Rheumatol Ther (2018) ORIGINAL RESEARCHCost-effectiveness Analysis of Golimumabin the Treatment of Non-Radiographic AxialSpondyloarthritis in ScotlandRebekah H. Borse . Sumesh Kachroo . Chloe Brown .Eilish McCann . Ralph P. InsingaReceived: January 26, 2018 / Published online: April 9, 2018Ó The Author(s) 2018ABSTRACTIntroduction: The aim of this study is to assessthe cost-effectiveness of golimumab for thetreatment of non-radiographic axial spondyloarthritis (nr-axSpA) vs. conventional therapyand other tumor necrosis factor inhibitors fromthe Scottish payer perspective.Methods: A model comprising a short-termdecision tree and a long-term Markov modelwas developed to compare cost-effectiveness(incremental costs per quality-adjusted life-year[QALY]) for patients in Scotland with nr-axSpAtreated by conventional therapy, adalimumab,certolizumab pegol, etanercept, or golimumabfor a lifetime period. A network meta-analysis(NMA) was conducted to identify clinical andsafety data for treatments and synthesize theavailable evidence into relative treatmenteffects between comparators. The probability ofpatients achieving an Assessment of SpondyloArthritis International Society 20/40%response criteria (ASAS20/ASAS40) or a 50%improvement in Bath Ankylosing SpondylitisEnhanced content To view enhanced content for thisarticle go to H. Borse (&) S. Kachroo R. P. InsingaMerck & Co., Inc., Kenilworth, NJ, USAe-mail: [email protected] Brown E. McCannMerck Sharp & Dohme, Hoddesdon, UKDisease Activity Index score (BASDAI50) at week12 was obtained from the NMA for each of thecomparators. Baseline health state utilities werebased on the EQ-5D questionnaire collected inthe golimumab GO-AHEAD study. The cost oftreatment was calculated based on drug acquisition, drug administration, and initiation/monitoring costs.Results: Golimumab resulted in an increase of2.06 QALYs and additional cost of 39,770compared with conventional therapy. Incremental cost per QALY gained was 19,280 forgolimumab, which was lower than adalimumab( 19,737), etanercept ( 20,089), and higherthan certolizumab pegol ( 18,710). Golimumabremained cost-effective throughout a range ofsensitivity analyses where key assumptions weretested.Conclusions: From a Scottish perspective, golimumab was a cost-effective treatment for nraxSpA compared with conventional therapy at awillingness-to-pay threshold of 30,000 perQALY.Funding: Merck & Co., Inc.Keywords: Ankylosingspondylitis;Costeffectiveness; Decision tree; oarthritis; Scotland; Tumor necrosisfactor inhibitors

58INTRODUCTIONAxial spondyloarthritis (axSpA) is a group ofchronic, inflammatory diseases, primarily characterized by inflammation of the spine andsacroiliac joints [1]. The most commonlyoccurring form of axSpA is ankylosingspondylitis (AS), with radiographic structuralchanges in sacroiliac joints being the mostimportant criterion for a clinical diagnosis [2].In 2009, the Assessment of SpondyloarthritisInternational Society (ASAS) developed newclassification criteria in order to better characterize the group of patients with typical featuresof axSpA, such as the presence of inflammatoryback pain and positive human leukocyte antigen B27 (HLA-B27), but the absence of anydefinitive radiographic evidence of sacroiliitis[3]. This subpopulation is classified as havingnon-radiographic axial spondyloarthritis (nraxSpA) [3]. Both European data and Scotlandspecific data on the prevalence of nr-axSpA arevery limited because of the different diagnosticcriteria adopted in identifying patients with nraxSpA based on symptom duration, and varyingavailability of diagnostic tools. In Europe, theproportion of nr-axSpA among axSpA patients isestimated to be 23–80% [4]. The use of the ASASclassification criteria in identifying nr-axSpApatients in clinical practice was reported to beapproximately 52% (ranging from 24% in Germany to 75% in Spain) by a recent Europeancross-sectional study [5].Some nr-axSpA patients will progress to ASlater in their disease course while approximately30% will never develop radiographic sacroiliitisdespite persistent inflammatory back pain [4, 6].Studies report that the factors associated withdisease progression include male sex, high levelof inflammation (as demonstrated by C-reactiveprotein level), and longer symptom duration[7–9]. Patients treated with a tumor necrosisfactor inhibitor (TNFi) also have slower rate ofdisease progression, compared with patientswho have never been treated with a TNFi [5].Regardless of whether progression to ASoccurs, nr-axSpA can have substantial negativeimpacts on patients’ daily activities and qualityof life due to disease-related symptoms andRheumatol Ther (2018) 5:57–73treatment-related adverse events, as well as thefunctional effects of living with the disease. Anumber of studies have shown that physicalfunction is considerably impaired amongpatients with nr-axSpA, as assessed with theBath Ankylosing Spondylitis