GSK Clinical Study Register

Transcription

1 In February 2013, GlaxoSmithKline (GSK) announced a commitment to further clinical transparency through the public disclosure of GSK Clinical Study Reports (CSRs) on the GSK Clinical Study Register. The following guiding principles have been applied to the disclosure: Information will be excluded in order to protect the privacy of patients and all named persons associated with the study Patient data listings will be completely removed* to protect patient privacy. Anonymized data from each patient may be made available subject to an approved research proposal. For further information please see the Patient Level Data section of the GSK Clinical Study Register. Aggregate data will be included; with any direct reference to individual patients excluded *Complete removal of patient data listings may mean that page numbers are no longer consecutively numbered

2 Evidence synthesis of UMEC/VI and comparators in COPD Report GK14290D Version 19.0 Prepared by Mapi for And Written by & 13 November, 2014 GK14290 version 1 8 th July

3 Version History Version Number Summary/Reason for Changes Date Issued 1-16 For details of the reasons for changes of the previous 16 versions, please see version 16 of the technical report 17 Update of the technical report with the addition of an evidence synthesis within the Bayesian framework August 20 th, 2014 September 5 th, Updated report based on comments from GSK October 10 th, Updated report based on comments from GSK November 13 th, 2014 GK14290 version 19 November 13 th, 2014 Page 2 of 178

4 Contents VERSION HISTORY... 2 CONTENTS... 3 ADMINISTRATIVE STRUCTURE... 5 ABBREVIATIONS... 6 EXECUTIVE SUMMARY INTRODUCTION OBJECTIVES METHODOLOGY Study identification and selection Data extraction Frequentist Indirect Treatment Comparison Bayesian Network Meta-Analysis Sample size / Power calculations RESULTS Study identification Evidence base FREQUENTIST INDIRECT TREATMENT COMPARISON ITC of UMEC/VI vs. IND + TIO ITC of UMEC/VI vs. TIO + SAL ITC of UMEC/VI vs. TIO + FOR ITC of UMEC/VI vs. FLUT/SAL 500/50 + TIO ITC of UMEC/VI vs. QVA BAYESIAN NETWORK META-ANALYSIS Trough FEV SGRQ total score TDI focal score Rescue medication use DISCUSSION AND CONCLUSIONS SUMMARY TABLES BAYESIAN NMA SUMMARY TABLES FREQUENTIST ITC APPENDIX 1 BIBLIOGRAPHIC SEARCH STRATEGIES APPENDIX 2 REGISTRY SEARCH STRATEGIES APPENDIX 3 SELECTION CRITERIA GK14290 version 19 November 13 th, 2014 Page 3 of 178

5 APPENDIX 4 - DECISION RULES FOR NMA APPENDIX 5 BAYESIAN NMA MODEL REFERENCES GK14290 version 19 November 13 th, 2014 Page 4 of 178

6 Administrative Structure Sponsor (Name) Director (Respiratory) Global Health Outcomes GlaxoSmithKline Stockley Park (B10) Uxbridge UB11 1BU Tel: Operators (Mapi) (Project Lead) (Senior Lead) HEOR & Strategic Market Access Mapi De Molen AX Houten The Netherlands Tel.: Fax: GK14290 version 19 November 13 th, 2014 Page 5 of 178

7 Abbreviations CFB CI CrI ES FEV 1 FLUT/SAL IND ITC L MCID NMA OR QVA149 RR SE SD SGRQ SLR TIO UMEC/VI Change from baseline Confidence Interval Credible Interval Effect size Forced expiratory volume in 1 second Fluticasone/salmeterol Indacaterol Indirect treatment comparison Liters Minimum clinically important difference Network meta-analysis Odds Ratio Indacaterol 110 mcg / glycopyrronium 50 mcg Relative Risk Standard error Standard deviation St. George's Respiratory Questionnaire Systematic literature review Tiotropium Umeclidinium plus vilanterol GK14290 version 19 November 13 th, 2014 Page 6 of 178

8 Executive Summary Objective: The objective of this study was to assess the relative efficacy of UMEC/VI 62.5/25 µg OD (UMEC/VI 62.5/25) versus indacaterol 150µg + tiotropium 18µg (IND TIO 18) OD, tiotropium 18 mcg OD + salmeterol 50 mcg BID (TIO+SAL), tiotropium 18 mcg OD + formoterol 10 or 12 mcg BID (TIO+FOR), fluticasone/salmeterol 500µg/50µg + Tiotropium 18µg OD (FLUT/SAL 500/50 + TIO 18) and indacaterol/glycopyrronium 110/50mcg OD (QVA149). Methods: A systematic literature review (SLR) was performed to identify randomized controlled trials (RCTs) of more than 10 weeks duration in COPD patients eligible to receive COPD maintenance therapy. The selected studies comparing IND TIO 18, TIO 18 + SAL 50, TIO 18 + FOR 10/12, FLUT/SAL 500/50 + TIO 18 or QVA149 to TIO 18 or placebo were used for the evidence synthesis project. Outcomes of interest are trough FEV 1, SGRQ total score, TDI focal score and rescue medication use at 12, 24 and 52 weeks. The evidence was synthesized by means of a frequentist indirect treatment comparison (ITC) and a Bayesian network meta-analysis (NMA), to indirectly compare UMEC/VI 62.5/25 to relevant comparators by using TIO 18 or placebo as reference treatment. For the frequentist ITC, two scenario analyses were performed. One scenario analysis pooled together the LABA+LAMA open dual combination studies to be able to make a class comparison. Another scenario analysis in- or excluding the SPARK study was performed in the ITC of UMEC/VI 62.5/25 vs. QVA149 because of its more severe patient population. Also two scenario analyses were performed in the Bayesian NMA. The base case included studies comparing LABA/LAMA treatments to each other or to placebo. Due to the lack of studies of LABA/LAMA studies including a placebo arm, the scenario analysis was designed to include TIO 18 as comparator to allow indirect comparisons to more relevant open and closed dual comparators. Results: 13 trials were identified in the SLR that could be included in the evidence synthesis. Results of the frequentist ITC UMEC/VI 62.5/25 was comparable to IND TIO 18 by means of trough FEV 1 and rescue medication use at 12 weeks, and was also comparable to TIO + FOR by means of trough FEV1 at 12 weeks, SGRQ total score at 24 weeks and TDI focal score at 12 weeks. Regarding trough FEV 1 at 12 weeks, UMEC/VI 62.5/25 was more efficacious than FLUT/SAL 500/50 + TIO 18 with a difference in change from baseline of 54.00ml (95%CI 12.80; 95.19, p=0.010). This effect was diminished at 24 weeks, because UMEC/VI 62.5/25 did not show a statistically significant improvement in trough FEV 1 over FLUT/SAL 500/50 + TIO18 anymore. UMEC/VI 62.5/25 showed similar efficacy versus TIO + SAL by means of trough FEV1, SGRQ total score and TDI focal score at 24 weeks. UMEC/VI 62.5/25 showed comparable results to QVA149 by means of trough FEV 1, SGRQ total score, TDI focal score GK14290 version 19 November 13 th, 2014 Page 7 of 178

9 and rescue medication use at 12, 24 and 52 weeks. Excluding SPARK did not have a major impact on these conclusions. Results of the Bayesian NMA UMEC/VI 62.5/25 demonstrated higher efficacy than placebo for every outcome analyzed. In the base case, UMEC/VI 62.5/25 showed similar efficacy compared with QVA149 by means of trough FEV1 at 12 and 24 weeks, SGRQ total score at 24 weeks, TDI focal score at 12 and 24 weeks and rescue medication use at 24 weeks. UMEC/VI 62.5/25 demonstrated also similar efficacy versus TIO 18 + FOR 10 by means of SGRQ total score at 24 weeks. The scenario analysis showed similar results for these comparators. More studies could be included in the scenario analysis for comparing a LABA/LAMA treatment to TIO 18. UMEC/VI 62.5/25 is expected to be more efficacious than TIO 18 on all available endpoints except for TDI focal score at 24 weeks. In addition, UMEC/VI demonstrated similar efficacy compared with IND TIO 18 by means of trough FEV1 and rescue medication use at 12 weeks, comparable efficacy relative to TIO 18 + SAL 50 by means of trough FEV1, SGRQ total score and TDI focal score at 24 weeks and to TIO 18 + FOR 12 by means of trough FEV1 and TDI focal score at 12 weeks. Conclusions: Based on an evidence synthesis of the available RCTs reporting on bronchodilation (as assessed by trough FEV 1 ), health status (as assessed by the SGRQ total score) and breathlessness (as assessed by TDI focal score), UMEC/VI 62.5/25 is comparable to other open dual or closed LABA/LAMA combination therapies. Both within the frequentist and the Bayesian framework, UMEC/VI demonstrated similar efficacy compared with QVA149, IND TIO 18, TIO 18 + SAL 50, TIO 18 + FOR 10 and TIO 18 + FOR 12 on available efficacy endpoints. The scenario analyses in the frequentist or Bayesian framework did not have a major impact on these conclusions. GK14290 version 19 November 13 th, 2014 Page 8 of 178

10 1. Introduction Chronic obstructive pulmonary disease (COPD) is a disease characterized by the development of airway obstruction, manifesting as a decline in lung function, breathlessness and exacerbations. The disease is usually progressive and associated with an enhanced chronic inflammatory response in the airways and lungs. It is the fourth leading cause of death worldwide and a major cause of chronic morbidity and mortality throughout the world. 1 The COPD burden is expected to increase over the coming decades because of continued exposure to COPD risk factors and aging of the population. 1 The goals of effective COPD management are to prevent disease progression, relieve symptoms, prevent and treat complications and exacerbations and to improve health status. Current guidelines 1, 2 recommend the use of long-acting bronchodilators, as they are more effective at producing maintained symptom relief than short-acting bronchodilators. The choice between long-acting β2- agonists (LABAs), long-acting anticholinergic (LAMA), and fixed combination of LABA plus ICS (inhaled corticosteroids) in one inhaler depends on the individual patient response in terms of symptom relief and side effects. Also, combining bronchodilators of different pharmacology classes may be considered as a better alternative to manage COPD compared to increasing the dose of a single bronchodilator. Currently GSK is developing a closed combination of an inhaled long-acting muscarinic antagonist (LAMA) and long-acting beta-2 agonist (LABA) for COPD, umeclidinium/vilanterol (UMEC/VI), delivered using the Ellipta dry-powder inhaler. The clinical efficacy and safety of UMEC/VI is being evaluated in a comprehensive Phase III program. The anticipated target population for UMEC/VI is mild to moderate COPD patients who have a history of no or very few exacerbations, including patients who are symptomatic on monotherapy. A clear value proposition, supported by robust comparative clinical evidence will be required to address the needs of payers in different markets. GSK asked Mapi to conduct a robust comparative assessment of UMEC/VI with all of its potential comparators. Therefore a comprehensive systematic review (SLR) was performed to identify randomized controlled trials (RCTs) in COPD patients eligible for maintenance therapy. A feasibility assessment was performed to evaluate the appropriateness of evidence synthesis for UMEC/VI with its comparators (based on the results of the systematic literature review). If appropriate, the trial evidence should be used to perform a frequentist indirect treatment comparison. The results presented in this report are based on the protocol - HO / etrack which explains the ways in which indirect comparison of UMEC/VI versus open LABA/LAMA combinations, open ICS+LABA+LAMA combinations and QVA149 (indacaterol/glycopyrronium) can be performed together with the appropriate methods. GK14290 version 19 November 13 th, 2014 Page 9 of 178

11 2. Objectives The objectives of the project described in this report were: Primary 1. To assess the relative efficacy of UMEC/VI 62.5mcg/25mcg OD vs. Indacaterol (IND) 150mcg + Tiotropium (TIO) 18mcg OD at 12 weeks Fluticasone/salmeterol 500mcg/50mcg BID + Tiotropium 18mcg OD (FLUT/SAL + TIO) at 12 and 24 weeks QVA149 (glycopyrronium/indacaterol 50mcg/110mcg OD) at 12 and 24 weeks by means of lung function (difference in change from baseline for trough FEV 1 ). Secondary 2. To assess the relative efficacy of UMEC/VI 62.5mcg/25mcg OD vs. IND + TIO by means of rescue medication use at 12 weeks 3. To assess the relative efficacy of UMEC/VI 62.5/25mcg OD vs tiotropium 18 mcg OD + salmeterol 50 mcg BID (TIO+SAL) by means of available efficacy endpoints (trough FEV1, SGRQ total score, TDI focal score, rescue medication use) at 12 or 24 weeks 4. To assess the relative efficacy of UMEC/VI 62.5/25mcg OD vs tiotropium 18 mcg OD + formoterol 10 or 12 mcg BID (TIO + FOR) by means of available efficacy endpoints (trough FEV1, SGRQ total score, TDI focal score, rescue medication use) at 12 or 24 weeks 5. To assess the relative efficacy of UMEC/VI 62.5/25mcg OD vs LABA+LABA open dual combination therapies (IND+TIO, TIO+FOR, TIO+SAL) by means of available efficacy endpoints (trough FEV1, SGRQ total score, TDI focal score, rescue medication use) at 12 or 24 weeks in a scenario analysis 6. To assess the relative efficacy of and UMEC/VI 62.5mcg/25mcg OD vs. QVA149 by means of: Health related quality of life (change from baseline for SGRQ total score) at 12 and 24 weeks. Dyspnoea (change from baseline for TDI focal score) at 12 and 24 weeks. Decrease in rescue medication use (mean number of puffs per day) at 24 weeks. Comparative assessment time points have been selected by GSK based on the availability of evidence as determined by the systematic literature review. Specific endpoints have been selected by GSK based on the requirements of payers. GK14290 version 19 November 13 th, 2014 Page 10 of 178

12 3. Methodology 3.1. Study identification and selection A systematic review of the literature following the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines was performed to identify published, randomized controlled clinical trials (RCTs) comparing selected therapies in patients with COPD. The comparators, outcomes and time points of interest were identified by GSK prior to the current study. The selection criteria were broader than the objective in this ITC, to identify also LABA and LAMA monotherapies. Searches were performed between April 14th 2014 and April 16 th 2014 in several databases and trial registry websites, using predefined search strategies specifically tailored to each platform (Appendix 1 and 2). The relevance of each identified citation was based on title and abstract according to predefined selection criteria (Appendix 3). Selection was performed by two researchers independently. Any discrepancies between researchers were resolved by consensus. For the abstracts that met the selection criteria, available publications were obtained and evaluated using the full-text selection criteria. The final selected citations were grouped per study (one study could have been published in several sources such as conference abstract, full text article, trial registration). Only studies reporting the required outcomes of interest were selected for the analysis. The predefined selection criteria for the SLR were based on the PICOS criteria, i.e. population, intervention, comparators, outcomes and study design of interest. Studies were selected if the following selection criteria were fulfilled: Population: Adult patients with COPD. Interventions of interest: All studies reporting on at least one of the interventions in Table 1 are of interest. Table 1. Interventions of interest in the SLR Anticipated comparator drugs Trade name LAMA monotherapies Umeclidinium 62.5 mcg OD Aclidinium 400 mcg BID Tiotropium 18 mcg OD Glycopyrronium 50 mcg OD Incruse Tudorza Pressair, Eklira Genuair Spiriva Seebri, Enurev Breezhaler LABA monotherapies Salmeterol 50 mcg BID Serevent GK14290 version 19 November 13 th, 2014 Page 11 of 178

13 Formoterol 12 mcg BID Indacaterol 75, 150, 300 mcg OD Olodaterol 5, 10 mcg OD Foradil, Oxis, Atimos-Modulite Onbrez, Arcapta Stiverdi Respimat LABA + LAMA Umeclidinium/vilanterol 62.5/25 mcg OD Zephyr, Anoro Ellipta Indacaterol + tiotropium Tiotropium + salmeterol Tiotropium + formoterol Indacaterol/glycopyrronium (indacaterol/nva237) QVA149 Any other combination of a LABA and LAMA LABA/ICS + LAMA Fluticasone/Salmeterol 500/50mcg + Tiotropium 18mcg Comparators of interest: Studies that compare agents as listed under the interventions of interest (table 1) to each other or to placebo were included. Outcomes: For all the outcomes evaluated at 12 and 24 weeks, a margin of time was deemed to be acceptable to strengthen the networks. If the endpoint of interest was not reported at 24 weeks, but instead at a time comprised between 8 to 16 weeks and 20 to 28 weeks, the proxy outcome was reported, with a comment specifying the substitute time. For the analysis presented in this report, the following outcomes were of interest: Trough FEV 1 TDI score SGRQ score Rescue medication Study design: Only randomized controlled trials (RCTs) with minimum duration of 10 weeks were included. Databases: The bibliographic systematic searches were conducted independently in the following databases: MEDLINE (through OVID platform), GK14290 version 19 November 13 th, 2014 Page 12 of 178

14 MEDLINE In-Process (through OVID platform), EMBASE (through OVID platform), The Cochrane Library: Cochrane Database of Systematic Review (CDSR) and Cochrane Central Register of Controlled Trials (CENTRAL), Database of Abstracts of Reviews of Effects (DARE), Health Technology Assessment (HTA) websites: HTA database, National Institute for Health Research (NIHR). Secondary systematic searches were performed in clinical trial registries: Clinicaltrials.gov, the US National Institutes of Health clinical trial register World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) Current Controlled Trials EU Clinical Trials Register (EU-CTR) Klinische Prüfungen PharmNet.Bund The International Prospective Register of Systematic Reviews (PROSPERO). Time horizon: No time restrictions were used for the searches. Language: The search was restricted to English and German language. The selection criteria for the SLR were broader than the selection criteria for the ITC. LABA or LAMA monotherapies were not of interest in the ITC and therefore an additional step was added to the SLR. After finalizing the full text screening, the studies were reviewed for inclusion in the ITC. Studies were included in the ITC if they evaluated one of the following combination therapies of LABA and LAMA and reported data on trough FEV1, TDI score, SGRQ score or rescue medication use. Umeclidinium/vilanterol 62.5/25 mcg OD Indacaterol + tiotropium Tiotropium + salmeterol Tiotropium + formoterol Indacaterol/glycopyrronium (indacaterol/nva237) GK14290 version 19 November 13 th, 2014 Page 13 of 178

15 3.2 Data extraction Data extraction was performed with a standard data extraction form (DEF). Key data from each eligible study was extracted by recording data from original reports into the DEF. Data extraction was performed by one researcher and reviewed by another. This means that based on the data extraction form, the second reviewer was tracing back every value/number/comment to the original report and checked the extracted data. The following study characteristics were extracted: author publication year compared interventions including drug name, dose and administration frequency number of randomized patients trial design centers and countries inclusion criteria background treatments trial duration ICS allowed (as background) LABA's allowed (as background) The following baseline patient characteristics were extracted: proportion of males age and standard deviation (sd) proportion of current smokers proportion of patients with severe or very severe COPD proportion of patients using ICS duration of COPD (sd) smoking history pack-years (sd) FEV 1 % pred (sd) FEV 1 /FVC percentage (sd) FVC mean (sd) BDI mean number of exacerbations in previous year percentage reversibility ethnicity The following outcomes were extracted at 12 (8-16 weeks) and 24 weeks (20-28 weeks) GK14290 version 19 November 13 th, 2014 Page 14 of 178

16 Trough FEV 1 TDI score SGRQ total score Rescue medication use (number of puffs per day) 3.3 Frequentist Indirect Treatment Comparison An indirect treatment comparison (ITC) of a treatment A versus a treatment B can be performed if it is anchored on a third treatment C (the term adjusted is also used in the literature). Figure 1. ITC network For the indirect comparison the method proposed by Bucher et al 1997 will be used. 3 To derive indirect comparisons of A versus B, this method compares the magnitude of treatment effects of these interventions relative to a common comparator (i.e. AC and BC). First, the feasibility to perform the ITC will be evaluated by closely reviewing the study design, patient characteristics and definition of outcomes of the included studies. If the trials and outcomes of interest are not similar enough, the ITC will not be feasible Continuous data Step 1: The pooled mean difference in change from baseline for each treatment of interest relative to a common comparator (i.e. AC and BC) will be estimated using standard random effects metaanalyses, as proposed by DerSimonian and Laird. 5 Step 2: Let μ AC, SE(μ AC ) and μ BC, SE(μ BC ) be the combined mean difference and corresponding standard error of the A vs. C and B vs. C meta-analytic comparisons, respectively. The indirect comparison of treatments A versus B may then be estimated as follows: μ AB = μ AC μ BC var(µ AB ) = var(µ AC ) + var(µ BC ) (4) SE(µ AB ) = SE(µ AC ) 2 + SE(µ BC ) 2 95% CI AB = μ AB ± 1.96 SE(µ AB ) For each meta-analysis, e.g. A vs. C, we can estimate the variance using the following formula: GK14290 version 19 November 13 th, 2014 Page 15 of 178

17 var(µ AC ) = ((UL 95%CI AC LL 95%CI AC )/3.92) 2 (5) (with UL and LL the upper en lower 95% confidence interval of CI AC ) Heterogeneity The statistical heterogeneity will be assessed by means of the Cochran Q, chi-square test and the I 2 statistic with 95% CI. The clinical heterogeneity will be assessed by means of study design inclusion criteria related to FEV 1, FEV 1 /FVC, exacerbations and smoking background treatment: ICS and/or LABA randomization blinding open label arms cross over design Baseline patient characteristics exacerbation history % patients per COPD severity level mean FEV 1 % predicted % current smokers mean pack-years % of male patients mean age If confounders are present in an indirect comparison, it is only possible to adjust for them via metaregression. This can be done by replacing Step 2 with a random effects meta-regression in order to estimate the comparative treatment effects between two drugs (e.g. by using the STATA function metareg). However, in our analysis due to the relatively low number of studies it is not feasible to perform adjustment via meta-regression (for example ICWG Report p. 26 suggests at least 10 studies per adjustment variable). 6 Therefore, we will conduct a separate scenario analysis excluding the evidence where such clinical heterogeneity exists and it s feasible to do so with regard to the number of studies included. Results of the meta-analyses GK14290 version 19 November 13 th, 2014 Page 16 of 178

18 Results of the individual trials (point estimates and 95% confidence intervals) together with the pooled results (point estimates and 95% confidence intervals), the Cochran Q value, the I 2 value and chisquare test results are presented. Results of ITC The results of the ITC will be presented as mean difference in CFB with 95% CI, and p-value. The random effects pairwise meta-analyses will be conducted with STATA using metan (StataCorp Stata Statistical Software: Release 11) or R (R Development Core Team, using the packages meta and metafor. 3.4 Bayesian Network Meta-Analysis Bayesian network meta-analysis (NMA) models were used to analyze the created data set for the outcomes of interest in order to simultaneously synthesize the results of the included studies and to obtain relative treatment effects. 7-9 NMA within the Bayesian framework involve data, a likelihood distribution, a model with parameters, and prior distributions. 9 The model relates the data from the individual studies to basic parameters reflecting the (pooled) relative treatment effect of each intervention compared to an overall reference treatment, e.g. placebo. Based on these basic parameters, the relative efficacy between each of the competing interventions was obtained. Different statistical models should be used for different types of data. Only continuous outcomes were of interest in the current project, therefore a generalized linear model with identity link and a normal likelihood distribution was used. The analysis was based on the difference between the least square mean at follow-up (LS mean at FU) or the change from baseline (CFB) for the active treatment versus the baseline comparator as well as the associated standard error (SE) of the difference. See appendix 4 for the decision rules regarding the continuous outcomes. For each outcome, a fixed and a random effects approach were evaluated. The fixed effects model assumes that the differences in true relative treatment effects across studies in the network of evidence are only caused by the differences in treatment comparisons (i.e. that there is no variation in relative treatment effects for a particular pair wise comparison). The random effects model, on the other hand, assumes that differences in observed treatment effects across the studies in the network are not only caused by the different treatment comparisons, but that there is also heterogeneity in the relative effects for a particular type of comparison caused by factors that modify that relative treatment effect. With the NMA models used, the heterogeneity is assumed constant for every treatment comparison. The random effects models are presented at the end of this section. With a NMA, randomization only holds within a trial and not across trials. As a result, there is the risk that patients who were studied in different comparisons are not similar which leads to consistency violations. In order to minimize confounding bias, treatment by covariate interactions can be incorporated in the models. However, as in the frequentist analysis, it was not feasible to perform GK14290 version 19 November 13 th, 2014 Page 17 of 178

19 adjustment via meta-regression due to the relatively low number of studies in the base case scenarios (for example ICWG Report p. 26 suggests at least 10 studies per adjustment variable). 6 As an alternative, scenario analyses can be developed to test the impact of certain studies on the relative treatment estimates. In order to identify the most appropriate model (i.e. fixed or random effects models with or without constant treatment by covariate interactions,) given the evidence base, the goodness-of-fit of model predictions to the observed data can be measured by calculating the posterior mean residual deviance, D. 10 The deviance information criterion (DIC) was used to compare the fixed and random effects model and provides a measure of model fit that penalizes model complexity according to DIC = D + pd, pd = D Dˆ. pd is the effective number of parameters and Dˆ is the deviance evaluated at the posterior mean of the model parameters. Given the dataset used, the fixed effect model was chosen over the random effect model unless there was enough evidence to suggest the random effect model was substantially different (i.e. DIC value was lower and Monte Carlo error was not out of proportion). In order to avoid prior beliefs influencing the results of the model, non-informative prior distributions were used. Prior distributions of the relative treatment effects were normal distributions with mean 0 and a variance of A uniform distribution with range of 0-2 was used for the prior distribution of heterogeneity for the random effects models. The posterior densities for the unknown parameters were estimated using Markov chain Monte Carlo (MCMC) simulations for each model. As a standard, the runs were based on 80,000 iterations on three chains, with a burn-in of 20,000. Convergence assessment was based on visual inspection of trace and autocorrelation plots and on the Gelman Rubin-Brooks diagnostic. The accuracy of the posterior estimates was assessed using the Monte Carlo error for each parameter (Monte Carlo error < 5% of the posterior sd). WinBUGS statistical software was used for the analyses Models The random effect models for normally distributed outcomes are presented below. A fixed effect 2 model is obtained if σ equals zero. All WinBUGS codes are based on those published by Dias et al in NICE Decision Support Unit Technical Document 2 12 and are in line with HAS recommendations. 13 An overview of the model is presented in Appendix 5. GK14290 version 19 November 13 th, 2014 Page 18 of 178

20 Random effects network meta-analysis model for normally distributed outcomes 2 diff jbk ~Normal(δ jbk,σ jbk ) (Likelihood) δ d jbk AA ~Normal(d = 0 bk,σ 2 ) = Normal(d Ak d Ab,σ 2 ) (Random effects model) d Ak ~Normal( 0, ) σ~uniform( 0, 10 ) (Prior distributions) Where j = study b = control group, can be treatment A, B, C k = treatment group, can be B, C, D diff jbk = Observed difference in change from baseline for a continuous outcome for intervention k versus reference treatment b in study j δ jbk = difference in change from baseline for treatment k relative to comparator b in study j σ jbk = d Ak = 2 σ = uncertainty ( sampling error ) in the difference in change from baseline for intervention k versus reference treatment b in study j pooled difference in change from baseline for treatment k relative to reference treatment A constant variance of relative effects for each treatment versus reference treatment A across studies as a reflection of heterogeneity GK14290 version 19 November 13 th, 2014 Page 19 of 178

21 Presentation of results The Bayesian network meta-analysis provides posterior distributions of the relative treatment effects between interventions for each outcome of interest. The posterior distributions are summarized with the median to reflect the most likely value of the estimate, and the 2.5th and 97.5th percentile to capture the 95% Credible Interval (95%CrI). The 95% CrI represents the range of true underlying effects with 95% probability. For each endpoint, the probability that each treatment is better than a certain comparator is presented. Table 2. Outcomes presentation for relative treatment effects Type of outcome Continuous Outcome FEV 1 Trough SGRQ total score Rescue medication Relative treatment effects Mean difference in CFB vs placebo at 24 weeks. TDI total score Mean difference in TDI total score Imputation of standard errors In some occasions, studies reported only a mean difference in CFB without any measure of uncertainty, such as a SE or SD. Within the Bayesian NMA framework, an imputation of the SE can be made using the uncertainty of other trials in the network following the steps below Within the Bayesian framework, the following steps are followed to impute SE values 14 : #1 The SD of the difference is calculated for every study reporting enough data, using: SD of difference = SE of diff * square root of N #2 The average SD of the trials in the network is calculated #3 The average SD of the trials in the network is imputed for the trial that did not report a SD/SE/95%CI #4 The SE of the difference = average SD / square root of N GK14290 version 19 November 13 th, 2014 Page 20 of 178

22 3.2. Sample size / Power calculations The study described in this report is based on the synthesis of evidence provided by randomized controlled trials, by means of meta-analysis, ITC or network meta-analysis (NMA). For this reason, the suggested methods for the power calculations are not identical to those used for the power calculations of clinical studies, although based on the same principles and tests. As a first, general comment, we would like to underline the fact that to our knowledge there are no sound recommendations and clear guidelines from the HTA authorities on the ITC and NMA power calculations. The suggested methods are described below for the (post-hoc) estimation of the power per approach. I. Frequentist approach 1. Meta-analyses There are some well-established approaches for the estimation of the required information or, alternatively, the power (see for example Cohn & Becker 2003 and references therein). 15 The information size N needed to reject an intervention effect μ with a type I error less than a and type II error less than β is given by the following formula (Wetterslev et al, 2009) 16 : N AC = C z 1 a/2 + z 1 β 2 v AC μ AB 2 (9) Where, z 1 a/2 and z 1 β are the (1 a/2)th and (1 β)th percentiles of a standard normal distribution, v is the variance of the meta-analysis, C is a constant depending on the randomization ratio and number of treatment arm (C = 4 with a randomization ratio of 1:1 and two treatment arms, Thorlund & Mills 2012). 17 If statistical heterogeneity exists across the included trials (between-trial heterogeneity) in a metaanalysis, the calculated sample size can be adjusted to account for the additional variation (i.e., increased uncertainty). N AC = N AC (1 I 2 AC ) (10) Where the sample size of each meta-analysis is penalized by multiplying by the lack of homogeneity (1 I 2 AC ). We can retrospectively estimate the power (1-β) of the current data by re-arranging the expression of the required sample in eq. 9 (or eq. 9 & 10 in case of heterogeneity). 2. ITC GK14290 version 19 November 13 th, 2014 Page 21 of 178

23 Building upon this approach, Thorlund and Mills 17 proposed a simple algorithm for the estimation of power in case of ITC (i.e. 3 treatments A, B, C with head to head trails available for A vs C and B vs C). The effective power in an indirect comparison is calculated as follows: 1. Calculate the effective indirect sample size The effective indirect sample size in case of low heterogeneity is given by the following formula: N AB = N AC N BC N AC + N BC (11) The heterogeneity-corrected effective sample size for indirect evidence is given by: N AB = N AC(1 I 2 AC ) N BC (1 I 2 BC ) N AC (1 I 2 AC ) + N BC (1 I 2 BC ) (12) An important limitation of the above-proposed sample size heterogeneity correction in metaanalyses with limited number of trials (as in our analysis for ANORO) is the fact that I 2 s are typically unreliable and unstable. In most cases, it will therefore be preferable to simply assume some plausible degree (percentage). Typically, an assumption of 25% or 50% heterogeneity will be reasonable in the context of sample size considerations (Wetterslev et al. 2009). In this report, the assumption of 25% heterogeneity was deemed reasonable. 2. Calculate the power of the available indirect evidence Insert the effective indirect sample size (eq. 11 or 12) in the formula used for the metaanalyses: N AB = C z 1 a/2 + z 1 β 2 v AB μ AB 2 (13) and estimate the power (1-β) of the current test procedure by re-arranging the expression of the required sample. 3. Network meta-analyses (NMA) Power estimation in NMA, especially in the case of complicated networks (e.g. multiple, closed loops) is not straightforward and - to our knowledge - there are no well-established and widely- acceptable methods. A simple method is suggested by Thorlund and Mills, 17 as an extension of their ITC power calculation method. This approach should have to be tested in more complicated networks to assess the results. GK14290 version 19 November 13 th, 2014 Page 22 of 178

24 4. Results 4.1. Study identification The bibliographic literature search was performed on 16 April 2014 in Medline, Medline In-process, EMBASE, Cochrane CENTRAL, Cochrane CDSR, and the University of York Centre for Review and Dissemination DARE and HTA databases. The number of identified abstracts per database is presented in Appendix 1. In total, 3,006 abstracts were identified in the different databases and were screened independently by two researchers (Figure 3). Of these, 309 (10.3%) abstracts were of interest based on the predefined selection criteria. The reason for excluding the other abstracts was study design (839 abstracts), population (64 abstracts), intervention (118 abstracts), comparator (65 abstracts) conference abstract published before 2009 (390 abstracts), outcomes (1) duplicates (1,195 abstracts) and language other than German or English (25 abstracts). Full text articles were ordered for the 309 abstracts of interest and these were screened for inclusion again by two researchers independently. Of these, 150 publications were included in the systematic literature review, while 159 articles were excluded. The reasons for exclusion were study design (57 articles), intervention (21 articles), comparison (18 articles), outcomes (53 articles), duplicates (4 articles), language (2 articles), or because they were non-retrievable. The 150 included publications reported on 97 different trials. In parallel, additional internet searches were performed in trial registries to identify completed and ongoing trials. The number of identified registries per website is presented in Appendix 2. Of the 4,720 identified registries, 159 were included in the SLR. Of the 4,560 excluded registries, 1,813 were excluded for study design, 845 for intervention, 504 for population, 78 for comparator out of scope, 17 for language not of interest and 1,303 were duplicates. The 159 trial registries reported on 138 separate trials. The selection criteria for the SLR were broader than for this ITC; LABA and LAMA monotherapies were excluded from the ITC. 109 abstracts were excluded from the abstract screening process, for reporting on LABA or LAMA monotherapies only. Then, the results of the registry screening and the database screening were merged, and different sources of the same trial were grouped together. Studies on aclidinium/formoterol, tiotropium/olodaterol and studies without any data on the efficacy endpoints were not of interest in the ITC. The evidence base of the ITC comprised of 59 citations related to 13 trials. Four CSRs were provided by GSK (DB , DB , DB , and ZEP117115) and in total the evidence base retrieved in the SLR comprised of the following studies: DB (1 conference abstract; two trial registries; 1 CSR) DB (1 conference abstract; two trial registries; 1 CSR) DB (1 full text paper; 1 conference abstract; 2 trial registries; 1 CSR) GK14290 version 19 November 13 th, 2014 Page 23 of 178

25 ZEP (2 trial registries; 1 CSR) SPARK (1 full text paper; 12 conference abstracts; 2 registries) ENLIGHTEN (1 full text paper; 2 conference abstracts; 2 trial registries) SHINE (1 full text paper; 9 conference abstracts; 2 trial registries) INTRUST-1 (1 full text paper (reporting on INTRUST-1 and -2); 2 registries) , 68, 69 INTRUST-2 (1 full text paper (reporting on INTRUST-1 and -2); 2 registries) 70, 71 Aaron 2007 full text publication and trial registry Cazzola 2007 full text publication 72 Vogelmeier 2008 full text publication 73 Tashkin 2009 publication 74 The study of Aaron et al (2007) compared FLUT/SAL+TIO with TIO at time points of 4, 20, 36 and 52 weeks. 70 A margin of time was deemed to be acceptable to strengthen the networks, endpoints at 20 weeks were similar enough to 24 week data to be included in an ITC. GK14290 version 19 November 13 th, 2014 Page 24 of 178

26 Search on April 2014 (n=4720) Clinicaltrials.gov (n=949), HTA (n=13), WHO ICTRP (n=2922), Current controlled trial (n=87), EU-CTR (n=307), PharmNet.Bund (n=320), PROSPERO (n=122) Trial screening n= 4720 Not relevant n= 4561 Study design out of scope (n=1813) Population out of scope (n=504) Intervention our of scope (n=845) Comparator out of scope (n=78) Outcomes not of interest (n=1) Duplicates (n=1303) Language (n=17) 159 Registries included for 138 trials clinicaltrials.gov (n=102), WHO ICTRP (n=5), Current controlled trials (n=1), EU- CTR (n=51) Not relevant n= 102 Intervention out of scope (n=102) 57 Registries included for 42 trials clinicaltrials.gov (n=28), Current controlled trials (n=1), EU-CTR (n=28) Figure 2. Flow chart of registry selection process. GK14290 version 19 November 13 th, 2014 Page 25 of 178

27 Search on16 April 2014 (n=3006) EMBASE (n=881), MEDLINE (n=610), CDSR (n=21) CENTRAL (n=1415) DARE (n=62) HTA (n=17) Title/ abstract screening n= 3006 Not relevant n= 2697 Study design out of scope (n=839) Population out of scope (n=64) Intervention out of scope (n=118) Comparison out of scope (n=65) Outcomes out of scope (n=1) Conference abstract <2009 (n=390) Duplicate (n=1195) Language (n=25) Full text screening n= 309 Not relevant n= 159 Study design out of scope (n=57) Population out of scope (n=0) Intervention out of scope (n=21) Comparison out of scope (n=18) Outcomes out of scope (n=53) Duplicate (n=4) Language (n=2) Non retrievable (n=4) Included in SLR: 150 publications related to 97 trials Full text articles, n=60 Conference abstracts, n=90 Included in registry screening n=57 These 57 registries reported on 42 individual studies Clinicaltrials.gov (n=28) EU-CTR (n=28) Current controlled trials (ISRCTN) (n=1) CSRs from GSK n=4 Not relevant n= 152 Monotherapies only (n=109) LABA/LAMA out of scope (n=21) No results reported (n=22) Figure 3. Flow chart of study selection process Included in ITC: 59 citations related to 13 trials Full text articles, n=9 Conference abstracts, n=27 Trial registries, n=19 CSRs, n=4 GK14290 version 19 November 13 th, 2014 Page 26 of 178

28 4.2. Evidence base Study characteristics An overview of study characteristics is presented in Table 3. All studies were randomized, double blind, multicenter trials. Three studies (Vogelmeier 2008, SPARK and SHINE) included tiotropium 18mcg as an open label arm, the other arms of these trials were double-blind. Cazzola 2007, DB , DB and ZEP were double-blind, double-dummy trials that had trough FEV1 as primary efficacy endpoint. DB and SHINE also had trough FEV1 as primary objective. Vogelmeier had FEV1 2 hours post-dose as primary outcome and Tashkin 2009 focused on the area under the curve for FEV1 measured 0-4 hours after dosing as primary efficacy variable. The primary outcome of INTRUST-1 and INTRUST-2 was the standardized area under the curve of FEV1 from 5 minutes to 8 hours post dose. The primary objective of SPARK and the study by Aaron et al (2007) was the exacerbation rate, and the ENLIGHTEN study had the number of treatment emergent adverse events as primary endpoint. The SPARK (64 weeks), Aaron 1007 (52 weeks) and ENLIGHTEN (52 weeks) studies were considerable longer than the other trials, that had a duration of 26 weeks (SHINE) or 24 weeks (Vogelmeier 2008, DB , DB , DB and ZEP117115) or 12 weeks (Cazzola 2007, INTRUST-1, INTRUST-2, Tashkin 2009) Patient characteristics All trials included patients with COPD older than 40 years and with a smoking history of more than 10 pack-years. Cazzola 2007 included well-controlled severe to very severe COPD patients with a smoking history of 20 pack-years or more, a baseline FEV1 of less than 50% of predicted and a FEV1/FVC ratio of less than 70%. Also the SPARK study included more severe COPD patients than the other studies. SPARK included patients with stage III or IV COPD (severe or very severe COPD according to GOLD guidelines) with a FEV1/FVC ratio of less than 70% and a FEV1 of less than 50% of predicted normal values, and at least one exacerbation in the previous year requiring treatment with systemic corticosteroids, antibiotics or both. In contrast, the other studies included moderate or severe (stage II or III) COPD patients with a FEV1/FVC ratio of less than 70% and a FEV1 of less than 70% of predicted normal values, and no inclusion criterion related to the number of exacerbations in the previous year. As a result of these different inclusion criteria, differences were observed in the Cazzola 2007 and SPARK study compared with the other trials in COPD severity at baseline. For the same reason, the proportion of patients using ICS at baseline in the SPARK study was higher (75%) than in other studies (51% on average). In addition, the FEV1 percentage predicted was lower at baseline in the SPARK study (37.2% predicted) and Cazzola 2007 (38.7%) versus other studies (FEV1 ranging from 42.1% predicted to 59.4% predicted across treatment arms). GK14290 version 19 November 13 th, 2014 Page 27 of 178

29 The other patient characteristics were homogeneous across trials. The percentage of males in each study arm ranged from 65% to 77.3% and the mean age was 61.9 to 65.0 years across treatment arms within the trials. Long acting bronchodilators had to be discontinued before entering all trials. As a result, INTRUST-1, INTRUST-2, Aaron 2007, Tashkin 2009, Vogelmeier 2008, DB , DB , DB , ZEP117115, SHINE and ENLIGHTEN are expected to be similar enough to be synthesized in an ITC. The more severe patient population of Cazzola 2007 and SPARK may introduce clinical heterogeneity in the ITC. However, since Cazzola 2007 was the only trial providing the link in the network from TIO 18 to FLUT/SAL+TIO by means of FEV1 trough at 12 weeks, it was not possible to design a scenario analysis excluding this trial. To evaluate the impact of the severe population from SPARK on the relative efficacy estimates of UMEC/VI 62.5/25 vs. QVA149 in the ITC, a sensitivity analysis in- or excluding this trial was performed. The low number of included trials in each direct comparison (i.e. three or less) did not allow for a meta-regression. GK14290 version 19 November 13 th, 2014 Page 28 of 178

30 Table 3. Key study characteristics for all studies included in the ITC (only arms of interest) Trial Study Treatment Inclusion criteria Duration Background treatment DB Tiotropium; 18µg; OD Vilanterol 25µg + Umeclidinium 62.5µg Vilanterol 25 µg + Umeclidinium 125µg 24 weeks Outpatient; >= 40 years old; diagnosed with COPD, post-salbutamol FEV 1 70% and post-salbutamol FEV 1/FVC ratio < 0.7. Smoking history 10 pack-years Allowed: ICS at a dose of up to1000 mcg/day of FP or equivalent, salbutamol/albuterol as rescue Not allowed: LABAs, short acting β2-agonists, short acting anticholinergics and SABA/ICS combination products DB Tiotropium; 18µg; OD Vilanterol 25µg + Umeclidinium 62.5µg Vilanterol 25 µg + Umeclidinium 125µg 24 weeks Outpatient; >=40 years old; diagnosed with COPD; post-salbutamol FEV 1/FVC ratio of <0.70 and a postsalbutamol FEV 1 of 70%; Smoking history 10 packyears Allowed: ICS at a dose of up to1000 mcg/day of FP or equivalent, salbutamol/albuterol as rescue Not allowed: LABAs, oral short acting and long acting β2-agonists, inhaled short acting β2-agonists, inhaled short acting anticholinergics and SABA/ICS combination products DB Placebo Vilanterol 25µg + Umeclidinium 62.5µg 24 weeks Outpatient; >=40 years old; diagnosed with COPD; post-salbutamol FEV 1/FVC ratio of <0.70 and a postsalbutamol FEV 1 of 70%; Smoking history 10 packyears Allowed: ICS at a dose of up to1000 mcg/day of FP or equivalent, salbutamol/albuterol as rescue Not allowed: LABAs, LABA/ICS combination products, short acting β2-agonists, short acting anticholinergics and SABA/ICS combination products ZEP Tiotropium; 18µg; OD Vilanterol 25µg + Umeclidinium 62.5µg 24 weeks Outpatient; >=40 years old; diagnosed with COPD; post-salbutamol FEV 1/FVC ratio of <0.70 and a postsalbutamol FEV 1 of 70%; Smoking history 10 packyears Allowed: ICS at a dose of up to1000 mcg/day of FP or equivalent, salbutamol/albuterol as rescue Not allowed: LABAs, LABA/ICS combination products, oral short acting and long acting β2- agonists, inhaled short acting β2-agonists, inhaled short acting anticholinergics and SABA/ICS combination products GK14290 version 19 November 13 th, 2014 Page 29 of 178

