (e.g., in the interval betweenthe DESIR visits scheduled for every 6 months). To evaluatethe changes occurring between pre- and postinitiation ofTNF� blockers, a definition of “baseline” (before initiation)and “followup” (after at least 8 weeks of TNF� blockertreatment) visits was required. The baseline visit was definedas the last DESIR cohort visit before the initiation of TNF�blockers or the visit taking place within 7 days of initiating suchtreatment. The followup visit was defined as the first DESIRcohort visit taking place after at least 8 weeks of treatment.

Variables. The variables collected at each DESIRcohort visit have been described in previous studies (18).Briefly, information on patients’ characteristics (age, sex, so-ciodemographic features, comorbid conditions, smoking sta-tus, alcohol intake, and familial comorbid conditions), diseasecharacteristics (axial disease, peripheral disease, criteria fulfill-ment [Amor (2), ESSG (3), or ASAS (4)], HLA–B27 status,modified New York criteria for sacroiliitis fulfillment, presenceof inflammatory lesions on MRI, abnormal findings on enthe-sis ultrasound, and bone densitometry status), disease activity(BASDAI [21] and Ankylosing Spondylitis Disease ActivityScore [ASDAS] [22]), and disease severity (Bath AnkylosingSpondylitis Metrology Index [BASMI] [23], BASFI [24],Health Assessment Questionnaire for Ankylosing Spondylitis[HAQ-AS] [25], and variables for calculating the ASAS non-steroidal antiinflammatory drug (NSAID) score [e.g., drugname, mean dosage, and percentage of days taken per weekduring the previous 6 months (26)]) were collected at each

TNF� BLOCKERS IN EARLY AXIAL SpA 1735

DESIR cohort visit or annually, according to the study proto-col (detailed protocol information available online atwww.lacohortedesir.fr/desir-in-english).

Effectiveness end points. Treatment effectiveness wasdefined as the equivalent of efficacy when the treatment effectwas observed in routine clinical practice (e.g., in pragmatictrials) (27,28).

Primary effectiveness end point. The primary effective-ness end point was the proportion of patients achieving anASAS40 response (29) after at least 8 weeks of TNF� blockertherapy. An ASAS40 response was defined as an improvementof at least 40% and an absolute improvement of at least 2 units(on a scale of 0–10) in 3 or more of the 4 areas: the BASFI,patient’s assessment of pain, patient’s global assessment ofdisease activity, and mean score from questions 5 and 6 on theBASDAI, with no worsening in the remaining areas.

Secondary effectiveness end points. The following sec-ondary effectiveness variables were evaluated at every DESIRcohort visit: the BASDAI, BASFI, and ASAS NSAID score forthe last week preceding the visit. Moreover, response wasevaluated according to changes between the baseline andfollowup visits on the following criteria: ASDAS-CRP as acontinuous outcome, ASAS20 response (defined as at least20% improvement and absolute improvement of at least 1 unit[on a scale of 0–10] versus baseline in 3 or more of thefollowing 4 areas: the BASFI, patient’s assessment of pain,patient’s global assessment of disease activity, and the meanscore from questions 5 and 6 on the BASDAI, with noworsening of �20% in the remaining areas as a binaryoutcome), BASDAI50 response (defined as a reduction of atleast 50% or 2 units versus baseline as a binary outcome), and3 binary outcomes based on 3 different ASDAS-CRP thresh-olds: ASDAS-CRP inactive disease (ID; defined as anASDAS-CRP score �1.3), ASDAS-CRP clinically importantimprovement (CII defined as a decrease from baseline of�1.1), and ASDAS-CRP major improvement (MI; defined asa decrease from baseline of �2.0) (30).

Study groups. Effectiveness was compared in 2 groupsof patients: the active drug group (e.g., DESIR patientsreceiving a TNF� blocker during the first 2 years of followup)and the control group (e.g., selected patients in the DESIRcohort receiving any other usual treatment and not TNF�blockers). For this purpose, each patient in the active druggroup was matched with a patient in the control group. Thismatching process took into account that the baseline charac-teristics of these 2 groups were essentially different. To over-come these differences, the matching process was based on ascore evaluating the probability of receiving a TNF� blocker.This methodology is called the propensity score method(31,32). The propensity score is the probability of receiving aparticular treatment conditioned on the individual baselinecharacteristics, here the probability of being treated with aTNF� blocker. All baseline data (except for bone mineraldensity status and abnormal findings on enthesis ultrasound,which were not available for all patients) were used to con-struct the multivariate logistic regression model predicting theprobability of being treated with a TNF� blocker (data notshown). The model performed adequately (area under thecurve 0.880).

Handling of missing data. If a patient discontinuedtreatment between 2 DESIR cohort visits, the informationcollected at the last DESIR cohort visit when the patient was

receiving treatment was collected and carried forward (e.g.,using the last observation carried forward technique). If novisit was available between the baseline visit and time ofdiscontinuation, the baseline observation was carried forward.

Statistical analysis. Estimating the frequency of use ofTNF� blockers. The percentages of patients initiating TNF�blockers during the first 2 years of followup, along with the95% confidence intervals (95% CIs), were estimated at eachintermediate DESIR cohort visit (e.g., 6, 12, 18, and 24months) using the life-table analysis technique according to theKaplan-Meier method.