Disease ActivityIndex (BASDAI) and the Bath AnkylosingSpondylitis Functional Index (BASFI), with theBath Ankylosing Spondylitis Metrology Index(BASMI) being used to assess spinal mobility[7, 8, 10, 11]. Studies have shown that, compared with axSpA, nr-axSpA has a greaterimpact on productivity, both at home and inthe workplace [12, 13]. As would be expected,these effects translate to an increased economicburden, with healthcare-resource utilizationand lost productivity resulting in approximatemean 3-year costs of between 5000 and 38,000 in a mixed cohort of SpA patients, themajority of whom had nr-axSpA [14].Non-steroidalanti-inflammatorydrugs(NSAIDs) are recommended as first-line treatment for nr-axSpA by ASAS and the EuropeanLeague Against Rheumatism (EULAR) [15].Some patients respond well to initial treatmentwith NSAIDs, but a considerable proportion ofpatients have an inadequate response and willneed subsequent treatment such as TNFi therapy to control persistently high disease activity.Currently, there are four TNFis (adalimumab,certolizumab pegol, etanercept, and golimumab) with marketing authorization for thetreatment of nr-axSpA in Europe, when patientshave high disease severity despite adequateNSAID treatment (a minimum of two NSAIDsfor a minimum of 4 weeks) [15]. In addition,golimumab, adalimumab, etanercept, and certolizumab pegol are recommended by theScottish Medicines Consortium (SMC) for thetreatment of patients with nr-axSpA [16, 17].Golimumab is a once-a-month, subcutaneously injected treatment for nr-axSpA thathas been approved for previously treated nraxSpA in Europe. GO-AHEAD, a 16-week, ted the efficacy and tolerability ofgolimumab in patients with nr-axSpA who werenot responding to NSAIDs [18]. Significantlymore patients in the golimumab groupachieved ASAS 20% response criteria (ASAS20),

Rheumatol Ther (2018) 5:57–73compared with the placebo group at week 16(71.1 vs. 40.0%, P \ 0.0001), when looking atthe full analysis set. In part 2 of GO-AHEAD,where patients in the golimumab group continued to receive golimumab, and patients inthe placebo group crossed over to receive golimumab, the proportion of ASAS20 respondersin the golimumab/golimumab group continuedto increase from week 16 to week 32, and in theplacebo/golimumab group, there was anotable increase in patients achieving ASAS20after switching to golimumab [18].The objective of this analysis was to assessthe cost-effectiveness of golimumab for treatment of nr-axSpA vs. conventional therapy(including NSAIDs, cyclooxygenase-2 inhibitors, steroids, gastro-protectants, bisphosphonates, and physiotherapy) and other TNFis,namely adalimumab, certolizumab pegol, andetanercept, from the perspective of payer inScotland.METHODSStudy DesignA static decision model comprising an initial,short-term decision tree and a Markov modelwas developed for Scotland to compare cost-effectiveness outcomes for patients with nr-axSpAtreated with conventional therapy, or TNFis.The decision tree modeled the potential forcontinuation or discontinuation of patientsfrom TNFis based on response to treatment at12 weeks. The first cycle of the Markov modelthen started, with the first transitions betweenMarkov health states being applied at 24 weeks.The model incorporated the BASDAI, BASDAI50, ASAS20, and ASAS40 scores. BASDAI is avalidated instrument completed by patients,consisting of six 10-cm horizontal visual analogue scales (VAS) to measure severity of fatigue,spinal and peripheral joint pain, localized tenderness, and morning stiffness, with the finalBASDAI score ranging from 0 to 10 [19]. BASDAI50 has been recommended by the ASAS asthe response criterion used to determine treatment success. The ASAS score is a compositemeasure, comprising a 10-cm horizontal VAS59for pain, inflammation, well-being, and function. Improvement in three modalities by 20%or more, without deterioration in the fourthmodality, constitutes an ASAS20 response, and40% or more an ASAS40 response [20].This study is a modeling study using dataderived from previously conducted studies, anddoes not contain any new studies with humanor animal subjects performed by any of theauthors.Model StructureDecision TreeThe decision tree provided estimates of expected costs and outcomes over the short term(Fig. 1a). Patients were initially managed witheither a TNFi or conventional therapy. Themodel assumed that no patient discontinuedconventional therapy and that no switchingbetween different types of conventional therapyoccurred. For patients who started treatmentwith a TNFi, a probability of remaining on TNFitreatment was assigned based on response totreatment (achieving ASAS20, ASAS40, or BASDAI50) informed by data from the GO-AHEADstudy. Very few patients (\ 5%) discontinuedgolimumab treatment prior to week 12 of GOAHEAD. It was therefore assumed that discontinuation of TNFis did not occur prior to week12. Discontinuation at week 12 occurred if thepatient did not have a treatment responseaccording to the efficacy results. As there are nopublished data on long-term persistence withthe modeled TNFis in a large Scottish or UKpopulation of patients with nr-axSpA, followingweek 12 an annual rate of discontinuation of5% was assumed in the model base case.Responders were assumed to stay on the sameTNFi treatment and not switch to a secondTNFi. Non-responders to TNFis who switched toconventional therapy remained in that state forthe remainder of the model.Markov ModelThe Markov model provided estimates ofexpected costs and outcomes over the long term(Fig. 1b). Patients entered the Markov modelfollowing the initial decision tree. The Markov