31 Study Treatment Trial Duration Inclusion criteria Background treatment INTRUST1 (Mahler 2012) Tiotropium; 18µg; OD Indacaterol 150µg + Tiotropium; 18µg; OD 12 weeks >= 40 years old; Post-bronchodilator FEV 1 65% and 30%. Post-bronchodilator FEV 1/FVC < 70%. Smoking history 10 pack years Allowed: ICS monotherapy, salbutamol/albuterol as rescue Not allowed: LABAs, short acting β2-agonists (except those prescribed in the study), theophylline, anticholinergics INTRUST2 (Mahler 2012) Tiotropium; 18µg; OD Indacaterol 150µg + Tiotropium; 18µg; OD 12 weeks >= 40 years old; Post-bronchodilator FEV 1 65% and 30%. Post-bronchodilator FEV 1/FVC < 70%. Smoking history 10 pack years Allowed: ICS monotherapy, salbutamol/albuterol as rescue Not allowed: LABAs, short acting β2-agonists (except those prescribed in the study), theophylline, anticholinergics Cazzola 2007 Tiotropium; 18μg; OD Salmeterol/Fluticasone; 50/500μg; BID + Tiotropium; 18μg; OD 12 weeks >= 40 years old ; FEV 1 <50%; post-bronchodilator FEV 1/FVC< 70%; 20 pack-year smoking history; excluded if unstable doses oral corticosteroid 4 weeks prior Allowed: stable regimens of theophylline, salbutamol for relief of breakthrough symptoms Not Allowed: regular treatment for COPD (LABAs, ICS), inhaled or oral bronchodilators, systemic corticosteroids, ipratropium, oxitropium, leukotriene modifiers Aaron 2007 Tiotropium 18µg OD + salmeterol 25µg 2puffs BID Tiotropium 18µg OD + placebo 2 puffs BID Salmeterol/Fluticasone; 25/250μg/puff; two puffs BID + Tiotropium; 18μg; OD 52 weeks >= 35 years old; diagnosis of moderate or severe COPD; >= 1 exacerbation of COPD requiring systemic steroids or antibiotics in previous 12 months; smoking history 10 pack years; Post-bronchodilator FEV 1 65%; FEV 1/FVC < 70%. Allowed: albuterol for relief of symptoms Not allowed: ICS, LABA, anticholinergics GK14290 version 19 November 13 th, 2014 Page 30 of 178

32 Study Treatment Trial Duration Inclusion criteria Background treatment ENLIGHTEN (Dahl 2013) Placebo QVA149 (110 µg indacaterol/50 µg glycopyrronium); OD 52 weeks >= 40 years old; diagnosis of moderate or severe COPD (stage II or III according to GOLD 2008 criteria); post-bronchodilator FEV 1 < 80% and 30%. Postbronchodilator FEV 1/FVC < Smoking history 10 pack years Allowed: Albuterol as rescue medication, ICS monotherapy Not allowed: long-acting bronchodilators (LABA, LAMA, theophylline), short-acting muscarinic antagonists SPARK (Wedzicha 2013) Tiotropium; 18μg; OD QVA149 (110 µg indacaterol/50 µg glycopyrronium); OD 64 weeks >= 40 years old; diagnosis of severe or very severe COPD (stage III or IV according to GOLD 2008 criteria); post-bronchodilator FEV 1 < 50%; FEV 1/FVC < 0.70; 1 exacerbation in the previous 12 months requiring systemic corticosteroids or antibiotics; smoking history 10 pack years Allowed: salbutamol, stable dose of ICS Not allowed: long-acting bronchodilators SHINE (Bateman 2013) Placebo Tiotropium; 18μg; OD QVA149 (110 µg indacaterol/50 µg glycopyrronium); OD 26 weeks >= 40 years old; diagnosis of moderate or severe COPD (stage II or III according to GOLD 2008 criteria); post-bronchodilator FEV 1 < 80% and 30%. Postbronchodilator FEV 1/FVC < Smoking history 10 pack years Allowed: Salbutamol/albuterol as rescue medication, inhaled or intranasal corticosteroids in constant doses Not allowed: LABA, LAMA, LABA/ICS Vogelmeier 2008 Formoterol; 10µg; BID + Tiotropium; 18µg; OD Tiotropium; 18µg; OD Placebo 24 weeks Stable COPD, aged 40 years at COPD onset and with a smoking history of 10 pack-years, forced expiratory volume in 1 second (FEV1)<70% of patient s predicted normal value (and 1.00 L), and FEV1/forced vital capacity (FVC)<70%. Allowed: Salbutamol, ICS monotherapy Tashkin 2009 Tiotropium 18µg OD + Formoterol 12µg BID Tiotropium bromide 18µg OD + Placebo BID 12 weeks aged 40 years; post-bronchodilator FEV1 <70% and>30% predicted normal or >0.75 L, whichever was less, at run-in; FEV1/FVC<0.70 Continued use of prior stable ICS regimens and systemic corticosteroids for the treatment of exacerbations was permitted throughout the study. All patients were provided with albuterol inhalers for use as rescue medication GK14290 version 19 November 13 th, 2014 Page 31 of 178

33 Study Treatment Trial Duration Inclusion criteria Background treatment Chan 2007 BI trial: Tiotropium 18µg; OD Placebo 48 weeks 40 years old; 10 pack-years; FEV1 65%; FEV1/FVC 70%; included if 1 exacerbation previous yr but not in 6 wks prior (later amended to incl 1 exacerbation in past 2 yrs) Allowed: Stable dose oral corticosteroids, ICS, theophylline preparations, mucolytic preparations (not containing bronchodilators), LABAs TIPHON (Tonnel 2008) Tiotropium 18µg; OD Placebo 36 weeks 40 years old; >10 pack-years; FEV %;FEV1/SVC 70%; Allowed: Stable doses of theophylline preparations (excluding 24-hour preparations), mucolytics, ICS, and oral steroids UPLIFT Tiotropium 18µg; OD Placebo 4 years 40 years old; >10 pack-years; FEV1 70%;FEV1/FVC 70%;excluded if exacerbation 4 wks prior Allowed: All respiratory medications, except other inhaled anticholinergic drugs (Tashkin 2008 and Celli 2009) Niewoehner 2005 Tiotropium 18µg; OD Placebo 6 months 40 years old; 10 pack-years; FEV1 60%; FEV1/FVC 70%;excluded if not recovered from exacerbation 30 days prior Allowed: All other respiratory medications (including ICS and LABAs) Not Allowed: Open-label anticholinergic bronchodilator GK14290 version 19 November 13 th, 2014 Page 32 of 178

34 Study Treatment Trial Duration Inclusion criteria Background treatment Tiotropium 18µg; OD Placebo 24 weeks >40 years old; >10 pack-years; FEV1 65%; FEV1/FVC 70%; NR Brusasco 2003 Tiotropium 18µg; OD Placebo 24 weeks 40 years old; >10 pack-years; FEV1 60%;FEV1/FVC 70%; Allowed: Usual ICS and oral steroids; Not Allowed: Inhaled anticholinergic LABAs Donohue 2002 Tiotropium 18µg; OD Placebo 56 weeks 40 yearls old; 10 pack years; FEV1 65%;FEV1/FVC 70%; Allowed: stable doses of theophylline, ICS, oral prednisone Casaburi 2002 Donohue 2010 Tiotropium 18µg; OD Placebo 26 weeks Patients aged 40 years or older with a smoking history of 20 pack-years or more and a diagnosis of moderateto-severe COPD (GOLD criteria) were enrolled. Postbronchodilator (within 30 min of inhaling albuterol 360 μg) forced expiratory volume in 1 second (FEV1) <80% and 30% predicted and FEV1/forced vital capacity (FVC) <70%. Patients could continue inhaled corticosteroid (ICS) monotherapy if stable for 1 month before screening; dose and regimen were to remain stable throughout the study. Before the start of the run-in period, treatment with anticholinergic bronchodilators or with β2-agonists was discontinued with appropriate washout, and patients receiving fixed-combination β2-agonist/ics were switched to ICS monotherapy at an equivalent dose. All patients were supplied with albuterol for use as needed. GK14290 version 19 November 13 th, 2014 Page 33 of 178

35 Study Treatment Trial Duration Inclusion criteria Background treatment GLOW 2 (Kerwin 2012) Glycopyrronium 50µg OD Tiotropium 18µg; OD Placebo 52 weeks Males and females 40 years, with a smoking history of 10 pack-yrs, a diagnosis of moderate-to-severe stable COPD, post-bronchodilator FEV1 30% and <80% of the predicted normal, and post-bronchodilator FEV1/FVC<0.70 were enrolled. Allowed: inhaled or intranasal corticosteroids and H1- antagonists; Salbutamol/albuterol as rescue medication. Not allowed: LAMAs (min 7 days before run-in); LABAs or LABA/ICS combinations (min 48h before run-in). Verkindre 2006 Tiotropium 18µg; OD Placebo 12 weeks FEV1 50%;FEV1/SVC 70%;residual volume 125%; excluded if unstable doses oral corticosteroid 6 wks prior Allowed: Stable doses oral corticosteroids, ICS, theophylline preparations, mucolytic agents; Not Allowed: Use of SABAs, oral ß2-agonists, or LABAs Casaburi 2000 Tiotropium 18µg; OD Placebo 13 weeks FEV1 65%;FEV1/FVC 70%; 40 years of age; diagnosis of COPD defined by ATS; smoking history of > 10 pack-years Allowed: stable doses of theophylline, ICS, oral prednisone Not Allowed: Other inhaled or oral bronchodilators Covelli 2005 Tiotropium 18µg; OD Placebo 12 weeks FEV1 60%;FEV1/FVC 70%;excluded if exacerbation in prior 6 wks Allowed: ICS, LABAs and theophyllines Not Allowed: Cromones, leukotriene antagonists, and inhaled anticholinergics GK14290 version 19 November 13 th, 2014 Page 34 of 178

36 Study Treatment Trial Duration Inclusion criteria Background treatment Garcia 2007 Tiotropium 18µg; OD Placebo 12 weeks Ambulatory patients of either sex; >40 years old, diagnosed with COPD (FEV1<60% of the predicted value and FEV1/FVC <70%); smokers or ex-smokers with a history of having smoked at least 10 pack-years NR Moita 2008 Tiotropium 18µg; OD Placebo 12 weeks FEV1 70%;FEV1/FVC 70%;excluded if 3 exacerbations previous year or exacerbation in 6wks prior Allowed: LABAs, theophylline, mucolytics, ICS, stable doses oral corticosteroids. Temporary increases in theophylline or oral steroids for exacerbations; Not Allowed: theophylline 24 h preparation GK14290 version 19 November 13 th, 2014 Page 35 of 178

37 Table 4. Key patient characteristics at baseline for all studies included in the ITC (only arms of interest) Age Severe or Male Current very ITT (sd) smokers severe (%) (%) in years (%) ICS use (%) COPD Duration mean (sd) in years Pack years (sd) FEV 1 % predicted (sd) DB DB DB ZEP INTRUST 1 (Mahler 2012) INTRUST 2 (Mahler 2012) Cazzola 2007 Tiotropium; 18μg; OD Vilanterol 25μg + Umeclidinium 62.5μg Vilanterol 25 μg + Umeclidinium 125μg 62.6 (9.39) NR 41.9 (24.44) 47.8 (13.36) (8.67) NR 44.8 (27.65) 48.0 (12.94) (8.87) NR 43.5 (24.98) 47.2 (12.79) Tiotropium; 18μg; OD (8.3) NR 54.0 (31.59) 47.4 (13.10) Vilanterol 25μg + Umeclidinium 62.5μg Vilanterol 25 μg + Umeclidinium 125μg (8.62) NR 47.8 (26.13) 47.7 (13.55) (8.51) NR 46.9 (24.90) 47.1 (12.88) Placebo (9.04) NR 47.2 (27.21) 46.7 (12.71) Vilanterol 25µg + Umeclidinium 62.5µg (8.71) NR 46.5 (25.80) 47.8 (13.19) Tiotropium; 18μg; OD (8.50) NR 44.4 (25.03) 46.5 (12.76) Vilanterol 25μg + Umeclidinium 62.5μg (8.41) NR 44.1 (24.44) 46.2 (13.02) Tiotropium; 18μg; OD (9.22) (6.45) 47.2 (26.58) 48.9 (11.46) Indacaterol; 150μg; OD + Tiotropium; 18μg; OD (9.07) (6.12) 47.2 (25.86) 48.3 (9.70) Tiotropium; 18μg; OD (8.98) (6.26) 46.3 (24.64) 48.6 (9.76) Indacaterol; 150μg; OD + Tiotropium; 18μg; OD (8.83) (6.48) 46.2 (25.52) 48.6 (9.74) Tiotropium; 18μg; OD a 66.1 (NR) c 0 d NR 50.7 (NR) 38.5 (NR) Salmeterol/Fluticasone; 50/500μg; BID b 66.9 (NR) c 0 d NR 46.9 (NR) 39.0 (NR) GK14290 version 19 November 13 th, 2014 Page 36 of 178

38 Aaron 2007 ENLIGHTEN (Dahl 2013) SPARK (Wedzicha 2013) SHINE (Bateman 2013) Vogelmeier 2008 Tiotropium; 18μg; OD ITT Male (%) Age (sd) in years Current smokers (%) Severe or very severe (%) ICS use (%) COPD Duration mean (sd) in years Pack years (sd) FEV 1 % predicted (sd) Tiotropium; 18μg; OD (8.9) 27 NR (8.8) e 51.8 (28.0) 42.1 (13.5) Tiotropium 18µg OD + salmeterol 25µg 2puffs BID Salmeterol/Fluticasone; 25/250μg/puff; two puffs BID + Tiotropium; 18μg; OD (8.2) 24.3 NR 34.9 NR 48.7 (27.1) 41.2 (13.0) (8.9) 32 NR (8.1) 50.3 (23.1) 42.2 (12.2) Placebo 113 f (8.14) (5.1) 38.1 (15.93) QVA149 (110 µg indacaterol/50 µg glycopyrronium); OD 226 f (8.81) (5.74) 36.3 (16.01) (12.5) (13.27) Tiotropium 18µg; OD 742 f (7.8) c (5.5) 47 (28) 37.4 (8.1) QVA149 (110 µg indacaterol/50 µg glycopyrronium); OD 741 f (8.1) c (5.8) 45 (23) 37.0 (8.1) Placebo 234 f (8.6) (5.7) NR 55.2 (12.7) Tiotropium 18µg; OD 483 f (8.7) (5.5) NR 55.1 (13.5) QVA149 (110 µg indacaterol/50 µg glycopyrronium); OD 475 f (8.9) (5.5) NR 55.7 (13.2) Tiotropium; 18µg; OD (9.5) NR NR NR 6.9 (6.3) 38.6 (19.3) 51.6 (11.2) Formoterol; 10µg; BID + Tiotropium; 18µg; OD (8.8) NR NR NR 7.2 (7.0) 37.9 (18.2) 50.4 (10.5) Placebo (8.6) NR NR NR 6.7 (6.1) 40.1 (22.8) 51.1 (11) GK14290 version 19 November 13 th, 2014 Page 37 of 178

39 Tashkin 2009 Chan 2007 BI trial: TIPHON (Tonnel 2008) UPLIFT (Tashkin 2008 and Celli 2009) Niewoehner 2005 Brusasco 2003 Donohue 2002 Casaburi 2002 Tiotropium 18µg OD + Formoterol 12µg BID Tiotropium bromide 18µg OD + Placebo BID ITT Male (%) Age (sd) in years Current smokers (%) Severe or very severe (%) ICS use (%) COPD Duration mean (sd) in years Pack years (sd) FEV 1 % predicted (sd) (8.7) 49 NR 27 NR NR NR (8.5) 46 NR 27 NR NR NR Tiotropium; 18µg; OD (8.7) 32 NR (8.1) 50.2 (22.6) 0.39 (0.13) Placebo (9.1) 30 NR (7.9) 51.0 (26.3) 0.39 (0.14) Tiotropium; 18µg: OD (9.7) (7.6) 44.4 (21.3) 0.47 (0.13) Placebo (10.1) (7.9) 43.0 (22.5) 0.46 (0.12) Tiotropium; 18µg; OD (8.4) (7.6) 49.0 (28.0) 0.40 (0.12) Placebo (8.5) (7.4) 48.4 (27.9) 0.39 (0.12) Tiotropium; 18µg; OD (8.7) 29 NR (10.4) 67.4 (35.4) 0.36 (0.13) Placebo (8.5) 30 NR (10.5) 69.4 (36.6) 0.36 (0.13) Tiotropium; 18µg; OD (8.0) NR NR NR 9.0 (7.3) 44.1 (22.9) 0.39 (0.12) Placebo (8.6) NR NR NR 9.8 (7.4) 42.4 (22.7) 0.39 (0.12) Tiotropium; 18µg; OD (7.9) NR NR (7.8) 47.0 (25.0) 0.41 (NR) Placebo (7.8) NR NR (7.9) 46.0 (24.0) 0.41 (NR) Tiotropium; 18µg; OD (9.0) NR NR (7.4) 63.0 (31.0) 0.39 (0.14) Placebo (9.0) NR NR (6.8) 59.0 (30.0) 0.38 (0.14) Donohue 2010 Tiotropium; 18µg; OD (8.8) NR NR 35 NR 50.0 (25.1) 0.54 (0.16) GK14290 version 19 November 13 th, 2014 Page 38 of 178

40 ITT Male (%) Age (sd) in years Current smokers (%) Severe or very severe (%) ICS use (%) COPD Duration mean (sd) in years Pack years (sd) FEV 1 % predicted (sd) Placebo (8.9) NR NR 40 NR 49.7 (23.9) 0.56 (0.14) GLOW2 (Kerwin 2012) Verkindre 2006 Casaburi 2000 Covelli 2005 Garcia 2007 Moita 2008 Tiotropium 18µg OD (8.2) 44 NR (6.6) 50.2 (28.0) 0.56 (0.13) Placebo (9.1) 46 NR (6.6) 48.0 (24.0) 0.56 (0.14) Tiotropium; 18µg; OD (9.5) 24 NR NR 9.7 (6.9) 45.6 (23.1) 0.35 (0.09) Placebo (10.2) 33 NR NR 8.8 (6.6) 41.8 (18.0) 0.36 (0.09) Tiotropium; 18µg; OD (8.6) NR NR NR 9.3 (8.0) 64.5 (33.1) 0.39 (0.14) Placebo (9.0) NR NR NR 8.6 (6.9) 60.5 (30.2) 0.38 (0.14) Tiotropium; 18µg; OD (8.9) 40 NR (8.1) 66 (35.6) 0.40 (0.13) Placebo (9.2) 37 NR (7.7) 65 (31.2) 0.39 (0.14) Tiotropium; 18µg; OD 123 NR NR NR NR NR NR NR NR Placebo 125 NR NR NR NR NR NR NR NR Tiotropium; 18µg; OD 147 NR NR 28 NR NR NR NR NR Placebo 164 NR NR 25 NR NR NR NR NR a calculated: 28 / 30 = 93% males; b calculated: 26 / 30 = 87% males c Severe or very severe COPD was an inclusion criterion. d ICS use had to be stopped at run-in, 23 patients in TIO18 and 20 patients in FLUT/SAL+TIO arms used ICS before that; e Duration of reported dyspnoea; f Randomized population GK14290 version 19 November 13 th, 2014 Page 39 of 178

41 5. Frequentist indirect treatment comparison First, the networks and results of the frequentist indirect treatment comparison are presented. The results of the Bayesian NMA will be discussed in section Network of studies Figure 4 to Figure 8 present the network diagrams for the ITC based on the 13 trials. Within the frequentist framework it is not possible to analyze the relative efficacy of UMEC/VI versus all relevant comparators simultaneously within one network. Instead, UMEC/VI will be indirectly compared to each of the relevant comparators by selecting the shortest path possible. By selecting a short path, the uncertainty within the relative efficacy estimates will be reduced as much as possible. An overview of the networks is presented below per comparator of interest. Note that the DB , DB and ZEP could be included in all networks. UMEC/VI 62.5/25 could only be compared indirectly to IND+TIO and to FLUT/SAL+TIO by using TIO 18 as common comparator. This means that there was only one path in these ITCs. However, two common comparators were available for the ITC of UMEC/VI 62.5/25 versus QVA149. An ITC was performed using each common comparator separately, and the results were pooled in a random effects metaanalysis. The study by Aaron et al did not report enough data to be included in the analysis. This study reported data at a 20 week endpoint for trough FEV1 at SGRQ total score, however only mean values were reported in a graph without any measure of uncertainty (i.e. standard error, standard deviation, confidence interval). Although it seems like TIO 18 + FOR 10 and TIO 18 + FOR 12 were pooled in the analysis, this was not the case. Tashkin 2009 (TIO 18 + FOR 12) could only be included in 12 week networks (trough FEV1 and TDI focal score), while Vogelemeier 2008 (TIO 18 + FOR 10) could only be included in SGRQ total score at 24 weeks. Figure 4. Network of studies included in the ITC of UMEC/VI 62.5/25 vs. IND+TIO Figure 5. Network of studies included in the ITC of UMEC/VI 62.5/25 vs. TIO+SAL. Dr. (2007) provided mean values and SD on request. GK14290 version 19 November 13 th, 2014 Page 40 of 178

42 Figure 6. Network of studies included in the ITC of UMEC/VI 62.5/25 vs. TIO+FOR Figure 7. Network of studies included in the ITC of UMEC/VI 62.5/25 vs. FLUT/SAL+TIO Figure 8. Network of studies included in the ITC of UMEC/VI 62.5/25 vs. QVA149 GK14290 version 19 November 13 th, 2014 Page 41 of 178

43 5.2. ITC of UMEC/VI vs. IND + TIO The outcomes of interest for the ITC of UMEC/VI 62.5/25 vs. IND + TIO were the CFB for trough FEV 1, and the mean daily number of puffs rescue medication at 12 weeks. Five studies (DB , DB , ZEP117115, INTRUST1 and INTRUST2) are included in the systematic literature review, reporting these outcomes for the treatments of interest Trough FEV 1 at 12 weeks Based on the available studies, the following network can be formed (Figure 9). Figure 9. Trough FEV 1 at 12 weeks: Network of evidence Individual study results All studies reported the least squares (LS) mean change and the difference in trough FEV 1 versus tiotropium, with the corresponding 95% CI. As tiotropium is the reference treatment and will be used as the anchor of the ITC, the differences in CFB were reversed for all trials and are reported in Table 5 and Table 6. The results are suggesting that the UMEC/VI 62.5/25 fixed combination and indacaterol + tiotropium open combination are more efficacious in improving trough FEV 1 at 12 weeks than tiotropium as monotherapy. Table 5. Least squares (LS) mean difference for trough FEV 1 (in ml) at 12 weeks (Day 84) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB ZEP Data source: DB , Table 6.05 & DB , Table 6.05 & ZEP117115, Table 28. GK14290 version 19 November 13 th, 2014 Page 42 of 178

44 Table 6. Difference for trough FEV 1 (ml) at 12 weeks. Difference LS mean TIO18 vs. IND + TIO18 95% CI low 95% CI high INTRUST INTRUST Data source: Mahler 2012, p Meta-analysis results The individual study results presented were synthesized by means of a random effects meta-analysis. As expected, the findings of the separate meta-analyses were in line with the findings of the individual trials, and show that UMEC/VI 62.5/25 and indacaterol + tiotropium are more efficacious than tiotropium as monotherapy, by means of trough FEV 1 at 12 weeks. The studies synthesized by each meta-analysis are highly homogeneous, as their patient characteristics as well as their study design are similar. Table 7. TIO18 vs. IND150 + TIO18 meta-analysis results, trough FEV 1 at 12 weeks Effect Study [95% Conf. Interval] % Weight Size INTRUST INTRUST Pooled effect size Heterogeneity chi-squared = 0.37, (df=1), p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 9.27, p < GK14290 version 19 November 13 th, 2014 Page 43 of 178

45 Study % ID ES (95% CI) Weight INTRUST ( , ) INTRUST (-90.00, ) Overall (I-squared = 0.0%, p = 0.540) (-89.52, ) NOTE: Weights are from random effects analysis Figure 10. Forest plot of TIO18 vs. IND150 + TIO18 meta-analysis results, trough FEV 1 at 12 weeks Table 8. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 at 12 weeks Study Effect Size [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 1.03 (df. = 2) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 8.71 p < GK14290 version 19 November 13 th, 2014 Page 44 of 178

46 Study % ID ES (95% CI) Weight DB ( , ) DB ( , ) ZEP ( , ) Overall (I-squared = 0.0%, p = 0.596) ( , ) NOTE: Weights are from random effects analysis Figure 11. Forest plot of TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 at 12 weeks A retrospective statistical power analysis of the meta-analyses presented in this paragraph suggested a power of >99% in estimating a MCID of 100ml ITC Results Using eq. 5, the variance for each meta-analysis was calculated and subsequently, using eq. 4 the variance, standard error and 95% confidence intervals for the ITC between UMEC/VI 62.5/25 and indacaterol + tiotropium was estimated. UMEC/VI 62.5/25mcg demonstrated comparable but numerically higher CFB versus indacaterol + tiotropium, with a difference in CFB of 25.10ml (95%CI: , 52.30, p=0.071) (Table 9). The difference was not statistically significant and was lower than the minimum clinically important difference (MCID) of trough FEV 1 (100mL). Table 9. UMEC/VI vs. Indacaterol + Tiotropium ITC results, trough FEV 1 at 12 weeks Mean Difference variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. IND + TIO A retrospective calculation of the ITC power suggested a statistical power of >99% in estimating a MCID of 100ml for the comparison between UMEC/VI 62.5/25 vs. IND + TIO. GK14290 version 19 November 13 th, 2014 Page 45 of 178

47 Rescue medication at 12 weeks Five studies (DB , DB , ZEP117115, INTRUST1 and INTRUST2) reported data on rescue medication use at 12 weeks and could be included in the ITC of UMEC/VI 62.5/25 vs. IND + TIO. Based on the available studies, the following network can be formed (Figure 12). Figure 12. Rescue medication use at 12 weeks: Network of evidence Individual study results Table 10 and Table 11 report the reversed difference in change from baseline (LS mean) rescue medication use between the treatments of interest and tiotropium 18µg. The results are suggesting that UMEC/VI 62.5/25 and indacaterol + tiotropium are more efficacious in decreasing the number of puffs rescue medication per day. Table 10. Difference in rescue medication use (number of puffs) at 12 weeks (vs. TIO) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB ZEP Data source: DB Week 1-12, Table 28, p.1, DB Week 1-12, Table 27, and ZEP Week 1-12, Table 4, all from an additional analyses provided by GSK Table 11. Difference in rescue medication use (number of puffs) at 12 weeks (vs. TIO) Difference TIO18 vs. IND + TIO18 LS mean 95% CI low 95% CI high INTRUST a 1.5 a INTRUST Data source: Mahler 2012, Table 3 p. 786 a: Mahler et al. (2012) reports a difference between treatments in LS mean change from baseline salbutamol use during treatment (puffs/day) for INTRUST1 of -1.1 (95%CI: -0.8; -1.5). It was assumed that the lower and upper bound of the GK14290 version 19 November 13 th, 2014 Page 46 of 178

48 confidence interval should have been reported the other way around. In this report, the lower bound of the 95%CI is represented by the lowest value reported (-1.5) and the upper bound by the highest value reported (-0.8) Meta-analysis results The individual study results presented were synthesized by means of a random effects meta-analysis (Table 12 and Table 13). As expected, the findings of the separate meta-analyses were in line with the findings of the individual trials, and show that UMEC/VI 62.5/25 and indacaterol + tiotropium are more efficacious than tiotropium as monotherapy, by means of rescue medication use at 12 weeks. Table 12. TIO18 vs. IND150 + TIO18 meta-analysis results, rescue medication use (puffs/day) at 12 weeks Study ES [95% Conf. Interval] % Weight INTRUST INTRUST Pooled effect size Heterogeneity chi-squared = 2.89 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) = 65.4% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 4.45 p < Study % ID ES (95% CI) Weight INTRUST (0.80, 1.50) INTRUST (0.40, 1.00) Overall (I-squared = 65.4%, p = 0.089) 0.89 (0.50, 1.28) NOTE: Weights are from random effects analysis Figure 13. Forest plot of TIO18 vs. IND + TIO meta-analysis results, rescue medication use at 12 weeks GK14290 version 19 November 13 th, 2014 Page 47 of 178

49 Table 13. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, rescue medication use (puffs/day) at 12 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 0.78 (d.f. = 2) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 5.70 p < Study % ID ES (95% CI) Weight DB (-0.11, 1.14) 8.41 DB (0.19, 1.37) 9.44 ZEP (0.30, 0.70) Overall (I-squared = 0.0%, p = 0.677) 0.53 (0.35, 0.71) NOTE: Weights are from random effects analysis Figure 14. Forest plot of TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, rescue medication use at 12 weeks ITC Results The variance for each meta-analysis was calculated and subsequently the standard error and 95% confidence intervals for the ITC between UMEC/VI 62.5/25 and indacaterol + tiotropium were estimated (Eq 4 and 5). UMEC/VI demonstrated comparable results versus indacaterol + tiotropium by means of rescue medication use (number of puffs per day) at 12 weeks. UMEC/VI 62.5/25 resulted in a difference in CFB of 0.36 puffs/day (95%CI: -0.07, 0.79, p=0.100) versus indacaterol + tiotropium (Table 14). GK14290 version 19 November 13 th, 2014 Page 48 of 178

50 Table 14. UMEC/VI vs. Indacaterol + Tiotropium ITC results, rescue medication use (puffs/day) at 12 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. IND + TIO GK14290 version 19 November 13 th, 2014 Page 49 of 178

51 5.3. ITC of UMEC/VI vs. TIO + SAL Only one study compared the open dual combination of TIO 18 and SAL 50 with TIO 18. This study, Aaron 2007, was a double blind, double dummy RCT, therefore patients in the TIO 18 arm were also provided with a placebo inhaler to be used twice daily (similar to treatment with SAL 50). The TIO 18 + PBO arm was assumed to be similar enough to the TIO 18 treatment arms of DB , DB and ZEP Aaron 2007 compared TIO 18 + SAL 50 with TIO 18. This study measured trough FEV1, SGRQ total score and TDI focal score. However, not enough data was provided on 20 week outcomes in the publication to perform an ITC. Therefore, the authors of the publication were contacted, and Dr. provided us with the mean and standard deviation (SD) of trough FEV1, SGRQ total score and TDI focal score at each visit, including the 20 week visit. Endpoints at 20 week data were similar enough to 24 week data to be included in an ITC. Aaron et al (2007) did not report rescue medication use, therefore no indirect comparison of UMEC/VI 62.5/25 versus TIO + SAL was possible for this endpoint Trough FEV1 at 24 weeks Based on the available studies, the following network could be formed. An overview of the network is presented below. Figure 15. UMEC/VI 62.5/25 vs. TIO + SAL; network of studies Individual study results DB , DB and ZEP presented the LS mean difference between UMEC/VI 62.5/25 and TIO 18 in trough FEV1 at 24 weeks. The results of Aaron 2007 were kindly provided by the first author of the paper (Dr. The mean trough FEV1 at 20 weeks and standard deviation were provided, together with the trough FEV1 and standard deviation at baseline. Based on the mean trough FEV1 at 20 weeks and the baseline value, we were able to calculate the CFB and the difference in CFB between TIO + SAL and TIO. However, since the baseline and 20 week values are expected to be correlated (they are measured in the same patient) it was not possible to calculate a standard error of the change from baseline using this data. Therefore, we applied the methods proposed by Cochrane (Cochrane review by Karner et al, ). We calculated a SE for the DCFB at 52 weeks and assumed this value was relatively constant over the study period, i.e. that the SE of the DCFB at 52 weeks would be similar with the SE of the DCFB at 20 weeks. GK14290 version 19 November 13 th, 2014 Page 50 of 178

52 Therefore, we calculated the SE of the DCFB at 52 weeks, using the data reported in the publication and provided by the authors. The results are presented below. The P-value of the difference in trough FEV1 at 52 weeks of TIO + SAL versus TIO + PBO was 0.87, as presented below Figure 3 in the publication. The DCFB was calculated using the data provided by Dr. and was ml. The number of patients in the TIO + SAL arm was 148 and the number of patients in the TIO arm was 156. We used the following formula to estimate the SE, and the result was 45.46: DCFB SE DCFB = p value TINV(, (N N 2 2) Table 15. Difference for trough FEV1 (ml) at 24 weeks UMEC/VI 62.5/25 versus TIO 18 Difference UMEC/VI 62.5/25 vs. TIO 18 (ml) Author or study acronym LS Mean 95% CI Low 95% CI High Source CSR - DB Day 169, table 29, p. 83 CSR - DB Day 169, table 29, p. 85 CSR - ZEP Day 169, Table 27, p. 68 Table 16. Difference for trough FEV1 (ml) at 20 weeks TIO18 + SAL 50 versus TIO 18 Trough FEV1 at 20 weeks CFB trough FEV1 at 20 weeks Difference TIO +SAL vs. TIO Mean SD SE Mean SE Mean 95% CI 95% CI low high Aaron TIO 18 + SAL * * 2007 TIO Calculated using the imputation method for the SE as described before. The SE was estimated at The 95%CI was estimated as mean ± 1.96 * SE Meta-analysis results A random effects meta-analysis was performed in STATA to estimate the pooled treatment effect of UMEC/VI 62.5/25 versus TIO 18. The results are presented below and indicate that UMEC/VI 62.5/25 is more efficacious than TIO 18 and the effect is statistically significant. No meta-analysis was needed for the comparison of TIO + SAL versus TIO, because only one study reported data for this comparison on this endpoint. GK14290 version 19 November 13 th, 2014 Page 51 of 178

53 Table 17. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 at 20 and 24 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 3.07 (df. = 2) p = I-squared (variation in ES attributable to heterogeneity) = 34.9% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 5.89 p < Study % ID ES (95% CI) Weight CSR - DB (10.00, ) CSR - DB (39.00, ) CSR - ZEP (81.00, ) Overall (I-squared = 34.9%, p = 0.215) (61.52, ) NOTE: Weights are from random effects analysis Figure 16. Forest plot of UMEC/VI 62.5/25 versus TIO meta-analysis results, trough FEV1 at 24 weeks ITC results The results of the random effects meta-analysis of UMEC/VI 62.5/25 versus TIO 18 and of the study by Aaron et al of TIO + SAL versus TIO were synthesized in an ITC. The results are presented below and show that UMEC/VI 62.5/25 is expected to be similar efficacious as TIO + SAL, although the results were numerically in favor of UMEC/VI 62.5/25. The mean difference in trough FEV1 was ml (95%CI: , , p=0.125) but these results were below the MCID for trough FEV1 and not statistically significant. GK14290 version 19 November 13 th, 2014 Page 52 of 178

54 Table 18. UMEC/VI vs. TIO + SAL ITC results, FEV1 total score at 24 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. TIO + SAL GK14290 version 19 November 13 th, 2014 Page 53 of 178

55 SGRQ total score at 24 weeks After Dr. provided us with the mean SGRQ total score at the 20 week visit and a standard deviation of this mean, the following network could be formed. Figure 17. UMEC/VI 62.5/25 vs. TIO + SAL; network of studies Individual study results Except for the Aaron 2007 study, all studies reported the (LS mean) difference in CFB for SGRQ total score at 24 weeks versus tiotropium. Aaron reported mean values in a graph and was kind to provide us with the mean values at 20 week and a standard deviation. Calculations were performed as described in section Table 19. Difference for SGRQ total score at 24 weeks UMEC/VI 62.5/25 versus TIO 18 Author or study acronym Difference UMEC/VI 62.5/25 versus TIO 18 LS Mean 95% CI Low 95% CI High Source CSR - DB Table 50, p. 125 CSR - DB Table 51, p. 126 CSR - ZEP Table 45, p. 99 Table 20. Difference for SGRQ total score at 20 weeks TIO18 + SAL 50 versus TIO 18 SGRQ total score at 20 weeks CFB SGRQ total score at 20 weeks Difference TIO +SAL vs. TIO Mean SD SE Mean SE Mean 95% CI 95% CI low high Aaron TIO 18 + SAL TIO DCFB SE DCFB = p value TINV(, (N N 2 2) GK14290 version 19 November 13 th, 2014 Page 54 of 178

56 DCFB at 52 weeks was reported to be -1.8 (calculated using data on p. 551) and the p-value of this difference was With 156 patients in the TIO arm and 148 in the TIO + SAL arm, the estimated SE at 52 weeks was This was assumed to be similar to 20 weeks and imputed for this endpoint. The 95%CI was calculated using formula: mean ± 1.96 * SE Meta-analysis results Aaron 2007 was the only study comparing TIO 18 + SAL 50 with TIO 18, therefore no meta-analysis was needed for this comparison. The GSK trials were synthesized in a random effects meta-analysis and the results are presented below. Table 21. UMEC/VI 62.5/25 versus TIO 18 meta-analysis results, SGRQ total score at 24 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 3.70 (d.f. = 2) p = I-squared (variation in ES attributable to heterogeneity) = 45.9% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 0.93 p = Figure 18. Forest plot of UMEC/VI 62.5/25 versus TIO 18 meta-analysis results, SGRQ total score at 24 weeks GK14290 version 19 November 13 th, 2014 Page 55 of 178

57 ITC results UMEC/VI 62.5/25 showed similar results versus TIO + SAL. The LS mean difference for UMEC/VI 62.5/25mcg versus TIO + SAL was 0.62 units (95%CI: -1.63, 2.87, p=0.590). The difference was not statistically significant and did not exceed the MCID level for the SGRQ total score (-4 units difference). Table 22. UMEC/VI vs. TIO + SAL ITC results, SGRQ total score at 24 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. TIO + SAL GK14290 version 19 November 13 th, 2014 Page 56 of 178

58 TDI focal score at 24 weeks Breathlessness was measured using the TDI focal score in DB , DB and Aaron Although Aaron et al. did not report this outcome, at 20 weeks in their publication, data was kindly provided by the authors on request. As a result, the network of studies was developed as presented below. Figure 19. UMEC/VI 62.5/25 vs. TIO + SAL; network of studies Individual study results DB and DB reported the LS mean difference for UMEC/VI 62.5/25 versus TIO 18, the data is presented below. This was not available for Aaron 2007, however the authors of this paper provided us with the mean TDI focal score at 20 weeks and a corresponding standard deviation. The analyses in the paper were performed using the intention to treat (ITT) principle and therefore the numbers of patients of the ITT population were used to calculate the standard error of the TDI focal score. These values were used to calculate the difference between the treatment arms. Both UMEC/VI 62.5/25 and TIO + SAL are expected to be similar efficacious by means of TDI focal score at 24 weeks. Table 23. LS mean difference for TDI focal score at 24 weeks (vs. TIO) Difference UMEC/VI 62.5/25 versus TIO LS mean 95% CI low 95% CI high DB DB Data source: DB day 168, Table 33, p. 96, and DB day 168, Table 33, p. 97 GK14290 version 19 November 13 th, 2014 Page 57 of 178

59 Table 24. LS mean difference for TDI focal score at 20 weeks TIO + SAL versus TIO TDI focal score at 20 weeks Difference Aaron 2007 TIO 18 + SAL 50 Mean SD SE Mean TIO +SAL vs. TIO 95% CI low TIO Data source: mean TDI and SD at 20 weeks provided by Dr. 95% CI high These values were used to calculate the SE of the TDI score at 20 weeks, and to calculate the difference of TIO+SAL versus TIO Meta-analysis results No meta-analysis was performed for the comparison of TIO + SAL versus TIO. Random effects metaanalysis was performed for the comparison of UMEC/VI 62.5/25 versus TIO 18, the results are presented below. In line with the individual study results, TIO18 resulted in similar TDI scores at 24 weeks compared with UMEC/VI 62.5/25 (Table 85 and Figure 50) Table 25. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, TDI focal score at 24 weeks Study ES [95% Conf. Interval] % Weight DB DB Pooled effect size Heterogeneity chi-squared = 0.41 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 0.12 p = GK14290 version 19 November 13 th, 2014 Page 58 of 178

60 Study % ID ES (95% CI) Weight DB (-0.50, 0.90) DB (-0.70, 0.50) Overall (I-squared = 0.0%, p = 0.524) 0.03 (-0.43, 0.48) NOTE: Weights are from random effects analysis Figure 20. Forest plot of UMEC/VI 62.5/25 versus TIO meta-analysis results, TDI at 24 weeks ITC results Based on the results of the random effects meta-analysis and the Aaron 2007 trial an ITC was performed. The results of the ITC showed that UMEC/VI 62.5/25 is expected to be similar efficacious as the open dual combination of TIO 18 and SAL 50. The mean difference is presented in Table 40 and was 0.45 (95%CI: -0.51, 1.41; p=0.360). This difference was not statistically significant and did not exceed the MCID for TDI focal score. Table 26. UMEC/VI 62.5/25 vs. TIO + SAL ITC results, TDI focal score at 24 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. TIO 18 + SAL GK14290 version 19 November 13 th, 2014 Page 59 of 178

61 5.4. ITC of UMEC/VI vs. TIO + FOR Tashkin 2009 reported the difference in change from baseline trough FEV1 of TIO+FOR compared with TIO at 12 weeks. Additionally, this study measured SGRQ total score, TDI focal score and rescue medication use at 12 weeks. However, no measure of uncertainty was reported for the SGRQ total score and the mean number of puffs rescue medication use. These outcomes were therefore not included in an analysis. For the TDI focal score, the score measured at the final visit was reported together with a 95% confidence interval of the difference. Although this outcome was measured using the last observation carried forward, it was assumed this endpoint was similar enough to be included in the 12 week networks. Vogelmeier 2008 only reported data for SGRQ total score at 24 weeks in a graph. The other outcomes were not reported, or were reported as median values. For the analysis, mean value with a measure of uncertainty are needed, therefore these outcomes were included in the analysis Trough FEV1 at 12 weeks Four trials reported on trough FEV1 at 12 weeks that could be included in the analysis. The network of trials is presented in Figure 21. Figure 21. UMEC/VI 62.5/25 vs. TIO + FOR; network of studies for trough FEV1 at 12 weeks Individual study results The difference in change from baseline (DCFB) and the 95% confidence interval (95% CI) for trough FEV1 at 12 weeks was extracted from all studies of interest. DB , DB and ZEP reported the LS mean difference versus TIO 18 and its 95% confidence interval. Tashkin 2009 reported the mean difference in change from baseline (DCFB) at 12 weeks, with a 95% confidence interval. The results are presented below for each direct comparison in the network. The individual study results show that UMEC/VI and TIO+FOR both were expected to be more efficacious than TIO 18 by means of trough FEV1 at 12 weeks. Table 27. Difference for trough FEV1 (ml) at 12 weeks UMEC/VI 62.5/25 versus TIO 18 Difference UMEC/VI 62.5/25 vs. TIO 18 (ml) GK14290 version 19 November 13 th, 2014 Page 60 of 178

62 Author or study acronym LS Mean 95% CI Low 95% CI High Source CSR - DB Table 30, p. 85 CSR - DB Table 30, p. 87 CSR - ZEP Table 28, p. 69 Table 28. Difference for trough FEV1 (ml) at 12 weeks TIO 18 + FOR 12 versus TIO 18 Difference TIO 18 + FOR 12 vs. TIO 18 (ml) Author or study acronym LS Mean 95% CI Low 95% CI High Source Tashkin Text p. 21 and Figure Meta-analysis results Only one trial reported the DCFB of TIO+FOR versus TIO 18, therefore no meta-analysis was needed for this comparison. On the other hand, three trials reported the DCFB of UMEC/VI 62.5/25 versus TIO 18 and were synthesized in a random effects meta-analysis. The results are presented in Table 29 and Figure 22. Table 29. UMEC/VI 62.5/25 vs. TIO 18 meta-analysis results, trough FEV1 (ml) at 12 weeks Study ES [95% Conf. Interval] % Weight CSR - DB CSR - DB CSR - ZEP Pooled effect size Heterogeneity chi-squared = 1.03 (d.f. = 2) p = I-squared (variation in ES attributable to heterogeneity) = 0.0% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 8.71 p = GK14290 version 19 November 13 th, 2014 Page 61 of 178