Evaluating the effectiveness of TNF� blockers. Theprimary effectiveness end point was estimated by assessing thepercentage of patients achieving an ASAS40 response after atleast 8 weeks of treatment among the patients who receivedTNF� blockers for at least 8 weeks and the control patients,who did not receive TNF� blockers but were given any othertreatment (usual care). The control patients were matchedwith patients receiving TNF� blockers in terms of not onlytheir baseline characteristics, but also their time of evaluation.Therefore, the intervals between the followup visit and base-line visit were exactly the same in both study groups. Theprediction of the ASAS40 response was computed by a logisticregression model and adjusted by scores on the BASDAI,BASFI, and ASDAS-CRP. P values (2-tailed) less than 0.05were considered statistically significant.

Secondary effectiveness end points were assessed inthe same way (e.g., ASAS20 response, changes in scores on theBASDAI, ASAS NSAID, and BASFI, etc.) as described above.

These outcome measures were also evaluated in termsof the individual TNF� blocker the patients in the active drugtreatment group took: etanercept, adalimumab, or infliximab.

Exploring the interaction between baseline variables andthe effectiveness of TNF� blockers. To explore the impact ofbaseline characteristics on the primary effectiveness end point,we first reported the percentage of ASAS40 responders in each

Figure 1. Estimate of the probability that treatment with a tumornecrosis factor � (TNF�) blocker will be initiated within the first 2years of followup in a cohort of patients with early axial spondyloar-thritis. Numbers across the bottom are the number of patients at riskat the beginning of the indicated interval.

1736 MOLTO ET AL

Figure 2. Flow chart showing the distribution of study patients included in the analysis of the Devenir des Spondylarthropathies IndifferencieesRecentes (DESIR) cohort data. Response to treatment with tumor necrosis factor � (TNF�) blocker versus any other treatment (usual care) wasdetermined as the presence or absence of a 40% improvement in disease activity according to the ASsessment of SpondyloArthritis InternationalSociety criteria (ASAS40). For the imaging arm, the presence or absence of sacroiliitis on magnetic resonance imaging (MRI) and radiography wasdetermined. For the clinical arm, the presence or absence of an elevated C-reactive protein (CRP) level was determined.

Table 1. Baseline characteristics of the 404 matched patients and controls from the DESIR cohort*

Patients receiving TNF� blockers(n � 202)

Patients receiving usual care(n � 202)

Age, median (IQR) years 33.8 (27.1–41.44) 33.7 (27.2–39.7)No. (%) male 88 (43.6) 80 (39.6)Disease duration, median (IQR) months 17.2 (10.0–27.6) 16.3 (8.6–24.2)No. (%) HLA–B27 positive 115 (56.9) 114 (56.4)Sacroiliitis determination, no. (%)

Radiographic (modified New York criteria) 67 (33.2) 59 (30.1)MRI 82 (40.6) 64 (33.0)

BASDAI, median (IQR) (range 0–100) 57.0 (45.0–67.0)† 53.0 (39.0–65.0)BASFI, median (IQR) (range 0–100) 42.0 (25.0–59.0)‡ 35.0 (18.0–55.0)CRP, median (IQR) mg/liter 5.0 (3.0–13.6) 4.4 (2.9–8.0)ASDAS-CRP, median (IQR) 3.1 (2.4–3.7)§ 2.8 (2.2–3.3)NSAIDs

No. (%) taking NSAIDs 174 (90.6) 168 (89.8)ASAS NSAID score in last week, median (IQR) 55.0 (0.0–100.0) 50.0 (0.0–100.0)

* Patients and controls from the Devenir des Spondylarthropathies Indifferenciees Recentes (DESIR) cohort were matched accordingto a propensity score method (see Patients and Methods for details). Total numbers of patients and controls contributing data wereas follows: 199 and 196 for the modified New York criteria, 198 and 194 for magnetic resonance imaging (MRI), 201 and 201 for theBath Ankylosing Spondylitis Disease Activity Index (BASDAI), 202 and 199 for the Bath Ankylosing Spondylitis Functional Index(BASFI), 196 and 198 for the C-reactive protein (CRP) level, 192 and 187 for the number taking nonsteroidal antiinflammatory drugs(NSAIDs), and 197 and 197 for the Assessment of SpondyloArthritis international Society (ASAS) NSAID score. TNF� � tumornecrosis factor �; IQR � interquartile range; ASDAS-CRP � Ankylosing Spondylitis Disease Activity Score using the CRP.† P � 0.012 versus usual care, by Student’s t-test.‡ P � 0.005 versus usual care, by Student’s t-test.§ P � 0.006 versus usual care, by Student’s t-test.

TNF� BLOCKERS IN EARLY AXIAL SpA 1737

subgroup of patients according to the presence of the baselinecharacteristics: definite radiographic sacroiliitis (e.g., struc-tural damage; yes/no), MRI sacroiliitis (e.g., inflammatorylesions of the SI joints; yes/no), abnormal CRP level (yes/no),HLA–B27 positivity (yes/no), history of psoriasis (yes/no),history of peripheral arthritis (yes/no), smoking status (currentsmoker, yes/no), presence of objective signs of structuraldamage to the SI joints (radiographic sacroiliitis) or inflamma-tion (abnormal findings on MRI or abnormal CRP level)(yes/no), and fulfilment of each arm of the ASAS criteria(imaging/clinical). For these analyses, we considered only thepatients who had available information by which to appropri-ately classify them according to the ASAS criteria. For exam-ple, among the 197 patients who received at least 1 TNF�blocker, data for evaluating the fulfilment of such criteria wereavailable for 194 patients (e.g., 3 patients had missing data),and among those 194 patients, 146 fulfilled the ASAS criteriaand 48 did not. Among the 146 patients who fulfilled the ASAScriteria, 96 had structural (radiographic; n � 64) and/orinflammatory (MRI; n � 79) abnormalities of the SI joints,and 50 had no abnormal findings on imaging of the SI joints(e.g., had negative findings on imaging studies) but wereHLA–B27 positive and 2 had features of SpA.