60Fig. 1 Short-term decision tree (a) and long-term Markovmodel (b). a Short-term decision treea. a The decision treemodel represents the clinical decision that can be taken at12 weeks (informed by treatment response) to eithercontinue or discontinue treatment with TNFis. Responders were assumed to stay on the same TNFi treatment.Non-responders who switched onto conventional therapyremained on conventional therapy for the remainder of themodel time horizon. Patients who started on conventionaltherapy remained on conventional therapy throughout themodel, irrespective of the probability of response at Week12. b Long-term Markov modela. aPatients entered theMarkov model, following the initial decision tree, afterweek 12. Patients treated with a TNFi either stayed onRheumatol Ther (2018) 5:57–73(the same) treatment or discontinued treatment. Forpatients who discontinued treatment, the transition fromthe ‘TNFi Treatment’ state to the ‘Discontinued Treatment’ state was assumed to take 24 weeks (two cycles).Discontinuing patients first moved to the ‘Just Discontinued’ tunnel state and remained in that state for one cycleand then moved to the ‘Discontinued’ tunnel state andremained in that state for one cycle. Patients in the‘Discontinued Treatment’ state were thereafter assumed toreceive conventional therapy alone. Patients could diewhile in any of the health states. nr-axSpA Nonradiographic axial spondyloarthritis; TNFi Tumor necrosisfactor inhibitor

Rheumatol Ther (2018) 5:57–73model comprised defined, mutually exclusivehealth states through which patients transitioned at a rate, which was dependent on therate of disease progression and the age-/sexspecific standardized mortality rate (SMR) (patients could die while in any of the healthstates) for people with nr-axSpA, which wasassumed to be the same as for patients with AS[21]. A 12-week model cycle was used. Cost andutility data were summed by treatment arm asthe model progressed cycle by cycle, allowingfor the calculation of incremental costs andeffectiveness per treatment.Patients treated with a TNFi either stayed ontreatment or discontinued. The proportions ofpatients discontinuing treatment were assumedto be the same for all TNFis. An annual discontinuation rate of 5% was applied for theremainder of the model time horizon, informedby the golimumab arm in the GO-AHEADstudy.If discontinuing treatment, patients firstmoved to the ‘Just Discontinued’ tunnel stateand remained in that state for one cycle andthen moved to the ‘Discontinued’ tunnel stateand remained in that state for one cycle. Tunnelstates are modeled to assist with modeling alinear resolution of treatment effect followingtreatment discontinuation over a 24-week period. Patients then moved to the ‘DiscontinuedTreatment’ state. It was assumed once enteringthe ‘Discontinued treatment’ state patients’disease progression rebounded to the level as ifthey had been treated with conventional carealone. Patients in the ‘Discontinued Treatment’state were thereafter assumed to receive conventional therapy alone. Patients could diewhile in any of the health states.Efficacy MeasurementsThere are no head-to-head trials investigatingthe efficacy and safety of golimumab comparedto other TNFis, and data informing comparativeefficacy between treatment options of nr-axSpAare very limited. Therefore, a systematic literature review (SLR) and a network meta-analysis(NMA) were conducted to identify all relevanttreatments and synthesize the available61evidence into relative treatment effects betweencomparators. Conventional therapy, adalimumab, certolizumab pegol, and etanerceptwere included as comparators in the modelbased on their relevance to current clinicalpractice in managing nr-axSpA.GO-AHEAD compared golimumab to conventional therapy, providing direct evidence onthe comparative efficacy of these two treatments.As well as GO-AHEAD, the SLR identified published data from the comparator trials, includingtwo phase 3 trials investigating adalimumab[22, 23], one phase 3 trial on certolizumab pegol[24], and two phase 3 trials on etanercept [25, 26].In each trial, the TNFi was compared to conventional therapy. The relative treatment effectswere established by the