63 Figure 22. Forest plot of UMEC/VI 62.5/25 vs. TIO 18 meta-analysis results, trough FEV1 (ml) at 12 weeks ITC results The results of the ITC show that UMEC/VI 62.5/25 is expected to be similar in terms of efficacy as the open dual combination of TIO 18 and FOR 12. The mean difference is presented in Table 30 and was 9.00 ml (95%CI: , 68.34; p=0.766). This difference was not statistically significant nor did it exceed the minimal clinically important difference (MCID) for trough FEV1. Table 30. UMEC/VI 62.5/25 vs. TIO + FOR ITC results, trough FEV 1 (ml) at 12 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. TIO 18 + FOR GK14290 version 19 November 13 th, 2014 Page 62 of 178

64 SGRQ total score at 24 weeks Five trials were identified by GSK that could be included in the ITC of UMEC/VI 62.5/25 versus TIO 18 + FOR 10 by means of SGRQ total score at 24 weeks. Since Vogelmeier 2008 included both a placebo and a TIO 18 arm, both interventions were used as common comparator. The network of trials is presented in Figure 23 Figure 23. UMEC/VI 62.5/25 vs. TIO + FOR; network of studies for SGRQ total score at 24 weeks Individual study results The difference in change from baseline (DCFB) and the 95% confidence interval (95% CI) for trough FEV1 at 12 weeks was extracted from all studies of interest. If needed, data was extracted from figures using DigitizeIt software. Vogelmeier et al. presented the mean DCFB SGRQ total score versus placebo in a graph. These values were extracted for the DCFB of TIO + FOR versus placebo and for TIO versus placebo. The extracted values were used to calculate the mean DCFB of TIO + FOR versus TIO. DB reported the mean DCFB of UMEC/VI 62.5/25 versus placebo, and DB , DB and ZEP reported the mean DCFB of UMEC/VI 62.5/25 versus TIO 18. The results are presented below for each direct comparison in the network. The individual study results show that UMEC/VI and TIO + FOR are both expected to be more efficacious than placebo. Negative scores represent better quality of life on the SGRQ total score. GK14290 version 19 November 13 th, 2014 Page 63 of 178

65 Table 31. Difference for SGRQ total score at 24 weeks UMEC/VI 62.5/25 versus TIO 18 Author or study acronym Difference UMEC/VI 62.5/25 versus TIO 18 LS Mean 95% CI Low 95% CI High Source CSR - DB Table 50, p. 125 CSR - DB Table 51, p. 126 CSR - ZEP Table 45, p. 99 Table 32. Difference for SGRQ total score at 24 weeks UMEC/VI 62.5/25 versus placebo Difference UMEC/VI 62.5/25 vs. PBO Author or study acronym LS Mean 95% CI Low 95% CI High Source CSR - DB Day 168. Table 50, p. 136 Table 33. Difference for SGRQ total score at 24 weeks TIO 18 + FOR 10 versus TIO 18 Author or study acronym Difference TIO 18 + FOR 10 versus TIO 18 LS Mean 95% CI Low Vogelmeier * % CI High DCFB TIO+FOR vs TIO = DCFB TIO+FOR vs PBO DCFB TIO vs PBO Source Calculated using extracted data from Figure 3a using DigitizeIt SE DCFB = SE TIO+FOR vs PBO 2 + SE TIO vs PBO 2 Table 34. Difference for SGRQ total score at 24 weeks TIO 18 + FOR 10 versus placebo Difference TIO 18 + FOR 10 versus placebo Author or study acronym LS Mean 95% CI Low 95% CI High Source Vogelmeier Figure 3a using DigitizeIt Meta-analysis results The trials comparing UMEC/VI 62.5/25 with TIO 18 were synthesized in a random effects metaanalysis and are presented in Figure 24 and Table 35. The results indicate that UMEC/VI 62.5/25 is expected to be numerically more efficacious than TIO 18 by means of SGRQ total score at 24 weeks, although the result was not statistically significant. No meta-analysis was needed for the other direct comparisons in the network, since only one trial reported data for every one of them. GK14290 version 19 November 13 th, 2014 Page 64 of 178

66 Table 35. UMEC/VI 62.5/25 versus TIO 18 meta-analysis results, SGRQ total score at 24 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 3.70 (d.f. = 2) p = I-squared (variation in ES attributable to heterogeneity) = 45.9% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 0.93 p = Figure 24. Forest plot of UMEC/VI 62.5/25 versus TIO 18 meta-analysis results, SGRQ total score at 24 weeks ITC results First, the ITC was performed for UMEC/VI 62.5/25 versus TIO + FOR using TIO 18 as common comparator, then the ITC was repeated using placebo as common comparator. The results of both ITCs are presented in Table 36. Heterogeneity was detected in the results; the result of the ITC was different depending on the common comparator used. However, since the 95% confidence intervals of both ITC overlapped, the results could be synthesized in a random effects meta-analysis. This analysis shows that UMEC/VI 62.5/25 is expected to be similar efficacious as TIO 18 + FOR 10 by means of SGRQ total score at 24 weeks, with a mean difference of (95%CI: -4.10, 1.20; p=0.284) (Table 36 and Figure 25). These results should be interpreted with caution due to statistical GK14290 version 19 November 13 th, 2014 Page 65 of 178

67 heterogeneity in the random effects meta-analysis. The I 2 statistic showed signs of heterogeneity. The results were below the MCID for SGRQ total score and not statistically significant. Table 36. UMEC/VI 62.5/25 vs. TIO + FOR ITC results, SGRQ total score at 24 weeks MD variance se [95% Conf. Interval] p- value TIO as common comparator PBO as common comparator UMEC/VI 62.5/25 vs. TIO 18 + FOR Heterogeneity chi-squared = 0.91 (d.f. = 1) p = I-squared (variation in ES attributable to heterogeneity) = 0.0% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 1.07 p = Figure 25. UMEC/VI 62.5/25 vs. TIO + FOR ITC results, SGRQ total score at 24 weeks GK14290 version 19 November 13 th, 2014 Page 66 of 178

68 TDI focal score The TDI focal score was reported in one study including an open dual combination of a LABA and a LAMA, Tashkin Tashkin et al. compared TIO 18 + FOR 12 to TIO 18, and as a result the latter was the only available common comparator with UMEC/VI 62.5/25. Two trials reported the DCFB of UMEC/VI 62.5/25 versus TIO 18 and were also included in the network (Figure 26). Figure 26. UMEC/VI 62.5/25 vs. TIO + FOR; network of studies for TDI focal score at 12 weeks Individual study results The difference in change from baseline (DCFB) and the 95% confidence interval (95% CI) for TDI focal score at 12 weeks was extracted from all studies of interest. DB , DB and Tashkin 2009 all reported the mean difference between the combination product and TIO 18, along with the 95% confidence interval. The results are presented below for each direct comparison in the network. The individual study results show that UMEC/VI and TIO+FOR both are expected to be similar efficacious with TIO 18 by means of TDI focal score at 12 weeks. Table 37. Difference for TDI focal score at 12 weeks UMEC/VI 62.5/25 versus TIO 18 Difference UMEC/VI 62.5/25 vs. TIO 18 (ml) Author or study acronym LS Mean 95% CI Low 95% CI High Source CSR - DB Day 84. Table 33, p. 95 CSR - DB Day 84. Table 33, p. 97 Table 38. Difference for TDI focal score at 12 weeks TIO 18 + FOR 12 versus TIO 18 Difference TIO 18 + FOR 12 vs. TIO 18 (ml) Author or study acronym LS Mean 95% CI Low 95% CI High Source Tashkin Table2 p23 and text p Meta-analysis results No meta-analysis was needed for the comparison of TIO+FOR versus TIO because only one trial reported on this outcome. Two trials reported the DCFB of UMEC/VI 62.5/25 versus TIO 18 and were synthesized in a random effects meta-analysis. The results are presented in Table 39. GK14290 version 19 November 13 th, 2014 Page 67 of 178

69 The results of the meta-analyses confirmed the results from the individual studies and showed that UMEC/VI 62.5/25 is expected to be similar efficacious as TIO 18 by means of trough FEV1 at 12 weeks. Table 39. UMEC/VI 62.5/25 vs. TIO 18 meta-analysis results, TDI focal score at 12 weeks Study ES [95% Conf. Interval] % Weight CSR - DB CSR - DB Pooled effect size Heterogeneity chi-squared = 1.57 (d.f. = 1) p = I-squared (variation in ES attributable to heterogeneity) = 36.5% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 1.70 p = Figure 27. Forest plot of UMEC/VI 62.5/25 vs. TIO 18 meta-analysis results, TDI focal score at 12 weeks ITC results Based on the results of the random effects meta-analysis and the Tashkin 2009 trial an ITC was performed. The results of the ITC showed that UMEC/VI 62.5/25 is expected to be similar efficacious as the open dual combination of TIO 18 and FOR 12, although the results are numerically in favour of UMEC/VI 62.5/25. The mean difference is presented in Table 40 and was 0.35 (95%CI: -0.55, 1.25; GK14290 version 19 November 13 th, 2014 Page 68 of 178

70 p=0.444). This difference was not statistically significant and did not exceed the MCID for trough FEV1. Table 40. UMEC/VI 62.5/25 vs. TIO + FOR ITC results, TDI focal score at 12 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. TIO 18 + FOR GK14290 version 19 November 13 th, 2014 Page 69 of 178

71 5.5. ITC of UMEC/VI vs. FLUT/SAL 500/50 + TIO18 DB , DB and ZEP reported CFB for trough FEV 1 at 12 and 24 weeks, Cazzola 2007 reported at 12 weeks and Aaron 2007 at 20 weeks. An overview of the study design is presented in Table 3. All studies are randomized, parallel, double-blind and multi-center. Table 4 provides an overview of the main patient characteristics Trough FEV 1 at 12 weeks Based on the available studies, the following network can be formed (Figure 28). Trough FEV 1 at 12 weeks Figure 28. Trough FEV 1 at 12 weeks: Network of evidence Individual study results DB , DB and ZEP reported the LS mean change and LS mean difference versus tiotropium with the corresponding 95% CI, these results were presented in Table 5. Cazzola 2007 only reported the CFB for each study arm, along with a 95% CI. These numbers were used to calculate the DCFB versus tiotropium (Table 41). As tiotropium is the common comparator and will be used as the anchor of the ITC, the differences in CFB were reversed and are also reported in Table 41. The results are suggesting that both UMEC/VI fixed combinations and indacaterol + tiotropium open combination are more efficacious than tiotropium as monotherapy, by means of trough FEV 1 at 12 weeks. Table 41. Difference in CFB for trough FEV 1 (ml) at 12 weeks. Mean CFB 95% CI low 95% CI high DCFB TIO18 vs. FLUT/SAL 500/50 + TIO18 Mean 95% CI low 95% CI high Cazzola 2007 TIO FLUT/SAL 500/50 + TIO Data source: Cazzola 2007, p CFB was used to calculate DCFB using the following formulas. SE of difference = (SE of CFB Treatment A) 2 + (SEof CFB Treatment B) 2 95% CI of difference = DCFB / SE to DCFB GK14290 version 19 November 13 th, 2014 Page 70 of 178

72 Meta-analysis results Since only one study compared TIO18 to FLUT/SAL 500/50 + TIO18, no meta-analysis could be performed. The results of the random effects meta-analysis for DB , DB and ZEP are presented in Table 8 and Figure 11. As expected, the results of the meta-analysis confirm the individual trial results suggesting that UMEC/VI 62.5/25 is more efficacious than tiotropium as monotherapy, by means of trough FEV 1 at 12 weeks. The studies synthesized by the metaanalysis are highly homogeneous, their patient characteristics and study design are similar ITC Results Using eq. 4 and 5, the variance for each meta-analysis was calculated and subsequently the standard error and 95% confidence intervals for the ITC between UMEC/VI and FLUT/SAL+TIO were estimated. UMEC/VI 62.5/25mcg demonstrated a statistically significant higher CFB versus FLUT/SAL+TIO, with a difference in CFB of 54.00ml (95%CI: 12.80; 95.19, p=0.010). (Table 42) The difference was lower than the MCID level of trough FEV 1 (100ml). Table 42. UMEC/VI vs. FLUT/SAL 500/50 mcg BID + TIO18 ITC results, trough FEV 1 at 12 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs FLUT/SAL+TIO A retrospective calculation of the ITC power suggested the same statistical power of 26.8% in estimating a difference of 100ml for the ITC of UMEC/VI 62.5/25 vs. FLUT/SAL+TIO. GK14290 version 19 November 13 th, 2014 Page 71 of 178

73 Trough FEV 1 at 24 weeks Aaron 2007 was included in the systematic literature review comparing TIO 18 + FLUT/SAL 250/25 mcg/puff, two puffs twice daily and TIO18 alone. This study reported CFB trough FEV 1 at 52 weeks, as well as a graph of the mean FEV 1 trough at 4, 20, 36 and 52 weeks (text p.551 and Figure 3 in publication). This graph did not report a standard deviation, standard error or confidence interval and could therefore not be used for this ITC. However, a Cochrane review by Karner et al. reported having FEV 1 data at 5 months, received from the authors. In a forest plot, Karner et al report the mean difference between TIO + FLUT/SAL vs. TIO at 5 months, along with a 1 year SE (Analysis 1.7, p.31 of the publication). As it is not clear if the 1 year SE is the right measure of uncertainty for the mean CFB reported, these results should be interpreted with caution. 20 week data from Aaron 2007 were pooled in the network with 24 week data from DB , DB and ZEP to strengthen the networks and be able to perform the ITC. The study and patient characteristics were reported in Table 3 and Table 4. The following network could be created (Figure 29). Figure 29. Trough FEV 1 at 24 weeks: Network of evidence Individual study results Individual study results of the DB , DB and ZEP trials are reported in Table 43. As tiotropium is the common comparator and will be used as the anchor of the ITC, the LS mean differences were reversed. Table 43. LS mean difference for trough FEV 1 (in ml) at 24 weeks (Day 169) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB ZEP Data source: DB , Table 29 & DB , Table 29; ZEP117115: slide deck 4Nov2013, slide 10. GK14290 version 19 November 13 th, 2014 Page 72 of 178

74 Table 44. Difference in CFB for trough FEV 1 (ml) at 20 weeks. DCFB TIO18 vs. FLUT/SAL+TIO LS mean 95% CI low 95% CI high Aaron Data source: Cochrane report (Karner et al, 2011) p Meta-analysis The individual study results of DB , DB and ZEP were synthesized by means of a random effects meta-analysis. As expected, the results of the meta-analysis are confirming the evidence that UMEC/VI 62.5/25 is more efficacious than tiotropium as monotherapy, by means of trough FEV 1 at 24 weeks. Meta-analysis was not needed for the comparison between TIO18 and FLUT/SAL+TIO, as only one study reported data for this comparison. Table 45. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 at 20 and 24 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 3.07 (df. = 2) p = I-squared (variation in ES attributable to heterogeneity) = 34.9% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 5.89 p < GK14290 version 19 November 13 th, 2014 Page 73 of 178

75 Study % ID ES (95% CI) Weight DB ( , ) DB ( , ) ZEP ( , ) Overall (I-squared = 34.9%, p = 0.215) ( , ) NOTE: Weights are from random effects analysis Figure 30. Forest plot of TIO18 vs. UMEC/VI 625/25 meta-analysis results, trough FEV 1 at 20 and 24 weeks ITC Results Equation 5 was used to calculate the variance of the meta-analysis, which was used to calculate the variance, standard error and 95% confidence intervals for the ITC (Equation 4) of UMEC/VI vs. FLUT/SAL+TIO. It should be noted that the comparison of TIO18 vs. FLUT/SAL+TIO was at 20 weeks, and the comparison of UMEC/VI vs. TIO was at 24 weeks. These data were pooled to be able to make an indirect comparison. At 24 weeks no statistically significant difference could be detected between UMEC/VI 62.5/25 and FLUT/SAL+TIO. UMEC/VI 62.5/25mcg demonstrated a difference in CFB versus FLUT/SAL+TIO of 21.17ml (95%CI: ; 88.55, p=0.538). The difference was lower than the MCID level of trough FEV 1 (100ml) (Table 46). Table 46. UMEC/VI vs. FLUT/SAL 500/50 mcg BID + TIO18 ITC results (trough FEV 1 at 20 and 24 weeks) MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. FLUT/SAL+TIO GK14290 version 19 November 13 th, 2014 Page 74 of 178

76 5.6. ITC of UMEC/VI vs. QVA149 The outcomes of interest for the ITC of UMEC/VI 62.5/25 vs. QVA149 were trough FEV 1, SGRQ total score, TDI focal score and the daily mean number of puffs rescue medication at 12 and 24 weeks. Seven studies were identified in the systematic review reporting data on at least one of the outcomes of interest (ENLIGHTEN, SPARK, SHINE, DB , DB , DB and ZEP117115). Details regarding their study design and patient characteristics are presented in Table 3 and Table 4. No network of studies was feasible for the daily number of puffs rescue medication at 12 weeks. The SPARK study (Wedzicha 2013) included patients with COPD stages III and IV with a postbronchodilator FEV 1 < 50% of predicted normal values, who had a documented history of at least one exacerbation in the previous 12 months. This is a more severe population than was included in the GSK trials, comprising of COPD patients with a post-salbutamol FEV 1 70% of predicted normal values, without any restriction to the exacerbation history. To evaluate if inclusion of the SPARK study has any impact on the results of the indirect treatment comparison, two scenarios were performed for each outcome; one including the SPARK study, and one excluding the SPARK study. Regarding the 24 week outcome, the GSK trials (DB , DB , DB and ZEP117115) reported data at 169 days (24 weeks), SPARK reported data at 24 weeks and SHINE (Bateman 2013) and ENLIGHTEN (Dahl 2013) reported data at 26 weeks. These time-points were pooled in the analysis at 24 weeks, to strengthen the networks. GK14290 version 19 November 13 th, 2014 Page 75 of 178

77 Trough FEV 1 at 12 weeks All seven studies reported data on trough FEV 1 at 12 weeks (ENLIGHTEN (Dahl 2013), SPARK (Wedzicha 2013), SHINE (Bateman 2013), DB , DB , DB and ZEP117115). The following network of studies could be formed. Figure 31. Through FEV 1 at 12 weeks: Network of evidence Individual study results DB , DB , DB and ZEP reported the (LS mean) difference in change from baseline in trough FEV 1 versus placebo or tiotropium at 12 weeks. ENLIGHTEN and SPARK reported a graph with LS mean trough FEV 1 values over the study period, and SHINE reported the LS mean trough FEV 1 at 12 weeks for placebo, and the differences between QVA149 and tiotropium versus placebo. These data were used to calculate LS mean differences between QVA149 and tiotropium. Data from graphs was extracted using DigitizeIt. The individual study results show that both UMEC/VI and QVA149 are more efficacious in improving trough FEV 1 at 12 weeks versus placebo and tiotropium. Table 47. LS mean difference for trough FEV 1 at 12 weeks (vs. TIO) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB GK14290 version 19 November 13 th, 2014 Page 76 of 178

78 ZEP Data source: DB Day 84, table 30, p. 85, DB Day 84, table 30, p. 87 and ZEP Day 84, table 28, p. 69 Table 48. LS mean difference for trough FEV 1 at 12 weeks (vs. PBO) Difference PBO vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB Data source: DB Day 84, Table 28, p. 91 Table 49. LS mean difference for trough FEV 1 at 12 weeks in the SPARK and ENLIGHTEN study Baseline LS mean at 12 weeks LS mean difference TIO18/PBO vs. QVA149 Mean SD LS mean SE LS mean 95% CI low 95% CI high ENLIGHTEN PBO a b -100 b QVA a SPARK TIO c -9 b -42 b QVA a 10 a Data source: Wedzicha 2013 (SPARK), Table 1, p. 3 and Figure 3, p. 5, Dahl 2013 (ENLIGHTEH), Figure 3, p. 7 a: Extracted from figure using software DigitizeIt (Enlighten: Figure 3, p. 7; Spark: Figure 3, p. 5) b: SE of LS mean at follow-up was to calculate the SE of the difference and its 95%CI. SE of difference = (SE of LS Mean Treatment A) 2 + (SEof LS mean Treatment B) 2 95% CI of difference = DCFB / SE to DCFB Table 50. LS mean difference for trough FEV 1 at 12 weeks (vs. PBO and TIO) in the SHINE study Difference PBO vs. QVA149 Difference TIO vs. QVA149 LS mean 95% CI low 95% CI High LS mean 95% CI low 95% CI high SHINE Data source: Bateman 2013 (SHINE), Table 2, p. 20. Table 2 reports the LSMean follow-up value for placebo, and the difference of TIO18 or QVA149 and placebo (and 95%CI). These values were used to calculate the difference of TIO vs. QVA149. GK14290 version 19 November 13 th, 2014 Page 77 of 178

79 Meta-analysis The individual study results presented for UMEC/VI 62.5/25 were synthesized by means of a random effects meta-analysis. As expected, the findings of the meta-analysis were in line with the findings of the individual trials, and show that both UMEC/VI and QVA149 are more efficacious in improving trough FEV 1 at 12 weeks than tiotropium as monotherapy or placebo. Please note the effect size for trough FEV 1 at 12 weeks for QVA149 vs. TIO18 is larger in the SHINE trial than in de SPARK trial. Table 51. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 at 12 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 1.03 (df. = 2) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 8.71 p < Study % ID ES (95% CI) Weight DB ( , ) DB ( , ) ZEP ( , ) Overall (I-squared = 0.0%, p = 0.596) ( , ) NOTE: Weights are from random effects analysis Figure 32. Forest plot of TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 at 12 weeks Table 52. TIO18 vs. QVA149 meta-analysis results, trough FEV 1 (ml) at 12 weeks Study ES [95% Conf. Interval] % Weight GK14290 version 19 November 13 th, 2014 Page 78 of 178

80 SPARK SHINE Pooled effect size Heterogeneity chi-squared = 1.98 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) = 49.5% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 5.62 p < Study % ID ES (95% CI) Weight SPARK (-98.00, ) SHINE ( , ) Overall (I-squared = 49.5%, p = 0.159) ( , ) NOTE: Weights are from random effects analysis Figure 33. Forest plot of TIO18 vs. QVA149 meta-analysis results, trough FEV 1 at 12 weeks Table 53. PBO vs. QVA149 meta-analysis results, trough FEV 1 at 12 weeks Study ES [95% Conf. Interval] % Weight ENLIGHTEN SHINE Pooled effect size Heterogeneity chi-squared = 3.32 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) = 69.9% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 6.10 p < GK14290 version 19 November 13 th, 2014 Page 79 of 178

81 Study % ID ES (95% CI) Weight ENLIGHTEN ( , ) SHINE ( , ) Overall (I-squared = 69.9%, p = 0.068) ( , ) NOTE: Weights are from random effects analysis Figure 34. Forest plot of PBO vs. QVA149 meta-analysis results, trough FEV 1 at 12 weeks A retrospective statistical power analysis of the meta-analyses presented in this paragraph suggested a power of >99% in estimating a MCID of 100ml ITC results UMEC/VI 62.5/25 demonstrated comparable results versus QVA149, with a difference in CFB of 10.45mL (95%CI: ; 43.41, p=0.535). The difference was not statistically significant and lower than the MCID level of trough FEV 1 (100mL). Table 54. UMEC/VI vs. QVA149 ITC results, trough FEV 1 at 12 weeks. ES [95% Conf. Interval] % Weight TIO as common comparator PBO as common comparator UMEC/VI 62.5/25 vs. QVA The retrospective calculation of ITC power suggested a statistical power of 61.20% for detecting the same difference of 100mL in the ITC of UMEC/VI 62.5/25 vs. QVA149 with placebo as reference GK14290 version 19 November 13 th, 2014 Page 80 of 178

82 treatment, and a statistical power of >99% when TIO18 is the reference treatment in the ITC of UMEC/VI 62.5/25 vs. QVA Trough FEV 1 excluding SPARK at 12 weeks To evaluate the influence of SPARK study in the relative efficacy of UMEC/VI 62.5/25 versus QVA149, this study was excluded in a scenario analysis. After exclusion of SPARK, only the SHINE trial provided direct comparative evidence of QVA149 vs. TIO18. (Figure 35). Figure 35. Through FEV 1 at 12 weeks: Network of evidence after excluding the SPARK study Individual study results and meta-analysis The individual study results of the GSK trials were described in Table 47 and Table 48. Individual study results of the ENLIGHTEN trial were shown in Table 49, and Table 50 presents the individual study results of the SHINE study. The meta-analysis of TIO vs. UMEC/VI 62.5/25 (Table 51), and PBO vs. QVA149 (Table 53) have been described previously. No meta-analysis was needed for the comparison of TIO18 and QVA149, as only the SHINE study reported on this outcome after exclusion of SPARK ITC results Both placebo and TIO18 were used as reference treatment, therefore the ITC was performed for each reference treatment separately and pooled in a meta-analysis. The results are presented in Table 55. Excluding SPARK from the analysis did not have a major influence on the results of the ITC. UMEC/VI 62.5/25 still showed comparable results with QVA149, with a mean difference of -2.22ml (95%CI ; 31.51, p=0.897). The difference was not statistically significant and was lower than the MCID level of trough FEV 1 (100ml). GK14290 version 19 November 13 th, 2014 Page 81 of 178

83 Table 55. UMEC/VI vs. QVA149 ITC results, trough FEV 1 at 12 weeks ES [95% Conf. Interval] % Weight TIO as common comparator PBO as common comparator UMEC/VI 62.5/25 vs. QVA Trough FEV 1 at 24 weeks SPARK and the GSK studies reported CFB for trough FEV 1 at 24 weeks. SHINE and ENLIGHTEN reported data for trough FEV 1 at 26 weeks, and were pooled with 24 week data to strengthen the networks. An overview of the study design is presented in Table 3. Table 4 provides an overview of the main patient characteristics. Based on the available studies, the following network can be formed (Figure 36). Figure 36. Through FEV 1 at 24 weeks: Network of evidence Individual study results Table 56. LS mean difference for trough FEV 1 (ml) at 24 weeks (vs. TIO) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB ZEP GK14290 version 19 November 13 th, 2014 Page 82 of 178

84 Data source: DB Day 169, table 29, p. 83, and DB Day 169, table 29, p. 85 and ZEP Day 169, Slide deck 4nov2013, slide 10 Table 57. LS mean difference for trough FEV 1 (ml) at 24 weeks (vs. PBO) Difference PBO vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB Data source: DB Day 169, Table 27, p. 89 Table 58. LS mean difference for trough FEV 1 (ml) at 24 weeks in the SPARK and ENLIGHTEN study Baseline Follow-up (24 weeks) Difference TIO18/PBO vs. QVA149 Mean SD LS LS 95% CI 95% CI SE mean mean low high ENLIGHTEN PBO a b -83 b QVA a SPARK TIO b -97 b -43 b QVA a 10 a Data source: Wedzicha 2013 (SPARK), Figure 3, p. 5, Dahl 2013 (ENLIGHTEH), Figure 3, p. 7 a: Extracted from figure using software DigitizeIt (Enlighten: Figure 3, p. 7; Spark: Figure 3, p. 5) b: SE of LS mean at follow-up was to calculate the SE of the difference and its 95%CI. SE of difference = (SE of LS Mean Treatment A) 2 + (SEof LS mean Treatment B) 2 95% CI of difference = DCFB /+1.96 SE to DCFB Table 59. LS mean difference for trough FEV 1 (ml) at 24 weeks (vs. PBO and TIO) in the SHINE study Difference PBO vs. QVA149 Difference TIO vs. QVA149 LS mean 95% CI low 95% CI High LS mean 95% CI low 95% CI high SHINE -200 a a Data source: Bateman 2013 (SHINE), Table 2, p. 20 and Appendix Figure 3a, p. 16 a: Table 3 reports the LS Mean at follow-up for placebo, and the difference of TIO18 or QVA149 and placebo (and 95%CI). These values were used to calculate the difference of TIO vs. QVA149. GK14290 version 19 November 13 th, 2014 Page 83 of 178

85 Meta-analysis The results of the random effects meta-analyses for TIO18 vs. UMEC/VI 62.5/25, TIO18 vs. QVA149 and PBO vs. QVA149 are presented below. As expected, the results of the meta-analyses are confirming the trial evidence that both UMEC/VI and QVA149 are more efficacious than placebo and tiotropium as monotherapy, by means of trough FEV 1 at 24 weeks. Table 60. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 (ml) at 24 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 3.07 (df. = 2) p = I-squared (variation in ES attributable to heterogeneity) = 34.9% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 5.89 p < Study % ID ES (95% CI) Weight DB ( , ) DB ( , ) ZEP ( , ) Overall (I-squared = 34.9%, p = 0.215) ( , ) NOTE: Weights are from random effects analysis Figure 37. Forest plot of TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, trough FEV 1 at 24 weeks GK14290 version 19 November 13 th, 2014 Page 84 of 178

86 Table 61. TIO18 vs. QVA149 meta-analysis results, trough FEV 1 (ml) at 24 weeks Study ES [95% Conf. Interval] % Weight SPARK SHINE Pooled effect size Heterogeneity chi-squared = 0.28 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 8.05 p < Study % ID ES (95% CI) Weight SPARK (-97.00, ) SHINE ( , ) Overall (I-squared = 0.0%, p = 0.594) (-93.73, ) NOTE: Weights are from random effects analysis Figure 38. Forest plot of TIO18 vs. QVA149 meta-analysis results, trough FEV 1 at 24 weeks Table 62. PBO vs. QVA149 meta-analysis results, trough FEV 1 at 24 weeks Study ES [95% Conf. Interval] % Weight ENLIGHTEN SHINE Pooled effect size Heterogeneity chi-squared = 1.48 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) = 32.4% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 8.44 p < GK14290 version 19 November 13 th, 2014 Page 85 of 178

87 Study % ID ES (95% CI) Weight ENLIGHTEN ( , ) SHINE ( , ) Overall (I-squared = 32.4%, p = 0.224) ( , ) NOTE: Weights are from random effects analysis Figure 39. Forest plot of PBO vs. QVA149 meta-analysis results, trough FEV 1 at 24 weeks The retrospective power calculations for the meta-analyses presented in this paragraph suggested that in all cases, the power for detecting a difference of 100ml is 99% or higher ITC results UMEC/VI 62.5/25mcg demonstrated a comparable CFB versus QVA149, with a LS mean difference of 6.99ml (95%CI: ; 38.02, p=0.659) (Table 63). The difference was not statistically significant and was lower than the MCID level of trough FEV 1 (100ml). Table 63. UMEC/VI vs. QVA149 ITC results, trough FEV 1 at 24 weeks. ES [95% Conf. Interval] % Weight TIO as common comparator PBO as common comparator UMEC/VI 62.5/25 vs. QVA Furthermore, the retrospective calculation of ITC power suggested a statistical power of 77.12% for detecting a difference of 100ml in the ITC of UMEC/VI 62.5/25 vs. QVA149 with placebo as reference treatment, and a statistical power of >99% when TIO18 is the reference treatment in the ITC of UMEC/VI 62.5/25 vs. QVA149. GK14290 version 19 November 13 th, 2014 Page 86 of 178

88 Trough FEV 1 excluding SPARK at 24 weeks After the exclusion of the SPARK study, six studies remained that reported data on trough FEV 1 at 24 weeks: ENLIGHTEN, SHINE, DB , DB , DB and ZEP SHINE and ENLIGHTEN reported data at 26 weeks and DB , DB , DB and ZEP reported data at 24 weeks. Figure 40 presents the network of evidence after excluding the SPARK study. Figure 40. Through FEV 1 at 24 weeks: Network of evidence after excluding the SPARK study Individual study results and meta-analysis Individual study results are presented in Table 56 to Table 59. The meta-analysis of TIO vs. UMEC/VI 62.5/25 (Table 60), and PBO vs. QVA149 (Table 62) have been described previously. No metaanalysis was needed for the comparison of TIO18 and QVA149, as only the SHINE study reported on this outcome after exclusion of SPARK ITC results Both placebo and TIO18 were used as reference treatments, therefore the ITC was performed for each reference treatment separately and pooled in a meta-analysis. The results are presented in Table 64. The results of the scenario excluding SPARK are in line with the scenario including SPARK. UMEC/VI 62.5/25 was still comparable to QVA149, with a mean difference of 2.52ml (95%CI ; 35.28, p=0.880). These differences were not statistically significant and were lower than the MCID level of trough FEV 1 (100ml). Table 64. UMEC/VI vs. QVA149 ITC results, trough FEV 1 (ml) at 24 weeks ES [95% Conf. Interval] % Weight TIO as common comparator GK14290 version 19 November 13 th, 2014 Page 87 of 178

89 PBO as common comparator UMEC/VI 62.5/25 vs. QVA SGRQ total score at 12 weeks Only tiotropium could be used as reference treatment for the ITC, since no study could be included that compared QVA149 to placebo by means of SGRQ total score at 12 weeks. The SHINE study reported (LS mean) difference in SGRQ total score at 12 weeks of QVA149 or tiotropium versus placebo. However, as no measure of uncertainty (SD, SE, 95%CI) was reported, this study could not be included in the analysis. Four studies were identified in the systematic literature review that reported change from baseline in the SGRQ total score at 12 weeks for UMEC/VI 62.5/25 or QVA149 compared with TIO 18: SPARK, DB , DB and ZEP The network of studies for SGRQ total score at 12 weeks is presented in Figure 41. Figure 41. SGRQ total score at 12 weeks: Network of evidence Individual study results DB , DB and ZEP reported the LS mean DCFB (Difference Change From Baseline) of UMEC/VI 62.5/25 versus tiotropium. The SPARK study (Wedzicha 2013) presented (LS mean) follow-up values for tiotropium and QVA149 at 12 weeks in a figure. Software program Digitizeit was used to extract values from the graphs, to be able to calculate the DCFB of tiotropium versus QVA149. Therefore, the LS mean differences were presented as tiotropium versus UMEC/VI in Table 65. The results are suggesting that UMEC/VI 62.5/25 and QVA149 are more efficacious in improving the SGRQ total score at 12 weeks. (Lower SGRQ total scores represent an improvement). No scenario analysis excluding the SPARK study was performed, as an ITC would no longer be possible. The individual study results show that treatment with TIO18 resulted in higher SGRQ total scores than treatment with QVA149 or UMEC/VI 62.5/25 (Table 65 and Table 66). Table 65. Difference in CBF for SGRQ total score at 12 weeks (Day 84) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high GK14290 version 19 November 13 th, 2014 Page 88 of 178

90 DB DB ZEP Data source: DB , Day 84 Table 51, p. 126, DB , Day 84, Table 50, p. 125, ZEP117115, Day 84 Table 45, p 99 Table 66. LS mean difference for SGRQ total score at 12 weeks. Baseline Follow-up Difference Mean SD LS mean (12 weeks) SE LS mean TIO18 vs. QVA149 95% CI low 95% CI high SPARK TIO b 1.27 b 4.73 b QVA a 0.60 a Data source: Wedzicha 2013, Table 1, p. 3 and Figure 4, p. 6 a: Extracted from figure using software DigitizeIt (Enlighten: Figure 3, p. 7; Spark: Figure 3, p. 5) b: LS mean at follow-up and SE were used to calculate the difference (with SE) and its 95%CI. SE of difference = (SE of LS Mean Treatment A) 2 + (SEof LS mean Treatment B) Meta-analysis results 95% CI of difference = DCFB /+1.96 SE to DCFB The individual study results presented for UMEC/VI 62.5/25 were synthesized by means of a random effects meta-analysis. As expected, the finding of the meta-analysis was in line with the findings of the individual trials, and show that UMEC/VI 62.5/25 is more efficacious than tiotropium as monotherapy, by means of SGRQ total score at 12 weeks. Table 67. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, SGRQ total score at 12 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 1.66 (df. = 2) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 3.13 p = GK14290 version 19 November 13 th, 2014 Page 89 of 178

91 Study ID ES (95% CI) % Weight DB DB (-0.47, 4.48) (-2.29, 2.76) ZEP Overall (I-squared = 0.0%, p = 0.437) 2.08 (0.72, 3.45) 1.73 (0.65, 2.81) NOTE: Weights are from random effects analysis Figure 42. Forest plot of TIO vs. UMEC/VI 62.5/25 for SGRQ total score at 12 weeks ITC Results Using eq. 5, the variance for each meta-analysis was calculated and subsequently, using eq. 4 we can estimate the variance, standard error and 95% confidence intervals for the ITC between UMEC/VI and QVA149. UMEC/VI 62.5/25 showed similar results versus QVA149. The LS mean difference for UMEC/VI 62.5/25mcg versus QVA149 was 1.27 units (95%CI: -0.77; 3.31, p=0.222). The difference was not statistically significant and did not exceed the MCID level for the SGRQ total score (-4 units difference). Table 68. UMEC/VI vs. QVA149 ITC results, SGRQ total score at 12 weeks MD variance se [95% Conf. Interval] p-value UMEC/VI 62.5/25 vs. QVA GK14290 version 19 November 13 th, 2014 Page 90 of 178

92 SGRQ total score at 24 weeks The endpoint of interest is the change from baseline in SGRQ total score at 24 weeks. Six trials were identified in the systematic literature review that reported data on this outcome, i.e. SPARK, SHINE, DB , DB , DB and ZEP The following network of trials could be developed (Figure 43). Figure 43. SGRQ total score at 24 weeks: Network of evidence Individual study results Except for the SPARK study, all studies reported the (LS mean) difference in CFB for SGRQ total score at 24 weeks versus tiotropium or placebo. The SPARK study (Wedzicha 2013) presented (LS mean) follow-up values for tiotropium and QVA149 at 24 weeks in a figure. A software program (Digitizeit) was used to extract values from this graph, to be able to calculate the LS mean difference of tiotropium versus QVA149. With the exception of UMEC/VI 62.5/25 in the DB trial, UMEC/VI 62.5/25 and QVA149 are expected to be numerically more efficacious in improving (reducing) the SGRQ total score at 24 weeks (Table 69, Table 70, Table 71 and Table 72). Table 69. LS mean difference for SGRQ total score at 24 weeks (vs. TIO) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB ZEP GK14290 version 19 November 13 th, 2014 Page 91 of 178

93 Data source: DB day 168, Table 50, p. 125, and DB day 168, Table 51, p. 126; ZEP117115, Day 168. Slide deck 4Nov2013, slide 20 Table 70. LS mean difference for SGRQ total score at 24 weeks (vs. PBO) Difference PBO vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB Data source: DB day 84, Table 50, p. 136 Table 71. LS mean difference for SGRQ total score at 24 weeks in the SPARK study Baseline Follow-up Difference Mean SD LS mean (24 weeks) SE LS mean TIO18 vs. QVA149 95% CI low 95% CI high SPARK TIO a a 3.40 a QVA a 0.65 a Data source: Wedzicha 2013 (SPARK), Table 1, p. 3 and Figure 4, p. 6 a: LS mean values at 24 weeks were extracted (using Digitizeit) and used to calculate the LS mean difference and its 95%CI. SE of difference = (SE of LS mean Treatment A) 2 + (SEof LS mean Treatment B) 2 95% CI of difference = LS mean 1.96 SE to LS mean SE Table 72. LS mean difference for SGRQ total score at 24 weeks (vs. PBO and TIO) in the SHINE study Difference PBO vs. QVA149 Difference TIO vs. QVA149 LS mean 95% CI low 95% CI High LS mean 95% CI low 95% CI High SHINE Data source: Bateman 2013 (SHINE), Appendix Table 3, p. 9 and Figure 4a, p. 20 a: Table 3 reports the LSMean follow-up value for placebo, and the difference of TIO18 or QVA149 and placebo (and 95%CI). These values were used to calculate the LS mean difference of TIO vs. QVA149. Data were checked with Figure 4a Meta-analysis results Using random effects meta-analysis, the individual study results for UMEC/VI 62.5/25 and QVA149 versus tiotropium were synthesized. As expected, the findings of the meta-analyses were in line with the findings of the individual trials, and show that UMEC/VI 62.5/25 and QVA149 are numerically GK14290 version 19 November 13 th, 2014 Page 92 of 178

94 more efficacious than tiotropium as monotherapy, by means of SGRQ total score at 24 weeks. These results were not statistically significant. Table 73. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, SGRQ total score at 24 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 3.70 (df. = 2) p = I-squared (variation in ES attributable to heterogeneity) = 45.9% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 0.93 p = Study % ID ES (95% CI) Weight DB (-2.52, 2.85) DB (-3.63, 2.12) ZEP (0.59, 3.61) Overall (I-squared = 45.9%, p = 0.158) 0.85 (-0.94, 2.65) NOTE: Weights are from random effects analysis Figure 44. Forest plot of TIO vs. UMEC/VI 62.5/25 for SGRQ at 24 weeks Table 74. TIO18 vs. QVA149 meta-analysis results, SGRQ total score at 24 weeks Study ES [95% Conf. Interval] % Weight SPARK SHINE Pooled effect size GK14290 version 19 November 13 th, 2014 Page 93 of 178

95 Heterogeneity chi-squared = 0.03 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 3.09 p = Study % ID ES (95% CI) Weight SPARK 1.60 (-0.20, 3.40) SHINE 2.13 (0.90, 3.36) Overall (I-squared = 0.0%, p = 0.634) 1.96 (0.94, 2.98) NOTE: Weights are from random effects analysis Figure 45. Forest plot of TIO vs. QVA149 for SGRQ at 24 weeks ITC Results Using eq. 5, the variance for each meta-analysis was calculated and subsequently, using eq. 4 the variance, standard error and 95% confidence intervals for the ITC between UMEC/VI 62.5/25 and QVA149 were calculated. Two reference treatments were used (placebo and TIO18), the results of the ITC are presented per common comparator. Although the 95% confidence intervals still slightly overlap in the ITCs, the difference between the point estimates of the ITC using placebo and TIO 18 were not consistent (1.11 vs respectively). The difference between these estimates was almost the MCID for SGRQ total score. This inconsistency was deemed too large and the ITCs were not synthesized in a random effects metaanalysis. GK14290 version 19 November 13 th, 2014 Page 94 of 178

96 Despite the inconsistency, both ITCs did not demonstrate any statistically significant differences between UMEC/VI 62.5/25 and QVA149. Table 75. UMEC/VI vs. QVA149 ITC results, SGRQ total score at 24 weeks Common comparator MD variance se [95% Conf. Interval] p- value UMEC/VI 62.5/25 vs. QVA149 TIO UMEC/VI 62.5/25 vs. QVA149 PBO SGRQ total score excluding SPARK at 24 weeks The outcome of interest is the SGRQ total score at 24 weeks. The SPARK study was excluded from this scenario analysis, therefore five studies were included in the analysis (i.e. SHINE, DB , DB , DB and ZEP117115). The network of studies is presented in Figure 46. Figure 46. SGRQ total score at 24 weeks: network of evidence after excluding the SPARK study Individual study results and meta-analysis The individual study results were presented in Table 69, Table 70 and Table 72. No meta-analysis was needed for the comparison between QVA149 and tiotropium or placebo, as only one study reported on this outcome. The meta-analysis results were presented in Table 73 and Figure ITC results Both placebo and TIO18 were used as reference treatments, therefore the ITC was performed for each reference treatment separately and pooled in a meta-analysis. The results are presented in the GK14290 version 19 November 13 th, 2014 Page 95 of 178

97 table below. Because of the large heterogeneity from a clinical perspective (the mean difference between the ITCs was almost the MCID for SGRQ total score) the results were not synthesized in a meta-analysis. The results of the individual ITCs are in line with the scenario including SPARK, no statistically significant differences were found between UMEC/VI 62.5/25 and QVA149. Table 76. UMEC/VI vs. QVA149 ITC results, SGRQ total score at 24 weeks Common comparator MD variance se [95% Conf. Interval] p- value UMEC/VI 62.5/25 vs. QVA149 TIO UMEC/VI 62.5/25 vs. QVA149 PBO GK14290 version 19 November 13 th, 2014 Page 96 of 178