Thereafter, we evaluated the possibility of interactionbetween the presence of these baseline characteristics and thetreatment effect. For this purpose, we performed a univariateanalysis by logistic regression, with a significant interactiondefined as P � 0.10. Finally, we performed a multivariateanalysis (including in the model all variables with P � 0.30 onunivariate analysis, but also on the interaction model) toexplore the most relevant variable.

RESULTS

Estimating the frequency of use of TNF� block-ers. Of the 708 patients included in our analysis, data forKaplan-Meier estimates were available for 685. Kaplan-Meier estimates of the proportion of patients initiating aTNF� blocker were 20.0% (95% CI 17.1–22.9), 24.4%(95% CI 21.1–27.7), 27.9% (95% CI 24.6–31.2), and30.2% (95% CI 26.7–33.7) at 6, 12, 18, and 24 monthsafter inclusion, respectively (Figure 1).

Overall, a total of 202 patients (28.5% of theinitial cohort) received at least 1 TNF� blocker duringfollowup. The population analysis flow chart is shown inFigure 2. Of the 202 patients initiating treatment withtheir first TNF� blocker, 103 (51.0%) received etaner-cept, 79 (39.1%) received adalimumab, and 20 (9.9%)received infliximab.

Table 1 summarizes the baseline characteristicsof the 202 patients and 202 controls included in oureffectiveness analysis after matching by the propensityscore method. Interestingly, in the group of patientsreceiving TNF� blockers, 72 patients (35.6%) presentedwithout any objective sign of inflammation or structuraldamage at baseline, while 127 patients (62.8%) pre-

sented with either radiographic sacroiliitis (67 [33.2%])or MRI sacroiliitis (82 [40.6%]) or with abnormal CRPlevels (83 [41.1%]).

In the group of 197 patients who received aTNF� blocker, 75.3%, 86.3%, and 88.3% fulfilled theASAS, Amor, and ESSG criteria, respectively. Thus,95.9% of the patients fulfilled at least 1 set of criteria.

We found no significant differences between thegroups, except for the scores on the BASDAI, BASFI,and ASDAS-CRP, with higher values in the groupreceiving TNF� blockers.

Evaluating the effectiveness of TNF� blockers.The followup visit took place after a mean � SD of21.6 � 7.3 weeks of study (range 8.4–45.1 weeks). Asignificantly greater percentage of patients receivingTNF� blockers than control patients receiving usualcare (62 of 197 [31.5%] versus 26 of 197 [13.2%])achieved the primary end point, an ASAS40 response,after at least 8 weeks of treatment (OR 2.99 [1.80–4.99],P � 0.0002 for intergroup difference) (Figure 3A).

Similar results were observed for the secondaryeffectiveness end points. For example, a significantlyhigher percentage of patients receiving TNF� blockersthan control patients receiving usual care achieved aresponse according to any other set of clinical responsecriteria (Table 2). It should be emphasized that resultson symptomatic outcome variables (e.g., ASAS responsecriteria) were observed concomitantly with a reductionin NSAID intake during the study (e.g., mean � SDchanges of –35.5 � 66.3 in those taking TNF� blockersand –22.5 � 52.3 in the controls on the ASAS NSAIDscore; P � 0.019).

No significant differences were observed in theprimary effectiveness end point (ASAS40 response) withregard to the TNF� blockers: 25 of 78 patients takingadalimumab (32.1%), 33 of 101 patients taking etaner-cept (32.7%), and 4 of 18 patients taking infliximab(22.2%) (P � 0.529 for interaction). Similar percentagesof response were observed for the secondary end points,irrespective of the type of drug (Table 3).

Exploring the interaction between baseline vari-ables and the effectiveness of the TNF� blockers. Nointeractions were found between the ASAS40 responseand the presence of radiographic sacroiliitis, abnormalCRP levels, HLA–B27 positivity, peripheral arthritis,psoriasis, or smoking status at baseline (Figure 3). Theonly statistically significant interaction found was for thepresence versus absence of sacroiliitis on MRI (OR 4.99[95% CI 2.17–11.51] versus OR 1.75 [95% CI 0.88–3.46],P � 0.057 for interaction) (Figure 3C).

However, a trend toward a greater treatment

1738 MOLTO ET AL

Figure 3. Subgroup interaction analyses of the treatment response (40% improvement in disease activity according to the ASsessment ofSpondyloArthritis International Society criteria [ASAS40]) after at least 8 weeks of treatment with a tumor necrosis factor � (TNF�) blocker or withany other treatment (usual care). Odds ratios (ORs) and 95% confidence intervals are shown for each comparison. � � P � 0.05 for intergroupdifference; �� � P � 0.10 for interaction difference. MRI � magnetic resonance imaging; CRP � C-reactive protein.