NMA based on the identified data in a fixed effects model (as due tomultiple treatment comparators only havingdata from a single trial available, there wereinsufficient trials to estimate the heterogeneityparameter for a random effects model).Model InputsTransition ProbabilitiesThe probability of patients achieving anASAS20, ASAS40, or BASDAI50 response at week12 was obtained from the NMA for each of theTNFis and for conventional therapy. BASDAI50was used as the efficacy measure in the modelbase case because this represents the measurefor treatment response most commonly used inScottish clinical practice as informed throughseeking expert input.Disease progression was demonstrated bychange in BASDAI and BASFI scores. BaselineBASDAI and BASFI scores were based on baselinescores from the GO-AHEAD study, for both conventional therapy and TNFis, and were assumedto be identical for all TNFis. The magnitude ofresponse to conventional therapy was informedby the change in BASDAI and BASFI scores atweek 12 in GO-AHEAD, and the magnitude ofresponse to TNFis was derived from the NMA.The NMA demonstrated that there were nostatistically significant differences in anyadverse events, any serious adverse events, orinfections (clinically significant opportunistic

Rheumatol Ther (2018) 5:57–7362infections) between TNFis or with respect toconventional therapy.UtilitiesBaseline health state utility values were basedon the EQ-5D health questionnaire collected inthe GO-AHEAD study alongside the BASDAIand BASFI scores. A regression equation [27]using those baseline EQ-5D scores observed inthe GO-AHEAD study was constructed to estimate the average change in EQ-5D score per 1unit change in BASDAI and BASFI scores. Inorder to estimate the EQ-5D score at each cycle,the following equation that accounted for ageand sex was used to elicit the change in EQ-5D.The average baseline utility score of the cohortwas 0.43.Outcomes in Ankylosing Spondylitis International Study (OASIS) [34]. This approach wasvalidated by clinical experts in Scotland.For adverse event costs, we included aweighted cost of infections and the weightswere based on previous Health TechnologyAssessment submissions [35]. The cost of aninfection was counted as a one-time cost andapplied for one cycle in the model. The probabilities of infections for each TNFi were derivedfrom the NMA. For simplicity, other adverseevents were not modeled, as they would not beexpected to have a substantive impact on modelresults and, even in the case of infection, therewas little change in the incremental cost-effectiveness ratio (ICER) in one-way deterministicsensitivity analysis, as described in the Resultssection.EQ 5D ¼ 0:43 þ ð0:10034 0:05735 BASDAI 0:03120 BASFIþ 0:00126 Sex 0:00440 AgeÞ:CostsThe cost of treatment was calculated based ondrug acquisition, drug administration, and initiation/monitoring costs. All treatment strategies included conventional therapy and thusthe costs associated with conventional therapywere not included in the model.The unit costs of TNFis were sourced fromthe British National Formulary [28]. Doses wereinaccordancewithproductlicences[16, 17, 29, 30]. TNFis were assumed to be selfadministered following instruction from anurse. The cost of the instruction was 50.00,based on 1 h of patient-related nurse time [31].This cost was only applied in the decision-treecomponent of the model (i.e., the first12 weeks). Initiation and monitoring costs forTNFis were sourced from the National Institutefor Health and Care Excellence (NICE) MultipleTechnology Appraisal (MTA) for nr-axSpA andwere inflated to 2013/14 prices using theHospital & Community Health Services Pay &Prices Index [32, 33]. Annual disease management costs for nr-axSpA, beyond the treatmentcosts associated with TNFis, were applied basedon a costing equation (NHS cost 1284.186 *EXP [0.213 9 BASFI]) derived from theModel AnalysesThe base-case analyses evaluated a lifetime timehorizon and applied an annual discount rate of3.5% to both costs and health benefits based onSMC’s recommendations. Clinical outcomeswere expressed as gains in quality-adjusted lifeyears (QALYs). Cost-effectiveness was measuredin terms of incremental costs per QALY gained,comparing the incremental clinical benefit fromgolimumab versus other TNFis or conventionaltherapy.A one-way deterministic sensitivity analysis(DSA) was used to identify the key drivers ofuncertainty in the estimation of the incremental cost-effectiveness of golimumab. The keyparameters varied in the DSA included timehorizon, age at baseline, proportion of males,12-week treatment response measure, annualdiscontinuation rate with TNFis, annual diseaseprogression (conventional therapy), annualdisease progression post-week 260 (to accountfor the uncertainty of the long-term diseaseprogression), data source for infections, longterm disease management costs, and utilitydecrement per one unit change in BASDAI/BASF