98 TDI focal score at 12 weeks The outcome of interest is the TDI focal score at 12 weeks. Four studies (DB , DB , DB and SHINE) are included in the systematic literature review, reporting this outcome for the treatments of interest at 12 weeks. Since the SPARK study did not report any data for TDI focal score, no scenario analysis was needed excluding this study. Based on the available studies, the following network can be formed (Figure 47). Figure 47. TDI at 12 weeks: Network of evidence Individual study results All studies reported the (LS mean) difference TDI at 12 weeks versus tiotropium or placebo, with the corresponding 95% CI. The differences in CFB were reversed for all trials and are reported in Table 77, Table 78 and Table 79. These tables show that treatment with tiotropium or placebo is expected to be associated with a lower change from baseline in TDI focal score at 12 weeks than UMEC/VI 62.5/25 or QVA149. Table 77. LS mean difference for TDI focal score at 12 weeks (vs. TIO) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB Data source: DB day 84, Table 33, p. 95, and DB day 84, Table 33, p. 97 GK14290 version 19 November 13 th, 2014 Page 97 of 178

99 Table 78. LS mean difference for TDI focal score at 12 weeks (vs. PBO) Difference PBO vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB Data source: DB day 84, Table 30, p. 96 Table 79. LS mean difference for TDI focal score at 12 weeks (vs. PBO and TIO) Difference Difference LS mean PBO vs. QVA149 95% CI low 95% CI High LS mean TIO vs. QVA149 95% CI low 95% CI high SHINE a a a a Data source: Appendix Figure 4b, p. 21 a: Standard errors were extracted from Figure 4b and used to calculate the 95%CI of the change from baseline Meta-analysis results The comparison of PBO vs. UMEC/VI 62.5/25 was not included in a meta-analysis, as only one trial reported on this comparison. A random effects meta-analysis was performed for the comparison of TIO18 vs. UMEC/VI 62.5/25. As expected, the results of the meta-analysis are in line with the individual study results, showing that UMEC/VI 62.5/25 is associated with a larger change from baseline in TDI score at 12 weeks than tiotropium (Table 80). Table 80. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, TDI focal score at 12 weeks. Study ES [95% Conf. Interval] % Weight DB DB Pooled effect size Heterogeneity chi-squared = 1.57 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) = 36.5% Estimate of between-study variance Tau-squared = Test of ES=0 : z= 1.70 p = GK14290 version 19 November 13 th, 2014 Page 98 of 178

100 Study % ID ES (95% CI) Weight DB (-1.30, -0.10) DB (-0.70, 0.30) Overall (I-squared = 36.5%, p = 0.210) (-0.91, 0.07) NOTE: Weights are from random effects analysis Figure 48. Forest plot of TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, TDI at 12 weeks ITC Results The variance for each meta-analysis was calculated using eq. 5 and subsequently, using eq. 4 the variance, standard error and 95% confidence intervals for the ITC between UMEC/VI 62.5/25 and QVA149 were calculated. Two reference treatments were used (placebo and tiotropium 18), the ITC was performed for each reference treatment separately and pooled in a meta-analysis. The results are presented in Table 81. UMEC/VI 62.5/25 showed similar results compared with QVA149. UMEC/VI 62.5/25 resulted in a difference of point (95%CI of -0.54; 0.40, p=0.764) compared with QVA149. The difference was not statistically significant and did not exceed the MCID level for the TDI focal score (1 point). Table 81. UMEC/VI vs. QVA149 ITC results, TDI focal score at 12 weeks. ES [95% Conf. Interval] % Weight TIO as common comparator PBO as common comparator UMEC/VI 62.5/25 vs. QVA TDI focal score at 24 weeks The TDI focal score at 24 weeks is the outcome of interest. Four studies (DB , DB , DB and SHINE) are included in the systematic literature review, reporting this outcome at 24 weeks. Based on the available studies, the following network can be formed (Figure 49). GK14290 version 19 November 13 th, 2014 Page 99 of 178

101 Figure 49. TDI at 24 weeks: Network of evidence Individual study results All studies reported the (LS mean) difference TDI at 24 weeks versus tiotropium or placebo, with the corresponding 95% CI. The differences in CFB were reversed for all trials and are reported in Table 82, Table 83 and Table 84. UMEC/VI 62.5/25 and QVA149 showed higher CFB TDI focal scores at 24 weeks than TIO18 and placebo, however these results were not statistically significant for the comparison between UMEC/VI 62.5/25 and TIO18. Table 82. LS mean difference for TDI focal score at 24 weeks (vs. TIO) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB Data source: DB day 168, Table 33, p. 96, and DB day 168, Table 33, p. 97 Table 83. LS mean difference for TDI focal score at 24 weeks (vs. PBO) Difference PBO vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high GK14290 version 19 November 13 th, 2014 Page 100 of 178

102 DB Data source: DB day 84, Table 30, p. 168 Table 84. LS mean difference for TDI focal score at 24 weeks (vs. PBO and TIO) Difference PBO vs. QVA149 Difference TIO vs. QVA149 LS mean 95% CI low 95% CI High LS mean 95% CI low 95% CI high SHINE Data source: Appendix Table 3, p. 9 and Figure 4a, p. 20 a: Table 3 reports the LSMean follow-up value for placebo, and the difference of TIO18 or QVA149 and placebo (and 95%CI). These values were used to calculate the LS mean difference of TIO vs. QVA149. Data were checked with Figure 4a Meta-analysis results No meta-analysis was performed for the comparison of PBO vs. UMEC/VI 62.5/25. Random effects meta-analysis was performed for the comparison of TIO18 vs. UMEC/VI 62.5/25, the results are presented below. In line with the individual study results, TIO18 resulted in similar TDI scores at 24 weeks compared with UMEC/VI 62.5/25 (Table 85 and Figure 50) Table 85. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, TDI focal score at 24 weeks Study ES [95% Conf. Interval] % Weight DB DB Pooled effect size Heterogeneity chi-squared = 0.41 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 0.12 p = GK14290 version 19 November 13 th, 2014 Page 101 of 178

103 Study % ID ES (95% CI) Weight DB (-0.90, 0.50) DB (-0.50, 0.70) Overall (I-squared = 0.0%, p = 0.524) (-0.48, 0.43) NOTE: Weights are from random effects analysis Figure 50. Forest plot of TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, TDI at 24 weeks ITC Results The ITC was performed using eq. 4 and 5. Two reference treatments were used (placebo and tiotropium 18), the ITC was performed for each reference treatment separately and pooled in a metaanalysis. The results are presented in Table 86. UMEC/VI 62.5/25 resulted in similar TDI focal scores at 24 weeks compared with QVA149. The difference between UMEC/VI 62.5/25 and QVA149 was unit (95%CI ; 0.338, p=0.423). The difference was not statistically significant and did not exceed the MCID level for the TDI focal score (1 unit). Table 86. UMEC/VI vs. QVA149 ITC results, TDI focal score at 24 weeks ES [95% Conf. Interval] % Weight TIO as common comparator PBO as common comparator UMEC/VI 62.5/25 vs. QVA Rescue medication at 24 weeks Four GSK trials reported data on the rescue medication use at 24 weeks (DB , DB , DB and ZEP117115) and one QVA149 trial (SHINE) reported data on this outcome. Based on the available studies, the following network can be formed (Figure 51). Since the SPARK study did GK14290 version 19 November 13 th, 2014 Page 102 of 178

104 not report any data for rescue medication use at 24 weeks, no scenario analysis was needed excluding this study. Figure 51. Rescue medication use at 24 weeks: Network of evidence Individual study results The GSK studies reported the (LS mean) difference in number of puffs over 1-24 weeks versus tiotropium or placebo, with the corresponding 95% CI. SHINE reported this outcome at 26 weeks, but this was considered to be similar enough to be included in the meta-analysis. The differences in CFB were reversed for all trials as tiotropium and placebo are the reference treatments and are reported in Table 87, Table 88 and Table 89. The individual study results show that treatment with tiotropium or placebo is associated with a higher need for rescue medication than treatment with UMEC/VI 62.5/25 and QVA149. Table 87. Difference in rescue medication use (number of puffs) at 24 weeks (vs. TIO) Difference TIO18 vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB DB ZEP Data source: DB Week 1-24, Table 42, p. 115, and DB Week 1-24, Table 42, p. 116 and ZEP Week 1-24, slide deck 4Nov2013, slide 18. GK14290 version 19 November 13 th, 2014 Page 103 of 178

105 Table 88. Difference in rescue medication use (number of puffs) at 24 weeks (vs. PBO) Difference PBO vs. UMEC/VI 62.5/25 LS mean 95% CI low 95% CI high DB Data source: DB Week 1-24, Table 42, p. 126 Table 89. Difference in rescue medication use (number of puffs) at 24 weeks (vs. PBO and TIO) Difference PBO vs. QVA149 Difference TIO vs. QVA149 LS mean 95% CI low 95% CI high LS mean 95% CI low 95% CI high SHINE a 0.31 a 0.79 a Data source: Appendix Table 3, p. 9. a: Table 3 reports the LSMean follow-up value for placebo, and the difference of TIO18 or QVA149 and placebo (and 95%CI). These values were used to calculate the LS mean difference of TIO vs. QVA Meta-analysis results The comparison of PBO vs. UMEC/VI 62.5/25 was not synthesized in a meta-analysis, as only one trial reported this comparison. A random effects meta-analysis was performed for the comparison of TIO18 vs. UMEC/VI 62.5/25. As expected, the results of the meta-analysis are in line with the individual study results, showing that patients treated with tiotropium or placebo need more rescue medication than patients treated with UMEC/VI 62.5/25 or QVA149. Table 90. TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, daily number of puffs rescue medication at 24 weeks Study ES [95% Conf. Interval] % Weight DB DB ZEP Pooled effect size Heterogeneity chi-squared = 0.06 (df. = 1) p = I-squared (variation in ES attributable to heterogeneity) < 0.001% Estimate of between-study variance Tau-squared < Test of ES=0 : z= 3.16 p = GK14290 version 19 November 13 th, 2014 Page 104 of 178

106 Study % ID ES (95% CI) Weight DB (0.00, 1.20) DB (0.10, 1.20) ZEP (0.20, 0.70) Overall (I-squared = 0.0%, p = 0.794) 0.54 (0.33, 0.76) NOTE: Weights are from random effects analysis Figure 52. Forest plot of TIO18 vs. UMEC/VI 62.5/25 meta-analysis results, rescue medication use (number of puffs) at 24 weeks ITC Results Using eq. 5, the variance for each meta-analysis was calculated and subsequently, using eq. 4 the variance, standard error and 95% confidence intervals for the ITC between UMEC/VI 62.5/25 and QVA149 were calculated. Two reference treatments were used (placebo and tiotropium 18), the ITC was performed for each reference treatment separately and pooled in a meta-analysis. The results are presented in Table 91. UMEC/VI 62.5/25 showed comparable results with QVA149, with a mean difference of 0.04 (95%CI: ; 0.33, p=0.765). Table 91. UMEC/VI vs. QVA149 ITC results, rescue medication use (number of puffs) at 24 weeks ES [95% Conf. Interval] % Weight TIO as common comparator PBO as common comparator UMEC/VI 62.5/25 vs. QVA GK14290 version 19 November 13 th, 2014 Page 105 of 178

107 6. Bayesian network meta-analysis For the Bayesian NMA, studies comparing a LABA/LAMA combination therapy (UMEC/VI, TIO+SAL, TIO+FOR, IND+TIO, QVA149) to each other or to placebo were of interest. As a result, DB , SHINE, ENLIGHTEN and Vogelmeier 2008 could be included in the networks. The overall network of studies for the base case scenario is presented in Figure 53. Figure 53. Overall network of studies in the base case - NMA of UMEC/VI versus LABA/LAMA combination therapies In the base case, UMEC/VI could not be compared to IND 150 +TIO 18, TIO 18 + SAL 50 and to TIO 18 + FOR 12 because no study was identified comparing these interventions to another LABA/LAMA combination therapy or to placebo. The main objective of the Bayesian NMA was to indirectly compare UMEC/VI 62.5/25 to other open dual or closed LABA/LAMA combination therapies. Therefore, studies were selected comparing a LABA/LAMA intervention of interest to another LABA/LAMA intervention of interest or to placebo. However, because many studies using TIO 18 as comparator arm instead of placebo, a scenario analysis was performed with studies comparing a LABA/LAMA intervention of interest to another LABA/LAMA intervention of interest or to TIO 18 or placebo. The inclusion of TIO 18 in the networks allowed for the utilization of data from three more studies on UMEC/VI (DB , DB , ZEP117115) and also allows for the inclusion of INTRUST-1, INTRUST-2, Aaron 2007 and Tashkin The overall network of studies in the scenario analysis is presented in Figure studies were identified in the SLR comparing TIO 18 mcg OD versus placebo by means of trough FEV1, SGRQ total score, TDI focal score or rescue medication use at 12 or 24 weeks, or within the margin of time around these time points. GK14290 version 19 November 13 th, 2014 Page 106 of 178

108 Figure 54. Overall network of studies in the scenario analysis - NMA of UMEC/VI versus LABA/LAMA combination therapies * Note: the 16 studies are in the reference list, but not spelled out in the network due to formatting restraints The specific trials for each analysis are presented in the outcome specific networks. GK14290 version 19 November 13 th, 2014 Page 107 of 178

109 6.1. Trough FEV Trough FEV1 at 12 weeks The network of studies for trough FEV1 at 12 weeks is presented in Figure 55 for the base case analysis and in Figure 58 for the scenario analysis. The results of the individual studies are presented in Table 92 for the base case and in Table 94 for the scenario analysis. For each study the difference in trough FEV 1 change from baseline for the treatment versus the comparator (placebo for all studies) at 24 weeks and the standard error of this difference are presented. If the difference in the LS mean at follow-up or the CFB was reported it was used in the analysis, otherwise it was calculated based on the difference between the values reported per treatment (see Appendix 4 for more details). Base case All studies report a statistically significant increase in trough FEV1 with a LABA/LAMA combination versus placebo (Table 93). Consequently, the results of the NMA for trough FEV1 at 12 weeks show that both active treatments increased trough FEV1 compared to placebo and the results were statistically significant. The probability of UMEC/VI 62.5/25 and QVA149 being better than placebo was over 99%. Figure 57 shows the mean difference in change from baseline of the active treatments versus placebo. UMEC/VI 62.5/25 resulted in similar trough FEV1 compared with QVA149, with a mean difference in CFB of ml (95%CrI: ; 27.20). These results were not statistically significant, since the 95%CrI crossed 0. The probability of UMEC/VI 62.5/25 being better than QVA149 was 21%. Scenario analysis The results of the scenario analysis are in line with the base case results, UMEC/VI demonstrated comparable results compared with QVA149, with a mean difference in CFB of 9.41 ml (95%CrI: 16.96; 35.77) and a probability of being better of 76%. The network in the scenario analysis included more pairwise comparisons because of the inclusion of TIO 18 as comparator. The probability of UMEC/VI 62.5/25 being a better treatment was 55% and 94% versus FOR12 + TIO 18 and IND TIO 18 respectively. The results of the NMA in the scenario analysis are presented in Table Base case Figure 55. Network of studies for trough FEV1 at 12 weeks (base case analysis) Table 92. Individual study results for trough FEV1 at 12 weeks in the base case analysis. Author Treatment Trough FEV1 in ml (Difference in CFB) GK14290 version 19 November 13 th, 2014 Page 108 of 178

110 Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 84. Table 28, p %CI used to calculate SE SHINE QVA149 vs. PBO ENLIGHTEN QVA149 vs. PBO Week 12. Table 2, p %CI used to calculate SE Week 26. LS mean and LS mean difference reported in Figure 3, SE at follow-up extracted using DigitizeIt Table 93. Results of the base case network meta-analysis for trough FEV1 (difference in CFB, ml), at 12 weeks Comparator Intervention Placebo QVA /50 estimate QVA /50 95% CrI P(better) >99% estimate UMEC/VI 62.5/25 95% CrI P(better) >99% 21% Figure 56. Forest plot for mean trough FEV1 of UMEC/VI 62.5/25 vs. comparators at 12 weeks GK14290 version 19 November 13 th, 2014 Page 109 of 178

111 Figure 57. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 12 weeks Scenario analysis Figure 58. Network of studies for trough FEV1 at 12 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 110 of 178

112 Table 94. Individual study results for trough FEV1 at 12 weeks in the scenario analysis. Author Treatment DB UMEC/VI 62.5/25 vs. PBO Trough FEV1 in ml (Difference in CFB) Mean SE Source Day 84. Table 28, p %CI used to calculate SE DB UMEC/VI 62.5/25 vs. TIO Day 84. Table 30, p. 85 DB UMEC/VI 62.5/25 vs. TIO Day 84. Table 30, p. 87 ZEP UMEC/VI 62.5/25 vs. TIO Day 84. Table 28, p. 69 SHINE QVA149 vs. PBO Week 12. Table 2, p %CI used to calculate SE SHINE TIO 18 vs. PBO Week 12. Table 2, p 20 SPARK QVA149 vs. TIO ENLIGHTEN QVA149 vs. PBO LS mean (se) extracted from Figure 3, p. 5 Week 26. LS mean and LS mean difference reported in Figure 3, SE at follow-up extracted using DigitizeIt INTRUST-1 IND TIO 18 vs. TIO Mahler 2012, Thorax, p.784 INTRUST-2 IND TIO 18 vs. TIO Mahler 2012, Thorax, p.784 Tashkin 2009 TIO 18 + FOR 12 vs. TIO Text p. 21 and Figure 5 Chan 2007 TIO 18 vs. PBO Niewoehner 2005 LS Mean at 11 weeks extracted from figure 1a TIO 18 vs. PBO Spirometry, text, p 323 Donohue 2010 TIO 18 vs. PBO Table 2, p. 157 GLOW2 TIO 18 vs. PBO Week 12. Table 2, p 1111 Verkindre 2006 TIO 18 vs. PBO Day 84. Table 2, p. 423 Casaburi 2000 TIO 18 vs. PBO Day 92. Table 2 reports CFB and SE. Covelli 2005 TIO 18 vs. PBO Day 84. Table 7, p Moita 2008 TIO 18 vs. PBO Table 3, p. 149 Table 95. Results of the scenario analysis network meta-analysis for trough FEV1 (difference in CFB, ml), at 12 weeks Comparator Intervention Placebo TIO 18 QVA /50 FOR 12 + TIO 18 IND TIO 18 estimate TIO 18 95% CrI P(better) >99% QVA /50 estimate % CrI P(better) >99% >99% GK14290 version 19 November 13 th, 2014 Page 111 of 178

113 FOR 12 + TIO 18 IND TIO 18 UMEC/VI 62.5/25 estimate % CrI P(better) >99% >99% 57% estimate % CrI P(better) >99% >99% 20% 29% estimate % CrI P(better) >99% >99% 76% 55% 94% Figure 59. Forest plot for mean trough FEV1 of UMEC/VI 62.5/25 vs. comparators at 12 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 112 of 178

114 Figure 60. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 12 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 113 of 178

115 Trough FEV1 at 24 weeks The network of studies for trough FEV1 at 24 weeks looks similar to the network of trough FEV1 at 12 weeks, since all studies reporting at 12 weeks also presented data at 24 weeks. Three studies were included in the base case analysis; DB , SHINE and ENLIGHTEN and the network is presented in Figure 61. The input data for the NMA of these studies is presented in Table 96. All studies showed a statistically significant improvement in trough FEV1 by the active treatment compared with placebo. Base case The results of the NMA are in line with the input data and show a statistically significant improvement in trough FEV1 at 24 weeks for UMEC/VI 62.5/25 and for QVA149 versus placebo (Table 97). The ranked results of the active treatments versus placebo are also presented in Figure 63. For both active treatments, the difference exceeded the MCID for trough FEV1. Compared with QVA149, UMEC/VI 62.5/25 is expected to result in a similar trough FEV1 at 24 weeks. The mean difference in CFB was ml (95%CrI: ; 26.78) (Table 97, Figure 62). The difference was not statistically significant and did not exceed the MCID of 100 ml. The probability of UMEC/VI 62.5/25 being the better treatment versus QVA149 was 18%. Scenario analysis The results of the scenario analysis including TIO 18 were in line with the results of the base case. UMEC/VI 62.5/25 was more efficacious than placebo by means of trough FEV1 at 24 weeks and similar efficacious as QVA149. The mean difference in CFB versus QVA149 was ml (95%CrI: ; 42.25). This was higher than in the base case, likely because the strong treatment effect in the ZEP trial, the largest of the UMEC/VI 62.5/25 trials. This trial was not included in the base case network because it compared UMEC/VI 62.5/25 to TIO 18 only. The difference in CFB of UMEC/VI 62.5/25 versus TIO 18 + SAL 50 was ml (95%CrI: , ). Although the probability of being better (92%) indicates that UMEC/VI 62.5/25 is likely to be more efficacious, the results were not statistically significant and did not exceed the MCID for trough FEV Base case Figure 61. Network of studies for trough FEV1 at 24 weeks (base case analysis) GK14290 version 19 November 13 th, 2014 Page 114 of 178

116 Table 96. Individual study results for trough FEV1 at 24 weeks in the base case analysis. Trough FEV1 in ml (Difference in CFB) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO SHINE QVA149 vs. PBO ENLIGHTEN QVA149 vs. PBO Day 169. Table 27, p %CI used to calculate SE Week 26. Table 2, p %CI used to calculate SE Week 26. LS mean difference in Figure 3, SE at follow-up extracted using DigitizeIt Table 97. Results of the base case network meta-analysis for trough FEV1 (difference in CFB, ml), at 24 weeks Comparator Intervention Placebo QVA /50 estimate QVA /50 95% CrI P(better) >99% estimate UMEC/VI 62.5/25 95% CrI P(better) >99% 18% Figure 62. Forest plot for mean trough FEV1 of UMEC/VI 62.5/25 vs. comparators at 24 weeks GK14290 version 19 November 13 th, 2014 Page 115 of 178

117 Figure 63. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 24 weeks Scenario analysis Figure 64. Network of studies for trough FEV1 at 24 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 116 of 178

118 Table 98. Individual study results for trough FEV1 at 24 weeks in the scenario analysis. Trough FEV1 in ml (Difference in CFB) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 169. Table 27, p. 89 DB UMEC/VI 62.5/25 vs. TIO Day 169. Table 29, p. 84 DB UMEC/VI 62.5/25 vs. TIO Day 169. Table 29, p. 85 ZEP UMEC/VI 62.5/25 vs. TIO Day 169. Table 27, p. 68 SHINE QVA149 vs. PBO Week 26. Table 2, p %CI used to calculate SE SHINE TIO 18 vs. PBO Week 26. Table 2, p 20 SPARK QVA149 vs. TIO LS mean (se) extracted from Figure 3, p. 5 ENLIGHTEN QVA149 vs. PBO Aaron 2007 TIO 18 + SAL 50 vs. TIO UPLIFT TIO 18 vs. PBO Niewoehner 2005 Week 26. LS mean and LS mean difference reported in Figure 3, SE at follow-up extracted using DigitizeIt Mean at 20 weeks (SD) provided by Dr. SE calculated using p-value at 52 weeks and assumed similar to 20 weeks (assumption by Cochrane, Karner et al.) LS Mean at 6 months extracted from figure 2B using DigitizeIt TIO 18 vs. PBO Spirometry, text, p 323 Brusasco 2003 TIO 18 vs. PBO Spirometric tests, text, p 402 Donohue 2002 TIO 18 vs. PBO Text, p 49 Donohue 2010 TIO 18 vs. PBO Table 2, p. 157 GLOW 2 TIO 18 vs. PBO Week 26. Table 2, p 1111 GK14290 version 19 November 13 th, 2014 Page 117 of 178

119 Table 99. Results of the scenario analysis network meta-analysis for trough FEV1 (difference in CFB, ml), at 24 weeks Comparator Intervention Placebo TIO 18 QVA /50 TIO 18 + SAL 50 estimate TIO 18 95% CrI P(better) >99% QVA /50 TIO 18 + SAL 50 UMEC/VI 62.5/25 estimate % CrI P(better) >99% >99% estimate % CrI P(better) >99% 65% 13% estimate % CrI P(better) >99% >99% 84% 92% Figure 65. Forest plot for mean trough FEV1 of UMEC/VI 62.5/25 vs. comparators at 24 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 118 of 178

120 Figure 66. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 24 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 119 of 178

121 6.2. SGRQ total score SGRQ total score at 12 weeks None of the studies comparing an active treatment to a placebo reported on SGRQ total score at 12 weeks, except for the DB study. Although the SHINE study reported a mean difference in SGRQ total score of QVA149 versus placebo, no measure of uncertainty (SE/SD/95%CI) was presented. An imputation of the average SD was performed, however with only two studies in the network (SHINE and DB ) this means that the SD of DB was used for SHINE, assuming the studies were similar enough. The results should be interpreted with caution. The network in the scenario analysis was bigger, although QVA149 was the only LABA/LAMA comparator that could be included in this network. Tashkin 2009 reported the mean difference also without measure of uncertainty. The results of the scenario analysis are first presented without imputations, and followed by the results with the imputation of the SE for SHINE and Tashkin Base case The mean difference in CFB of SGRQ total score at 12 weeks between QVA149 and placebo was , the imputed SE was After this imputation, the mean difference in CFB of UMEC/VI 62.5/25 versus QVA149 is expected to be (95%CrI: -3.57; 2.12). These results should be interpret with caution, the network is small and the imputation therefore only includes the uncertainty in the DB trial. Scenario analysis In the scenario analysis, UMEC/VI 62.5/25 could be indirectly compared with QVA149 by using the trials in the network presented in Figure 70. The results show that UMEC/VI 62.5/25 is expected to be more efficacious than placebo or TIO 18, and similar efficacious to QVA149. The mean difference in CFB versus QVA149 was 1.17 (95%CrI: -0.81; 3.15). The results are presented in Table 103, Figure 71 and Figure 72. Tashkin 2009 and SHINE did not report a standard error for the difference in CFB of TIO 18 + FOR 10 versus TIO 18 and for QVA149 vs. TIO 18 or placebo respectively. These values were imputed using the average SD in the evidence base and the NMA of the scenario analysis was rerun including these studies. Since Tashkin 2009 was the only study comparing TIO 18 + FOR 12 with TIO 18, this analysis allowed for an indirect comparison between UMEC/VI 62.5/25 and TIO 18 + FOR 12. The results show similar efficacy by means of SGRQ total score at 12 weeks. The difference in the results of the comparison between UMEC/VI 62.5/25 and QVA149 is due to the imputation of the SHINE study, which provides direct evidence between QVA149 and placebo. The treatment effect of UMEC/VI 62.5/25 relative to placebo in the DB study was larger than the treatment effect of QVA149 versus placebo in the SHINE study, therefore the inclusion of SHINE had this relatively large impact on the relative efficacy estimates. The other results were in line with the scenario analysis without imputation of the SE. GK14290 version 19 November 13 th, 2014 Page 120 of 178

122 Base case Figure 67. Network of studies for SGRQ total score at 12 weeks (base case). Studies in blue represent studies that report only mean values without SE/SD/CI. Table 100. Individual study results for SGRQ total score at 12 weeks in the base case SGRQ total score (Difference in CFB) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 84. Table 50, p. 136 SHINE QVA149 vs. PBO * Appendix, p. 4 *Imputed value Table 101. Results of the base case network meta-analysis for SGRQ total score (difference in CFB), at 12 weeks with imputation of SE for SHINE Intervention Comparator Placebo QVA /50 estimate QVA /50 95% CrI P(better) >99% estimate UMEC/VI 62.5/25 95% CrI P(better) >99% 69% GK14290 version 19 November 13 th, 2014 Page 121 of 178

123 Figure 68. Forest plot for mean SGRQ total score of UMEC/VI 62.5/25 vs. comparators at 12 weeks (base case, with imputation for SHINE) Figure 69. Mean difference in change from baseline between active treatments and placebo, SGRQ total score at 12 weeks (base case with imputation for SHINE) GK14290 version 19 November 13 th, 2014 Page 122 of 178

124 Scenario analysis Figure 70. Network of studies for SGRQ total score at 12 weeks (scenario analysis) Table 102. Individual study results for SGRQ total score at 12 weeks in the scenario analysis. SGRQ total score (Difference in CFB) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 84. Table 50, p. 136 DB UMEC/VI 62.5/25 vs. TIO Day 84. Table 50, p. 125 DB UMEC/VI 62.5/25 vs. TIO Day 84. Table 51, p. 126 ZEP UMEC/VI 62.5/25 vs. TIO Day 84.Table 45, p. 99 SPARK QVA149 vs. TIO LS mean (SE) extracted from Figure 4a, p.6 SHINE QVA149 vs. PBO * Appendix, p. 4 SHINE TIO 18 vs. PBO * Appendix, p. 4 TIPHON TIO 18 vs. PBO CFB extracted from figure 3a using DigitizeIt Donohue 2010 TIO 18 vs. PBO Table E5, p. 42 GLOW2 TIO 18 vs. PBO Week 12. Table 2, p 1111 Verkindre 2006 TIO 18 vs. PBO Day 84. Table 3, p. 425 GK14290 version 19 November 13 th, 2014 Page 123 of 178

125 Tashkin 2009 TIO 18 + FOR 12 vs. TIO * Table 2, p. 23 *Imputed value Table 103. Results of the scenario analysis network meta-analysis for SGRQ total score (difference in CFB), at 12 weeks Intervention Comparator Placebo TIO 18 QVA /50 estimate TIO 18 95% CrI P(better) >99% estimate QVA /50 95% CrI P(better) >99% >99% estimate UMEC/VI 62.5/25 95% CrI P(better) >99% >99% 12% Figure 71. Forest plot for mean SGRQ total score of UMEC/VI 62.5/25 vs. comparators at 12 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 124 of 178

126 Figure 72. Mean difference in change from baseline between active treatments and placebo, SGRQ total score at 12 weeks (scenario analysis) With imputation for SHINE and Tashkin 2009 Figure 73. Forest plot for mean SGRQ total score of UMEC/VI 62.5/25 vs. comparators at 12 weeks (scenario analysis with imputation of SE for SHINE and Tashkin 2009) GK14290 version 19 November 13 th, 2014 Page 125 of 178

127 SGRQ total score at 24 weeks In the base case scenario, a network could be formed including UMEC/VI 62.5/25, QVA149 and the open dual combination of TIO 18 + FOR 10. The network was very small, only one study could be included for each comparator. The network of studies and the individual study results are presented below. The table with individual study results presents the difference in CFB between the active treatment and placebo and shows that all dual combination therapies are expected to be more efficacious than placebo by means of SGRQ total score at 24 weeks. These data were used as input data in the model. Base case The results of the NMA are presented below, both in graphs and in a table. All interventions are expected to be more efficacious than placebo by means of SGRQ total score at 24 weeks. UMEC/VI 62.5/25 is expected to be numerically more efficacious than QVA149 and the open dual combination of TIO 18 + FOR 10 by means of SGRQ total score at 24 weeks (Table 105). The difference in CFB of UMEC/VI 62.5/25 versus QVA149 was (95%CrI: -5.61; 0.63) and the difference in CFB versus TIO 18 + FOR 10 was (95%CrI: -6.10; 0.95). The probability of UMEC/VI 62.5/25 being the better treatment was 94% versus QVA149 and 92% compared with TIO 18 + FOR 10. These results were not statistically significant and did not exceed the MCID for SGRQ total score. Scenario analysis The results of the scenario analysis NMA are similar with the base case, although less pronounced. UMEC/VI 62.5/25 resulted in statistically significant higher efficacy than placebo and TIO 18 and was expected to be similar to QVA149, TIO 18 + SAL 50 and TIO 18 + FOR 10. The results are presented in tables and graphs in section 128. The results after imputation of a SE for Donohue 2002 (TIO 18 vs. PBO) were very similar to the results of the scenario analysis; differences were only at the second decimal place Base case Figure 74. Network of studies for SGRQ total score at 24 weeks (base case analysis) GK14290 version 19 November 13 th, 2014 Page 126 of 178

128 Table 104. Individual study results for SGRQ total score at 24 weeks in the base case analysis. Author Treatment DB UMEC/VI 62.5/25 vs. PBO SGRQ total score (Difference in CFB) Mean SE Source Week 26. Appendix Table 3, p. 9 SHINE QVA149 vs. PBO Figure 3a using DigitizeIt Vogelmeier 2008 TIO 18 + FOR 10 vs. PBO Day 168. Table 50, p. 136 Table 105. Results of the base case network meta-analysis for SGRQ total score (difference in CFB), at 24 weeks Comparator Intervention Placebo QVA /50 TIO 18 + FOR 10 estimate QVA /50 95% CrI P(better) >99% estimate TIO 18 + FOR 10 95% CrI P(better) % estimate UMEC/VI 62.5/25 95% CrI P(better) >99% 94% 92% Figure 75. Forest plot for mean SGRQ total score of UMEC/VI 62.5/25 vs. comparators at 24 weeks GK14290 version 19 November 13 th, 2014 Page 127 of 178

129 Figure 76. Mean difference in change from baseline between active treatments and placebo, SGRQ total score at 24 weeks Scenario analysis Figure 77. Network of studies for SGRQ total score at 24 weeks (scenario analysis). Studies in blue represent studies that report only mean values without SE/SD/CI. GK14290 version 19 November 13 th, 2014 Page 128 of 178

130 Table 106. Individual study results for SGRQ total score at 24 weeks in the scenario analysis. Author Treatment SGRQ total score (Difference in CFB) Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 168. Table 50, p. 136 DB UMEC/VI 62.5/25 vs. TIO Day 168. Table 50, p. 125 DB UMEC/VI 62.5/25 vs. TIO Day 168. Table 51, p. 126 ZEP UMEC/VI 62.5/25 vs. TIO Day 168. Table 45, p. 99 SHINE QVA149 vs. PBO SHINE TIO 18 vs. PBO SPARK QVA149 vs. TIO Aaron 2007 TIO 18 + SAL 50 vs. TIO Vogelmeier 2008 Vogelmeier 2008 Vogelmeier 2008 Week 26. Appendix Table 3, p. 9 Week 26. Appendix Table 3, p. 9 LS mean (SE) extracted from Figure 4a, p.6 Mean at 20 weeks (SD) provided by Dr. SE calculated using p-value at 52 weeks and assumed similar to 20 weeks (assumption by Cochrane, Karner et al.) TIO 18 + FOR 10 vs. PBO Figure 3a using DigitizeIt TIO 18 + FOR 10 vs. TIO Calculated using extracted numbers from Figure 3a TIO 18 vs. PBO Figure 3a using DigitizeIt TIPHON TIO 18 vs. PBO UPLIFT TIO 18 vs. PBO Brusasco 2003 TIO 18 vs. PBO Change from baseline extracted from figure 3a LS Mean at 6 months extracted from figure 2D using DigitizeIt Change from baseline reported in text. HRQoL, p 401 Donohue 2010 TIO 18 vs. PBO Table E5, p. 42 GLOW2 TIO 18 vs. PBO Week 26. Table 2, p 1111 Donohue 2002 TIO 18 vs. PBO * Text, p. 51 *Imputed value Table 107. Results of the scenario analysis network meta-analysis for SGRQ total score (difference in CFB), at 24 weeks Comparator Intervention Placebo TIO 18 QVA /50 TIO 18 + FOR 10 TIO 18 + SAL 50 estimate TIO 18 95% CrI P(better) >99% GK14290 version 19 November 13 th, 2014 Page 129 of 178

131 QVA /50 TIO 18 + FOR 10 TIO 18 + SAL 50 UMEC/VI 62.5/25 estimate % CrI P(better) >99% >99% estimate % CrI P(better) >99% >99% 4% estimate % CrI P(better) >99% 98% 37% 80% estimate % CrI P(better) >99% >99% 41% 89% 55% Figure 78. Forest plot for mean SGRQ total score of UMEC/VI 62.5/25 vs. comparators at 24 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 130 of 178

132 Figure 79. Mean difference in change from baseline between active treatments and placebo, SGRQ total score at 24 weeks (scenario analysis) With imputation for Donohue 2002 Figure 80. Forest plot for mean SGRQ total score of UMEC/VI 62.5/25 vs. comparators at 24 weeks (scenario analysis with imputation of SE for Donohue 2002) GK14290 version 19 November 13 th, 2014 Page 131 of 178

133 6.3. TDI focal score TDI focal score at 12 weeks Only two studies were identified that could be included in the network that reported on TDI focal score at 12 weeks (DB and SHINE). Both reported a statistically significant difference in CFB for the active treatment versus placebo. The individual trial results are presented in Table 108. Base case The results of the NMA show that UMEC/VI 62.5/25 and QVA149 are expected to be better than placebo by means of TDI focal score at 12 weeks. The probability of being better was over 99% for both treatments. UMEC/VI 62.5/25 is expected to be similar efficacious to QVA149 by means of TDI focal score at 12 weeks. The difference was 0.08 (95%CrI: -0.60; 0.76). This difference was not statistically significant and below the MCID for TDI focal score. The probability that UMEC/VI 62.5/25 is better than QVA149 by means of TDI focal score was 59%. Scenario analysis The results of the scenario analysis for TDI focal score at 12 weeks were in line with the base case results. UMEC/VI 62.5/25 showed improved TDI focal scores at 12 weeks compared with placebo and TIO 18 and this result was statistically significant. UMEC/VI is expected to be similar efficacious to QVA149 and TIO 18 + SAL 50, with mean differences in TDI of (95%CrI: -0.59; 0.38) and 0.39 (95%CrI: -0.43; 1.22) respectively. All studies reported a measure of uncertainty, therefore no imputation was needed Base case Figure 81. Network of studies for TDI focal score at 12 weeks (base case analysis) Table 108. Individual study results for TDI focal score at 12 weeks in the base case analysis. TDI focal score (Difference in TDI) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 84. Table 30, p. 96. SE calculated using 95%CI SHINE QVA149 vs. PBO Extracted from Appendix Figure 4b, p. 21 using DigitizeIt GK14290 version 19 November 13 th, 2014 Page 132 of 178

134 Table 109. Results of the base case network meta-analysis for TDI focal score (difference in TDI), at 12 weeks Intervention Comparator Placebo QVA /50 estimate 1.22 QVA /50 95% CrI P(better) >99% estimate UMEC/VI 62.5/25 95% CrI P(better) >99% 59% Figure 82. Forest plot for mean TDI focal score of UMEC/VI 62.5/25 vs. comparators at 12 weeks Figure 83. Mean difference in between active treatments and placebo, TDI focal score at 24 weeks GK14290 version 19 November 13 th, 2014 Page 133 of 178

135 Scenario analysis Figure 84. Network of studies for TDI focal score at 12 weeks (scenario analysis) Table 110. Individual study results for TDI focal score at 12 weeks in the scenario analysis. TDI focal score (Difference in TDI) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 84. Table 30, p. 96 DB UMEC/VI 62.5/25 vs. TIO Day 84. Table 33, p. 95 DB UMEC/VI 62.5/25 vs. TIO Day 84. Table 33, p. 97 SHINE QVA149 vs. PBO SHINE TIO 18 vs. PBO Extracted from Appendix Figure 4b, p. 21 using DigitizeIt Extracted from Appendix Figure 4b, p. 21 using DigitizeIt Tashkin 2009 TIO 18 + FOR 10 vs. TIO Table2 p23 and text p. 22 Casaburi 2002 TIO 18 vs. PBO TDI focal score at 3 months extracted from Figure 3 using DigitizeIt Donohue 2010 TIO 18 vs. PBO Table E3, p. 41 GK14290 version 19 November 13 th, 2014 Page 134 of 178

136 GLOW 2 TIO 18 vs. PBO Week 12. Table 2, p 1111 Verkindre 2006 TIO 18 vs. PBO Day 84. Table 3, p. 425 Table 111. Results of the scenario analysis network meta-analysis for TDI focal score (difference in TDI), at 12 weeks Intervention Comparator Placebo TIO 18 QVA /50 TIO 18 + FOR 12 estimate 0.76 TIO 18 95% CrI P(better) >99% estimate QVA /50 95% CrI P(better) >99% >99% estimate TIO 18 + FOR 12 95% CrI P(better) % 13% estimate UMEC/VI 62.5/25 95% CrI P(better) >99% >99% 34% 83% GK14290 version 19 November 13 th, 2014 Page 135 of 178

137 Figure 85. Forest plot for mean TDI focal score of UMEC/VI 62.5/25 vs. comparators at 12 weeks (scenario analysis) Figure 86. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 12 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 136 of 178

138 TDI focal score at 24 weeks The same studies that reported TDI focal score at 12 weeks also presented data at the 24 week endpoint: DB and SHINE. As a result, UMEC/VI could only be compared to QVA149 in the base case analysis. The individual trial results are presented below. Base case The results of the NMA show the difference in TDI focal score at 24 weeks including 95%CrI and the probability of being the better treatment. Both QVA149 and UMEC/VI 62.5/25 are expected to be more efficacious than placebo by means of TDI focal score at 24 weeks (Table 113). UMEC/VI 62.5/25 resulted in a similar TDI focal score with QVA149 at 24 weeks. The difference was expected to be 0.11 (95%CrI -0.58; 0.80). The difference between UMEC/VI 62.5/25 and QVA149 was not statistically significant and did not exceed the MCID for TDI focal score. Scenario analysis In the scenario analysis, UMEC/VI 62.5/25 demonstrated favourable results compared with placebo and similar efficacy versus TIO 18, QVA149 and TIO 18 + SAL 50. The treatment effect versus placebo was statistically significant, but not against the other comparators. Donohue 2002 reported a mean difference in CFB of TIO 18 vs. placebo of 1.02, without reporting a SE, SD or 95%CI. This was imputed using the trials in the network and the NMA was updated, however this did not have a large impact on the results. The results of the NMA in the scenario analysis are presented in a forest plot in Figure Base case Figure 87. Network of studies for TDI focal score at 24 weeks (base case analysis) Table 112. Individual study results for TDI focal score at 24 weeks in the base case analysis. TDI focal score (Difference in TDI) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 84. Table 30, p. 96. SE calculated using 95%CI SHINE QVA149 vs. PBO Extracted from Appendix Figure 4b, p. 21 using DigitizeIt GK14290 version 19 November 13 th, 2014 Page 137 of 178

139 Table 113. Results of the base case network meta-analysis for TDI focal score (difference in TDI), at 24 weeks Intervention Comparator Placebo QVA /50 estimate 1.09 QVA /50 95% CrI P(better) >99% estimate UMEC/VI 62.5/25 95% CrI P(better) >99% 62% Figure 88. Forest plot for mean TDI focal score of UMEC/VI 62.5/25 vs. comparators at 24 weeks GK14290 version 19 November 13 th, 2014 Page 138 of 178

140 Figure 89. Mean difference between active treatments and placebo, TDI focal score at 24 weeks Scenario analysis Figure 90. Network of studies for TDI focal score at 24 weeks (scenario analysis). Studies in blue represent studies that report only mean values without SE/SD/CI. GK14290 version 19 November 13 th, 2014 Page 139 of 178

141 Table 114. Individual study results for TDI focal score at 24 weeks in the scenario analysis. Author Treatment TDI focal score (Difference in TDI) Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Day 168. Table 30, p. 96 DB UMEC/VI 62.5/25 vs. TIO Day 168. Table 33, p. 96 DB UMEC/VI 62.5/25 vs. TIO Day 168. Table 33, p. 97 SHINE QVA149 vs. PBO Week 26. Appendix Table 3, p. 9 SHINE TIO 18 vs. PBO Week 26. Appendix Table 3, p. 9 Aaron 2007 TIO 18 + SAL 50 vs. TIO Mean TDI at 20 weeks (SD) provided by Dr. Brusasco 2003 TIO 18 vs. PBO Text, p. 402 Casaburi 2002 TIO 18 vs. PBO TDI focal score at 6 months extracted from Figure 3 using DigitizeIt Donohue 2010 TIO 18 vs. PBO Table E3, p. 41 GLOW 2 TIO 18 vs. PBO Week 26. Table 2, p 1111 Donohue 2002 TIO 18 vs. PBO * Text, p. 50 *Imputed value Table 115. Results of the scenario analysis network meta-analysis for TDI focal score (difference in TDI), at 24 weeks Intervention Comparator Placebo TIO 18 QVA /50 TIO 18 + SAL 50 estimate 0.87 TIO 18 95% CrI P(better) >99% QVA /50 TIO 18 + SAL 50 UMEC/VI 62.5/25 estimate % CrI P(better) >99% >99% estimate % CrI P(better) % 2% estimate % CrI P(better) >99% 82% 9% 90% GK14290 version 19 November 13 th, 2014 Page 140 of 178