TNF� BLOCKERS IN EARLY AXIAL SpA 1739

effect was observed in the subgroup of patients fulfillingthe imaging arm and/or in the subgroup of patients withany objective sign of inflammation (e.g., MRI sacroiliitisor raised CRP level) or structural damage of the SIjoints (e.g., radiographic sacroiliitis) (Figures 3I and J).

The multivariate analysis confirmed the presenceof MRI sacroiliitis at baseline as a predisposing factorfor the treatment effect of the TNF� blockers (OR 3.45[95% CI 1.81–6.60], P � 0.0002).

DISCUSSION

The findings of our study confirm that TNF�blockers are widely used in clinical practice to treatpatients with early axial SpA. As many as 30% ofpatients received TNF� blockers as their initial treat-ment during the first 2 years of followup. In addition, weconfirm the effectiveness of TNF� blockers under real-life conditions (with an average of 32% of patientsachieving an ASAS40 response after 22 weeks of TNF�blocker treatment). Findings of the exploratory sub-group interaction analyses suggest that early axial SpApatients presenting with MRI evidence of sacroiliitis aremore likely to experience an ASAS40 response.

The percentage of patients initiating TNF�blockers in our study was high, and a majority receivedthis treatment after only 6 months of followup. This hightreatment rate may be explained by the study design(e.g., patients attending their DESIR cohort visits at a

tertiary care center would have the opportunity to meetrheumatology research nurses and other patients andwould more easily obtain information about their dis-ease and the potential therapeutic options, such asTNF� blocker therapy). However, patient recruitmentin the DESIR study was performed in close connectionwith local community rheumatologists to ensure that thecohort was as representative as possible of all patientswith inflammatory back pain suggestive of axial SpA.Although the percentage of patients initiating TNF�blockers might seem high, similar rates have been re-ported in previous studies (33,34).

Our findings concerning the changes in the symp-tomatic outcome variables need to be compared to thosereported for RCTs of nonradiographic axial SpA. Inthose trials, the reported magnitude of treatment effect(e.g., the difference in ASAS40 response rates betweenthe active treatment group and the control group) was�22% (21% for adalimumab [10], 16% for etanercept[11], and 28% for certolizumab [35]). In our study, themagnitude of treatment effect was 18.3% and is there-fore comparable to the results of the other trials.

Moreover, a huge difference between the find-ings of our study and the findings of conventional RCTslies in NSAID intake during the study. In conventionalRCTs, it is recommended that NSAID intake remainstable throughout the study (36). In our study, however,NSAID intake was at the discretion of the patient and

Table 2. Effectiveness end points in the 394 patients and controls who completed the study*

All patients(n � 394)

Patients receiving TNF�blockers (n � 197)

Patients receiving usualcare (n � 197) OR (95% CI)† P

ASAS improvementASAS40 90 (23.1) 62 (31.8) 28 (13.5) 2.99 (1.80–4.99) 0.0003ASAS20 130 (33.5) 90 (46.2) 40 (20.7) 3.23 (2.10–5.13) �0.0001

Bath AS instruments�BASFI �8.6 � 19.5 �12.9 � 20.3 �4.3 � 17.3 – 0.014�BASDAI �11.6 � 20.4 �18.2 � 21.7 �5.0 � 17.5 – �0.0001BASDAI50 248 (63.3) 151 (77.0) 97 (49.5) 2.58 (1.62–4.11) �0.0001

ASDAS-CRP improvementASDAS-CRP �0.7 � 1.2 �1.0 � 1.3 �0.3 � 1.0 – 0.0002

ID 90 (23.7) 58 (29.4) 35 (17.9) 3.20 (1.79–5.72) �0.0001MI 51 (13.3) 39 (20.3) 12 (6.3) 3.52 (1.57–7.90) 0.017CII 125 (32.6) 89 (46.4) 36 (18.9) 3.72 (2.23–6.21) �0.0001

Taking NSAIDs 278 (70.6) 141 (71.6) 137 (69.5) – 0.658�ASAS NSAID score in last week �29.0 � 60.0 �35.5 � 66.3 �22.5 � 52.3 – 0.019

* Odds ratios (ORs) were adjusted for the Bath Ankylosing Spondylitis Disease Activity Index (BASDAI), the Bath Ankylosing SpondylitisFunctional Index (BASFI), and the Ankylosing Spondylitis Disease Activity Score using the C-reactive protein (ASDAS-CRP). Assessment ofSpondyloArthritis international Society (ASAS) improvement was determined as both 20% (ASAS20) and 40% (ASAS40) improvement. Bath ASimprovement was determined according to the change (�) from baseline in the BASFI and BASDAI values, as well as the BASDAI criteria for 50%improvement (BASDAI50). ASDAS-CRP improvement was categorized as inactive disease (ID), major improvement (MI), and clinically inactivedisease (CII). Values are the number (%) or the mean � SD. P values less than 0.05 were considered significant. 95% CI � 95% confidence interval.NSAIDs � nonsteroidal antiinflammatory drugs.

1740 MOLTO ET AL

Tab

le3.