142 Figure 91. Forest plot for mean TDI focal score of UMEC/VI 62.5/25 vs. comparators at 24 weeks (scenario analysis) Figure 92. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 12 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 141 of 178

143 With imputation for Donohue 2002 Figure 93. Forest plot for mean TDI focal score of UMEC/VI 62.5/25 vs. comparators at 24 weeks (scenario analysis with imputation for SE of Donohue 2002) GK14290 version 19 November 13 th, 2014 Page 142 of 178

144 6.4. Rescue medication use Rescue medication use at 12 weeks Only one study reported the rescue medication use at 12 weeks of a LABA/LAMA treatment versus placebo (DB ). None of the other studies compared to placebo or reported data for this outcome. Therefore, no network could be formed in the base case analysis, however there were studies comparing a LABA/LAMA combination therapy with TIO 18. UMEC/VI 62.5/25 could be compared to IND TIO 18 by means of rescue medication use at 12 weeks in the scenario analysis. Scenario analysis UMEC/VI 62.5/25 is expected to be more efficacious than placebo and TIO 18 by means of rescue medication use at 12 weeks. The mean difference in change from baseline in the number of puffs rescue medication per day was (95%CrI: -1.31; -0.55) versus placebo and puffs/day (95%CrI: -1.08; -0.35) versus TIO 18. UMEC/VI 62.5/25 showed similar efficacy compared with IND TIO 18, with a mean difference in CFB of 0.15 puffs/day (95%CrI: -0.28; 0.58). After imputation of the SE for Tashkin 2009 it was possible to estimate the relative efficacy of UMEC/VI 62.5/25 versus TIO 18 + FOR 12. The mean difference in CFB was 0.13 puffs per day (95%CrI: -0.65; 0.90) Base case No network available Scenario analysis Figure 94. Network of studies for rescue medication use at 12 weeks (scenario analysis). Studies in blue represent studies that report only mean values without SE/SD/CI. GK14290 version 19 November 13 th, 2014 Page 143 of 178

145 Table 116. Individual study results for rescue medication use (puffs/day) at 12 weeks in the scenario analysis. Author Treatment DB UMEC/VI 62.5/25 vs. PBO DB UMEC/VI 62.5/25 vs. TIO DB UMEC/VI 62.5/25 vs. TIO ZEP UMEC/VI 62.5/25 vs. TIO Rescue medication use (Difference in CFB) Mean SE Source Week 1-12, additional rescue medication analyses. Table 26, p. 2 Week Table 28 of additional rescue medication analyses, p. 1 Week Table 27 of additional rescue medication analyses, p. 1 Additional analysis rescue use 1-12 weeks INTRUST-1 IND TIO 18 vs. TIO Mahler 2012, Table 3, p.786 INTRUST-2 IND TIO 18 vs. TIO Mahler 2012, Table 3, p.786 Verkindre 2005 TIO 18 vs. PBO Rescue medication reported per group in text, p. 424 Tashkin 2009 TIO 18 + FOR 12 vs. TIO * Table 2, p. 23 *Imputed value Table 117. Results of the scenario analysis network meta-analysis for rescue medication use (difference in CFB), at 12 weeks Comparator Intervention Placebo TIO 18 IND TIO 18 TIO 18 IND TIO 18 UMEC/VI 62.5/25 estimate % CrI P(better) 0.86 estimate % CrI P(better) >99% >99% estimate % CrI P(better) >99% >99% 25% GK14290 version 19 November 13 th, 2014 Page 144 of 178

146 Figure 95. Forest plot for mean rescue medication use of UMEC/VI 62.5/25 vs. comparators at 12 weeks (scenario analysis) Figure 96. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 12 weeks (scenario analysis) With imputation for Tashkin 2009 GK14290 version 19 November 13 th, 2014 Page 145 of 178

147 Figure 97. Forest plot for mean rescue medication use of UMEC/VI 62.5/25 vs. comparators at 12 weeks (scenario analysis with imputation of SE for Tashkin 2009) GK14290 version 19 November 13 th, 2014 Page 146 of 178

148 Rescue medication use at 24 weeks Only two studies could be included in the base case analysis for rescue medication use at 24 weeks. SHINE reported the difference in CFB of QVA149 versus placebo and DB reported the difference in CFB of UMEC/VI 62.5/25 versus placebo. The network of studies was bigger in the scenario analysis, although QVA149 was still the only other combination therapy in the network next to UMEC/VI. The network and results of these trials are presented below. Base case The results of the base case indicate that UMEC/VI 62.5/25 is expected to be more efficacious than placebo and similar efficacious compared with QVA149. The mean difference of UMEC/VI 62.5/25 versus QVA149 was 0.16 puffs per day (95%CrI: -0.45; 0.77). Scenario analysis The results of the scenario analysis are in line with the base case results and demonstrate favourable efficacy for UMEC/VI 62.5/25 versus placebo and similar efficacy versus QVA149. In addition, UMEC/VI 62.5/25 demonstrated to be more efficacious than TIO 18 by means of rescue medication use at 24 weeks. Only very minor changes in the point estimates were found after the imputation of a SE for Donohue Base case Figure 98. Network of studies for rescue medication use at 24 weeks (base case analysis) Table 118. Individual study results for rescue medication use (puffs/day) at 24 weeks in the base case analysis. Rescue medication use (Difference in CFB) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Week 26. Appendix Table 3, p. 9 SHINE QVA149 vs. PBO Week Table 42, p. 126 Table 119. Results of the base case network meta-analysis for rescue medication use (difference in CFB), at 24 weeks Comparator Intervention Placebo QVA /50 estimate QVA /50 95% CrI GK14290 version 19 November 13 th, 2014 Page 147 of 178

149 UMEC/VI 62.5/25 P(better) >99% estimate % CrI P(better) >99% 30% Figure 99. Forest plot for mean rescue medication use of UMEC/VI 62.5/25 vs. comparators at 24 weeks Figure 100. Mean difference in change from baseline between active treatments and placebo, rescue medication use at 24 weeks GK14290 version 19 November 13 th, 2014 Page 148 of 178

150 Scenario analysis Figure 101. Network of studies rescue medication use at 24 weeks (scenario analysis). Studies in blue represent studies that report only mean values without SE/SD/CI. Table 120. Individual study results for rescue medication use (puffs/day) at 24 weeks in the scenario analysis. Rescue medication use (Difference vs. PBO) Author Treatment Mean SE Source DB UMEC/VI 62.5/25 vs. PBO Week Table 42, p. 126 DB UMEC/VI 62.5/25 vs. TIO Week Table 42, p. 115 DB UMEC/VI 62.5/25 vs. TIO Week Table 42, p. 116 ZEP UMEC/VI 62.5/25 vs. TIO Weeks Table 38, p. 92 SHINE QVA149 vs. PBO Week 26. Appendix Table 3, p. 9 SHINE TIO 18 vs. PBO Week 26. Appendix Table 3, p. 9 Donohue 2010 TIO 18 vs. PBO weeks. Table 3, p. 25 Donohue 2002 TIO 18 vs. PBO * Text, p. 52 *Imputed value GK14290 version 19 November 13 th, 2014 Page 149 of 178

151 Table 121. Results of the scenario analysis network meta-analysis for rescue medication use (difference in CFB), at 24 weeks Comparator Intervention Placebo TIO 18 QVA /50 TIO 18 QVA /50 UMEC/VI 62.5/25 estimate % CrI P(better) >99% estimate % CrI P(better) >99% >99% estimate % CrI P(better) >99% >99% 43% Figure 102. Forest plot for mean rescue medication use of UMEC/VI 62.5/25 vs. comparators at 24 weeks (scenario analysis) GK14290 version 19 November 13 th, 2014 Page 150 of 178

152 Figure 103. Mean difference in change from baseline between active treatments and placebo, trough FEV1 (ml) at 12 weeks (scenario analysis) With imputation for Donohue 2002 Figure 104. Forest plot for mean rescue medication use of UMEC/VI 62.5/25 vs. comparators at 24 weeks (scenario analysis with imputation of SE for Donohue 2002) GK14290 version 19 November 13 th, 2014 Page 151 of 178

153 7. Discussion and Conclusions The aim of this study was to evaluate the relative efficacy of UMEC/VI 62.5/25 versus IND150 + TIO18, TIO+SAL, TIO+FOR, FLUT/SAL 500/50 + TIO18 and QVA149 in patients with COPD eligible to receive COPD maintenance therapy. The evidence synthesis was performed both within a frequentist and Bayesian framework. The frequentist indirect treatment comparison was required for PBAC in Australia, while the Bayesian network meta-analysis was required for SMC in Scotland. The primary objective was to assess the relative efficacy by means of trough FEV 1 at 12 and 24 weeks for the comparison of UMEC/VI 62.5/25 with IND + TIO, FLUT/SAL + TIO, QVA149. The secondary objective was to assess the relative efficacy of UMEC/VI 62.5/25 versus TIO+FOR, TIO+SAL and QVA149 by means of SGRQ total score, TDI focal score and use of rescue medication at 12 and 24 weeks. A frequentist indirect treatment comparison (required for PBAC in Australia) was performed. Differences were observed between the SPARK study (Wedzicha 2013) and other studies. SPARK included a more severe patient population than other studies. To evaluate the extent of the effect this difference in patient characteristics had on the relative effect estimates of UMEC/VI 62.5/25 vs. QVA149, two scenarios were performed for each outcome, i.e. including and excluding the SPARK study. Conclusions of the frequentist indirect treatment comparison UMEC/VI 62.5/25 is comparable to IND 150 mcg + TIO 18 mcg OD and to TIO 18 mcg OD + FOR 12 mcg BID by means of trough FEV 1 at 12 weeks, and at 24 weeks also versus TIO 18 mcg OD + SAL 50 mcg BID. In addition, UMEC/VI 62.5/25 showed comparable results to TIO + FOR and TIO + SAL by means of SGRQ total score at 24 weeks, and TDI focal score at 12 weeks and 24 weeks respectively. Regarding trough FEV 1 at 12 weeks UMEC/VI 62.5/25 showed statistically significant better results with FLUT/SAL + TIO. At 24 weeks this effect was diminished, UMEC/VI 62.5/25 showed comparable results with FLUT/SAL + TIO by means of trough FEV 1 at that time point. Furthermore, UMEC/VI 62.5/25 showed comparable results to QVA149 with regard to trough FEV1 at 12 and 24 weeks, SGRQ total score at 12 and 24 weeks, TDI focal score at 12 and 24 weeks and rescue medication use at 24 weeks. The SPARK study reported on trough FEV 1 and SGRQ total score at 12 and 24 weeks, therefore scenarios in- and excluding this study were developed for these outcomes. Including SPARK did not impact the conclusions; however differences between the relative efficacy estimates were detected, mostly for trough FEV 1 outcomes. Regarding trough FEV 1 at 12 weeks, the exclusion of SPARK decreased the relative efficacy estimate of UMEC/VI 62.5/25 vs. QVA149 from 2.39 ml to ml. Since both 95% confidence intervals included zero, the relative efficacy of UMEC/VI 62.5/25 and QVA149 can be considered similar by means of trough FEV 1 at 12 weeks. Patients in SPARK had less benefit from QVA149 over tiotropium by means of trough FEV 1 (50 ml vs. 100 ml in SHINE), which influenced the relative efficacy of UMEC/VI 62.5/25 vs. QVA149. GK14290 version 19 November 13 th, 2014 Page 152 of 178

154 Including SPARK only had a minor effect on the relative efficacy estimates by means of SGRQ total score at 12 and 24 weeks. In addition, a retrospective calculation of the meta-analysis and ITC power was performed. It should be noted that there are currently no sound recommendations and clear guidelines from the HTA authorities on the ITC and NMA power calculations. The results should therefore be interpreted with caution. Conclusions of the Bayesian network meta-analysis The main objective of the Bayesian NMA was to indirectly compare UMEC/VI 62.5/25 to other open dual or closed LABA/LAMA combination therapies. Therefore, studies were selected comparing a LABA/LAMA intervention of interest to another LABA/LAMA intervention of interest or to placebo. However, because many studies using TIO 18 as comparator arm instead of placebo, a scenario analysis was performed with studies comparing a LABA/LAMA intervention of interest to another LABA/LAMA intervention of interest or to TIO 18 or placebo. The networks in the scenario analysis consisted of many more studies and allowed for indirect comparisons of UMEC/VI 62.5/25 versus LABA/LAMA treatments that could not be included in the base case networks. In the base case, UMEC/VI 62.5/25 demonstrated higher efficacy than placebo for every outcome analyzed. In addition, UMEC/VI 62.5/25 showed similar efficacy compared with QVA149 by means of trough FEV1 at 12 and 24 weeks, SGRQ total score at 24 weeks, TDI focal score at 12 and 24 weeks and rescue medication use at 24 weeks. UMEC/VI 62.5/25 demonstrated also similar efficacy versus TIO 18 + FOR 10 by means of SGRQ total score at 24 weeks. The scenario analysis showed similar results for these comparators. More studies could be included in the scenario analysis for comparing a LABA/LAMA treatment to TIO 18. UMEC/VI 62.5/25 is expected to be more efficacious than TIO 18 on all available endpoints except for TDI focal score at 24 weeks. In addition, UMEC/VI demonstrated similar efficacy compared with IND TIO 18 by means of trough FEV1 and rescue medication use at 12 weeks, comparable efficacy relative to TIO 18 + SAL 50 by means of trough FEV1, SGRQ total score and TDI focal score at 24 weeks and to TIO 18 + FOR 12 by means of trough FEV1 and TDI focal score at 12 weeks. Potential limitations A limitation of the analysis was the potential influence of confounders. A meta-regression to adjust for potential confounding was not feasible due to the low number (less than 10) of included studies in each meta-analysis. Heterogeneity was expected in the analysis for UMEC/VI vs. QVA149, because of the more severe patient population of the SPARK study. A separate scenario analysis was performed to test the impact of this heterogeneity on the relative efficacy estimates. Excluding SPARK had only a marginal impact of the effect estimates (changing the estimated mean difference in change from baseline without changing the conclusions) and the severe COPD patient population of SPARK is therefore not believed to be a likely source of bias in the analysis. GK14290 version 19 November 13 th, 2014 Page 153 of 178

155 Although the results of the separate scenarios analysis did not differ substantially, it was not possible to evaluate or adjust for all potential effect modifiers because of the low number of included studies. GSK trials were highly homogeneous, but differences exist between GSK trials and other studies. Even though the studies were similar enough to be included in a meta-analysis and indirect treatment comparison, there may still be residual confounding in these aggregated data. Conclusion The results of Bayesian NMA were in line with those from the Frequentist ITC. In both cases, UMEC/VI demonstrated similar efficacy compared with QVA149, IND TIO 18, TIO 18 + SAL 50, TIO 18 + FOR 10 and TIO 18 + FOR 12 on available efficacy endpoints. Within both frameworks, the scenario analyses did not have a major impact on these conclusions. GK14290 version 19 November 13 th, 2014 Page 154 of 178

156 8. Summary tables Bayesian NMA Table 122. Comparative efficacy of UMEC/VI 62.5/25 vs. comparators in the NMA (base case without imputations) UMEC/VI 62.5/25 OD vs. Placebo Mean (95% CrI) TIO 18 Mean (95% CrI) QVA149 Mean (95% CrI) IND TIO 18 Mean (95% CrI) TIO 18 + SAL 50 Mean (95% CrI) TIO 18 + FOR 12 Mean (95% CrI) TIO 18 + FOR 10 Mean (95% CrI) Trough FEV 1 at 12 weeks (DCFB, ml) (159.50, ) NA (-65.49, 27.20) NA NA NA NA Trough FEV 1 at 24 weeks (DCFB, ml) SGRQ total score at 12 weeks (DCFB) SGRQ total score at 24 weeks (DCFB) TDI focal score at 12 weeks (difference) TDI focal score at 24 weeks (difference) Rescue medication use at 12 weeks (DCFB) Rescue medication use at 24 weeks (DCFB) (127.00, ) NA (-73.57, 26.78) NA NA NA NA NA NA NA NA NA NA NA (-7.87, -3.14) 1.30 (0.85, 1.75) 1.20 (0.69, 1.71) NA NA NA (-5.61, 0.63) 0.08 (-0.60, 0.76) 0.11 (-0.58, 0.80) NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA NA (-1.31, -0.29) NA 0.16 (-0.45, 0.77) NA NA NA NA (-6.10, 0.95) GK14290 version 19 November 13 th, 2014 Page 155 of 178

157 Table 123. Comparative efficacy of UMEC/VI 62.5/25 vs. comparators in the NMA (scenario analysis including TIO as comparator, no imputation of SE) UMEC/VI 62.5/25 OD vs. Placebo Mean (95% CrI) TIO 18 Mean (95% CrI) QVA149 Mean (95% CrI) IND TIO 18 Mean (95% CrI) TIO 18 + SAL 50 Mean (95% CrI) TIO 18 + FOR 12 Mean (95% CrI) TIO 18 + FOR 10 Mean (95% CrI) Trough FEV 1 at 12 weeks (DCFB, ml) (187.90, ) (74.53, ) 9.41 (-16.96, 35.77) (-4.90, 44.40) NA 3.86 (-54.14, 62.18) NA Trough FEV 1 at 24 weeks (DCFB, ml) (169.70, ) (65.74, ) (-14.18, 42.25) NA (-25.25, ) NA NA SGRQ total score at 12 weeks (DCFB) SGRQ total score at 24 weeks (DCFB) (-5.58, -3.11) (-5.23, -2.95) (-2.81, -0.84) (-2.66, -0.52) 1.17 (-0.81, 3.15) 0.18 (-1.28, 1.63) NA NA NA NA NA (-1.84, 1.61) NA (-1.77, 0.39) TDI focal score at 12 weeks (difference) 1.22 (0.90, 1.54) 0.46 (0.16, 0.77) (-0.59, 0.38) NA NA 0.39 (-0.43, 1.22) NA TDI focal score at 24 weeks (difference) 1.04 (0.68, 1.39) 0.16 (-0.19, 0.52) (-0.73, 0.13) NA 0.58 (-0.33, 1.50) NA NA Rescue medication use at 12 weeks (DCFB) Rescue medication use at 24 weeks (DCFB) (-1.31, -0.55) (-1.25, -0.69) (-1.08, -0.35) (-0.71, -0.31) NA 0.02 (-0.27, 0.32) 0.15 (-0.28, 0.58) NA NA NA NA NA NA NA GK14290 version 19 November 13 th, 2014 Page 156 of 178

158 9. Summary tables frequentist ITC Table 124. Comparative efficacy of UMEC/VI 62.5/25 vs. IND + TIO Difference in change from baseline Trough FEV 1 at 12 weeks (-2.11; 52.30) Rescue medication use at 12 weeks 0.36 (-0.07; 0.79) Table 125. Comparative efficacy of UMEC/VI 62.5/25 vs. TIO + SAL Difference in change from baseline Trough FEV 1 at 24 weeks (-20.56, ) SGRQ total score at 24 weeks 0.62 (-1.63, 2.87) TDI focal score at 24 weeks 0.45 (-0.51, 1.41) Table 126. Comparative efficacy of UMEC/VI 62.5/25 vs. TIO + FOR Difference in change from baseline Trough FEV 1 at 12 weeks 9.00 (-50.34; 68.34) SGRQ total score at 24 weeks (-4.10; 1.20) TDI focal score at 12 weeks 0.35 (-0.55; 1.25) Table 127. Comparative efficacy of UMEC/VI 62.5/25 vs. FLUT/SAL 500/50 + TIO18 Difference in change from baseline Trough FEV 1 at 12 weeks (12.80; 95.19) Trough FEV 1 at 24 weeks (-46.20; 88.55) GK14290 version 19 November 13 th, 2014 Page 157 of 178

159 Table 128. Comparative efficacy of UMEC/VI 62.5/25 vs. QVA149 Individual trial results (including SPARK) Individual trial results (excluding SPARK) Trough FEV 1 at 12 weeks (-22.51; 43.41) (-35.96; 31.51) Trough FEV 1 at 24 weeks 6.99 (-24.05; 38.02) 2.52 (-30.25; 35.28) Trough FEV 1 at 52 weeks No network No network SGRQ total score at 12 weeks 1.27 (-0.77; 3.31) No network SGRQ total score at 24 weeks No meta-analysis performed No meta-analysis performed SGRQ total score at 52 weeks No network No network TDI focal score at 12 weeks n.a (-0.54; 0.40) TDI focal score at 24 weeks n.a (-0.80; 0.34) TDI focal score at 52 weeks No network No network Rescue medication use at 12 weeks No network No network Rescue medication use at 24 weeks n.a (-0.24; 0.33) Rescue medication use at 52 weeks No network No network Values are reported as difference in change from baseline. n.a. SPARK did not report data on this outcome GK14290 version 19 November 13 th, 2014 Page 158 of 178

160 Appendix 1 Bibliographic Search Strategies Table 129: Search strategy for the systematic review: MEDLINE and MEDLINE in-process Database Platform MEDLINE(R) In-Process & Other Non-Indexed Citations and MEDLINE(R) Ovid Date of search Time limits 1946 to 2014 week 15 Filters Line 6-13 are from the search filter: BMJ Clinical Evidence Strategy (MEDLINE randomized controlled trials strategy using Ovid). Available from: (accessed on ) # Searches Results 1 (formoterol or eformoterol or foradil or oxis or atimos modulite or atock or perforomist or salmeterol or serevent or tiotropium or spiriva or Ba 679 BR or indacaterol or onbrez or arcapta or NVA-237 or NVA237 or (NVA adj "237") or glycopyrronium bromide or glycopyrrolate or seebri or enurev breezhaler or aclidinium bromide or tudorza pressair or eklira genuair or symbicort or advair or seretide or olodaterol or striverdi or umeclidinium or GSK or vilanterol or GW or QVA149 or relovair or zephyr or anoro ellipta).ti,ab,nm. 2 exp Pulmonary Disease, Chronic Obstructive/ or exp Chronic obstructive lung disease/ 3 (COPD or chronic obstructive pulmonary disease or COAD or chronic obstructive airway disease or chronic obstructive lung disease or chronic bronchitis or emphysema).ti,ab or and "randomized controlled trial".pt (random$ or placebo$ or single blind$ or double blind$ or triple blind$).ti,ab (retraction of publication or retracted publication).pt or 7 or (animals not humans).sh ((comment or editorial or meta-analysis or practice-guideline or review or letter or journal correspondence) not "randomized controlled trial").pt. 12 (random sampl$ or random digit$ or random effect$ or random survey or random regression).ti,ab. not "randomized controlled trial".pt not (10 or 11 or 12) and limit 14 to (English or German) 610 ab, nm, pt, sh, ti: searches performed in abstract, name of substance, publication type, subject heading and title fields, respectively GK14290 version 19 November 13 th, 2014 Page 159 of 178

161 Table 130: Search strategy for the systematic review: EMBASE Database Platform EMBASE Ovid Date of search Time limits 1988 to 2014 week 15 Filters Line 6-12 are from the search filter: BMJ Clinical Evidence Strategy (EMBASE randomized controlled trials strategy using Ovid). Available from: (accessed on ) # Searches Results 1 (formoterol or eformoterol or foradil or oxis or atimos modulite or atock or perforomist or salmeterol or serevent or tiotropium or spiriva or Ba 679 BR or indacaterol or onbrez or arcapta or NVA-237 or NVA237 or (NVA adj "237") or glycopyrronium bromide or glycopyrrolate or seebri or enurev breezhaler or aclidinium bromide or tudorza pressair or eklira genuair or symbicort or advair or seretide or olodaterol or striverdi or umeclidinium or GSK or vilanterol or GW or QVA149 or relovair or zephyr or anoro ellipta).ti,ab. 2 exp Pulmonary Disease, Chronic Obstructive/ or exp Chronic obstructive lung disease/ 3 (COPD or chronic obstructive pulmonary disease or COAD or chronic obstructive airway disease or chronic obstructive lung disease or chronic bronchitis or emphysema).ti,ab or and (random$ or placebo$ or single blind$ or double blind$ or triple blind$).ti,ab RETRACTED ARTICLE/ or (animal$ not human$).sh,hw (book or conference paper or editorial or letter or review).pt. not exp randomized controlled trial/ 11 (random sampl$ or random digit$ or random effect$ or random survey or random regression).ti,ab. not exp randomized controlled trial/ not (9 or 10 or 11) and limit 13 to (English or German) 881 Table 131: Search strategy for the systematic review: Cochrane Library - CDSR, CENTRAL Database CENTRAL and CDSR Platform Cochrane Date of search Time limits 1988 to 2014 Filters n.a. # Searches Results GK14290 version 19 November 13 th, 2014 Page 160 of 178

162 1 (formoterol or eformoterol or foradil or oxis or atimos modulite or atock or perforomist or salmeterol or serevent or tiotropium or spiriva or Ba 679 BR or indacaterol or onbrez or arcapta or NVA-237 or NVA237 or (NVA near/3 237) or glycopyrronium bromide or glycopyrrolate or seebri or enurev breezhaler or aclidinium bromide or tudorza pressair or eklira genuair or symbicort or advair or seretide or olodaterol or striverdi or umeclidinium or GSK or vilanterol or GW or QVA149 or relovair or zephyr or anoro ellipta):ti,ab,kw MeSH descriptor: [Pulmonary Disease, Chronic Obstructive] explode all trees (COPD or chronic obstructive pulmonary disease or COAD or chronic obstructive airway disease or chronic obstructive lung disease or chronic bronchitis or emphysema):ti,ab,kw #1 and (#2 or #3) in Trials #1 and (#2 or #3) (in Cochrane Reviews [Reviews and Protocols]) 21 ab, kw, ti: searches performed in abstract, keyword and title fields, respectively; n.a: not applicable Line 4 is corresponds to the CENTRAL database, line 5 to the CDSR database. Both results were exported. Table 132: Search strategy for the systematic review: DARE Database Database DARE Platform Date of search Time limits Filters CRD ( No time limits n.a. # Searches Results 1 (formoterol or eformoterol or foradil or oxis or atimos modulite or atock or perforomist or salmeterol or serevent or tiotropium or spiriva or Ba 679 BR or indacaterol or onbrez or arcapta or NVA-237 or NVA237 or (NVA and "237") or glycopyrronium bromide or glycopyrrolate or seebri or enurev breezhaler or aclidinium bromide or tudorza pressair or eklira genuair or symbicort or advair or seretide or olodaterol or striverdi or umeclidinium or GSK or vilanterol or GW or QVA149 or relovair or zephyr or anoro ellipta) [ANY FIELD] 2 (COPD or chronic obstructive pulmonary disease or COAD or chronic obstructive airway disease or chronic obstructive lung disease or chronic bronchitis or emphysema) [ANY FIELD] and 2 in DARE 62 n.a.: not applicable Table 133: Search strategy for the systematic review: Health Technology Assessment Database Database HTA Platform CRD ( Date of search Time limits No time limits Filters n.a. # Searches Results 1 (formoterol or eformoterol or foradil or oxis or atimos modulite or atock or perforomist or salmeterol or serevent or tiotropium or spiriva or Ba 679 BR or indacaterol or onbrez or arcapta or NVA-237 or NVA237 or (NVA and "237") or glycopyrronium bromide or glycopyrrolate or seebri or enurev breezhaler or 226 GK14290 version 19 November 13 th, 2014 Page 161 of 178

163 aclidinium bromide or tudorza pressair or eklira genuair or symbicort or advair or seretide or olodaterol or striverdi or umeclidinium or GSK or vilanterol or GW or QVA149 or relovair or zephyr or anoro ellipta) [ANY FIELD] 2 (COPD or chronic obstructive pulmonary disease or COAD or chronic obstructive airway disease or chronic obstructive lung disease or chronic bronchitis or emphysema) [ANY FIELD] 3 1 and 2 in HTA HTA: HTA in progress and HTA published GK14290 version 19 November 13 th, 2014 Page 162 of 178

164 Appendix 2 Registry Search Strategies Table 134: Search strategy for Clinicaltrials.gov Trial registry clinicaltrials.gov URL Date of search Search strategy COPD OR COAD OR Chronic obstructive pulmonary disease OR Chronic obstructive lung disease OR chronic obstructive airway disease OR chronic bronchitis OR emphysema Phase 2, 3, 4 Results 949 Table 135: Search strategy for World Health Organization International Clinical Trials Registry Platform Trial registry URL Date of search WHO International Clinical Trials Registry Platform (ICTRP) Search strategy COPD OR chronic obstructive pulmonary disease OR COAD OR chronic obstructive airway disease OR chronic obstructive lung disease OR chronic bronchitis OR emphysema Results 3852 records for 2922 trials found* * The WHO ICTRP imports records from several registries. Trials are sometimes recorded in more than one registry. These records can refer to each other using a secondary identification number. The search portal uses this secondary identification number to group records about the same trial together in the search results. All results were reported in an excel database. However WHO ICTRP also collects data from Asian registries. As noncaucasian population is an exclusion criterion, trials listed on national non-caucasian registries were excluded for population not of interest. (i.e. Chinese Clinical Trial Registry; Clinical Trials Registry India; Iranian Registry of Clinical Trials; Japan Primary Registries Network) Table 136: Search strategy for Current Controlled Trials Trial registry Current controlled trials URL Date of search Search strategy (COPD or chronic obstructive pulmonary disease or COAD or chronic obstructive airway disease or chronic obstructive lung disease or chronic bronchitis or emphysema) in Databases: ISRCTN Register (International) - copy of ISRCTN Register; Action Medical Research (UK) - subset from ISRCTN Register; The Wellcome Trust (UK) - subset from ISRCTN Register; UK trials (UK) - subset from ISRCTN Register, UK trials only Results 87 ClinicalTrials.gov was removed from the list of resources searched in this aggregated database, as clinicaltrials.gov was searched directly in a separate search. Table 137: Search strategy for EU Clinical trials register Trial registry EU Clinical Trials Register (EU-CTR) URL Date of search Search strategy (COPD OR chronic obstructive pulmonary disease OR COAD OR chronic obstructive airway disease OR chronic obstructive lung disease OR chronic bronchitis OR emphysema) AND (Phase II OR Phase III or Phase IV [Select GK14290 version 19 November 13 th, 2014 Page 163 of 178

165 Results 307 trial phase]) Table 138: Search strategy for Klinische Prüfungen PharmNet.Bund Trial registry Klinische Prüfungen PharmNet.Bund URL Date of search Search strategy COPD in Textfelder AND Limit to phase II or phase III or phase IV [Trial phase] AND Limit to therapy or safety or efficacy [Trial scope] AND Limit to patients [Trial population] Results 320 Table 139: Search strategy for International Prospective Register of Systematic Reviews* Trial registry International Prospective Register of Systematic Reviews (PROSPERO) URL Date of search Search strategy Separate searches for: COPD [ALL FIELDS] or chronic obstructive pulmonary disease [ALL FIELDS] or COAD [ALL FIELDS] or chronic obstructive airway disease [ALL FIELDS] or chronic obstructive lung disease [ALL FIELDS] or chronic bronchitis [ALL FIELDS] or Emphysema [ALL FIELDS] Review status: Any review status Results 122 *Please note that search terms have to be searched for manually each and every one of them and them de-duplicated at the end. Table 140: Search strategy for NIHR Health Technology Assessment Trial registry National Institute for Health Research - Health Technology Assessment (NIHR HTA) URL Date of search Search strategy COPD [Keywords] and HTA [programme] in the advanced search Results 13 GK14290 version 19 November 13 th, 2014 Page 164 of 178

166 Appendix 3 Selection Criteria Table 141: Inclusion and exclusion criteria for selection of studies Criteria Inclusion Exclusion STUDY DESIGN Abstract selection Randomized controlled trials Cross-over studies, if cross over before 10 weeks in each arm. Post-hoc or retrospective analyses Cost-effectiveness analyses Observational studies Reviews or meta-analyses Methodology studies or protocols N of 1 trials (sample size of 1 patient) Studies lasting less than 10 weeks Conference abstracts < 2009 Studies not in the English or German language Full-text selection Randomized controlled trials Studies where patients were required to spend time in a sleep laboratory GK14290 version 19 November 13 th, 2014 Page 165 of 178

167 Criteria Inclusion Exclusion TREATMENT / INTERVENTION Abstract selection LAMA monotherapies Umeclidinium 62.5 mcg OD Aclidinium 400 mcg BID Tiotropium 18 mcg OD Glycopyrronium 50 mcg OD LABA monotherapies Salmeterol 50 mcg BID Formoterol 12 mcg BID Indacaterol 75, 150, 300 mcg OD Olodaterol 5, 10 mcg OD LABA + LAMA combinations Umeclidinium/vilanterol; Tiotropium/indacaterol; Salmeterol/tiotropium; Formoterol/tiotropium; Indacaterol/glycopyrronium (QVA149); Any other combination of a LABA and a LAMA LABA/ICS + LAMA combinations Fluticasone/salmeterol + tiotropium Beta-agonists (Bambuterol; Fonoterol; Tulobuterol) Short-acting anticholinergics (Ipratropium; Oxitropium) Methylxanthines (Theophylline) Inhaled corticosteroids Inhaled glucocorticosteroids (Beclomethasone; Budesonide; Fluticasone) leukotriene receptor antagonists (montelukast) Combinations of long-acting anticholinergics or LABAs with an inhaled corticosteroid (ICS) formoterol plus budesonide or fluticasone plus salmeterol that are administered separately COPD drugs in development or targeting other pathways (roflumilast; polyvalent mechanical bacterial lysate; lipopolysaccharide) All other pharmaceutical interventions not treating COPD (enoxaparin sodium) Non-pharmaceutical interventions such as pulmonary rehabilitation GK14290 version 19 November 13 th, 2014 Page 166 of 178

168 Criteria Inclusion Exclusion LAMA monotherapies Umeclidinium 62.5 mcg OD Aclidinium 400 mcg BID Tiotropium 18 mcg OD Glycopyrronium 50 mcg OD LABA monotherapies Salmeterol 50 mcg BID Formoterol 12 mcg BID Indacaterol 75, 150, 300 mcg OD Olodaterol 5, 10 mcg OD Full-text selection LABA + LAMA combinations Umeclidinium/Vilanterol 62.5/25mcg OD; Tiotropium/indacaterol 18/150mcg OD; Salmeterol/tiotropium 50/18mcg Formoterol/tiotropium 10/18 or 12/18mcg Indacaterol/glycopyrronium 110/50mcg OD (QVA149); Any other combination of a LABA and a LAMA Any treatments of interest with doses that are different to the doses of interest Studies of arformoterol (the (R,R) isomer of formoterol) LABA/ICS + LAMA combinations Fluticasone/salmeterol 500/50mcgt BID + tiotropium 18mcg OD Administered using any inhalation device COMPARATOR Abstract and full-text selection Studies that compare treatments of interest (above) with the following treatment arms are of interest: - Placebo - Tiotropium 18mcg OD Studies that only compare treatments that are not of interest Studies that only include the treatments of interest in combination with treatments not of interest (i.e. prednisolone + formoterol) Studies that only include the partial combinations of treatments of interest (i.e. tiotropium+ ICS) GK14290 version 19 November 13 th, 2014 Page 167 of 178

169 Criteria Inclusion Exclusion POPULATION OUTCOMES Abstract selection Full-text selection Abstract selection Patients with COPD as defined by GOLD guidelines (i.e. airflow limitation that is not fully reversible) Studies that include asthma patients and COPD patients and report data for COPD patients separately Adults Studies that include adults and children and report data for adults separately Patients with COPD as defined by GOLD guidelines (i.e. airflow limitation that is not fully reversible) Studies that include asthma patients and COPD patients and report data for COPD patients separately Adults Studies that include adults and children and report data for adults separately No selection based on outcomes Studies with only healthy patients without COPD Studies with patients who have reversible airway or obstructive lung disease Studies with only patients with asthma Studies that include asthma patients and COPD patients but do not report data for COPD patients separately Studies with only patients who have alpha-1-antitrypsindefinciency-realted COPD Studies that include only children Studies that include adults and children but do not report data for adults separately Trials in non-caucasian populations e.g. Chinese, Japanese patients (including non-caucasian registries) Studies with only healthy patients without COPD Studies with patients who have reversible airway or obstructive lung disease Studies with only patients with asthma Studies that include asthma patients and COPD patients but do not report data for COPD patients separately Studies with only patients who have alpha-1-antitrypsindefinciency-realted COPD Studies that include only children Studies that include adults and children but do not report data for adults separately Trials in non-caucasian populations E.g. Chinese, Japanese patients No selection based on outcomes GK14290 version 19 November 13 th, 2014 Page 168 of 178

170 Criteria Inclusion Exclusion Full-text selection Report results for at least one of the following outcomes (for all treatments): o Trough FEV 1 o St. George s Respiratory Questionnaire total score (SGRQ) o Transition dyspnoea index (TDI) o Rescue medication use None of the relevant outcomes (as listed in the inclusion criteria) is reported; Only report the following outcomes (without any outcomes of interest): mortality; bioactivity outcomes or biomarkers of inflammation; lung mucociliary clearance; arterial blood gases or degree of pulmonary hyper-inflation; plethysmography and oscillometry; nocturnal hypoxemia; quality of life in EuroQol; GK14290 version 19 November 13 th, 2014 Page 169 of 178

171 Appendix 4 - Decision rules for NMA Continuous outcomes For continuous outcomes the analysis will be based on the difference between the least square mean at follow-up (LS mean at FU) or the change from baseline (CFB) for the active treatment versus the baseline comparator as well as the associated standard error (SE) of the difference. If the difference in the LS mean at FU or the CFB was reported it will be used in the analysis and otherwise it will be calculated based on the difference between the values reported per treatment. If the SE of the difference was reported it will be used in the analysis and otherwise it will be calculated based on the following decision rules: #1 SE of the difference = SD of diff*sqrt(1/n active+1/n comparator); or #2 SE of the difference = (95% CI High of diff - 95% CI Low of diff)/3.92; or #3 SE of the difference = sqrt[(se of LS Mean at FU or CFB active tx)^2+(se of LS Mean at FU or CFB comparator)^2]; or #4 SE of the difference= difference for analysis/(tinv(p/2,(n randomized active+ N randomized comparator- 2))) If the SE of LS Mean at FU or CFB per treatment was reported it was used in calculation #3 above, and otherwise it was calculated based on the 95% confidence intervals: SE of LS Mean at FU or CFB per treatment= [(CI high of LS Mean at FU or CFB per tx reported)-(ci low of LS Mean at FU or CFB per tx reported)]/3.92 In the case that the SE of the difference is not reported and there is insufficient information reported to calculate the SE of the difference based on the decision rules, the SE of the difference will be imputed using the following steps: Step 1. Calculate the SD of the difference per comparison where the SE of the difference is available: SD of difference per comparison= (SE of difference in LS mean at FU or the CFB)/SQRT(1/N randomized active +1/N randomized comparator) Step 2. Calculate the average SD of the difference across the comparisons where available: Average SD of the difference = AVERAGE(SD of diff across comparisons where available) Step 3. Impute the SE of the difference per comparison where it is missing: Imputed SE of the difference in LS mean at FU or the CFB=Average SD of the difference *SQRT(1/N randomized active+1/n randomized comparator) GK14290 version 19 November 13 th, 2014 Page 170 of 178

172 Appendix 5 Bayesian NMA model Models are available in NICE DSU Technical document 2. (available on: ) The model below is from program 7(a) and a description of the data structure is provided in this document. # Normal likelihood, identity link, trial-level data given as treatment differences # Random effects model for multi-arm trials model{ for(i in 1:ns2) { y[i,2] ~ dnorm(delta[i,2],prec[i,2]) resdev[i] <- (y[i,2]-delta[i,2])*(y[i,2]-delta[i,2])*prec[i,2] # *** PROGRAM STARTS # LOOP THROUGH 2-ARM STUDIES # normal likelihood for 2-arm trials #Deviance contribution for trial i } for(i in (ns2+1):(ns2+ns3)) { for (k in 1:(na[i]-1)) { # LOOP THROUGH 3-ARM STUDIES # set variance-covariance matrix for (j in 1:(na[i]-1)) { Sigma[i,j,k] <- V[i]*(1-equals(j,k)) + var[i,k+1]*equals(j,k) } } Omega[i,1:(na[i]-1),1:(na[i]-1)] <- inverse(sigma[i,,]) #Precision matrix # multivariate normal likelihood for 3-arm trials y[i,2:na[i]] ~ dmnorm(delta[i,2:na[i]],omega[i,1:(na[i]-1),1:(na[i]-1)]) #Deviance contribution for trial i for (k in 1:(na[i]-1)){ # multiply vector & matrix ydiff[i,k]<- y[i,(k+1)] - delta[i,(k+1)] z[i,k]<- inprod2(omega[i,k,1:(na[i]-1)], ydiff[i,1:(na[i]-1)]) 93 } resdev[i]<- inprod2(ydiff[i,1:(na[i]-1)], z[i,1:(na[i]-1)]) } for(i in 1:(ns2+ns3)){ # LOOP THROUGH ALL STUDIES GK14290 version 19 November 13 th, 2014 Page 171 of 178

173 w[i,1] <- 0 control arm delta[i,1] <- 0 for (k in 2:na[i]) { var[i,k] <- pow(se[i,k],2) prec[i,k] <- 1/var[i,k] # adjustment for multi-arm trials is zero for # treatment effect is zero for control arm # LOOP THROUGH ARMS # calculate variances # set precisions } resdev[i] <- sum(dev[i,2:na[i]]) this trial for (k in 2:na[i]) { delta[i,k] ~ dnorm(md[i,k],taud[i,k]) md[i,k] <- d[t[i,k]] - d[t[i,1]] + sw[i,k] multi-arm trial correction) taud[i,k] <- tau *2*(k-1)/k multi-arm trial correction) w[i,k] <- (delta[i,k] - d[t[i,k]] + d[t[i,1]]) sw[i,k] <- sum(w[i,1:k-1])/(k-1) # summed residual deviance contribution for # LOOP THROUGH ARMS # trial-specific treat effects distributions # mean of treat effects distributions (with # precision of treat effects distributions (with # adjustment for multi-arm RCTs # cumulative adjustment for multi-arm trials } } totresdev <- sum(resdev[]) d[1]<-0 treatment for (k in 2:nt){ d[k] ~ dnorm(0,.0001) ] sd ~ dunif(0,5) tau <- pow(sd,-2) variance) #Total Residual Deviance # treatment effect is zero for reference # vague priors for treatment effects # vague prior for between-trial SD # between-trial precision = (1/between-trial } # *** PROGRAM ENDS GK14290 version 19 November 13 th, 2014 Page 172 of 178