Eff

ectiv

enes

sof

each

ofth

e3

tum

orne

cros

isfa

ctor

�bl

ocke

rs*

Gro

up

ASA

Sim

prov

emen

tB

ath

AS

inst

rum

ents

ASD

AS-

CR

Pim

prov

emen

t

ASA

S40

ASA

S20

�B

ASF

I�

BA

SDA

IB

ASD

AI5

0A

SDA

S-C

RP

IDM

IC

II

Ada

limum

abPa

tient

s25

(32.

1)34

(43.

6)�

14.1

�23

.2�

16.7

�22

.459

(75.

6)31

(39.

7)�

0.95

�1.

4122

(28.

2)16

(21.

3)C

ontr

ols

10(1

3.0)

12(1

5.6)

�2.

6�

16.8

�3.

6�

17.7

36(4

6.8)

16(2

0.5)

�0.

23�

0.96

12(1

5.4)

6(7

.9)

OR

(95%

CI)

3.07

(1.2

2–7.

73)

4.47

(1.9

0–10

.52)

––

3.46

(1.6

3–7.

36)

–�

0.6

�0.

25.

32(1

.93–

14.6

9)3.

37(1

.08–

10.4

7)P

0.01

70.

001

0.02

70.

001

0.00

10.

0002

0.00

10.

001

0.03

6E

tane

rcep

tPa

tient

s33

(32.

7)50

(50.

0)�

13.0

�19

.0�

19.4

�21

.278

(78.

0)40

(39.

6)�

1.06

�1.

2831

(30.

7)18

(18.

0)C

ontr

ols

14(1

4.1)

22(2

2.2)

�5.

0�

18.0

�5.

5�

18.1

50(4

9.5)

35(3

4.7)

�0.

36�

1.07

20(2

0.0)

6(6

.2)

OR

(95%

CI)

3.42

(1.5

8–7.

38)

3.22

(1.6

6–6.

23)

––

3.02

(1.5

8–5.

79)

–�

0.6

�0.

22.

69(1

.24–

5.82

)2.

46(0

.82–

7.42

)P

0.00

20.

001

0.02

70.

002

0.00

10.

012

0.00

020.

012

0.10

9In

flixi

mab

Patie

nts

4(2

2.2)

6(3

5.3)

�7.

3�

12.9

�18

.2�

22.3

14(7

7.8)

7(3

8.9)

�1.

27�

1.29

5(2

7.8)

5(2

9.4)

Con

trol

s2

(11.

8)6

(35.

3)�

7.8

�15

.5�

8.4

�13

.211

(61.

1)8

(44.

4)�

0.4

�0.

73

(16.

7)0

(0.0

)O

R(9

5%C

I)1.

80(0

.18–

18.0

8)0.

64(0

.09–

4.40

)–

–0.

64(0

.11–

3.91

)–

�0.

3�

0.3

5.15

(0.2

5–10

5.0)

�99

9(�

0.00

1–�

999)

P0.

619

0.64

90.

864

0.83

80.

700

0.89

30.

185

0.91

20.

959

*O

dds

ratio

s(O

Rs)

wer

ead

just

edfo

rth

eB

ASD

AI,

the

BA

SFI,

and

the

ASD

AS-

CR

P.V

alue

sar

eth

enu

mbe

r(%

)or

the

mea

n�

SD.P

valu

esle

ssth

an0.

05w

ere

cons

ider

edsi

gnifi

cant

.See

Tab

le2

for

defin

ition

s.

TNF� BLOCKERS IN EARLY AXIAL SpA 1741

his or her rheumatologist. TNF� blocker treatment inour study resulted in a statistically significant reductionin the intake of NSAIDs. Therefore, the symptomatictreatment effect of TNF� blockers observed in our study(as compared to that in other RCTs) in the context ofreduced NSAID intake might be considered clinicallyrelevant. Our findings suggest a similar treatment effectfor all 3 of the TNF� blockers evaluated.

Many factors have been suggested to predict agreater therapeutic response to TNF� blockers in axialSpA (e.g., age, disease duration, elevated CRP levels,BASDAI, BASFI, and HLA–B27 positivity) (11–14).More recently, RCTs (10) have reported that signs ofinflammation (MRI sacroiliitis and/or elevated CRPlevels) are predictors of a better response in patientswith nonradiographic axial SpA. In order to address thisquestion in our study, we considered that the appropri-ate statistical test was the interaction test and not theintertreatment group test (e.g., TNF� blocker versususual care) in each subgroup of patients (e.g., thosefulfilling the imaging arm [one test] and thereafter thosenot fulfilling the imaging arm [another test]).

The single baseline characteristic with a signifi-cant interaction in terms of the treatment effect was thepresence of sacroiliitis on MRI. The threshold P valueused in our subgroup analysis, which is commonly usedin subgroup analyses, was �0.10, in order to increase thepower of the statistical test.

A greater treatment effect, however, was ob-served in the subgroups of patients who fulfilled theimaging arm (25%) as compared to the subgroup ofpatients who fulfilled the clinical arm (9%), suggestingthe existence of a difference even in the absence of astatistically significant interaction test (P � 0.194). Sucha conclusion might be reinforced by the fact that the95% CI of the OR for predicting an ASAS40 response inthe subgroup of patients fulfilling the imaging arm was�1 (OR 3.77 [95% CI 1.80–7.90]), resulting in a signif-icant intertreatment group difference (P � 0.05) in theimaging arm, while the 95% CI of the OR for predictingan ASAS40 response in the clinical arm contained 1 (OR1.72 [95% CI 0.68–4.34]), resulting in a nonsignificantintertreatment group difference (P � 0.05) in the clinicalarm.