174 References 1. (GOLD) GIfCOLD. GOLD guidelines from the Global Strategy for the Diagnosis, Management and Prevention of COPD2013. Available from: 2. (NICE) NIfHaCE. Chronic obstructive pulmonary disease: Management of chronic obstructive pulmonary disease in adults in primary and secondary care (partial update)2010. Available from: 3. Bucher HC, Guyatt GH, Griffith LE, Walter SD. The results of direct and indirect treatment comparisons in meta-analysis of randomized controlled trials. Journal of clinical epidemiology. 1997;50(6): Hoaglin DC, Hawkins N, Jansen JP, Scott DA, Itzler R, Cappelleri JC, et al. Conducting indirect-treatment-comparison and network-meta-analysis studies: report of the ISPOR Task Force on Indirect Treatment Comparisons Good Research Practices: part 2. Value Health. 2011;14(4): DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986;7(3): (ICWG) ICWG. Report of the Indirect Comparisons Working Group to the Pharmaceutical Benefits Advisory Committee: assessing indirect comparisons2008. Available from: 7. Caldwell DM, Ades AE, Higgins JP. Simultaneous comparison of multiple treatments: combining direct and indirect evidence. Bmj. 2005;331(7521): Lu G, Ades A. Combination of direct and indirect evidence in mixed treatment comparisons. Statistics in medicine. 2004;23(20): Jansen JP, Crawford B, Bergman G, Stam W. Bayesian meta-analysis of multiple treatment comparisons: an introduction to mixed treatment comparisons. Value Health. 2008;11(5): Dempster AP. The direct use of likelihood for significance testing. Statistics and Computing. 1997;7(4): Lunn DJ, Thomas A, Best N, Spiegelhalter D. WinBUGS-a Bayesian modelling framework: concepts, structure, and extensibility. Statistics and computing. 2000;10(4): Dias S, Welton NJ, Sutton AJ, Ades A. NICE DSU Technical Support Document 2: a generalised linear modelling framework for pairwise and network meta-analysis of randomised controlled trials. National Institute for Health and Clinical Excellence, London, UK Santé HAd. Indirect comparisons - Methods and validity2009. Available from: sante.fr/portail/upload/docs/application/pdf/ /summary_report indirect_comparisons_methods_and_validity_january_2011_2.pdf. 14. Higgins JPT GSe. Cochrane Handbook for Systematic Reviews of Interventions Version [updated March 2011]. The Cochrane Collaboration. Available from: Cohn LD, Becker BJ. How meta-analysis increases statistical power. Psychological methods. 2003;8(3): Wetterslev J, Thorlund K, Brok J, Gluud C. Estimating required information size by quantifying diversity in random-effects model meta-analyses. BMC medical research methodology. 2009;9: Thorlund K, Mills EJ. Sample size and power considerations in network meta-analysis. Systematic reviews. 2012;1: Anzueto AD, M.~Kaelin,T.~Richard,N.~Tabberer,M.~Harris,S. The efficacy and safety of umeclidinium/vilanterol compared with tiotropium or vilanterol over 24 weeks in subjects with COPD. American journal of respiratory and critical care medicine. 2013;187:A GlaxoSmithKline. 24-week Trial Comparing GSK573719/GW With GW and With Tiotropium in Chronic Obstructive Pulmonary Disease. NCT [Internet] Available from: GlaxoSmithKline. DB : A multicenter trial comparing the efficacy and safety of GSK573719/GW with GW and with tiotropium over 24 weeks in subjects with COPD [Internet] Available from: GlaxoSmithKline. A Multicenter Trial Comparing the Efficacy and Safety of GSK573719/GW with GW and with Tiotropium over 24 Weeks in Subjects with COPD (DB ). Clinical Study Report Decramer MA, A.~Kerwin,E.~Richard,N.~Crater,G.~Tabberer,M. Efficacy and safety of umeclidinium/vilanterol compared with umeclidinium or tiotropium in COPD [Abstract]. European Respiratory Society Annual Congress, 2013 Sept711, Barcelona, Spain. 2013;42:751s. GK14290 version 19 November 13 th, 2014 Page 173 of 178

175 23. GlaxoSmithKline. 24-week Trial Comparing GSK573719/GW With GSK and With Tiotropium in Chronic Obstructive Pulmonary Disease. NCT [Internet] Available from: GlaxoSmithKline. A multicenter trial comparing the efficacy and safety of GSK573719/GW with GSK with tiotropium over 24 weeks in subjects with COPD [Internet] Available from: GlaxoSmithKline. A Multicenter Trial Comparing the Efficacy and Safety of GSK573719/GW with GSK and withtiotropium over 24 Weeks in Subjects with COPD (DB ). Clinical Study Report Donohue JFM-Y, M.R.~Kilbride,S.~Mehta,R.~Kalberg,C.J.~Church,A. Efficacy and safety of once-daily umeclidinium/vilanterol 62.5/25 mcg in COPD: a randomized, placebo-controlled, 24-week study. American journal of respiratory and critical care medicine. 2013;187:A Donohue JFM-Y, M.R.~Kilbride,S.~Mehta,R.~Kalberg,C.~Church,A. Efficacy and safety of once-daily umeclidinium/vilanterol 62.5/25 mcg in COPD. Respiratory medicine. 2013;107: GlaxoSmithKline. A 24-week Evaluation of GSK573719/Vilanterol (62.5/25mcg) and Components in COPD. NCT [Internet] Available from: GlaxoSmithKline. A 24-Week, Randomized, Double-Blind, Placebo-Controlled Study to Evaluate the Efficacy and Safety of GSK573719/GW Inhalation Powder and the Individual Components Delivered Once-Daily via a N [Internet] Available from: GlaxoSmithKline. The Purpose of This Study is to Evaluate the Spirometric Effect (Trough FEV1) of Umeclidinium/Vilanterol 62.5/25 Mcg Once Daily Compared With Tiotriopium 18 Mcg Once Daily Over a 24-week Treatment Period in Subjects With COPD. NCT [Internet] Available from: GlaxoSmithKline. A Multicenter Trial Comparing the Efficacy and Safety of Umeclidinium/Vilanterol 62.5/25 mcg Once Daily with Tiotropium 18 mcg Once Daily over 24 Weeks in Subjects with Chronic Obstructive Pulmonar [Internet] Available from: GlaxoSmithKline. A Multicenter Trial Comparing the Efficacy and Safety of Umeclidinium/Vilanterol 62.5/25 mcg Once Daily with Tiotropium 18 mcg Once Daily over 24 Weeks in Subjects with Chronic Obstructive Pulmonary Disease (COPD) (ZEP117115). Clinical Study Report Wedzicha JAD, M.~Ficker,J.H.~Niewoehner,D.E.~Sandstrà m,t.~taylor,a.f. Analysis of chronic obstructive pulmonary disease exacerbations with the dual bronchodilator QVA149 compared with glycopyrronium and tiotropium (SPARK): a randomised, double-blind, parallel-group study. The Lancet Respiratory Medicine. 2013;1: Wedzicha JAD, M.~Ficker,J.H.~Niewohner,D.~Sandström,T.~Fowler,Taylor A. Dual bronchodilation with once-daily QVA149 reduces exacerbations, improves lung function and health status versus glycopyrronium and tiotropium in severe-to-very severe COPD patients: The SPARK study [Abstract]. European Respiratory Society Annual Congress, 2013 Sept711, Barcelona, Spain. 2013;42:3s. 35. Wedzicha JAD, M.~Ficker,J.~Niewoehner,D.E.~Sandstrom,T.~Fowler,Taylor A. Efficacy and safety of QVA149 versus glycopyrronium and tiotropium in severe to very severe COPD: the Spark study. American journal of respiratory and critical care medicine. 2013;187:A Decramer MW, J.A.~Sandstrom,T.~Niewoehner,D.~Fowler,Taylor A.~D'Andrea,P. Safety and tolerability of QVA149, glycopyrronium and tiotropium in patients with severe to very severe COPD: the Spark study. American journal of respiratory and critical care medicine. 2013;187:A Decramer MW, J.A.~Ficker,J.H.~Fowler,Taylor A.~D'Andrea,P.~Arrasate,C. Once-daily QVA149 reduces exacerbations and improves health status in comparison with glycopyrronium and tiotropium in patients with severe-to-very severe COPD: The SPARK study [Abstract]. European Respiratory Society Annual Congress, 2013 Sept711, Barcelona, Spain. 2013;42:143s. 38. Ficker JHW, J.A.~Decramer,M.~Fowler,Taylor A.~D'Andrea,P.~Arrasate,C. QVA149 improves lung function and reduces exacerbations compared to tiotropium in patients with severe-to-very severe COPD: The SPARK study [Abstract]. European Respiratory Society Annual Congress, 2013 Sept711, Barcelona, Spain. 2013;42:751s. GK14290 version 19 November 13 th, 2014 Page 174 of 178

176 39. Decramer MW, J.A.~Ficker,J.~Fowler,Taylor A.~D'Andrea,P.~Arrasate,C. Once-daily QVA149 improves health-related quality of life in patients with severe to very severe COPD: the Spark study. American journal of respiratory and critical care medicine. 2013;187:A Ficker JHW, J.A.~Decramer,M.~Fowler,Taylor A.~D'Andrea,P.~Arrasate,C. QVA149 reduces COPD exacerbation rate in various subgroups of patients: The SPARK study [Abstract]. European Respiratory Society Annual Congress, 2013 Sept711, Barcelona, Spain. 2013;42:3s. 41. Wedzicha JAD, M.~Ficker,J.~Fowler,Taylor A.~D'Andrea,P.~Arrasate,C. Dual bronchodilation with QVA149 reduces COPD exacerbations: the Spark study. American journal of respiratory and critical care medicine. 2013;187:A Wedzicha JAD, M.~Ficker,J.H.~Fowlertaylor,A.~D'Andrea,P.~Arrasate,C.~Chen,H.~Banerji,D. Glycopyrronium and tiotropium demonstrate similar improvements in lung function and reductions in exacerbations in severe-to-very severe COPD: The spark study. RespirologyConference: 18th Congress of the Asian Pacific Society of Respirology Yokohama JapanConference Start: Conference End: Conference Publication: (varpagings). 2013;18(pp 69):November. 43. Ficker JW, J.~Fowlertaylor,A.~D'Andrea,P.~Arrasate,C.~Chen,H.~Banerji,D. QVA149 Improves Lung Function and Reduces Exacerbations Compared With Tiotropium in Patients With Severe-to-Very Severe COPD: The SPARK Study. Chest. 2014;145(3:Suppl):Suppl. 44. Wedzicha JF, J.~Fowlertaylor,A.~D'Andrea,P.~Arrasate,C.~Chen,H.~Banerji,D. Once-Daily QVA149 Reduces Exacerbations and Improves Health Status in Comparison With Glycopyrronium and Tiotropium in Patients With Severe-to-Very Severe COPD: The SPARK Study. Chest. 2014;145(3:Suppl):Suppl. 45. Wedzicha JF, J.~Niewohner,D.~Sandstrom,T.~Fowlertaylor,A.~Banerji,D. Dual Bronchodilation With Once-Daily QVA149 Reduces Exacerbations, Improves Lung Function and Health Status Versus Glycopyrronium and Tiotropium in Severe-to-Very Severe COPD Patients: The SPARK Study. Chest. 2014;145(3:Suppl):Suppl. 46. Novartis. Effect of QVA149 Versus NVA237 and Tiotropium on Chronic Obstructive Pulmonary Disorder (COPD) Exacerbations. NCT [Internet] Available from: Novartis. A 64-week treatment, multi-center, randomized, double-blind, parallel-group, active controlled study to evaluate the effect of QVA149 (110/50 µg o.d.) vs NVA237 (50 µg o.d.) and openlabel tiotropi [Internet] Available from: Dahl RC, K.~Rudolf,M.~Mehta,R.~Kho,P.~Alagappan,V. QVA149 administered once daily provides significant improvements in lung function over 1 year in patients with COPD: The ENLIGHTEN study [Abstract]. European Respiratory Journal: European Respiratory Society Annual Congress, Vienna, Austria, September ;40:529s. 49. Dahl RC, K.~Rudolf,M.~Mehta,R.~Kho,P.~Alagappan,V. QVA149 once daily provides significant improvements in lung function over 1 year in patients with COPD: the Enlighten study [Abstract]. Thorax. 2012;67:A Dahl RC, K.R.~Rudolf,M.~Mehta,R.~Kho,P.~Alagappan,V.K.T.~Chen,H.~Banerji,D. Safety and efficacy of dual bronchodilation with QVA149 in COPD patients: The ENLIGHTEN study. Respiratory medicine. 2013;107: Novartis. A Study to Assess the Long-term Safety of QVA149. NCT [Internet] Available from: Novartis. A multicenter, randomized, double-blind, placebocontrolled study, to assess the long term safety of 52 weeks treatment with QVA149 (110µg indacaterol / 50µg glycopyrrolate) in patients with moderat [Internet] Available from: Frith PT, P.~Wark,P.~Lindstrom,S.~Bateman,E. Benefits of dual bronchodilation with QVA149 once daily versus placebo, indacaterol, NVA237 and tiotropium in patients with COPD: the Shine study [Abstract]. Respirology. 2013;18: Bateman EF, G.T.~Barnes,N.~Gallagher,N.~Green,Y.~Horton,R. Benefits of dual bronchodilation with QVA149 once daily versus placebo, indacaterol, NVA237 and tiotropium in patients with COPD: The SHINE study [Abstract]. European Respiratory Journal: European Respiratory Society Annual Congress, Vienna, Austria, September ;40:509s. 55. Bateman EDW, T.~Hashimoto,S.~Gallagher,N.~Green,Y.~Horton,R. Dual bronchodilation with once-daily QVA149 provides superior bronchodilation compared to its mono-components and GK14290 version 19 November 13 th, 2014 Page 175 of 178

177 tiotropium in all subgroups of patients with COPD: the SHINE study. American journal of respiratory and critical care medicine. 2013;187:A Welte TB, N.~Nagase,T.~Gallagher,N.~Green,Y.~Horton,R. QVA149 once daily is safe and well tolerated in patients with COPD: The SHINE Study. American journal of respiratory and critical care medicine. 2013;187:A Bateman EDF, G.T.~Barnes,N.~Gallagher,N.~Green,Y.~Henley,M.~Banerji,D. Dual bronchodilation with QVA149 versus single bronchodilator therapy: The SHINE study. European respiratory journal. 2013;42: Barnes NB, E.~Gallagher,N.~Green,Y.~Horton,R.~Henley,M.~Banerji,D. QVA149 once daily provides superior bronchodilation versus indacaterol, glycopyrronium, tiotropium and placebo: The shine study. ThoraxConference: British Thoracic Society Winter Meeting 2012 London United KingdomConference Start: Conference End: Conference Publication: (varpagings). 2012;67(pp A147-A148):December. 59. Bateman EH, S.~Gallagher,N.~Green,Y.~Horton,R.~Henley,M.~Banerji,D. Glycopyrronium once daily provides rapid and sustained bronchodilation and is well tolerated in patients with COPD: The shine study. ThoraxConference: British Thoracic Society Winter Meeting 2012 London United KingdomConference Start: Conference End: Conference Publication: (varpagings). 2012;67(pp A148-A149):December. 60. Chapman KRB, E.D.~Gallagher,N.~Hu,H.~Kim,H.-J.~Banerji,D. Once-daily QVA149 improves lung function, dyspnea and health status regardless of disease severity and prior medications: The shine study. RespirologyConference: 18th Congress of the Asian Pacific Society of Respirology Yokohama JapanConference Start: Conference End: Conference Publication: (varpagings). 2013;18(pp 31):November. 61. Chapman KRB, E.D.~Gallagher,N.~Hu,H.~Banerji,D. QVA149 once daily improves lung function, dyspnoea and health status independent of prior medications and disease severity: The SHINE study. ThoraxConference: British Thoracic Society Winter Meeting 2013 Westminster United KingdomConference Start: Conference End: Conference Publication: (varpagings). 2013;68(pp A182-A183):December. 62. Chapman KBEGNHHBD. Once-Daily QVA149 Improves Lung Function, Dyspnea, and Health Status Independent of Disease Severity and Prior Medications: The SHINE Study. Chest. 2014;145:Suppl. 63. Novartis. A Study to Assess the Efficacy, Safety and Tolerability of Once-daily (q.d.) QVA149 in Patients With Moderate to Severe Chronic Obstructive Pulmonary Disease (COPD). NCT [Internet] Available from: Novartis. A 26-week treatment multi-center, randomized, doubleblind, parallel-group, placebo and active controlled (open label) study to assess the efficacy, safety and tolerability of QVA149 (110/50 µg q.d [Internet] Available from: Mahler DA, D'Urzo A, Bateman ED, Ozkan SA, White T, Peckitt C, et al. Concurrent use of indacaterol plus tiotropium in patients with COPD provides superior bronchodilation compared with tiotropium alone: a randomised, double-blind comparison. Thorax. 2012;67: Novartis. Efficacy and Safety of Indacaterol Plus Tiotropium Versus Tiotropium Alone in Patients With Chronic Obstructive Pulmonary Disease (INTRUST1). NCT [Internet] Available from: Novartis. A randomized, double-blind, controlled, parallel group, 12-week treatment study to compare the efficacy and safety of the combination of indacaterol 150 μg once daily with open label tiotropium [Internet] Available from: Novartis. Efficacy and Safety of Indacaterol Plus Tiotropium Versus Tiotropium Alone in Patients With Chronic Obstructive Pulmonary Disease (INTRUST2). NCT [Internet] Available from: Novartis. A randomized, double-blind, controlled, parallel group, 12-week treatment study to compare the efficacy and safety of the combination of indacaterol 150 μg once daily with open label tiotropium [Internet] Available from: Aaron SD, Vandemheen KL, Fergusson D, Maltais F, Bourbeau J, Goldstein R, et al. Tiotropium in combination with placebo, salmeterol, or fluticasone-salmeterol for treatment of chronic obstructive pulmonary disease: a randomized trial. Annals of internal medicine. 2007;146: GK14290 version 19 November 13 th, 2014 Page 176 of 178

178 71. (Canada) OHRI. Optimal therapy of chronic obstructive pulmonary disease to prevent exacerbations and improve quality of life: a randomised, double-blind, placebo-controlled trial. ISRCTN [Internet] /05/2014. Available from: Cazzola M, Andò F, Santus P, Ruggeri P, Marco F, Sanduzzi A, et al. A pilot study to assess the effects of combining fluticasone propionate/salmeterol and tiotropium on the airflow obstruction of patients with severe-to-very severe COPD. Pulmonary pharmacology & therapeutics. 2007;20: Vogelmeier CK, P.~Harari, S.~Gans, S. J.~Stenglein, S.~Thirlwell, J. Formoterol mono- and combination therapy with tiotropium in patients with COPD: a 6-month study. Respir Med. 2008;102(11): Tashkin DPP, J.~Iezzoni, D.~Varghese, S. T. Formoterol and tiotropium compared with tiotropium alone for treatment of COPD. Copd. 2009;6(1): Karner CC, C. J. Combination inhaled steroid and long-acting beta(2)-agonist in addition to tiotropium versus tiotropium or combination alone for chronic obstructive pulmonary disease. The Cochrane database of systematic reviews. 2011(3):Cd Bateman ED, Ferguson GT, Barnes N, Gallagher N, Green Y, Henley M, et al. Dual bronchodilation with QVA149 versus single bronchodilator therapy: the SHINE study. Eur Respir J. 2013;42(6): Brusasco V, Hodder R, Miravitlles M, Korducki L, Towse L, Kesten S. Health outcomes following treatment for six months with once daily tiotropium compared with twice daily salmeterol in patients with COPD.[Erratum appears in Thorax Feb;60(2):105]. Thorax. 2003;58(5): Casaburi R, Briggs DD, Jr., Donohue JF, Serby CW, Menjoge SS, Witek TJ, Jr. The spirometric efficacy of once-daily dosing with tiotropium in stable COPD: a 13-week multicenter trial. The US Tiotropium Study Group. Chest. 2000;118(5): Casaburi R, Mahler DA, Jones PW, Wanner A, San PG, ZuWallack RL, et al. A long-term evaluation of once-daily inhaled tiotropium in chronic obstructive pulmonary disease. European Respiratory Journal. 2002;19(2): Chan CK, Maltais F, Sigouin C, Haddon JM, Ford GT, Group SS. A randomized controlled trial to assess the efficacy of tiotropium in Canadian patients with chronic obstructive pulmonary disease. Canadian Respiratory Journal. 2007;14(8): Covelli H, Bhattacharya S, Cassino C, Conoscenti C, Kesten S. Absence of electrocardiographic findings and improved function with once-daily tiotropium in patients with chronic obstructive pulmonary disease. Pharmacotherapy:The Journal of Human Pharmacology & Drug Therapy. 2005;25(12): Donohue JF, Fogarty C, Lotvall J, Mahler DA, Worth H, Yorgancioglu A, et al. Once-daily bronchodilators for chronic obstructive pulmonary disease: indacaterol versus tiotropium. American Journal of Respiratory & Critical Care Medicine. 2010;182(2): Donohue JF, Van Noord JA, Bateman ED, Langley SJ, Lee A, Witek TJ, Jr., et al. A 6-month, placebo-controlled study comparing lung function and health status changes in COPD patients treated with tiotropium or salmeterol. Chest. 2002;122(1): Garcia RF. A randomised, double-blind, placebo-controlled, 12 weeks trial to evaluate the effect of Tiotropium Inhalation Capsules on the magnitude of exercise, measured using an accelerometer, in patients with Chronic Obstructive Pulmonary Disease (COPD). Boehringer Ingelheim TrialResults Kerwin E, Hebert J, Gallagher N, Martin C, Overend T, Alagappan VK, et al. Efficacy and safety of NVA237 versus placebo and tiotropium in patients with COPD: the GLOW2 study. European Respiratory Journal. 2012;40(5): Moita J, Barbara C, Cardoso J, Costa R, Sousa M, Ruiz J, et al. Tiotropium improves FEV1 in patients with COPD irrespective of smoking status. Pulmonary Pharmacology & Therapeutics. 2008;21(1): Niewoehner DE, Rice K, Cote C, Paulson D, Cooper JA, Jr., Korducki L, et al. Prevention of exacerbations of chronic obstructive pulmonary disease with tiotropium, a once-daily inhaled anticholinergic bronchodilator: a randomized trial.[summary for patients in Ann Intern Med Sep 6;143(5):I20; PMID: ]. Annals of Internal Medicine. 2005;143(5): Tashkin DP, Celli B, Senn S, Burkhart D, Kesten S, Menjoge S, et al. A 4-year trial of tiotropium in chronic obstructive pulmonary disease. New England Journal of Medicine. 2008;359(15): GK14290 version 19 November 13 th, 2014 Page 177 of 178

179 89. Tonnel AB, Perez T, Grosbois JM, Verkindre C, Bravo ML, Brun M, et al. Effect of tiotropium on health-related quality of life as a primary efficacy endpoint in COPD. International Journal of Copd. 2008;3(2): Verkindre C, Bart F, Aguilaniu B, Fortin F, Guerin JC, Le MC, et al. The effect of tiotropium on hyperinflation and exercise capacity in chronic obstructive pulmonary disease. Respiration. 2006;73(4): Vogelmeier C, Kardos P, Harari S, Gans SJ, Stenglein S, Thirlwell J. Formoterol mono- and combination therapy with tiotropium in patients with COPD: a 6-month study. Respir Med. 2008;102(11): GK14290 version 19 November 13 th, 2014 Page 178 of 178

180 HEALTH OUTCOMES PROTOCOL UNIQUE IDENTIFIER ABBREVIATED TITLE FINAL PROTOCOL APPROVED FULL TITLE SPONSORSHIP DIVISION/BUSINESS UNIT DEPARTMENT HO STUDY ACCOUNTABLE PERSON(S) CONTRIBUTING AUTHORS RETENTION CATEGORY INFORMATION TYPE KEY WORDS / MESH HEADINGS / META DATA trackho HO / etrack Evidence synthesis of UMEC/VI and comparators in COPD 19-JUL-2013 Protocol for evidence synthesis to assess comparative efficacy of UMEC/VI versus other treatments for COPD adult patients GSK Global Research & Development Global Health Outcomes Evidence synthesis by means of indirect comparison & network meta-analysis COPD Evidence synthesis Systematic review Meta-analysis Indirect treatment comparison Mixed treatment comparison Network meta-analysis Efficacy Effectiveness ASSET ID GSK ASSET Umeclidinium/Vilanterol INDICATION COPD

181 SPONSOR SIGNATORY Upload to isign, enter names of signatories (must have BioSafe Digital Credential), submit for electronic signature. Once signatures are complete, store in IMMS & archive to PIER. HO Study Accountable Person Date Group HO VP/Director Date Medical (GML/MDL/PPL/PMAL/RMD/Equivalent) Date Intellectual Property Date Trademarks Date Global Clinical Safety & Pharmacovigilence Date Statistics Date Enter other signatory Name/Role as needed Date HO PROTOCOL REVIEW COMMITTEE APPROVAL HO PRC CHAIR Date Approved 2

182 PROTOCOL SYNOPSIS Unique Identifier Abbreviated Title GSK Product Rationale Objectives (Primary, Secondary & Exploratory) trackho HO / etrack Evidence synthesis of UMEC/VI and comparators in COPD Anoro Ellipta; Umeclidinium (GSK573719)/Vilanterol (GW642444) Payers, HTA bodies and other decision makers in many countries require a formal comparison of clinical efficacy of UMEC/VI with other available treatments. GSK is currently developing a closed combination of umeclidinium (LAMA) + vilanterol (LABA) for COPD patients. Currently, UMEC/VI has direct evidence compared with tiotropium (LAMA) and is expected to have direct evidence compared with Seretide (ICS+LABA). However, no such evidence exists for open combinations of LABA+LAMA (open dual), open combinations of ICS+LABA+LAMA (open triple) and QVA149 (indacaterol + glycopyrronium), a novel LABA+LAMA therapy being developed by Novartis which are deemed to be relevant comparators in franchise LOCs for local and national market access. This protocol explains the ways in which indirect evidence can be generated to address this evidence gap. The information from a recently completed systematic literature review (etrack /TrackHO HO ; PRC approved protocol attached in Appendix 2) will be used to perform a frequentist indirect treatment comparison (required for PBAC in Australia) and a Bayesian network meta-analysis (preferred by other franchise LOC payers). Primary 1. To assess the relative efficacy of UMEC/VI 125mcg/25mcg OD and UMEC/VI 62.5mcg/25mcg OD separately vs. Indacaterol 150mcg + Tiotropium 18mcg OD at 12 weeks Fluticasone/salmeterol 500mcg/50mcg + Tiotropium 18mcg OD at 12 weeks QVA149 at 12, 24 and 52 weeks. by means of lung function (difference in change from baseline for trough FEV 1 ). Secondary 2. To assess the relative efficacy of UMEC/VI 125mcg/25mcg OD and UMEC/VI 62.5mcg/25mcg OD separately vs. QVA149 by means of Health related quality of life (change from baseline for SGRQ total score) at 12 and 24 weeks. 3

183 Dyspnoea (change from baseline for TDI focal score) at 12 and 24 weeks. Decrease in rescue medication use (mean number of puffs per day) at 24 and 52 weeks. 3. To assess the relative efficacy of UMEC/VI 125mcg/25mcg OD and UMEC/VI 62.5mcg/25mcg OD separately vs. Indacaterol 150mcg + Tiotropium 18mcg OD by means of decrease in rescue medication use (mean number of puffs per day) at 12 weeks. Comparative assessment time points have been selected based on the availability of evidence as determined by the systematic literature review. Specific endpoints have been selected based on the requirements of payers. Study Design A systematic review of the literature follows the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines to identify published, randomized controlled clinical trials (RCTs) comparing selected therapies in patients with COPD. Efficacy outcomes for the treatments of interest, obtained by a systematic literature review will be synthesized using two approaches: a. The first is based on an anchored (adjusted) indirect treatment comparison within a frequentist framework (Bucher et al 1997). Study Population and Sampling Methods b. As the networks formed are allowing for direct and indirect comparison, a Bayesian network meta-analysis will be also performed to evaluate the relative efficacy of all the drugs in the network. For continuous outcomes a generalized linear model with identity link and a normal likelihood distribution will be used, while a logit link with binomial likelihood distribution will be used for dichotomous outcomes (Dias et al 2011, Hoaglin et al 2011). Studies were included in the evidence synthesis networks if the following criteria were fulfilled. Patient characteristics: COPD patients 35 years of any race and gender, eligible to receive COPD maintenance therapy. Comparators: Only studies reporting results for indacaterol + tiotropium, salmeterol/fluticasone 500/50mcg + tiotropium 18mcg or QVA149 will be included. Outcomes: Reporting clinical efficacy by means of change in trough FEV 1, SGRQ total score, TDI focal score or rescue medication use (at least one outcome) Study design: Only randomised controlled trials (RCTs) will be included, 4

184 Data Source Data Analysis Methods irrespective of blinding status, either Phase III or phase IV, parallel-group with a minimum study duration of 12 weeks. In Appendix 2, the systematic lit review protocol provides a detailed list of study designs for inclusion and exclusion. The data used in the evidence synthesis obtained from a systematic literature review (SLR) that was conducted separately. An SLR protocol is available and has been included in Appendix 2. The continuous outcomes will be analyzed as differences in change from baseline between treatments: Trough FEV 1 (in L) SGRQ total score (in Units) TDI focal score (in Points) Rescue medication (in puffs per day) Dichotomous outcomes will be analyzed as RR. The outcomes for the treatments of interest, will be synthesized using two approaches: Indirect treatment comparison The ITC will be based on an anchored (adjusted) indirect treatment comparison within a frequentist framework (Bucher et al 1997). Two separate (random effects) meta-analyses will be performed with the common comparator and their results will be synthesized. The results of the meta-analyses will be presented in a standard forest plot graph including individual study results, pooled estimate (mean difference or RR) and 95%CI, Q, I 2 and chi-square. The results of the ITC (mean difference or RR) will be reported together with the 95% CI. The analysis will be conducted in STATA or R. Network meta-analysis As closed loop networks are allowing for direct and indirect comparison, a Bayesian network meta-analysis will be also performed to evaluate the relative efficacy of treatments of interest in the network. For continuous outcomes a generalized linear model with identity link and a normal likelihood distribution will be used, while a logit link with binomial likelihood distribution was used for dichotomous outcomes (Dias et al 2011). The results of the NMA will be presented as differences in CFB versus comparators and RR. Point estimates of both will be derived from the median of the posterior distribution while their 95% credible intervals (CrI) will be estimated from the 2.5th and 97.5th percentiles of the posterior distribution. The analysis will be conducted in WinBUGS or OpenBUGS. 5

185 Sample Size and Power This analysis is based on the synthesis of randomized controlled trials, thus the power analysis is retrospective. For the frequentist metaanalyses and ITC this power analysis will be based on the approach suggested by Thorlund & Mills No power analysis will be performed for the NMA and for the Bayesian ITC/NMA approach. Limitations The limitations of an evidence synthesis are well known in bibliography. Publication bias is an important limitation, as some clinical trials either fail to be published or are published in abstract form. Inhomogeneity is another limitation as in many cases the reported data are incomplete or not enough to ensure homogeneity. The systematic literature review has highlighted that there is limited evidence for the comparisons of interest. Availability of such limited evidence, with potential heterogeneity in patient populations, may lead to significant uncertainty in the interpretation of the results of this analysis. 6

186 TABLE OF CONTENTS 7

187 ABBREVIATIONS CFB CI CrI FEV1 ITC L NMA OR QVA149 RR SE sd SGRQ SLR UMEC/VI Change from baseline Confidence Interval Credible Interval Forced expiratory volume in 1 second Indirect treatment comparison Liters Network meta-analysis Odds Ratio indacaterol 110 mcg / glycopyrronium 50 mcg Relative Risk Standard error Standard deviation St. George's Respiratory Questionnaire Systematic literature review umeclidinium plus vilanterol 8

188 1. INTRODUCTION/BACKGROUND Payers, HTA bodies and other decision makers in many countries require a formal comparison of clinical efficacy of UMEC/VI with other available treatments. GSK is currently developing a closed combination of umeclidinium (LAMA) + vilanterol (LABA) for COPD patients as first line therapy. However, in some markets it is likely to be restricted to second line. The appropriate comparators for first line therapy include LAMA and ICS+LABA whilst those for second line include open combinations of LABA+LAMA (open dual), open combinations of ICS+LABA+LAMA (open triple) and potentially QVA149 (indacaterol + glycopyrronium), a novel LABA+LAMA therapy being developed by Novartis. Currently, UMEC/VI has direct evidence compared with market leading LAMA, tiotropium and is expected to have direct evidence compared with market leading ICS+LABA, salmeterol/fluticasone propionate. However, no such evidence exists for open dual, open triple and QVA149. This protocol explains the ways in which indirect evidence can be generated to address this evidence gap. The information from a recently completed systematic literature review (etrack /TrackHO HO ; PRC approved protocol attached in Appendix 2) will be used to perform a frequentist indirect treatment comparison (required for PBAC in Australia) and a Bayesian network meta-analysis (preferred by other franchise LOC payers). 2. OBJECTIVES 2.1. Primary To assess the relative efficacy of UMEC/VI 125mcg/25mcg OD and UMEC/VI 62.5mcg/25mcg OD separately vs. Indacaterol 150mcg + Tiotropium 18mcg OD at 12 weeks Fluticasone/salmeterol 500mcg/50mcg + Tiotropium 18mcg OD at 12 weeks QVA149 at 12, 24 and 52 weeks. by means of lung function (difference in change from baseline for trough FEV 1 ). 9

189 2.2. Secondary To assess the relative efficacy of UMEC/VI 125mcg/25mcg OD and UMEC/VI 62.5mcg/25mcg OD separately vs. QVA149, by means of: Health related quality of life (change from baseline for SGRQ total score) at 12 and 24 weeks Dyspnoea (change from baseline for TDI focal score) at 12 and 24 weeks Decrease in rescue medication use (mean number of puffs per day) at 12, 24 and 52 weeks To assess the relative efficacy of UMEC/VI 125mcg/25mcg OD and UMEC/VI 62.5mcg/25mcg OD separately vs. Indacaterol 150mcg + Tiotropium 18mcg OD by means of decrease in rescue medication use (mean number of puffs per day) at 12 weeks. N/A 2.3. Exploratory 3. RESEARCH METHODOLOGY 3.1. STUDY DESIGN Two approaches will be used for the indirect comparison of UMEC/VI with other treatments: a) Bucher s method (adjusted indirect comparison) and b) Bayesian network meta-analysis (mixed treatment comparison) 3.2. STUDY POPULATION Studies were included in the evidence synthesis networks if the following criteria were fulfilled. Patient characteristics: COPD patients 35 years of any race and gender, eligible to receive COPD maintenance therapy. Comparators: Only studies reporting results for indacaterol + tiotropium, salmeterol/fluticasone + tiotropium or QVA149 will be included. Outcomes: Reporting clinical efficacy by means of change in trough FEV 1, SGRQ total score, TDI focal score or rescue medication use (at least one outcome) Study design: Only randomised controlled trials (RCTs) will be included, irrespective of blinding status, either Phase III or phase IV, parallel-group with a minimum study duration of 12 weeks. In Appendix 2, the systematic lit review protocol provides a detailed list of study designs for 10

190 inclusion and exclusion ELIGIBILITY CRITERIA N/A N/A Inclusion Criteria Exclusion Criteria N/A SAMPLING 3.3. DATA SOURCE / DATA COLLECTION The data used in the evidence synthesis are obtained from a systematic literature that was conducted separately. An SLR protocol is available and has been included in Appendix 2. The main methodological features of the literature review are listed in the table below as part of this protocol. PICO and methods for the systematic literature review Objectives and research questions Objective To assess the evidence supporting the relative treatment effect of the inhaled therapies for the maintenance treatments of COPD. To assess the evidence supporting the safety and tolerability profiles of different inhaled therapies for the maintenance treatments of COPD. Studies to include Study designs Study design Randomised controlled trials, Phase 3 and Phase 4 The following study/publication types will be excluded: Studies with < 10 participants Pre-clinical and animal Studies including Short-acting beta agonist (SABAs) and Shortacting muscarinic antagonist (SAMAs) aimed at symptoms control Population Participants shall include individuals aged 12 years and older with an established diagnosis of chronic obstructive pulmonary disease (COPD) at any level of severity warranting the treatment with bronchodilators or corticosteroids (FEV1%p<=80%). Interventions Interventions Interventions of interest for this Comparators Comparators of interest for this 11

191 systematic review are: Relvar (fluticasone furoate + vilanterol) Anoro (umeclidinium + vilanterol) Umeclidinium monotherapy Closed triple (vilanterol, umeclidinium and fluticasone furoate) systematic review are: Inhaled corticosteroid (ICS) Long-acting beta agonist (LABA) Long-acting muscarinic antagonist (LAMA) Any combination therapy of ICS + LABA, in combined or separate inhalers Any combination therapy of LABA + LAMA, in combined or separate inhalers Any combination therapy of ICS + LABA + LAMA, in combined or separate inhalers Phosphodiesterase type 4 (PDE4) inhibitors Language No restriction on language List of non-english studies of interest to be provided to GSK Publication Database: Database inception to present date timeframe Conference proceedings: (last 3 years) Data sources Databases The following databases will be searched without date limits: Medline (OvidSP) Medline In-Process Citations & Daily Update (OvidSP) Embase (OvidSP) Cochrane Database Of Systematic Reviews (CDSR) (Wiley) Cochrane Central Register of Controlled Trials (CENTRAL) (Wiley) Database of Abstracts of Reviews of Effects (DARE) (Wiley) Health Technology Assessment Database (HTA) (Wiley) NIHR Health Technology Assessment Programme (Internet) PROSPERO (International Prospective Register of Systematic Reviews) (Internet) Completed and ongoing trials will be identified by searches of clinical trials registers. The following resources will be searched with no date limit: NIH Clinicaltrials.gov (Internet) Current Controlled Trials (Internet) 12

192 WHO International Clinical Trials Registry Platform (ICTRP) (Internet) European Medicines Agency European Public Assessment Reports (EMA EPARs) ( FDA website Conference abstracts from COPD conferences (American Thoracic Society (ATS), European Respiratory Society (ERS), American College of Chest Physicians (ACCP), from 2010). References in retrieved articles and relevant systematic reviews will be checked for further studies that might fulfil the inclusion criteria. Information to extract* Study information Name of first study author Date study publication Study location and setting Study group investigators (if known) Source of study funding (Government/Industry/Non-profit organisation/ not reported) Study design characteristics (and quality assessment) Study aim Period of participant recruitment Study duration and Length of follow-up Patient baseline characteristics (including GOLD classification of COPD severity) Summary of participants inclusion/exclusion criteria Types of interventions and comparators Total number of patients included (numbers per group) Outcomes of interest Total number of patients lost to followup/withdrawals/exclusions/refusals (numbers per study group) Reasons for withdrawals and dropouts per study group Study conclusions Any additional reviewer s comments Bibliographic details of all reports (full papers, abstracts, trial reports etc) relating to the study Efficacy COPD exacerbations (based on resource utilisation, symptoms, or 13

193 outcomes any other definition) o Moderate COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing a moderate COPD exacerbation; Total number of moderate COPD exacerbations experienced over the duration of the study; Mean rate of moderate COPD exacerbations per patient per year; Time to first moderate COPD exacerbation) o Severe COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing a severe COPD exacerbation; Total number of severe COPD exacerbations experienced over the duration of the study; Mean rate of severe COPD exacerbations per patient per year; Time to first severe COPD exacerbation o Composite of moderate/severe COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing moderate/severe COPD exacerbations; Total number of moderate/severe COPD exacerbations experienced over the duration of the study; Mean rate of moderate/severe COPD exacerbations per patient per year; Time to first moderate/severe COPD exacerbation) Change from baseline in FEV 1 o Prebronchodilator FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial); Percentage of patients achieving an increase in postbronchodilator FEV 1 of 12% and 200 ml above baseline at any time during 0-6 hours post-dose on treatment day 1 o Postbronchodilator FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial)) o Peak FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value)) o Trough FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial); Percentage of patients achieving an increase in trough FEV1 of 100mL above baseline, Time to onset (increase 12%+ from baseline FEV1)) o AUC FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value)) Exercise capacity e.g. six minute walk test,12 minutes walk test, incremental shuttle walk test, cycle ergometry test, treadmill test Inspiratory capacity, Residual volume (measure of lung hyperinflation), Functional residual capacity, Forced expiratory vital capacity (FVC), Exertional dyspnoea intensity (10-point Borg scale) during exercise, etc Inhaled rescue medication (Type of inhaled rescue medication used; Percentage of patients using inhaled rescue medication) 14

194 Safety outcomes Critical appraisal Change from baseline and Clinically relevant improvement in Patient reported Outcomes (PRO) measures 1 : o St. George s Respiratory Questionnaire (SGRQ) (MCID SGRQ 4 point change) o St. George s Respiratory Questionnaire for COPD patients (SGRQ-C) o Baseline Dyspnea Index Transition Dyspnea Index (BDI-TDI) (MCID TDI 1 point change) o Chronic Respiratory Questionnaire (CRQ) o Self-administered Chronic Respiratory Questionnaire (CRQ-SAS) o Shortness of Breath Questionnaire (SOBQ) o Shortness of Breath with Daily activities (SOBDA) o Exacerbations of Chronic Pulmonary Disease Tool Patient Reported Outcome (EXACT-PRO) o Modified Medical Research Council (mmrc) questionnaire o COPD Assessment Test (CAT) o EuroQol 5D (EQ-5D) o SF-36 o Rescue medication: Number of doses taken in a specified time period, % rescue medication use free days, Decrease in number of rescue med puffs per day, % symptom free days o Healthcare resource utilisation o Any other relevant patient reported outcome in COPD, to be specified in study protocol Safety and tolerability endpoints, including: Adverse events (any reported, any severity, by type, i.e. MI, arrhythmia, congestive heart failure, pneumonia, bone fractures, bone mineral density, specifying incidence, severity / grade, aortic pulse velocity, etc..) Withdrawals (all, due to AEs and any other specified, time to discontinuation) NICE checklist (CRD) Assessment of the quality of reporting of RCTs will be carried out using recommendations from the NICE single technology appraisal (STA) manufacturer s template (NICE STA 2009; Cochrane Handbook 2009) 1 Assessing disease-related symptoms and/or health-related quality of life 15

195 ENDPOINTS Primary Endpoint 1. Change from baseline in trough FEV 1 at 12 weeks Secondary Endpoint(s) 1. Change from baseline for trough FEV 1 at 24 and 52 weeks. 2. Change from baseline for St. George's Respiratory Questionnaire (SGRQ) total score at 12 and 24 weeks. 3. Change from baseline for Transitional dyspnea index focal score (TDI) focal score at 12 and 24 weeks. 4. Decrease in rescue medication use (mean number of puffs per day) at 12, 24 and 52 weeks. N/A Exploratory Endpoint(s) 3.4. SAMPLE SIZE / POWER CALCULATIONS The study proposed in this protocol is based on the synthesis of evidence provided by randomized controlled trials, by means of meta-analysis, ITC or network meta-analysis. For this reason, the methods that we suggest for the power calculations are not identical to those used for the power calculations of clinical studies, although based on the same principles and tests. As a first, general comment, we would like to underline the fact that to our knowledge there are no sound recommendations and clear guidelines from the HTA authorities on the ITC and NMA power calculations. Below, we suggest methods for the (post-hoc) estimation of the power per approach: I. Frequentist approach 1. Meta-analyses There are some well-established approaches for the estimation of the required information or, alternatively, the power (see for example Cohn & Becker 2003 and references therein). The information size N needed to reject an intervention effect μ with a type I error less than and a and type II error less than β is given by the following formula (Wetterslev et al, 2009): N AC = C z 1 a/2 z 1 β 2 v AC μ AC (1) Where, z 1 a/2 and z 1 β are the (1 a/2)th and (1 β)th percentiles of a standard normal distribution, v is the variance of the meta-analysis, C is a constant depending on the 16

196 randomization ratio and number of treatment arm (C = 4 with a randomization ratio of 1:1 and two treatment arms, Thorlund & Mills 2012). If statistical heterogeneity exists across the included trials (between-trial heterogeneity) in a meta-analysis, the calculated sample size can be adjusted to account for the additional variation (i.e., increased uncertainty). N AC = N AC (1 I 2 AC ) (2) Where the sample size of each meta-analysis is penalized by multiplying by the lack of homogeneity (1 I 2 AC ). We can retrospectively estimate the power (1-β) of the current data by re-arranging the expression of the required sample in eq. 1 (or eq. 1 & 2 in case of heterogeneity). 2. ITC Building upon this approach, Thorlund and Mills (Thorlund & Mills 2012) proposed a simple algorithm for the estimation of power in case of ITC (i.e. 3 treatments A, B, C with head to head trails available for A vs C and B vs C). The effective power in an indirect comparison is calculated as follows: 1. Calculate the effective indirect sample size The effective indirect sample size in case of low heterogeneity is given by the following formula: N AB = N AC N BC N AC + N BC (3) The heterogeneity-corrected effective sample size for indirect evidence is given by: N AB = N AC(1 I 2 AC ) N BC (1 I 2 BC ) N AC (1 I 2 AC ) + N BC (1 I 2 BC ) (4) An important limitation of the above-proposed sample size heterogeneity correction in meta-analyses with limited number of trials (as in our analysis for ANORO) is the fact that I 2 s are typically unreliable and unstable. In most cases, it will therefore be preferable to simply assume some plausible degree (percentage). Typically, an assumption of 25% or 50% heterogeneity will be reasonable in the context of sample size considerations (Wetterslev et al 2009). 2. Calculate the power of the available indirect evidence Insert the effective indirect sample size (eq. 3 or 4) in the formula used for the metaanalyses: N AB = C z 1 a/2 z 1 β 2 v AB μ AB (5) and estimate the power (1-β) of the current data by re-arranging the expression of the required sample. 3. Network meta-analyses (NMA) Power estimation in NMA, especially in the case of complicated networks (e.g. multiple, closed loops) is not straightforward and - to our knowledge - there are no well-established and widelyacceptable methods. A simple method is suggested by Thorlund and Mills (Thorlund & Mills 2012), as an extension of their ITC power calculation method. We could try this approach in more complicated networks and assess the results. 17

197 II. Bayesian approach In a Bayesian framework, the classical power estimation approach is not valid, as the (frequentist) statistical tests and the (frequentist) hypothesis testing are not applicable - see for example Spiegelhalter, Abrams, Myles Bayesian Approaches To Clinical Trials and Health-Care Evaluation, Wiley In the Bayesian approach, the inference is not based on one value (i.e. p-value) but on the posterior distribution from which the probabilities of interest can be derived. To our knowledge, there is no alternative to the (frequentist) power estimation methods listed above developed for the Bayesian ITC and NMA approach. 18

198 3.5. HYPOTHESES 1. UMEC/VI 125mcg/25mcg OD is comparable to indacaterol 150mcg plus tiotropium 18mcg OD, salmeterol/fluticasone 50mcg/500mcg and QVA149 by means of lung function (trough FEV1) and rescue medication use at the time points of interest. 2. UMEC/VI 62.5mcg/25mcg OD is comparable to indacaterol 150mcg plus tiotropium 18mcg OD, salmeterol/fluticasone 50mcg/500mcg and QVA149 by means of lung function (trough FEV1) at the time points of interest. 3. UMEC/VI 125mcg/25mcg OD is comparable to QVA149 by means of health related quality of life (SGRQ score), dyspnea (TDI score) and rescue medication use at the time points of interest. 4. UMEC/VI 62.5mcg/25mcg OD is comparable to QVA149 by means of health related quality of life (SGRQ score), dyspnea (TDI score) and rescue medication use at the time points of interest DATA ANALYSIS CONSIDERATIONS Frequentist Indirect Treatment comparison An indirect treatment comparison (ITC) of a treatment A versus a treatment B can be performed if it is anchored on a third treatment C (the term adjusted is also used in the literature). Fig. 1: ITC network For the indirect comparison the method proposed by Bucher et al 1997 will be used. To derive indirect comparisons of A versus B, this method compares the magnitude of treatment effects of these interventions relative to a common comparator (i.e. AC and BC). Binary data Step 1: The pooled relative risks (RR) for each treatment of interest relative to a common comparator (i.e. RR AC for A vs C and RR BC for B vs C) will be estimated using standard random effects meta-analyses, as proposed by DerSimonian and Laird. Step 2: The relative risks for the indirect comparison of treatments A versus B may then be estimated by taking the ratio of the two relative risk ratios: RR AB = RR AC RR BC (6) so, the indirect comparison can be calculated as: 19

199 ln (RR AB ) = ln (RR AC ) ln (RR BC ) (7) Given that the relative risks are estimated based on different studies and therefore are independent, the variance for this effect is the pooled variance, same for the standard error: var(ln (RR AB )) = var(ln (RR AC )) + var(ln (RR BC )) SE(ln (RR AB )) = SE(ln (RR AC )) 2 + SE(ln (RR BC )) 2 (8) Continuous data Step 1: The pooled mean difference in change from baseline for each treatment of interest relative to a common comparator (i.e. AC and BC) will be estimated using standard random effects meta-analyses, as proposed by DerSimonian and Laird. Step 2: Let μ AC, SE(μ AC ) and μ BC, SE(μ BC ) be the combined mean difference and corresponding standard error of the A vs C and B vs C meta-analytic comparisons, respectively. The indirect comparison of treatments A versus B may then be estimated as follows: μ AB = μ AC μ BC var(μ AB ) = var(μ AC ) + var(μ BC ) (9) SE(μ AB ) = SE(μ AC ) 2 + SE(μ BC ) 2 Heterogeneity The statistical heterogeneity will be assessed by means of the Cochran Q, chi-square test and the I 2 statistic with 95% CI. The clinical heterogeneity will be assessed by means of a) study design inclusion criteria related to FEV 1, FEV 1 /FVC, exacerbations and smoking background treatment: ICS and/or LABA randomization blinding open label arms cross over design b) baseline patient characteristics exacerbation history % patients per COPD severity level mean FEV 1 and mean FEV 1 % predicted % current smokers mean pack-years % of male patients 20

200 mean age If confounders are present in an indirect comparison, it is only possible to adjust for them via meta-regression. This can be done by replacing Step 2 with a random effects meta-regression in order to estimate the comparative treatment effects between two drugs (e.g. by using the STATA metareg). However, in our analysis due to the relatively low number of studies it is not feasible to perform adjustment via meta-regression. We will therefore conduct a separate analysis excluding the evidence where such clinical heterogeneity exists and its feasible to do so. Results of the meta-analyses Results of the individual trials (continuous: point estimates and 95% confidence intervals, dichotomous: number with the event, number in the group, RR) together with the pooled results (point estimates and 95% confidence intervals) the Cochran Q, the I 2, and chi-square test results, will be presented on a forest plot. Results of ITC The results of the ITC will be presented as RR (or mean difference in CFB) with 95% CI, and p- value. The random effects pairwise meta-analyses will be conducted with STATA using metan (StataCorp Stata Statistical Software: Release 11) or R (R Development Core Team, using the packages meta and metafor. Bayesian Network Meta-Analysis Network meta-analysis have been presented as an extension of traditional meta-analysis (where all included studies compare the same intervention with the same comparator) by including multiple different pair-wise comparisons across a range of different interventions (Caldwell et al., 2005; Lu & Ades, 2004). With network meta-analysis the efficacy of a particular intervention versus competing interventions can be obtained in the absence of head-to-head comparisons; indirect treatment comparison (ITC) of two interventions is made via a common comparator. Figure 2 represents the situation when, in addition to interventions B and C, the intervention D is of interest as well. For this latter intervention, direct estimates from BD trials are available. Given the network of direct comparisons across the range of interventions, indirect estimates can be obtained for d AB, d AC, d BC, d CD, and d AD. Given the mathematical relations between the true underlying estimates of the different comparisons in the network, we have both direct and indirect evidence available for all the pair wise comparisons, except for the CD comparisons (only indirect evidence) and the BD comparisons. The evidence network consists of loops of evidence and the analysis is called a Mixed Treatment Comparison (MTC). Hence, the advantages of the simultaneous analysis with MTC are that 1) estimates for indirect comparisons are obtained and 2) indirect comparisons can support evidence for direct estimates. If there are inconsistencies between direct and indirect evidence for particular loops there will be bias in the estimates. 21

201 A B d AB d AC C D d BC d BD A B C D Direct estimate Indirect estimate Figure 1. Network of evidence reflecting mixed treatment comparisons of trials Potential Bias in Network Meta-Analysis With a (network) meta-analysis, the value of randomization only holds within a trial, and not across trials. As a result there is the risk that patients assigned to the different trials are not comparable. If the distribution of treatment-effect modifiers (e.g. % of very severe patients) is not similar across trials comparing different interventions in the network of studies, transitivity is violated, the studies are heterogeneous, and the results will be biased. Potential effect modifiers in our analysis are the following: exacerbation history % patients per COPD severity level mean FEV 1 and mean FEV 1 % predicted at baseline % current smokers % patients using ICS Use of LABA as background treatment The similarity of studies included in each network will be assessed by means of the effect modifiers listed above. Furthermore, the risk of bias for the selected studies will be assessed using the Cochrane Risk of Bias Tool (Higgins et al., 2011). Evidence inconsistency Consistency relates to the loops of evidence. If the evidence networks consist of closed loops, the results of independent means models (i.e. only pooling direct evidence) can be compared with those obtained from the network meta-analysis models (i.e. consistency models). If there is a discrepancy between direct and indirect evidence for a particular contrast, possible sources of inconsistency could be explored. Models 22

202 Below, the models for network meta-analysis are presented that will be evaluated for the evidence set. The models, and the corresponding WinBUGS codes, are based on those published by Dias et al The random effects models are presented in details, the fixed effect model is obtained by setting 2 σ to zero. Bayesian Random effects network meta-analysis model for normally distributed outcomes diff 2 jbk ~Normal(δ jbk,σ jbk ) (Likelihood) 2 2 jbk~normal(dbk,σ ) = Normal(d Ak d Ab,σ ) δ d AA = 0 (Random effects model) d Ak ~Normal( 0, ) σ~uniform( 0, 10 ) (Prior distributions) Where j = study b = control group, can be treatment A, B, C k = treatment group, can be B, C, D diff jbk = δ jbk = σ jbk = d Ak = 2 σ = Observed difference in change from baseline for a continuous outcome for intervention k versus reference treatment b in study j difference in change from baseline for treatment k relative to comparator b in study j uncertainty ( sampling error ) in the difference in change from baseline for intervention k versus reference treatment b in study j pooled difference in change from baseline for treatment k relative to reference treatment A constant variance of relative effects for each treatment versus reference treatment A across studies as a reflection of heterogeneity Bayesian Random effects network-meta-analysis model for binary outcomes r jk ~Binomial logit (p δ d μ d jbk AA jb Ak = 0 jk (p jk,n jk ) μ jb b = A,B,C,... if k = b ) = μ jb + δ jbk if k alphabetically after ~Normal(d bk σ~uniform( 0, 2 ),σ ~Normal( 0, ) ~Normal( 0, ) 2 ) = Normal(d Ak d Ab,σ 2 ) b (Likelihood) (Random effects model) (Prior distributions) Where j = study b = control group, can be treatment A, B, C k = treatment group, can be B, C, D r jk = number of events observed for the treatment k in study j p jk = probability of an event for treatment k in study j n jk = number of subjects for treatment k in study j µ jb = log odds of an event for baseline treatment b in study j δ jbk = log odds ratio for treatment k relative to treatment b in study j 23

203 d bk = pooled log odds ratios of an event with treatment k relative to treatment b 2 σ = between study variance or heterogeneity parameter d Ak = pooled log odds ratio of an event with treatment k versus A d Ab = pooled log odds ratio of an event with treatment b versus A Bayesian Approach vs. Traditional (frequentist) Approach Network meta-analysis can be performed within a frequentist or Bayesian framework. Within the frequentist framework, estimation of model parameters and inference are based on some form of maximum likelihood. The output of the analysis is a point estimate along with a p- value or a 95% confidence interval as a measure of uncertainty. The 95% confidence intervals cannot be interpreted in terms of probabilities. The 95% confidence interval does not mean there is a 95% probability that the true value is between the boundaries of the interval. Within the Bayesian framework, parameters are considered random variables. The belief regarding the possible values of a model parameter before looking at data can be summarized with a probability distribution: the prior distribution. This probability distribution will be updated after having observed the data, resulting in the posterior distribution summarizing the updated probabilities of the values for this parameter. Hence, within the Bayesian framework, the analysis not only involves data, a likelihood distribution, and a model with parameters reflecting the treatment effects and impact of covariates, but also prior probability distributions for these parameters. In other words: Bayesian methods involve formal combination of a prior probability distribution with a (likelihood) distribution to obtain a posterior probability distribution of model parameters. The output of the Bayesian network meta-analysis is a complex posterior distribution of all relative treatment effects between interventions included in the network. The posterior distribution for each relative treatment effect can be summarized with a mean or median to reflect the most likely value for the effect size, as well as the 2.5th and 97.5th percentile: the 95% credible interval. Unlike the 95% confidence interval of the frequentist framework, the 95% credible interval can be interpreted in terms of probabilities: there is a 95% chance that the true parameter value falls between the boundaries of the credible interval. A major advantage of the Bayesian approach is that the method naturally leads into a decision framework to support decision-making. With the Bayesian approach for network meta-analysis, the multiple inferences based on confidence intervals or p-values can be replaced with probability statements: For example there is x% probability that treatment C is better than B, or there is a y% probability that treatment D is the most efficacious out of treatment A, B, C, D, and E regarding this outcome. Likelihood and Link-Functions To perform the network meta-analyses within a Bayesian framework, likelihood distributions need to be defined to relate the data to the parameters of the models. An overview of the likelihood and link function for different types of outcome data in the available evidence base is provided below. Given the different nature of the outcomes of interest, different statistical models will be used 24

204 for the network meta-analysis: c) For continuous outcomes (e.g trough FEV 1, SGRQ total score, TDI focal score) a generalized linear model with identity link and a normal likelihood distribution will be used. The analysis will be based on the difference between the change from baseline (CFB) for each treatment versus the CFB for the comparator at the time point of interest, as well as the associated standard error (SE) of the difference. Priors d) For dichotomous outcomes a logit link with binomial likelihood distribution will be used. The analysis will be based on the number of patients included in the analysis and the number of patient who experience an event. If the number of patients with an event is reported it will be used directly and alternatively it will be calculated as a function of the proportion of patients randomized. Vague (uninformative) priors are expected to be used for all calculations. A normal distribution with zero mean and variance equal to 10 4 will be used for treatment effects and a uniform distribution with range zero to 5 for the between-trial standard deviation. Convergence The posterior densities for the unknown parameters will be estimated using Markov chain Monte Carlo (MCMC) simulations for each model. As a standard, the runs will be based on 80,000 iterations on three chains, with a burn-in of 20,000. Convergence assessment will be based on visual inspection of trace and autocorrelation plots and on the Gelman Rubin-Brooks diagnostic. The accuracy of the posterior estimates will be assessed using the Monte Carlo error for each parameter (Monte Carlo error < 5% of the posterior sd). Goodness of fit For each outcome, fixed and random effects model will be evaluated. In order to identify the most appropriate model given the evidence base the goodness-of-fit of model predictions to the observed data can be measured by calculating the posterior mean residual deviance, D. The deviance information criterion (DIC) (Spiegelhalter et al 2002) will be used to compare the fixed and random effects model. The model with the lowest DIC, and therefore the model providing the best fit will be considered the base-case model given the dataset used. Software The Bayesian network meta-analyses will be conducted with WinBUGS or OpenBUGS. Presentation of the results The Bayesian network meta-analysis will provide posterior distributions of the relative treatment effects between interventions for each outcome of interest. The posterior distributions will be summarized with the median to reflect the most likely value of the estimate, and the 2.5th and 97.5th percentile to capture the 95% Credible Interval (95%CrI). Furthermore, the posterior distribution will be summarized with the probability that a treatment results in greater outcomes than the comparator. These results will be reported in a tabular format per outcome 25

205 (see Table 1). The probability that each treatment is ranked at a certain position (e.g. 60% probability that treatment X is ranked 1 st ) out of all compared interventions will be calculated and the expected rank will be reported. Table 1. Mock-up results table for relative effects Intervention Comparator Placebo Drug X Drug Y estimate 0.73 Drug X 95% CrI (0.66, 0.82) 1.00 P(better) >99% estimate Drug Y 95% CrI (0.68, 0.93) (0.91, 1.29) 1.00 P(better) >99% 17% estimate Drug Z 95% CrI (0.51, 1.09) (0.69, 1.52) (0.63, 1.42) P(better) 94% 46% 62% 26

206 4. LIMITATIONS The limitations of an evidence synthesis are well known in bibliography (see e.g. Salanti 2012). Publication bias: The evidence synthesis will be based on the data reported in the published studies identified by the SLR. There is a degree of publication bias present since some clinical trials fail to be published while others are published in abstract form, but not as full reports and thus will present limited information. Trials published in languages other than English were also excluded. Publication of outcome data relevant to this evidence synthesis may be incomplete or absent in a report, in which case, the publication must be excluded as well. These processes may lead to reporting bias. Inhomogeneity: In many cases the assessment of clinical homogeneity is limited by incomplete reporting of baseline patient characteristics, study design, outcomes definitions etc. 5. STUDY CONDUCT, MANAGEMENT & ETHICS 5.1. ETHICS/IRB APPROVAL N/A - Non interventional study 5.2. INFORMED CONSENT N/A - Non interventional study 5.3. DATA PRIVACY Only published trial data used 5.4. PERSONALLY IDENTIFIABLE INFORMATION (PII) N/A No PII used 5.5. AE REPORTING Only previously published or disclosable data used 5.6. DATA STORAGE/ARCHIVAL All data will be stored according to GSK policy & archived to PIER with appropriate meta-data. 6. EXTERNAL INVOLVEMENT 6.1. Third Party Supplier Mapi De Molen 84, 3995 AX Houten, The Netherlands Tel.: Fax: PhD 27

207 Associate Director HEOR & Strategic Market Access Mobile: PhD Senior Research Manager HEOR & Strategic Market Access Hamilton House, Mabledon Place, London WC1H 9BB, United Kingdom Tel.: Mobile: 6.2. External Expert/Health Care Professionals (Consultants & Research PIs) N/A 28

208 MILESTONES MILESTONE Forecast Final Protocol Approval GUIDANCE OR POLICY REQUIREMENT FORECAST DATE MM-YYYY XX Forecast GSK CSR Protocol Summary FPA Actual + 30 days XX Forecast Statistical Analysis Plan Approved XX Forecast Statistical Analysis Complete Forecast Final Study Report Complete Forecast GSK CSR Results Summary Posting Forecast Manuscript Submission SAC Actual + 6 months SAC Actual + 8 months SAC Actual + 18 months 29

209 DATA DISSEMINATION PLAN Attach the Manuscript & Congress Presentation Data Dissemination Plan (DDP) in the MCQP DDP Powerpoint template. Should include the following information: MANUSCRIPT PUBLICATION(S) Note: Only 1 primary manuscript per study is permitted unless approval from Medical. STUDY ID PUBLICATION SHORT TITLE LEAD AUTHOR STUDY ACCOUNTABLE PERSON ESTIMATED SUBMISSION DATE (< SAC + 18 months) TARGET JOURNAL CONGRESS PRESENTATION(S) STUDY ID ABSTRACT SHORT TITLE PRESENTER POSTER OR ORAL PRESENTATION CONGRESS CONGRESS LOCATION CONGRESS DATE DISCLOSURE PLAN GSK CSR PROTOCOL SUMMARY (FPA + 30 DAYS) GSK CSR RESULTS SUMMARY (SAC + 8 MONTHS) FULL PROTOCOL POSTING DATE (Manuscript submission actual + 30 days) FORECAST 30

210 Appendix 1: Statistical Analysis Plan Details Analysis #1: UMEC/VI vs Indacaterol + Tiotropium Outcome Difference in CFB for trough FEV 1 in L. Difference in CFB in rescue medication use in puffs/day Time point: 12 weeks Treatments UMEC/VI 125mcg/25mcg OD UMEC/VI 62.5mcg/25mcg OD Indacaterol 150mcg + Tiotropium 18mcg OD List of included studies INTRUST1 (Mahler 2012) INTRUST2 (Mahler 2012) GSK/DB GSK/DB INTRUST studies are not including very severe patients, while GSK studies are including 8-12%. As a scenario, the ISE (Integrated Summary of Efficacy) results will be used for ANORO. Network Models Figure 3. Network of evidence a. Frequentist ITC UMEC/VI 125/25 will be compared indirectly with IND TIO 18 via TIO 18. Two separate random effects meta-analyses will be performed: o UMEC/VI 125/25 vs. TIO 18 o IND TIO 18 vs. TIO 18 Subsequently the ITC will be performed using formulas (2),(3). 31

211 UMEC/VI 62.5/25 will be compared indirectly with IND TIO 18 via TIO 18, following the same steps. Results b. Bayesian network meta-analysis All treatments presented in the network (Fig. 3) will be included in a network meta-analysis. A generalized linear model with identity link, normal likelihood distribution and uninformative priors will be used. Random and fixed effects versions of the model will be evaluated and the goodness of fit will be assessed by the DIC value. The results of the model with the lower DIC will be presented. a. Frequentist ITC Mean difference in CFB for trough FEV 1 between UMEC/VI and IND150 + TIO18 at 12 weeks, 95% CI and p-value. b. Network Meta-Analysis All pairwise comparisons with the corresponding CrI, together with the probability of being a better treatment (see Table 1). The probability that each treatment is ranked at a certain position out of all interventions in compared (Fig. 3) will be calculated and the expected rank will be reported. Analysis #2: UMEC/VI vs Fluticasone/Salmeterol + Tiotropium 18 Outcome Difference in CFB in Trough FEV 1 in L. Time point: 12 weeks, 24 weeks Treatments UMEC/VI 125mcg/25mcg OD UMEC/VI 62.5mcg/25mcg OD Fluticasone / Salmeterol 500mcg/50mcg BD + Tiotropium 18 OD List of included studies Aaron 2007 Cazzola 2007 GSK/DB GSK/DB As a scenario, the ISE (Integrated Summary of Efficacy) results will be used for ANORO. The second scenario included trough FEV1 at 20 weeks (Aaron 2007) with trough FEV1 at 24 weeks (GSK trials). Cazzola 2007 includes high percentage of current smokers 83%-93% compared to GSK studies (42%- 58%), reporting also higher percentage of ICS use (70%-77%) compared to GSK studies (40%-56%). Aaron 2007 only reported trough FEV1 in a graph without standard error, standard deviation or a confidence interval. Instead, the Cochrane report of Karner et al (2011) was used to extract trough FEV1 at 20 weeks. 32

212 Networks Figure 4. Network of evidence at 12 weeks. Figure 5. Trough FEV1 at 20/24 weeks: Network of evidence Model a. Frequentist ITC UMEC/VI 125/25 will be compared indirectly with FLUT/SAL 500/50 via TIO 18. See Analysis #1 for details on the approach. UMEC/VI 62.5/25 will be compared indirectly with FLUT/SAL 500/50 via TIO 18. See Analysis #1 for details on the approach. b. Network Meta-Analysis See Analysis #1 for details on the approach. Results a. Frequentist ITC Mean difference in CFB for FEV1 between UMEC/VI and comparators, 95% CI and p-value. b. Network Meta-Analysis 33

213 As for Analysis #1. Analysis #3: UMEC/VI vs QVA149 Outcomes Difference in CFB for Trough FEV 1 in L Difference in CFB for SGRQ total score Difference in CFB for TDI focal score Difference in mean number of puffs of rescue medication Time points: 12, 24 and 52 weeks Treatments UMEC/VI 125mcg/25mcg OD UMEC/VI 62.5mcg/25mcg OD Indacaterol 110 mcg/glycopyrronium 50 mcg (QVA149) OD List of included studies SHINE SPARK ENLIGHTEN GSK/DB GSK/DB GSK/DB GSK/DB GSK/DB Three scenarios will be performed: Scenario including the SPARK study Scenario excluding the SPARK study Scenario using the ISE (Integrated Summary of Efficacy) results for ANORO Overview of networks Excluded studies: ILLUMINATE: Comparator BRIGHT: Study design (three way cross-over, each medication used for 3 weeks) Table 2. Overview of outcomes of interest per trial SHINE SPARK ENLIGHTEN FEV1 trough at 12 weeks FEV1 trough at 12 weeks FEV1 trough at 12 weeks FEV1 trough at 26 weeks FEV1 trough at 24 weeks FEV1 trough at 26 weeks SGRQ at 26 weeks FEV1 trough at 52 weeks FEV1 trough at 52 weeks TDI at 12 weeks SGRQ at 12 weeks Rescue med. at 52 weeks TDI at 26 weeks SGRQ at 24 weeks Rescue med. at 26 weeks SGRQ at 52 weeks Trough FEV1 at 12 weeks ENLIGHTEN (graph) 34

214 SPARK (graph) SHINE DB DB DB DB DB Trough FEV1 at 24 weeks ENLIGHTEN (graph) SPARK (graph) SHINE DB DB DB DB DB

215 Trough FEV1 at 52 weeks ENLIGHTEN (graph) SPARK (graph) DB SGRQ at 12 weeks SPARK (graph) DB DB DB DB SGRQ at 24 weeks SPARK (graph) SHINE DB DB DB

216 DB SGRQ at 52 weeks SPARK (graph) No ITC possible, only one trial TDI at 12 weeks SHINE (graph) DB DB DB DB TDI at 24 weeks SHINE DB DB DB

217 DB TDI at 52 weeks No trial data identified Rescue medication at 12 weeks No trial data identified Rescue medication at 24 weeks SHINE DB DB DB DB Rescue medication at 52 weeks ENLIGHTEN (graph) DB

218 Model a. Frequentist ITC UMEC/VI 125/25 will be compared indirectly with QVA149 via TIO 18 and via Placebo. Four separate random effects meta-analyses will be performed: o UMEC/VI 125/25 vs. TIO 18 o QVA149 vs. TIO 18 o UMEC/VI 125/25 vs. Placebo o QVA149 vs. Placebo Subsequently the ITC using formulas (2),(3) will be performed. The results of the separate ITCs (via Tio and via Placebo) could be synthesized by means of a meta-analysis UMEC/VI 62.5/25 will be compared indirectly with QVA149 via TIO 18 and via Placebo, following the same steps. b. Network Meta-Analysis As for Analysis #1. Results a. Frequentist ITC Mean difference in CFB for FEV1 between UMEC/VI and QVA149 at 12, 24 and 52 weeks, 95% CI and p-value. b. Network Meta-Analysis As in Analysis #1. 39

219 Appendix 2: Systematic Literature Review Protocol HEALTH OUTCOMES PROTOCOL UNIQUE IDENTIFIER HO / etrack # ABBREVIATED TITLE FINAL PROTOCOL APPROVED 31 May 2013 FULL TITLE Evidence Synthesis and Mixed Treatment Comparisons of COPD Treatments Protocol for Evidence Synthesis and Meta Analysis Framework for Evaluating Treatments for COPD Patients GSK ASSET DISEASE AREA SPONSORSHIP DIVISION/BUSINESS UNIT DEPARTMENT HO STUDY ACCOUNTABLE PERSON(S) CONTRIBUTING AUTHORS RETENTION CATEGORY INFORMATION TYPE KEY WORDS / MESH HEADINGS / META DATA Relvar (Fluticasone Furoate and Vilanterol (FF/VI). COPD GSK Sponsored Global Research & Development Global Health Outcomes Mixed Treatment Comparisons COPD Evidence synthesis Systematic review Meta analysis Mixed treatment comparison Efficacy Effectiveness Safety Tolerability GlaxoSmithKline 1

220 SPONSOR SIGNATORY NAME HERE HO Study Accountable Person DD-MMM-YYYY Date Group HO VP/Director Date Medical (GML/MDL/PPL/PMAL/RMD/Equivalent) Date Intellectual Property Date Trademarks Date Global Clinical Safety & Pharmacovigilence Date Statistics Date Enter other signatory Name/Role as needed Date HO PROTOCOL REVIEW COMMITTEE APPROVAL HO PRC CHAIR Date Approved GlaxoSmithKline 2

221 PROTOCOL SYNOPSIS Unique Identifier HO / etrack # Abbreviated Title GSK Product Evidence Synthesis and Mixed treatment comparisons of COPD Treatments Relvar (Fluticasone Furoate and Vilanterol (FF/VI) GW GW Rationale Payers, HTA bodies and other decision makers in many countries require a formal comparison of clinical efficacy of Relvar/Breo (FF/VI) with other available ICS/LABAs. This cannot be provided from direct head to head studies alone. This protocol is for an evidence synthesis of therapies for the treatment of COPD with the goal of identifying and synthesizing clinical trial evidence for Relvar/Breo (FF/VI) versus other available ICS/LABAs. The information from a recently completed literature review will be used to develop an integrated Bayesian evidence synthesis model. Objectives (Primary, Secondary & Exploratory) Primary Objective To determine the relative treatment efficacy of Relvar compared with the class of other ICS/LABAs in the treatment of COPD, based on improvement in lung function measured as the mean difference between treatments in FEV1 change from baseline Secondary Objectives A. To determine the relative treatment efficacy of Relvar on lung function (FEV1) compared with each of 1. Seretide (fluticasone propionate + salmeterol) 2. Symbicort (budesonide + formoterol) used in fixed dose treatment regimens B. To determine the relative treatment efficacy of Relvar on the yearly or annualized rate of the composite of moderate and severe exacerbations, compared with 1. Seretide (fluticasone propionate + salmeterol) 2. Symbicort (budesonide + formoterol) used in fixed dose treatment regimens C. To determine the relative treatment efficacy of Relvar on the yearly or annualized rate of any exacerbations, compared with the alternative treatments listed in points A, B, data permitting; D. To assess the relative treatment efficacy of Relvar on SGRQ compared with the alternative treatments listed in points A, B, in the general study populations E. To assess the probability that Relvar is non inferior to the class of GlaxoSmithKline 3

222 other ICS/LABAs in the treatment of COPD, based on improvement in lung function measured as the relative risk of FEV1 change from baseline; the non inferiority thresholds considered will be a range selected based on clinical opinion and on non inferiority thresholds from other non inferiority studies identified in the literature review. F. To determine the relative treatment efficacy of Relvar on the yearly or annualized rate of the composite of moderate and severe exacerbations, compared with the alternative treatments listed in points A, B above in individuals with COPD and a history of exacerbations, as documented in the supporting clinical trial publications. This scenario therefore will include studies in populations for which an exacerbation related criterion has been used as trial inclusion criterion and where this criterion has been explicitly reported, therefore an alternative set to the general population of studies included in objective B. GlaxoSmithKline 4

223 Study Design Study Population and Sampling Methods A set of Bayesian, hierarchical, mixed treatment comparisons (MTCs) will be developed. The models will incorporate known predictors of treatment outcome and the observed efficacy of treatment strategies evaluated in randomized clinical trials. The models will be used to make inferences on multiple measures of efficacy included in this protocol associated with existing therapies for COPD. Studies included in the MTC are studies of clinical efficacy and effectiveness in adolescents and adults ( 12 years) patients of any race and gender. The disease area of interest for the current review is COPD irrespective of the disease stage. Studies will also be included whether or not they have an explicit exclusion criterion based on history of exacerbations to determine eligibility of patients into the study; other than for objective F. Studies included are randomised controlled trials (RCTs) irrespective of blinding status, either Phase III or phase IV parallel group RCTs. Additionally, systematic reviews pertinent to the review objective will also be included for bibliographic searching of relevant RCTs not retrieved through primary database searching. Table 3 in the Appendix (systematic lit review protocol) provides a detailed list of study designs for inclusion and exclusion. History of exacerbations at baseline Most COPD studies prespecify a restriction of eligibility at enrolment based on exacerbations. There are two types of restrictions normally mentioned in trial reports: Exclusion of individuals that have or have had an exacerbation in the previous near period to recruitment, often one month or 2 4 weeks Inclusion or exclusion of individuals with a documented history of past exacerbations at least over 6 months or more prior to recruitment. These two criteria have different aims. Criterion 1 is applied to exclude individual that have an ongoing exacerbation at entry to the study, therefore it can be sen as an operational requirement, and in general is not aimed at selecting participants on the basis of risk. Criterion 2 on the contrary aims at restricting the study population based on the risk of exacerbations, i.e. either selecting a low risk (in case of no history of exacerbations) or high risk population (in case of history of exacerbations). COPD populations however are in general at high risk of exacerbations and therefore this criterion, when GlaxoSmithKline 5

224 applied, aims to identify the truly high risk populations within a specific period in time of (generally) one year. In the MTC, studies that aim to recruit patients from the general COPD population will be the study population for the primary and secondary research questions. The main analyses in this MTC will therefore be in populations with unqualified risk of exacerbations, i.e in populations for which an explicit exclusion criterion based on history of exacerbations in the year prior to recruitment was used to determine non eligibility of patients into the study (so, criterion 2 not applied for the year preceding recruitment). It should be noted that this is compatible with the application of criterion 1. A subanalysis will be conducted specifically in studies conducted in exacerbating populations (i.e. criterion 2 applied) for Objective F above. Data Source Data Analysis Methods The data used in the MTCs will be obtained from a systematic literature review of efficacy and effectiveness that was conducted separately. An independent protocol for the systematic review of the literature is available and has been included in Appendix to this Protocol Primary efficacy endpoint: FEV1: value at baseline and any specified time point, change from baseline, difference in change from baseline between groups, mean difference in FEV change from baseline, AUC FEV1; Secondary efficacy endpoints: Exacerbations: Mean rate of exacerbations per patient per year or annualised total number of exacerbations experienced over the course of the study (moderate; composite of moderate and severe; severe); Mean rate of exacerbations per patient per year or annualised total number of exacerbations experienced over the course of the study (any exacerbation as per study definition) Health status: St. Georges Respiratory Questionnaire (SGRQ) Note on Exacerbations Contrarily to other endpoints, the definitions of exacerbations vary across studies. The definition of exacerbations in the MTC reflects the various definitions used in the trials; in other words, the inclusion of each GlaxoSmithKline 6

225 study in the MTC will not be based on the use of a particular specific definition of exacerbations. Studies differ in their recording practice for exacerbation endpoints. Definitions used in the trials include the following: GlaxoSmithKline 7 deterioration in COPD leading to hospitalisation/emergency (or equivalent) treatment and/or the need for oral corticosteroid treatment for at least three days deterioration in COPD requiring treatment with an oral corticosteroid or an emergency department visit or hospitalization clinical exacerbation requiring emergency treatment or hospitalization exacerbation requiring emergency treatment, hospitalization, or use of a COPD medication not allowed by the protocol. Such definitions assume that exacerbation, in fact, is a composite endpoint comprising of any exacerbation of any level of severity. In other studies, exacerbations are explicitly classified or reported as severe, moderate or mild. Importantly, some studies use explicit definition of the severity of exacerbation endpoints as a restriction to the type of exacerbation data collected. So in some studies, the composite endpoint includes exacerbations of any severity, i.e. mild, moderate or severe; in other studies, the composite includes exacerbations that are moderate or severe. Classification criteria may also vary by study. For example, severe exacerbations are defined as deterioration in COPD requiring emergency hospital treatment or alternatively, any hospitalizations. Moderate exacerbations required administration of additional inhaled corticosteroids, bronchodilators and/or oral corticosteroids, or alternatively, as worsening of COPD leading to a prescription for a short use of oral corticosteroids. Mild exacerbations are deterioration in COPD requiring an increase in the use of relief medication which the physician considered to be clinically relevant. For further details r.e. the specification and definition of exacerbation endpoints used in the MTC, pls refer to the Appendix. Definition of outputs of the MTC The results of the mixed treatment comparison will be expressed on the original scale of the data extracted from the systematic review. MTC endpoints are FEV1, exacerbations, and health status scores: FEV1 is expressed in Litres and is a continuous outcome. Treatment effects calculated in the MTC will be expressed as treatment differences (for treatment X compared to treatment Y) in mean difference from baseline to last measurement. Exacerbations are a count variable and are expressed either

226 as total number of exacerbations (with treatment X and Y), yearly exacerbation rates (exacerbations per patient year, for treatment X and Y) or as relative risk ratio or odds ratios for treatment X vs treatment Y. The total number of exacerbations will be recalculated from the data extracted from the literature. Treatment effects calculated in the MTC will be expressed as risk ratios for treatment X compared with treatment Y. SGRQ results will be summarized as mean difference in score changes from baseline. Credible limits (CrL, 95%) and posterior distribution will be estimated for each endpoint. Sample Size and Power Limitations While the use of Bayesian hierarchical models require a number of assumptions (e.g., exchangeability of information at each level of the hierarchy, prior information), the ability of the model to incorporate the structure and limitations of the available information allows greater model validity than other approaches (e.g., traditional frequentist multi level modeling). These models are very robust, because they are so flexible. This mitigates to some degree typical worries regarding model dependencies and the validity of initial assumptions. When using such a model to make mixed treatment comparisons, one must often make assumptions about the additivity of treatment effects across studies. For example, if treatment A is compared to treatment B in one study and treatment B is compared to treatment C in another study, then an assumption of the additivity of treatment effects is needed to make any comparison between A and C. This can be a difficult assumption to test because of the lack of repeated single trials. All assumptions will be laid out in detail and explained thoroughly in the final study report. GlaxoSmithKline 8

227 TABLE OF CONTENTS GlaxoSmithKline 9

228 ABBREVIATIONS GlaxoSmithKline 10

229 1. INTRODUCTION/BACKGROUND Payers, HTA bodies and other decision makers in many countries require a formal comparison of clinical efficacy of Relvar/Breo (FF/VI) with other available ICS/LABAs. This cannot be provided from direct head to head studies alone. This protocol is for an evidence synthesis of therapies for the treatment of COPD with the goal of identifying and synthesizing clinical trial evidence for Relvar/Breo (FF/VI) versus other available ICS/LABAs. The information from a recently completed literature review will be used to develop an integrated Bayesian evidence synthesis model. 2. OBJECTIVES 2.1. Primary To determine the relative treatment efficacy of Relvar compared with the class of other ICS/LABAs in the treatment of COPD, based on improvement in lung function measured as the relative risk of FEV1 change from baseline 2.2. Secondary A. To determine the relative treatment efficacy of Relvar on lung function (FEV1) compared with 1. Seretide (fluticasone propionate + salmeterol) 2. Symbicort (budesonide + formoterol) used in fixed dose treatment regimens B. To determine the relative treatment efficacy of Relvar on the yearly or annualized rate of the composite of moderate and severe exacerbations, compared with 1. Seretide (fluticasone propionate + salmeterol) 2. Symbicort (budesonide + formoterol) used in fixed dose treatment regimens C. To determine the relative treatment efficacy of Relvar on the yearly or annualized rate of any exacerbations, compared with the alternative treatments listed in points A, B; D. To assess the relative treatment efficacy of Relvar on SGRQ compared with the alternative treatments listed in points A, B, in the general study populations E. To assess the probability that Relvar is non inferior to the class of other ICS/LABAs in the treatment of COPD, based on improvement in lung function measured as the relative risk of FEV1 change from baseline; the non inferiority thresholds considered will be a range selected based on clinical opinion and on non inferiority thresholds from other non inferiority studies identified in the literature review. F. To determine the relative treatment efficacy of Relvar on the yearly or annualized rate of the composite of moderate and severe exacerbations, compared with the alternative treatments listed in points A, B above and in individuals with COPD and a history of exacerbations, as documented in the supporting clinical trial publications. This scenario therefore will include studies in populations for which an exacerbation related criterion has been used as trial inclusion criterion and where this criterion has been explicitly GlaxoSmithKline 11

230 reported, therefore an alternative set to the general population of studies included in objective B. GlaxoSmithKline 12

231 3. RESEARCH METHODOLOGY 3.1. STUDY DESIGN A set of Bayesian, hierarchical, mixed treatment comparisons (MTCs) will be developed. The models will incorporate known predictors of treatment outcome and the observed efficacy of treatment strategies evaluated in randomized clinical trials. The models will be used to make inferences on multiple measures of efficacy included in this protocol associated with existing therapies for COPD STUDY POPULATION Studies included in the MTC are studies of clinical efficacy and effectiveness in adolescents and adults ( 12 years) patients of any race and gender. The disease area of interest for the current review is COPD irrespective of the disease stage. Studies will also be included whether or not they have an explicit exclusion criterion based on history of exacerbations to determine eligibility of patients into the study; other than for objective F. Studies included are randomised controlled trials (RCTs) irrespective of blinding status, either Phase III or phase IV parallel group RCTs. Additionally, systematic reviews pertinent to the review objective will also be included for bibliographic searching of relevant RCTs not retrieved through primary database searching. Table 3 in the Appendix (systematic lit review protocol) provides a detailed list of study designs for inclusion and exclusion. History of exacerbations at baseline Most COPD studies prespecify a restriction of eligibility at enrolment based on exacerbations. There are two types of restrictions normally mentioned in trial reports: Exclusion of individuals that have or have had an exacerbation in the previous near period to recruitment, often one month or 2 4 weeks Inclusion or exclusion of individuals with a documented history of past exacerbations at least over 6 months or more prior to recruitment. These two criteria have different aims. Criterion 1 is applied to exclude individual that have an ongoing exacerbation at entry to the study, therefore it can be sen as an operational requirement, and in general is not aimed at selecting participants on the basis of risk. Criterion 2 on the contrary aims at restricting the study population based on the risk of exacerbations, i.e. either selecting a low risk (in case of no history of exacerbations) or high risk population (in case of history of exacerbations). COPD populations however are in general at high risk of exacerbations and therefore this criterion, when applied, aims to GlaxoSmithKline 13

232 identify the truly high risk populations within a specific period in time of (generally) one year. In the MTC, studies that aim to recruit patients from the general COPD population will be the study population for the primary and secondary research questions. The main analyses in this MTC will therefore be in populations with unqualified risk of exacerbations, i.e in populations for which an explicit exclusion criterion based on history of exacerbations in the year prior to recruitment was used to determine non eligibility of patients into the study (so, criterion 2 not applied for the year preceding recruitment). It should be noted that this is compatible with the application of criterion 1. A subanalysis will be conducted specifically in studies conducted in exacerbating populations (i.e. criterion 2 applied) for Objective F above. Interventions for inclusion Interventions included in the MTC are the following: Fluticasone furoate in open or closed combination with vilanterol (FF/VI, Relvar): doses included 100/25; 200/25 Budesonide in open or closed combination with formoterol (BUD/FOR, Symbicort): doses included 100/6 (80/4.5), 200/6 (160/4.5) and 400/12 (320/9) Fluticasone propionate in open or closed combination with salmeterol (FP/SAL, Seretide): 100/50 (88/42); 250/50 (220/42), 500/50 (440/42) Additional interventions that will be part of the network are those that will be included in studies of relevant interventions as comparators. A non exhaustive list of additional interventions included is: BDP (Beclomethasone dipropionate): 200, 400, 500 FP (Fluticasone propionate): 100(88), 250(220), 500(440) MMF (momethasone Furoate) 100, 200, 400 CIC (Ciclesonide): 100(80), 200(160), 400(320) FOR (Formoterol): 6(4.5), 10(9), 12(9) SAL (Salmeterol) 50(42) Tiotropium 5, DATA SOURCE / DATA COLLECTION The data used in the MTCs will be obtained from a systematic literature review of efficacy and effectiveness that was conducted separately. An independent protocol for the systematic review of the literature is available and has been included in Appendix to this Protocol., The main methodological features of the literature review are listed in the table below as part of this protocol. PICO and methods for the systematic literature review Objectives and research questions GlaxoSmithKline 14

233 Objective To assess the evidence supporting the relative treatment effect of the inhaled therapies for the maintenance treatments of COPD. To assess the evidence supporting the safety and tolerability profiles of different inhaled therapies for the maintenance treatments of COPD. Studies to include Study designs Study design Randomised controlled trials, Phase 3 and Phase 4 The following study/publication types will be excluded: Studies with < 10 participants Pre clinical and animal Studies including Short acting beta agonist (SABAs) and Shortacting muscarinic antagonist (SAMAs) aimed at symptoms control Population Participants shall include individuals aged 12 years and older with an established diagnosis of chronic obstructive pulmonary disease (COPD) at any level of severity warranting the treatment with bronchodilators or corticosteroids (FEV1%p<=80%). Interventions Interventions Interventions of interest for this systematic review are: Relvar (fluticasone furoate + vilanterol) Anoro (umeclidinium + vilanterol) Umeclidinium monotherapy Closed triple (vilanterol, umeclidinium and fluticasone furoate) Comparators Comparators of interest for this systematic review are: Inhaled corticosteroid (ICS) Long acting beta agonist (LABA) Long acting muscarinic antagonist (LAMA) Any combination therapy of ICS + LABA, in combined or separate inhalers Any combination therapy of LABA + LAMA, in combined or separate inhalers Any combination therapy of ICS + LABA + LAMA, in combined or separate inhalers GlaxoSmithKline 15