Similarly, a greater treatment effect was observedin the subgroup of patients presenting with objectivesigns of either inflammation (MRI sacroiliitis or raisedCRP level) or structural damage of the SI joints (radio-graphic sacroiliitis) compared to the subgroup of pa-tients without any objective signs (16% versus 10%),here again suggesting a difference even in the absence of

a statistically significant interaction (P � 0.489). More-over, the 95% CI of the OR for predicting an ASAS40response in the subgroup of patients with objective signswas �1 (OR 3.46 [95% CI 1.81–6.60]) (statisticallysignificant difference between treatment groups), whilethe 95% CI of the OR for predicting an ASAS40response in the subgroup of patients without any objec-tive signs contained 1 (OR 2.31 [95% CI 0.90–5.93]) (nota statistically significant difference between treatmentgroups).

However, we would like to emphasize that theinterpretation of such P values is very hazardous, sincesuch analyses are post hoc analyses performed in sub-groups of patients.

Abnormal CRP levels have been reported to be apredisposing factor for a better response. In our study,no statistically significant differences were observed withregard to abnormal CRP levels (results of the interac-tion test were not significant; the 95% CI of the OR forpredicting an ASAS40 response was �1 in the subgroupof patients with raised CRP levels, as well as in thesubgroup of patients with normal CRP levels) (Figure3D). The latter finding might be explained by the designof our study. The baseline parameters in our analysis arethose recorded at the DESIR cohort visits, and theseparameters might be different from those observed theday before initiating the TNF� blockers, as the patientcould be stable at the DESIR visit (e.g., at month 6),present with a flare 2 months later, and be started ontreatment with TNF� blockers between 2 DESIR visits(e.g., at month 10). Moreover, because of multipletesting, the observed statistically different treatmenteffect with regard to the patients’ characteristics mighthave occurred by random chance alone. Therefore, ourresults should be considered with caution (e.g., interac-tion with the presence of MRI sacroiliitis or the absenceof interaction with elevated CRP levels). To avoid thesebiases, well-powered explanatory trials of TNF� block-ers in patients with early axial SpA should be conducted,particularly in the subgroup of patients with no objectivesigns of structural damage or inflammation (e.g., pa-tients fulfilling the clinical arm of the ASAS axial SpAcriteria with normal CRP levels).

Some limitations of this study need to be consid-ered. First, the DESIR cohort included patients present-ing with chronic inflammatory back pain who had a�50% likelihood of being diagnosed by a rheumatolo-gist as having axial SpA. Although rheumatologists wereasked at the end of each visit whether another diagnosiswas more likely so that patients with an inappropriatediagnosis could be excluded from the study, it is not

1742 MOLTO ET AL

impossible that the cohort included patients with condi-tions other than axial SpA (e.g., inflammatory back painlinked to degenerative discopathies). However, the ma-jority of patients in the study (95.9%) fulfilled at least 1set of criteria for SpA.

Second, DESIR is an observational study withpotential biases and confounding, mainly due to the lackof randomization. As has been recommended for such asituation (31,37–39), we addressed this potential biasconcerning the estimation of treatment effects by match-ing the patients receiving active drug with the controlpatients based on a propensity score, bringing our studyclose to a pragmatic trial and providing valuable data onreal-life prescription rates and therapeutic effects inclinical practice.

Some additional strengths of this study should beemphasized. First, it is the largest prospective observa-tional cohort of patients with inflammatory back painsuggestive of early axial SpA to describe and analyze indetail the use of TNF� blockers in a real-life clinicalsetting. Second, because of the quality of the datacollection both at study inclusion and every 6 monthsthereafter, the data used to assess the primary outcomewas complete with regard to clinical, physical, labora-tory, and imaging in the large majority of patients.

Our study confirms the effectiveness of TNF�blockers in patients with early axial SpA in a real-lifeclinical setting in terms of clinical response, particularlyin the subgroup of patients with inflammation of the SIjoints seen on MRI. Further explanatory RCTs focusingon different subgroups of patients, especially thosefulfilling the clinical arm of the ASAS criteria withnormal CRP levels, are required in order to explore therole of TNF� blockers in this context.

ACKNOWLEDGMENTS

The authors would like to thank the scientific commit-tee of the DESIR cohort for accepting our application for thisstudy, the investigators for adequately collecting the data thatpermitted such analyses, and the patients for agreeing toparticipate in the study.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising itcritically for important intellectual content, and all authors approvedthe final version to be published. Dr. Molto had full access to all of thedata in the study and takes responsibility for the integrity of the dataand the accuracy of the data analysis.Study conception and design. Molto, Dougados.Acquisition of data. Molto, Dougados.Analysis and interpretation of data. Molto, Paternotte, Claudepierre,Breban, Dougados.

ROLE OF THE STUDY SPONSOR

Pfizer Ltd. France had no role in the study design or in thecollection, analysis, or interpretation of the data, the writing of themanuscript, or the decision to submit the manuscript for publication.Publication of this article was not contingent upon approval by PfizerLtd. France.

REFERENCES

1. Van der Linden S, Valkenburg HA, Cats A. Evaluation of thediagnostic criteria for ankylosing spondylitis: a proposal for mod-ification of the New York criteria. Arthritis Rheum 1984;27:361–8.

2. Amor B, Dougados M, Mijiyawa M. Criteres de classification desspondylarthropathies. Rev Rhum Mal Osteoartic 1990;57:85–9.

3. Dougados M, van der Linden S, Juhlin R, Huitfeldt B, Amor B,Calin A, et al, and the European Spondylarthropathy StudyGroup. The European Spondylarthropathy Study Group prelimi-nary criteria for the classification of spondylarthropathy. ArthritisRheum 1991;34:1218–27.

4. Rudwaleit M, van der Heijde D, Landewe R, Listing J, Akkoc N,Brandt J, et al. The development of Assessment of SpondyloAr-thritis international Society classification criteria for axial spondy-loarthritis (part II): validation and final selection. Ann Rheum Dis2009;68:777–83.

5. Baraliakos X, van den Berg R, Braun J, van der Heijde D. Updateof the literature review on treatment with biologics as a basis forthe first update of the ASAS/EULAR management recommenda-tions of ankylosing spondylitis. Rheumatology (Oxford) 2012;51:1378–87.

6. Braun J, Brandt J, Listing J, Zink A, Alten R, Burmester G, et al.Long-term efficacy and safety of infliximab in the treatment ofankylosing spondylitis: an open, observational, extension study of athree-month, randomized, placebo-controlled trial. ArthritisRheum 2003;48:2224–33.

7. Calin A, Dijkmans BA, Emery P, Hakala M, Kalden J, Leirisalo-Repo M, et al. Outcomes of a multicentre randomised clinical trialof etanercept to treat ankylosing spondylitis. Ann Rheum Dis2004;63:1594–600.

8. Sieper J, van der Heijde D, Dougados M, Brown LS, Lavie F,Pangan AL. Early response to adalimumab predicts long-termremission through 5 years of treatment in patients with ankylosingspondylitis. Ann Rheum Dis 2012;71:700–6.

9. Haibel H, Rudwaleit M, Listing J, Heldmann F, Wong RL,Kupper H, et al. Efficacy of adalimumab in the treatment of axialspondylarthritis without radiographically defined sacroiliitis: re-sults of a twelve-week randomized, double-blind, placebo-controlled trial followed by an open-label extension up to weekfifty-two. Arthritis Rheum 2008;58:1981–91.

10. Sieper J, van der Heijde D, Dougados M, Mease PJ, Maksymo-wych WP, Brown MA, et al. Efficacy and safety of adalimumab inpatients with non-radiographic axial spondyloarthritis: results of arandomised placebo-controlled trial (ABILITY-1). Ann RheumDis 2013;72:815–22.

11. Dougados M, van der Heijde D, Sieper J, Braun J, MaksymowychWP, Citera G, et al. Safety and efficacy of etanercept in earlynon-radiographic axial spondyloarthritis: a randomized, double-blind, placebo-controlled trial at 24 weeks [abstract]. ArthritisRheum 2013;65 Suppl:S642–3.

12. Arends S, Brouwer E, van der Veer E, Groen H, Leijsma MK,Houtman PM, et al. Baseline predictors of response and discon-tinuation of tumor necrosis factor-� blocking therapy in ankylosingspondylitis: a prospective longitudinal observational cohort study.Arthritis Res Ther 2011;13:R94.

13. Lord PA, Farragher TM, Lunt M, Watson KD, Symmons DP,Hyrich KL, on behalf of the BSR Biologics Register. Predictors ofresponse to anti-TNF therapy in ankylosing spondylitis: results

TNF� BLOCKERS IN EARLY AXIAL SpA 1743

from the British Society for Rheumatology Biologics Register.Rheumatology (Oxford) 2010;49:563–70.

14. Vastesaeger N, van der Heijde D, Inman RD, Wang Y, DeodharA, Hsu B, et al. Predicting the outcome of ankylosing spondylitistherapy [published erratum appears in Ann Rheum Dis 2012;71:1434]. Ann Rheum Dis 2011;70:973–81.

15. Rudwaleit M, Claudepierre P, Wordsworth P, Cortina EL, SieperJ, Kron M, et al. Effectiveness, safety and predictors of goodclinical response in 1250 patients treated with adalimumab foractive ankylosing spondylitis. J Rheumatol 2009;36:801–8.

16. Ciurea A, Scherer A, Exer P, Bernhard J, Dudler J, Beyeler B, etal, on behalf of the Rheumatologists of the Swiss Clinical QualityManagement Program for Axial Spondyloarthritis. Tumor necrosisfactor � inhibition in radiographic and non-radiographic axialspondyloarthritis: results from a large observational cohort. Ar-thritis Rheum 2013;65:3096–106.

17. DESIR in English. La cohorte DESIR. URL: http://www.lacohortedesir.fr/desir-in-english/.

18. Dougados M, d’Agostino MA, Benessiano J, Berenbaum F, Bre-ban M, Claudepierre P, et al. The DESIR cohort: a 10-yearfollow-up of early inflammatory back pain in France: study designand baseline characteristics of the 708 recruited patients. JointBone Spine 2011;78:598–603.

19. Calin A, Porta J, Fries JF, Schurman DJ. Clinical history as ascreening test for ankylosing spondylitis. JAMA 1977;237:2613–4.

20. Rudwaleit M, Metter A, Listing J, Sieper J, Braun J. Inflammatoryback pain in ankylosing spondylitis: a reassessment of the clinicalhistory for application as classification and diagnostic criteria.Arthritis Rheum 2006;54:569–78.

21. Garrett S, Jenkinson T, Kennedy LG, Whitelock H, Gaisford P,Calin A. A new approach to defining disease status in ankylosingspondylitis: the Bath Ankylosing Spondylitis Disease ActivityIndex. J Rheumatol 1994;21:2286–91.

22. Lukas C, Landewe R, Sieper J, Dougados M, Davis J, Braun J, etal, for the Assessment of SpondyloArthritis international Society.Development of an ASAS-endorsed disease activity score(ASDAS) in patients with Ankylosing Spondylitis. Ann Rheum Dis2009;68:18–24.

23. Jones SD, Porter J, Garrett SL, Kennedy LG, Whitelock H, CalinA. A new scoring system for the Bath Ankylosing SpondylitisMetrology Index (BASMI). J Rheumatol 1995;22:1609.

24. Calin A, Garrett S, Whitelock H, Kennedy LG, O’Hea J, MallorieP, et al. A new approach to defining functional ability in ankylos-ing spondylitis: the development of the Bath Ankylosing Spondy-litis Functional Index. J Rheumatol 1994;21:2281–5.

25. Daltroy LH, Larson MG, Roberts NW, Liang MH. A modificationof the Health Assessment Questionnaire for the spondyloar-thropathies. J Rheumatol 1990;17:946–50.

26. Dougados M, Paternotte S, Braun J, Burgos-Vargas R, Mak-symowych WP, Sieper J, et al. ASAS recommendations for col-lecting, analysing and reporting NSAID intake in clinical trials/epidemiological studies in axial spondyloarthritis. Ann Rheum Dis2011;70:249–51.

27. Price D, Hillyer EV, van der Molen T. Efficacy versus effective-ness trials: informing guidelines for asthma management. CurrOpin Allergy Clin Immunol 2013;13:50–7.

28. Roland M, Torgerson D. Understanding controlled trials: what arepragmatic trials? BMJ 1998;316:285.

29. Van der Heijde D, Dougados M, Davis J, Weisman MH, Mak-symowych W, Braun J, et al. ASsessment in Ankylosing SpondylitisInternational Working Group/Spondylitis Association of Americarecommendations for conducting clinical trials in ankylosing spon-dylitis. Arthritis Rheum 2005;52:386–94.

30. Machado P, Landewe R, Lie E, Kvien TK, Braun J, Baker D, et al,for the Assessment of SpondyloArthritis international Society.Ankylosing Spondylitis Disease Activity Score (ASDAS): definingcut-off values for disease activity states and improvement scores.Ann Rheum Dis 2011;70:47–53.

31. Rosenbaum PR, Rubin DB. The central role of the propensityscore in observational studies for causal effects. Biometrika 1983;70:41–55.

32. Rubin DB. Estimating causal effects from large data sets usingpropensity scores. Ann Intern Med 1997;127:757–63.

33. Vander Cruyssen B, Ribbens C, Boonen A, Mielants H, de VlamK, Lenaerts J, et al. The epidemiology of ankylosing spondylitisand the commencement of anti-TNF therapy in daily rheumatol-ogy practice. Ann Rheum Dis 2007;66:1072–7.

34. Pham T, Landewe R, van der Linden S, Dougados M, Sieper J,Braun J, et al. An international study on starting tumour necrosisfactor-blocking agents in ankylosing spondylitis. Ann Rheum Dis2006;65:1620–5.

35. Landewe R, Braun J, Deodhar A, Dougados M, MaksymowychWP, Mease PJ, et al. Efficacy of certolizumab pegol on signs andsymptoms of axial spondyloarthritis including ankylosing spondy-litis: 24-week results of a double-blind randomised placebo-con-trolled Phase 3 study. Ann Rheum Dis 2014;73:39–47.

36. Dougados M, Braun J, Szanto S, Combe B, Geher P, Leblanc V,et al. Nonsteroidal antiinflammatory drug intake according to theAssessment of SpondyloArthritis international Society Score inclinical trials evaluating tumor necrosis factor blockers: example ofetanercept in advanced ankylosing spondylitis. Arthritis Care Res(Hoboken) 2012;64:290–4.

37. Concato J, Shah N, Horwitz RI. Randomized, controlled trials,observational studies, and the hierarchy of research designs.N Engl J Med 2000;342:1887–92.

38. Soubrier M, Lukas C, Sibilia J, Fautrel B, Roux F, Gossec L, et al.Disease activity score-driven therapy versus routine care in pa-tients with recent-onset active rheumatoid arthritis: data from theGUEPARD trial and ESPOIR cohort. Ann Rheum Dis 2011;70:611–5.

39. Gomez-Reino JJ, Maneiro JR, Ruiz J, Rosello R, Sanmarti R,Romero AB, on behalf of the MIRAR Study Group. Comparativeeffectiveness of switching to alternative tumour necrosis factor(TNF) antagonists versus switching to rituximab in patients withrheumatoid arthritis who failed previous TNF antagonists: theMIRAR Study. Ann Rheum Dis 2012;71:1861–4.

1744 MOLTO ET AL