234 Phosphodiesterase type 4 (PDE4) inhibitors Language No restriction on language List of non English studies of interest to be provided to GSK Publication Database: Database inception to present date timeframe Conference proceedings: (last 3 years) Data sources Databases The following databases will be searched without date limits: Medline (OvidSP) Medline In Process Citations & Daily Update (OvidSP) Embase (OvidSP) Cochrane Database Of Systematic Reviews (CDSR) (Wiley) Cochrane Central Register of Controlled Trials (CENTRAL) (Wiley) Database of Abstracts of Reviews of Effects (DARE) (Wiley) Health Technology Assessment Database (HTA) (Wiley) NIHR Health Technology Assessment Programme (Internet) PROSPERO (International Prospective Register of Systematic Reviews) (Internet) Completed and ongoing trials will be identified by searches of clinical trials registers. The following resources will be searched with no date limit: NIH Clinicaltrials.gov (Internet) Current Controlled Trials (Internet) trials.com/ WHO International Clinical Trials Registry Platform (ICTRP) (Internet) European Medicines Agency European Public Assessment Reports (EMA EPARs) ( FDA website Conference abstracts from COPD conferences (American Thoracic Society (ATS), European Respiratory Society (ERS), American College of Chest Physicians (ACCP), from 2010). References in retrieved articles and relevant systematic reviews will be checked for further studies that might fulfil the inclusion criteria. Information to extract* Study information Name of first study author Date study publication GlaxoSmithKline 16

235 Study location and setting Study group investigators (if known) Source of study funding (Government/Industry/Non profit organisation/ not reported) Study design characteristics (and quality assessment) Study aim Period of participant recruitment Study duration and Length of follow up Patient baseline characteristics (including GOLD classification of COPD severity) Summary of participants inclusion/exclusion criteria Types of interventions and comparators Total number of patients included (numbers per group) Outcomes of interest Total number of patients lost to followup/withdrawals/exclusions/refusals (numbers per study group) Reasons for withdrawals and dropouts per study group Study conclusions Any additional reviewer s comments Bibliographic details of all reports (full papers, abstracts, trial reports etc) relating to the study Efficacy outcomes COPD exacerbations (based on resource utilisation, symptoms, or any other definition) o Moderate COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing a moderate COPD exacerbation; Total number of moderate COPD exacerbations experienced over the duration of the study; Mean rate of moderate COPD exacerbations per patient per year; Time to first moderate COPD exacerbation) o Severe COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing a severe COPD exacerbation; Total number of severe COPD exacerbations experienced over the duration of the study; Mean rate of severe COPD exacerbations per patient per year; Time to first severe COPD exacerbation GlaxoSmithKline 17

236 o Composite of moderate/severe COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing moderate/severe COPD exacerbations; Total number of moderate/severe COPD exacerbations experienced over the duration of the study; Mean rate of moderate/severe COPD exacerbations per patient per year; Time to first moderate/severe COPD exacerbation) Change from baseline in FEV 1 o Prebronchodilator FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial); Percentage of patients achieving an increase in postbronchodilator FEV 1 of 12% and 200 ml above baseline at any time during 0 6 hours post dose on treatment day 1 o Postbronchodilator FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial)) o Peak FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value)) o Trough FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial); Percentage of patients achieving an increase in trough FEV1 of 100mL above baseline, Time to onset (increase 12%+ from baseline FEV1)) o AUC FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value)) Exercise capacity e.g. six minute walk test,12 minutes walk test, incremental shuttle walk test, cycle ergometry test, treadmill test Inspiratory capacity, Residual volume (measure of lung hyperinflation), Functional residual capacity, Forced expiratory vital capacity (FVC), Exertional dyspnoea intensity (10 point Borg scale) during exercise, etc Inhaled rescue medication (Type of inhaled rescue medication used; Percentage of patients using inhaled rescue medication) Change from baseline and Clinically relevant improvement in Patient reported Outcomes (PRO) measures 1 : o St. George s Respiratory Questionnaire (SGRQ) (MCID SGRQ 4 point change) o St. George s Respiratory Questionnaire for COPD patients (SGRQ C) o Baseline Dyspnea Index Transition Dyspnea Index (BDI TDI) (MCID TDI 1 point change) 1 Assessing disease-related symptoms and/or health-related quality of life GlaxoSmithKline 18

237 Safety outcomes Critical appraisal o o o o o o o o o o o o Chronic Respiratory Questionnaire (CRQ) Self administered Chronic Respiratory Questionnaire (CRQ SAS) Shortness of Breath Questionnaire (SOBQ) Shortness of Breath with Daily activities (SOBDA) Exacerbations of Chronic Pulmonary Disease Tool Patient Reported Outcome (EXACT PRO) Modified Medical Research Council (mmrc) questionnaire COPD Assessment Test (CAT) EuroQol 5D (EQ 5D) SF 36 Rescue medication: Number of doses taken in a specified time period, % rescue medication use free days, Decrease in number of rescue med puffs per day, % symptom free days Healthcare resource utilisation Any other relevant patient reported outcome in COPD, to be specified in study protocol Safety and tolerability endpoints, including: Adverse events (any reported, any severity, by type, i.e. MI, arrhythmia, congestive heart failure, pneumonia, bone fractures, bone mineral density, specifying incidence, severity / grade, aortic pulse velocity, etc..) Withdrawals (all, due to AEs and any other specified, time to discontinuation) NICE checklist (CRD) Assessment of the quality of reporting of RCTs will be carried out using recommendations from the NICE single technology appraisal (STA) manufacturer s template (NICE STA 2009; Cochrane Handbook 2009) GlaxoSmithKline 19

238 ENDPOINTS Primary Endpoint FEV1: value at baseline and any specified time point, change from baseline, difference in change from baseline between groups, mean difference in FEV change from baseline, AUC FEV1; Secondary Endpoint(s) Exacerbations: Mean rate of exacerbations per patient per year or annualised total number of exacerbations experienced over the course of the study (moderate/composite of moderate and severe/severe); Mean rate of exacerbations per patient per year or annualised total number of exacerbations experienced over the course of the study (any exacerbation as per study definition) Note on Exacerbations Contrarily to other endpoints, the definitions of exacerbations vary across studies. The definition of exacerbations in the MTC reflects the various definitions used in the trials; in other words, the inclusion of each study in the MTC will not be based on the use of a particular specific definition of exacerbations. Studies differ in their recording practice for exacerbation endpoints. Definitions used in the trials include the following: deterioration in COPD leading to hospitalisation/emergency (or equivalent) treatment and/or the need for oral corticosteroid treatment for at least three days deterioration in COPD requiring treatment with an oral corticosteroid or an emergency department visit or hospitalization clinical exacerbation requiring emergency treatment or hospitalization exacerbation requiring emergency treatment, hospitalization, or use of a COPD medication not allowed by the protocol. Such definitions assume that exacerbation, in fact, is a composite endpoint comprising of any exacerbation of any level of severity. In other studies, exacerbations are explicitly classified or reported as severe, moderate or mild. Importantly, some studies use explicit definition of the severity of exacerbation endpoints as a restriction to the type of exacerbation data collected. So in some studies, the composite endpoint includes exacerbations of any severity, i.e. mild, moderate or severe; in other studies, the composite includes exacerbations that are moderate or severe. Classification criteria may also vary by study. For example, severe exacerbations are defined as deterioration in COPD requiring emergency hospital treatment or alternatively, any hospitalizations. Moderate exacerbations required administration of additional inhaled corticosteroids, bronchodilators and/or oral corticosteroids, or alternatively, as worsening of COPD leading to a prescription for a short use of oral corticosteroids. Mild exacerbations are deterioration in COPD requiring an increase in the use of relief medication which the physician considered to be clinically relevant. GlaxoSmithKline 20

239 Health status: St. Georges Respiratory Questionnaire (SGRQ) GlaxoSmithKline 21

240 Exploratory Endpoint(s) 3.4. SAMPLE SIZE / POWER CALCULATIONS 3.5. HYPOTHESES Definition of outputs of the MTC The results of the mixed treatment comparison will be expressed on the original scale of the data extracted from the systematic review. MTC endpoints are FEV1, exacerbations, and health status scores: FEV1 is expressed in Litres and is a continuous outcome. Treatment effects calculated in the MTC will be expressed as treatment differences (for treatment X compared to treatment Y) in mean difference from baseline to last measurement. Exacerbations are a count variable and are expressed either as total number of exacerbations (with treatment X and Y), yearly exacerbation rates (exacerbations per patient year, for treatment X and Y) or as relative risk ratio or odds ratios for treatment X vs treatment Y. The total number of exacerbations will be recalculated from the data extracted from the literature. Treatment effects calculated in the MTC will be expressed as risk ratios for treatment X compared with treatment Y. SGRQ results will be summarized as mean difference in score changes from baseline. Credible limits (CrL, 95%) and posterior distribution will be estimated for each endpoint. 4. DATA ANALYSIS CONSIDERATIONS Evidence Synthesis Models The role of the modeling is to build a network model which allows the analysis of multiple studies with overlapping drugs. Within this single model many different questions can be addressed. Individual drugs may be compared, or a group of drugs may be compared, i.e. determining the best drug or the likelihood a drug is in the top 5, etc. Therefore, this single model is the generator of many of the analysis results. For all comparators in the analysis, information on treatment outcomes is only available at the study level. Therefore, the information available will be averaged over the trial and treatment arm (e.g., primary and secondary outcomes, patient characteristics and other covariates). The general formulation of the model is made up of multiple parts. The first is the outcome variable, the second is the study structure, the third is the treatment characterization, and the fourth is the treatment of covariates and populations. The generic analytical mathematical structure used in the MTC is: Y Study Treat Covariate The outcome variable, Y, represents the clinical outcome, FEV, Exacerbations, QOL. This can be a continuous variable, dichotomous variable, a time to event variable, or even a repeated measures analysis. A common mathematical structure will be created to handle GlaxoSmithKline 22

241 the type of intrinsic uncertainty created by the outcome variable. Typically the modeling assumptions for the remaining aspects of the evidence synthesis will remain constant for example, the same assumptions about study effects will be made in the modeling of a continuous or dichotomous variable. Model Study Effects The key to most mixed treatment comparisons are the possibly disparate studies utilized. The mathematical handling of the studies plays a critical role in the results of the analysis. Two standard types of models will be used in this MTC, borrowing the terminology used for meta analyses: fixed and random effects models. Fixed effect models assume an independent parameter for each study in the dataset. In practice, this means that a distribution will be assigned to each study and these studies effects will be allowed to vary independently. Because this type of model is naïve and because of the overlapping nature of studies, this model may yield numerically unstable results. For the reason above, in this MTC, the base case will be estimated using a statistical approach referred to as hierarchical model or random effects model. This is a powerful way of treating the studies as coming from a common distribution of studies, hence there the common link to the studies is explicitly modeled, assigning one distribution to the population of parameters; this approach allows understanding of the variability from studyto study, and using this in the analysis of each study. A hierarchical model is the standard model used in mixed treatment comparisons and evidence syntheses as it is expected to have better performance and properties that the simple fixed effects model. A hierarchical model will be built as in the following paragraphs: Model studies Each study will be characterized by a study parameter, Study S for S=1,, N S. Each study is then modeled with a distribution, creating the multi level, or hierarchical, approach: S ~ N, 2 The population of studies is then modeled with the above distribution, which can be used to draw conclusions about possible new studies. The two parameters, µ and 2, are then modeled with a second level (hyper) prior. The analysis creates a posterior distribution for these hyper parameters, µ and 2. The posterior also uses the population of studies to individually estimate each study, resulting in better estimates, and smaller standard errors. Model Treatment Effects The modeling of the treatment effect can also vary. The most straightforward assumption is that each treatment has its own parameter. This is straightforward because it allows comparisons between treatments that are not dependent on the study or population. Model 1. Will be estimated assuming Independent prior distributions. A single parameters for each treatment is selected and they are modeled independently However, depending on heterogeneity that will be identified in the data, more elements may be taken into account. If heterogeneity of the study data identified is large, then the treatment effect can depend on population, study, or time point. For methods to assess GlaxoSmithKline 23

242 heterogeneity, see below (Section 6.2). Assumptions used in the base case are as follows: 1. Treatment doses will be assumed as independent treatments, i.e. the model will not be used to model a dose response relationship. This is because in COPD, treatment doses are part of clinical escalation protocols and are prescribed to patients with different characteristics at baseline 2. Combination treatments, both open and closed combinations, will be considered one treatment If required, an alternative modeling approach will be designed, including the following scenarios: Model 2: Interaction with population or study. Source studies would be potentially conducted in heterogeneous populations of slightly different severity and FEV entry criteria, as well as history of exacerbations. Hisotory of exacerbation at baseline will be taken into account as predictor of the analysis if associated with heterogeneity. The parameters for each treatment will be modeled as specific parameters to a study or a population subset, i.e. for exacerbators and for non exacerbators. This approach is in fact equivalent to stratification in a standard metaanalytical approach. A model the effect of treatment 1 with µ 11, µ 12,, µ 1S, where each µ is the effect for Treatment 1 in each study (explicitly allowed to vary). Within this model the effects for each treatment within each study are modeled hierarchically. The same type of model will be used to incorporate study effects being dependent on the population. Model 3: Treatment class hierarchical model. Treatment will also be modeled as a member of a treatment class (drug class for example). So, the individual treatment effect, µ T, will be modeled with a hierarchical distribution depending on the drug class: 2 T ~ N Class, Class GlaxoSmithKline 24. This allows the analysis of each treatment to be modeled within its drug class. This model aids in the estimating of each treatment effect, but also allows drug class level variables to be estimated, i.e. estimates of the mean and standard deviation for a drug class. Model 4: Treatment and covariates model. The modeling of the covariates is usually done in two ways. The first is the fixed effect approach and the second is to model the interaction with treatment (described in the treatment section above). The standard approach to modeling covariates is to create a coefficient for each covariate. The following covariates will be considered: Percentage of patients <18 years of age COPD severity at baseline (eligibility, moderate, severe, very severe) Prior therapy: Inhaled bronchodilators Prior therapy: Short acting beta agonist When available, history of prior exacerbations The interaction between covariates (population) and treatment is rarely possible without the availability of subject level data therefore interaction terms will not be considered in this MTC. A general test of model fit (standard model fitting diagnostics, deviance information

243 criterion (DIC) and Bayesian Information Criterion (BIC)) will be applied to assess and capture sensitivity to assumptions, i.e. evaluating the to determine if a hierarchical aspect is necessary, and provide sensitivity analyses when warranted. As part of the exploration of model performance, an assessment will be made of the performance of the fixed vs random effects moel based on different scenarios, to consider whether random or fixed parameters ensure the best fit. As part of this assessment, random effects will be tested for drug classes vs each treatment separately considered, as well as treatments by study duration and population baseline characteristics. The inclusion of these terms is in practice equivalent to assuming or removing fixed effects on drivers of study heterogeneity. General Modeling Strategy We will use the above structure to address the modeling of the studies, treatment, and covariates. Before addressing any individual analysis we structure the standard modeling. We will refer to scenarios as the modeling of an endpoint, treatment subsets, and specific covariates. The default modeling approach for each scenario is: 1. Standard modeling of each variable: a. Normal distribution model for continuous outcomes b. Logistic link function for dichotomous outcomes c. Exponential modeling for time to event d. Poisson Model for count data 2. Hierarchical additive study effects 3. Fixed effects for each treatment 4. Coefficients for each covariate. This modeling will allow general conclusions on the comparisons, ranking, and ordering of different treatments. Specifically the modeling will allow us to address the following research questions: The relative treatment effect (non inferiority) of Relvar vs. other ICS/LABA combination therapies (primary). The rank of Relvar in comparison with other ICS/LABAs (secondary): In addition to the posterior quantities used to assess non inferiority, the model will provide the probability that each of the treatments is the best, second best, etc., among the population of treatments allowing ranking of Relvar in comparison with other ICS/LABAs. The role of heterogeneity in study design (sample size, differences in endpoints definitions, trial duration, population characteristics) in the observed variability of trial results (secondary) Several approaches to the assessment of heterogeneity in the study design will be attempted. First, as part of the systematic review, a qualitative summary of the variability of demographics at the study level (population characteristics) will be provided (not explained in detail here as not reported yet). The second aspect is how much variability in efficacy/safety is realized across studies. In GlaxoSmithKline 25

244 Bayesian frameworks, there is no explicit assessment of heterogeneity other than that based on the properties and convergence of the model. An index of heterogeneity is not available for more complex frameworks than the head to head comparisons assuming fixed effects. In a Bayesian hierarchical model, a more appropriate measure criterion is the explicit modeling of the heterogeneity across studies, summarized by the between study variability (with relevant posterior summaries, such as the posterior mean and standard deviation of the study to study heterogeneity). Handling different duration of follow up in source studies The evidence base for this MTC is made up of several studies of potentially different duration, ranging from 4 to 156 weeks. Differences in follow up will be handled as follows: 1. All studies will be consodlidated in the base case analysis regardless of study duration, with the exception of all studies of less than 6 weeks, as the very shortterm treatment effect may be heavily time dependent. As this analysis does not use individual level patient data, it may be difficult to assess invariance of treatment effect with respect to time for these short term studies; 2. Stratification of the analysis by study duration will be used to assess heterogeneity of the remaining studies with respect to duration of follow up. The analysis will explore whether the treatment effect in studies of shorter duration (i.e weeks) differs (i.e. derives from a different distribution) than that in studies of longer duration (24 weeks and above). Depending on the finding, the model will be updated account for the potential source of heterogeneity Handling treatment dosages Treatment protocols in respiratory diseases generally recommend treatment dosage escalation in the case where patients intiated on low dose therapy do not gain benefit. The paradigm of treat to target implies that in real practice, more severe patients may receive higher doses of the same therapeutic compounds. For this reason, it is not appropriate to analyse Phase 3 trials in such a way that a relationship between dose and efficacy should be expected. Handling heterogeneity from differential loss to follow up The evidence base for this MTC is made up primarity by studies conducted in the context of regulatory approval and therefore a high compliance with treatment and low loss to followup are expected. Nevertheless, it is possible that loss to follow up may drive heterogeneity in the model. All analyses will be based on the ITT dataset reported in the studies to ensure potential biases pertinent to the exclusion of cases in PP analyses is handled. Further, wwe will conduct an exploration of the impact of loss to follow up in the exacerbation endpoint as the ITT denominators enter the calculation of the exacerbation rates directly. A qualitative inspection of rates of loss to follow up as well as a comparison of treatment effects for studies with high or low attrition will be attempted. Selection of Priors Priors will be elicited based on the views from the clinical COPD experts, and if appropriate, from large epidemiological studies conducted by GSK or other appropriate sources. GlaxoSmithKline 26

245 The strategy of prior development in this MTC is the following. First priors are separated into two components one is a modeling choice, like a hierarchical model, which allows for the needed heterogeneity. This aspect of prior development will be incorporated in the model development. The second component is the selection of priors for the hierarchical hyperparameters. For these prior specifications we will use the following staged approach: 1. Base case analysis: non informative priors for all parameters will be developed. These priors will be selected to allow the empirical information to shape the posterior completely. This approach achieves an added level of external validity, in that it eschews subjective information, and has a level of objectivity to it. The priors selected in this strategy may play an important role (data dependent through the hierarchical model). A sequence of priors will be created, different from those selected above, to assess the role of sensitivity analyses. These priors will span the space of priors from strong to weak. We will present the results of the analyses for the range of priors in the sensitivity analyses to explore the sensitivity of the results to the prior distributions. 2. Expert opinion elicitation: there are a number of parameters for which prior information will play an important role and potentially affect decisions. These will typically be heterogeneity parameters. These variance components are important parts of the model and allow it to range from a pooling model to a completely separate model. Because of this importance, we will develop internal GSK priors prospectively on these critical parameters. This will be done using a sequence of qualitative discussions, simulation results, and leveraging our prior construction experience. The range of priors in this analysis will be used for measuring prior sensitivity. SEE APPENDIX 1 for DETAILED STATISTICAL ANALYSIS PLAN GlaxoSmithKline 27

246 5. LIMITATIONS While the use of Bayesian hierarchical models require a number of assumptions (e.g., exchangeability of information at each level of the hierarchy, prior information), the ability of the model to incorporate the structure and limitations of the available information allows greater model validity than other approaches (e.g., traditional frequentist multi level modeling). These models are very robust, because they are so flexible. This mitigates to some degree typical worries regarding model dependencies and the validity of initial assumptions. When using such a model to make mixed treatment comparisons, one must often make assumptions about the additivity of treatment effects across studies. For example, if treatment A is compared to treatment B in one study and treatment B is compared to treatment C in another study, then an assumption of the additivity of treatment effects is needed to make any comparison between A and C. This can be a difficult assumption to test because of the lack of repeated single trials. All assumptions will be laid out in detail and explained thoroughly in the final study report. 6. STUDY CONDUCT, MANAGEMENT & ETHICS 6.1. ETHICS/IRB APPROVAL Non interventional study 6.2. INFORMED CONSENT Non interventional study 6.3. DATA PRIVACY Only published trial data used 6.4. PERSONALLY IDENTIFIABLE INFORMATION (PII) No PII used 6.5. AE REPORTING Only previously published or disclosable analyses used 6.6. DATA STORAGE/ARCHIVAL GlaxoSmithKline 28

247 7. EXTERNAL INVOLVEMENT 7.1. Third Party Supplier (Company Name, Address & Staff Names/ /Phone) 7.2. External Expert/Health Care Professionals (Consultants & Research PIs) CID ## consultant, not government employee GlaxoSmithKline 29

248 MILESTONES MILESTONE GUIDANCE OR POLICY REQUIREMENT FORECAST DATE MM-YYYY Forecast Final Protocol Approval Forecast GSK CSR Protocol Summary FPA Actual + 30 days Forecast Statistical Analysis Plan Approved Forecast Statistical Analysis Complete Forecast Final Study Report Complete SAC Actual + 6 months Forecast GSK CSR Results Summary Posting SAC Actual + 8 months Forecast Manuscript Submission SAC Actual + 18 months GlaxoSmithKline 30

249 DATA DISSEMINATION PLAN Attach the Manuscript & Congress Presentation Data Dissemination Plan (DDP) in the MCQP DDP Powerpoint template. Should include the following information: MANUSCRIPT PUBLICATION(S) STUDY ID PUBLICATION SHORT TITLE LEAD AUTHOR STUDY ACCOUNTABLE PERSON ESTIMATED SUBMISSION DATE (< SAC + 18 months) TARGET JOURNAL CONGRESS PRESENTATION(S) STUDY ID ABSTRACT SHORT TITLE PRESENTER POSTER OR ORAL PRESENTATION CONGRESS CONGRESS LOCATION CONGRESS DATE DISCLOSURE PLAN GSK CSR PROTOCOL SUMMARY (FPA + 30 DAYS) GSK CSR RESULTS SUMMARY (SAC + 8 MONTHS) FULL PROTOCOL POSTING DATE (Manuscript submission actual + 30 days) FORECAST GlaxoSmithKline 31

250 GlaxoSmithKline 32

251 Appendix 1: Study Definitions of exacerbations used in the MTC Most studies were deemed to have generally comparable definitions. The major sources of difference across studies in the exacerbation endpoints definitions were around whether or not a deterioration of COPD symptoms was required in order to classify a particular healthcare resource use as exacerbation, and whether mild exacerbations were defined and reported as study endpoints. For the association of HRU with deterioration of symptoms, the following studies did *not* require that a record of worsening symptoms: Calverley, 2003 (10268 and 10138); GlaxoSmithKline, 2005; Dal Negro, 2003; Wedzicha, 2008; Calverley, Mild exacerbations were not defined or reported in the following studies: 10268, Calverley, 2003 (10268 and 10138); Calverley 2007; Zheng 2007; Rennard 2009; GlaxoSmithKline/HZC113107, 2012; Wedzicha, 2008; Zhong 2012; Caverley 2010; Sharafkhaneh The overall rate of exacerbation in these studies was therefore assumed comparable with the composite of moderate + severe exacerbations in other studies. The definitions of moderate and severe exacerbations in these studies were, again, comparable with the remaining studies. Studies including a definition of Overall exacernations endpoint Overall definition 10268, Calverley, 2003 Antibiotics, oral corticosteroids, or both + hospital admissions (described as acute in original source) 23277, GlaxoSmithKline, 2005 Mild, moderate and severe 9508, Zheng, 2007 Antibiotics or oral corticosteroids and/or hospitalization 9621, Calverley, 2007 Antibiotic agents, systemic corticosteroids, hospitalization, or a combination 10138, Calverley, 2003 Oral antibiotics and/or corticosteroids or hospitalisation 2966, Rennard, 2009 oral corticosteroid or hospitalization 9439, Wedzicha, 2008 Oral/intravenous corticosteroids and/or antibiotics and/or emergency room treatment/hospitalization 1487, Zhong, 2012 Oral/intravenous corticosteroids and/or antibiotics and/or emergency room treatment/hospitalization 2497, Calverley, 2010 Oral/intravenous corticosteroids and/or antibiotics and/or emergency room treatment/hospitalization 28520, Sharafkhaneh, 2012 Oral/intravenous corticosteroids and/or antibiotics and/or emergency room treatment/hospitalization 28525, GlaxoSmithKline/HZC113107, 2012 Any treatment other than study medication or rescue salbutamol, including antibiotics, systemic corticosteroids and/or emergency treatment or hospitalisation. All studies that reported exacerbations of different severity reported a composite of moderate + severe exacerbations. In the study by Szafranski, 2003, severe exacerbations were defined as associated with the use of oral steroids and/or antibiotics and/or hospitalisation and therefore were assumed to be comparable to the composite of moderate / severe used in other studies. GlaxoSmithKline 33

252 The composite of moderate and severe for these studies was therefore consolidated with the overall exacerbations endpoint reported in the group of studies above, as the two definitions are comparable. Studies including definitions of exacerbations by severity Study Severe exacerbations Moderate exacerbations Mild exacerbations 10205, Dal Negro, 2003 Emergency treatment or hospitalization Antibiotics and/or oral corticosteroids Salbutamol >=2 occasions/24 h, >=2 consecutive days vs. RRV (run in period) 10273, Szafranski, 2003 Severe exacerbations: oral steroids and/or antibiotics and/or hospitalisation 1445, Tashkin, 2012 Emergency treatment or hospitalization 28529, GlaxoSmithKline/HZ C102970, , HCZ102871, , GlaxoSmithKline/HZ C113107, , Wouters, , Anzueto, , Ferguson, 2008 (SCO40043) In patient hospitalisation In patient hospitalisation In patient hospitalisation In patient hospitalisation Moderate/severe exacerbations requiring treatment with antibiotics, oral corticosteroids, and/or hospitalization. Moderate/severe exacerbations requiring treatment with antibiotics, oral corticosteroids, and/or hospitalization. moderate (required antibiotic and/or oral steroid treatment), oral corticosteroids and/or antibiotics oral corticosteroids and/or antibiotics systemic corticosteroids and/or antibiotics. course of oral corticosteroids with/ without course of antibiotics One day with 4 inhalations of reliever medication above the mean run in use managed with increased shortacting bronchodilator use on any two consecutive days self management (not including antib, Ocs) no further specifics provided self management (not including antib, Ocs) no further specifics provided salbutamol, >=two consecutive days, >=3+ inhalations per 24 hours above reference rescue value (daily mean use in run in) Mild xacerbations not requiring treatment with antibiotics or oral corticosteroids. Mild xacerbations not requiring treatment with antibiotics or oral corticosteroids. We identified two studies with a different definition of exacerbation compared with the rest of the evidence base. The difference consisted in the use of change in the dose of the maintenance medication associated with worsening symptoms to identify exacerbations. To assess heterogeneity in the analysis, a sensitivity analysis will be conducted including and excluding these studies from the model. Study Severe exacerbations Moderate exacerbations 27400, GlaxoSmithKline, 2010 (Hagedorn 2013) Severe (grade III) hospitalization or emergency room treatment. (Rodriguez Roisin (21) moderate (grade II): change of respiratory medication (increased dose or addition of antibiotics, mucolytics, systemic steroids, theophylline) and medical assistance GlaxoSmithKline 34

253 4112, Kardos, 2007 Severe (grade III) hospitalization or emergency room treatment. (Rodriguez Roisin (21) moderate (grade II): change of respiratory medication (increased dose or addition of antibiotics, mucolytics, systemic steroids, theophylline) and medical assistance GlaxoSmithKline 35

254 Appendix 2: Statistical Analysis Plan Details FEV (primary endpoint) An MTC model will be designed to estimate the relative treatment effect on FEV1 / FEV AUC for the following comparators: 1. FF/VI 50/25 2. FF/VI 100/25 3. FF FF/VI 200/25 5. Placebo 6. SFC 250/25 7. SFC 100/50 8. SFC 250/50 9. SFC 500/ FP BUD BUD MOM MOM MOM SAL FORM FORM FORM FORM BUD/FOR 80/4.5 QD 22. BUD/FOR 80/4.5 BID 23. BUD/FOR 160/4.5 QD 24. BUD/FOR 160/4.5 BID 25. BUD/FOR 320/9 BID 26. BUD/FOR 320/4.5 BID 27. BUD/FOR 160/4.5 BID 28. MOM/FOR 200/ MOM/FOR 400/ VI FF200 A total of 33 studies are included in this scenario, therefore, a distribution will be assigned to each study, Study S for S=1,, 33 With S ~ N, 2 GlaxoSmithKline 36

255 Scenario 1 (Base case): The model will be estimated assuming the following: 1. Normal model for continuous data (likelihood) 2. Hierarchical additive study effects 5. Random effects for each treatment Treatment t, for t=1, 31 T(x s,t ) N(µ s,t, 2 t ) With µ s,t, N(µ t, 2 t ) With s=study The following model outputs will be generated, based on the posterior distributions: The relative treatment effect (non inferiority) of Relvar vs. other ICS/LABA combination therapies (primary) The rank of Relvar in comparison with other ICS/LABAs (secondary) The effect of drug class ICS/LABAs compared with ICSs (secondary) The role of heterogeneity in study design (sample size, use of surrogate endpoints, trial duration, population characteristics) in the observed variability of trial results (secondary) Evidence networks for lung function Table XX below shows the evidence network as identified based on the results for the systematic review. The network is well connected and links via SFC 250/50 and via placebo. There are no breaks in the network therefore no rejoining strategies are required. GlaxoSmithKline 37

256 Exacerbations An MTC model will be designed to estimate the relative treatment effect on exacerbation rates for the following comparators: 1. FF/VI 50/25 2. FF/VI 100/25 3. FF/VI 200/25 4. Placebo 5. SFC 250/50 6. SFC 500/50 7. BUD/FOR 160/4.5 BID 8. BUD/FOR 320/9 BID 9. MOM/FOR 200/10 A total of 33 studies are included in this scenario, therefore, a distribution will be assigned to each study, Study S for S=1,, 33 With S ~ N, 2 Scenario 1 (Base case): The model will be estimated assuming the following: 3. Poisson Model for count data (likelihood) 4. Hierarchical additive study effects 6. Random effects for each treatment Treatment t, for t=1, 9 P(x s,t ) Pois ( s,t ) With t N(µ, 2 ) With s=study The following model outputs will be generated, based on the posterior distributions: The relative treatment effect (non inferiority) of Relvar vs. other ICS/LABA combination therapies (primary) The rank of Relvar in comparison with other ICS/LABAs (secondary) The effect of drug class ICS/LABAs compared with ICSs (secondary) The role of heterogeneity in study design (sample size, definition of exacerbations, trial duration, population characteristics) in the observed variability of trial results (secondary) Evidence networks for exacerbations Here below is a summary of the network links between interventions included in the MTC for the exacerbations endpoint. The primary network is the one made up by studies that report a composite of moderate / severe exacerbations. Based on total yearly exacerbation rates (as reported): GlaxoSmithKline 38

257 Placebo: links to SFC 250/50 (n=1); SFC 500/50 (n=3) BUD/FOR 320/9 (n=2), BUD/FOR 160/9 (n=2) FOR. FP, SAL, single agents etc.. SFC 500/50 links to FF/VI 100/25 (n=1) FF/VI 100 links to FF/VI 200/50, FF/VI 50/50 SFC 500/50 and PBO link to TIO18 SAL 50 links to SAL/Daxas (comparison not of interest) A secondary network will be used to assess the impact of changing the definition of exacerbations to include rate of any reported exacerbation (Table below) GlaxoSmithKline 39

258 Both networks are integer and therefore there is no need to develop a network rejoining strategy. GlaxoSmithKline 40

259 Health Status Health status data appear heterogeneous. The measure taken into consideration in this protocol is the SGRQ score. Other measures are collected in a subset of studies and are therefore unsuitable for evidence synthesis based on extensive comparisons SGRQ An MTC model will be designed to estimate the relative treatment effect on the SGRQ for the following comparators: 1. Placebo 2. FP FP BUD 160 QD 5. BUD 160 BID 6. BUD 400 BID 7. BUD 800 QD 8. MOM 200 BID 9. Salmeterol 50mcg bid 10. Formoterol 4.5mcg od 11. Formoterol 4.5mcg bid 12. Formoterol 5.0mcg bid 13. Formoterol 12mcg od 14. Formoterol 10mcg bid 15. Formoterol 12mcg bid 16. FF/VI 100/ SFC 100/ SFC 200/ SFC 250/ SFC 250/ SFC 250/ SFC 250/ SFC 500/ BUD/FOR 80/4.5 QD 25. BUD/FOR 80/4.5 BID 26. BUD/FOR 160/4.5 QD 27. BUD/FOR 160/9 BID 28. BUD/FOR 320/9 BID 29. BUD/FOR 320/9 BID 30. BUD/FOR 160/4.5 QD A total of 55 studies are included in this scenario, therefore, a distribution will be assigned to each study, Study S for S=1,, 55 With S ~ N, 2 GlaxoSmithKline 41

260 Scenario 1 (Base case): The model will be estimated assuming the following: 5. Normal Model for continuous data (likelihood) 6. Hierarchical additive study effects 7. Random effects for each treatment Treatment t, for t=1, 30, n T(x s,t ) N(µ s,t, 2 t ) With µ s,t, N(µ t, 2 ) Where s=study The following model outputs will be generated, based on the posterior distributions: The relative treatment effect of Relvar vs. other ICS/LABA combination therapies (secondary endpoint) The rank of Relvar in comparison with other ICS/LABAs (secondary) The role of heterogeneity in study design (sample size, trial duration, population characteristics) in the observed variability of trial results (secondary) Evidence networks for SGRQ Table XX below shows the evidence network as identified based on the results for the systematic review. GlaxoSmithKline 42

261 GlaxoSmithKline 43

262 The network links via SFC 250/50, FF/VI 100/25 and placebo. In addition, despite the network is broken, it is so for interventions not relevant to this study therefore the relevant studies will be excluded from the analysis.

263 Appendix 3: Systematic review of the clinical effectiveness of inhaled therapies for the maintenance treatment of COPD Protocol December 2012 Kleijnen Systematic Reviews Ltd 45

264 table of Contents 1. BACKGROUND OBJECTIVES OF THE PROJECT RESEARCH QUESTIONS Methods LITERATURE SEARCHES METHODS OF STUDY SELECTION, QUALITY ASSESSMENT AND DATA EXTRACTION ANALYSIS Optional Syntheses (not included in costings) Error! Bookmark not defined. 7. TIMETABLE references 63 Appendix 1: example search strategies 64 Appendix 2: Included treatments 65 Kleijnen Systematic Reviews Ltd 46

265 BACKGROUND Chronic obstructive pulmonary disease (COPD) is characterised by airflow obstruction. The airflow obstruction is usually progressive, not fully reversible and does not change markedly over several months. The airflow obstruction is due to a combination of airway and parenchymal damage as the result of chronic inflammation that differs from that seen in asthma and is usually the result of tobacco smoke. Other factors, particularly occupational exposures, may also contribute to the development of COPD. 1 COPD is now the preferred term for the conditions in patients with airflow obstruction who were previously diagnosed as having chronic bronchitis or emphysema. There is no diagnostic test for COPD. Making a diagnosis relies on clinical judgement based on a combination of history, physical examination and confirmation of the presence of airflow obstruction using spirometry. 1 COPD produces symptoms, disability and impaired quality of life which may respond to pharmacological and other therapies that have limited or no impact on the airflow obstruction. The rate of COPD in the population is estimated at between 2% and 4%, representing between 982,000 and 1.96 million people in England. The diagnosed prevalence of COPD was 1.5% of the population in 2007/08 according to the Quality Outcome Framework (QOF) statistical bulletin. 2 Approximately 835,000 people in England have been diagnosed with COPD in However it is currently estimated that over 3 million people have the disease and that an estimated 2 million have undiagnosed COPD, among whom it is considered that 5.5% will have COPD at the mild end of the spectrum. 3 Currently COPD is managed by encouraging people to stop smoking, providing effective inhaled therapies, vaccinations, non invasive ventilation and rehabilitation for those seriously disabled by the disease. Maintenance treatment in COPD is to control inflammation, breathlessness and other respiratory symptoms, acute exacerbations and longer term deterioration in lung function. The current range of maintenance treatment options with inhaled therapies include: inhaled corticosteroids (ICSs) bronchodilators; long acting β2 agonists (LABA), long acting muscarinic antagonists (LAMAs), new generation of bi functional muscarinic antagonist Beta2 agonist (MABA). combination therapies Bronchodilators and corticosteroids generally rely on distinct therapeutic objectives and for this reason, drug agents can be administered alone or in a combination, typically, of one ICS with one or two bronchodilators. The most important combination therapies currently in use are ICS+LABA, LAMA+LABA and ICS+LABA+LAMA. The administration of combination therapies is typically part of treatment pathways in which diagnosed patients receive progressively more intensive therapy as health status deteriorates in time. An indicative treatment pathway for inhaled maintenance therapy in COPD is illustrated in Figure 1. The scope of this review also covers all combinations therapies, which can be administered via a inhaler containing multiple agents (i.e. closed combinations) or via the administration of multiple inhalers containing one agent each ( open combinations). The most important combination therapies currently in use are ICS+LABA, LAMA+LABA and ICS+LABA+LAMA. The administration of combination therapies is typically part of treatment pathways in which diagnosed patients receive progressively more intensive therapy as health status deteriorates in time. An indicative treatment pathway for inhaled maintenance therapy in COPD is illustrated in Figure 1. Kleijnen Systematic Reviews Ltd 47

266 GSK has a number of inhaled therapies in development for the maintenance treatment of COPD. These include: Relvar, a newly developed daily dose of fluticasone furoate (an ICS) + vilanterol (a LABA) Project Zephyr, a daily dual bronchodilator of vilanterol + umeclidinium umeclidinium monotherapy a closed triple therapy of vilanterol + umeclidinium + fluticasone furoate. As part of this programme, a systematic literature review will be undertaken to identify and summarise the clinical trial evidence for a range of comparators. The systematic review will cover all maintenance treatment options relevant for COPD, over the entire treatment pathway. Kleijnen Systematic Reviews Ltd 48

267 Figure 1: Treatment pathway, inhaled maintenance therapies in COPD (NICE COPD Guideline 1 ) SABA = Short acting beta agonist; SAMA = Short acting muscarinic antagonist; LABA = Long acting beta agonist; LAMA = Long acting muscarinic antagonist; ICS = Inhaled corticosteroid; FEV= forced expiratory volume. Kleijnen Systematic Reviews Ltd 49

268 OBJECTIVES OF THE PROJECT The aim of this project is to conduct a systematic review of the clinical effectiveness (including safety and tolerability) of maintenance treatments in COPD. The main objectives are: To assess the evidence supporting the relative treatment effect of the inhaled therapies for the maintainance treatments of COPD. To assess the evidence supporting the safety and tolerability profiles of different inhaled therapies for the maintainance treatments of COPD. Kleijnen Systematic Reviews Ltd 50

269 RESEARCH QUESTIONS The systematic literature review aims to collate all relevant studies to inform the assessment of clinical evidence around the benefits of inhaled therapies in the maintenance treatment of COPD. The main research questions for this project are: What is the evidence supporting the assessment of the relative treatment effect of Relvar versus other ICS/LABAs combination therapies? What is the evidence supporting the assessment of the relative treatment effect of Relvar versus LABAs monotherapy? What is the evidence supporting the assessment of the treatment effect of UMEC/VI versus LAMA or LABA monotherapy? What is the evidence supporting the assessment of the relative treatment effect of UMEC/VI versus other LABA/LAMA combinations (open and closed combinations)? What is the evidence supporting the assessment of UMEC monotherapy versus other LAMA or LABA monotherapy? What is the evidence supporting the assessment of effectiveness of triple combination therapy versus other triple and dual therapies? Kleijnen Systematic Reviews Ltd 51

270 Methods This systematic review will follow the Centre for Reviews and Dissemination Guidance for undertaking reviews in health care 4 and the Cochrane Collaboration Handbook LITERATURE SEARCHES Attempts will be made to identify all relevant studies regardless of language or publication status (published, unpublished, in the press or in progress). The search strategies (keywords) will be developed specifically for each database and variety of synonyms for COPD will be utilised. Only studies conducted in humans will be sought. Specific search filters for randomised controlled trials will be used to retrieve studies of clinical effectiveness. An example Embase strategy is presented in Appendix 1. Files from the manufacturer will also be examined to identify unpublished and ongoing studies. The following databases will be searched without date limits: Medline (OvidSP) Medline In Process Citations & Daily Update (OvidSP) Embase (OvidSP) Cochrane Database Of Systematic Reviews (CDSR) (Wiley) Cochrane Central Register of Controlled Trials (CENTRAL) (Wiley) Database of Abstracts of Reviews of Effects (DARE) (Wiley) Health Technology Assessment Database (HTA) (Wiley) NIHR Health Technology Assessment Programme (Internet) PROSPERO (International Prospective Register of Systematic Reviews) (Internet) Completed and ongoing trials will be identified by searches of clinical trials registers. The following resources will be searched with no date limit: NIH Clinicaltrials.gov (Internet) Current Controlled Trials (Internet) trials.com/ WHO International Clinical Trials Registry Platform (ICTRP) (Internet) European Medicines Agency European Public Assessment Reports (EMA EPARs) ( FDA website Conference abstracts from COPD conferences (American Thoracic Society (ATS), European Respiratory Society (ERS), American College of Chest Physicians (ACCP), from 2010). References in retrieved articles and relevant systematic reviews will be checked for further studies that might fulfil the inclusion criteria. Handling of citations Identified references will be downloaded into Endnote X4 software for further assessment and handling (we also have Reference Manager if GSK wants us to use that, however we recommend using Endnote because we understand that in the near future Reference Manager will not be sold or further supported). Rigorous records will be maintained as part of the searching process. Individual records within the Endnote reference libraries will be tagged with searching information, such as searcher, date searched, database host, database searched, strategy name and iteration, theme or Kleijnen Systematic Reviews Ltd 52

271 search question. This enables the information specialist to track the origin of each individual database record, and its progress through the screening and review process. Quality assurance within the search process For all searches undertaken by Kleijnen Systematic Reviews Information team, the main Embase strategy for each set of searches will be independently peer reviewed by a second Information Specialist, using the PRESS EBC checklist. 6, 7 Update searches At a timepoint to be determined by GSK (e.g. June 2013), update searches will be performed, and any relevant studies found incorporated into an updated report. 4.2 METHODS OF STUDY SELECTION, QUALITY ASSESSMENT AND DATA EXTRACTION Study selection Two reviewers will independently inspect the abstract of each reference identified by the search and determine the potential relevance of each article. For potentially relevant articles, or in cases of disagreement, the full article will be obtained, independently inspected, and inclusion criteria will be applied. Any disagreement will be resolved through discussion and will be checked by a third reviewer. Justification for excluding studies from the review will be documented. Studies will be included if they are randomised trials in patients as defined below, reporting at least one outcome from the outcomes listed below, and compare either one of the interventions listed below with any other treatment; or two or more of any of the drugs listed under either interventions or comparators. Any new comparator found in trials of one of the four interventions listed below, that is not listed under comparators, will automatically be added to the list of comparators. The inclusion critera are as follows: Patients/Participants Participants shall include individuals aged 12 years and older with an established diagnosis of chronic obstructive pulmonary disease (COPD) at any level of severity warranting the treatment with bronchodilators or corticosteroids (FEV1<=80%). Interventions Interventions of interest for this systematic review are: Relvar (fluticasone furoate + vilanterol) Zephyr (umeclidinium + vilanterol) Umeclidinium monotherapy Closed triple (vilanterol, umeclidinium and fluticasone furoate) Comparators Comparators of interest for this systematic review are: Inhaled corticosteroid (ICS) Long acting beta agonist (LABA) Long acting muscarinic antagonist (LAMA) Any combination therapy of ICS + LABA, in combined or separate inhalers Any combination therapy of LABA + LAMA, in combined or separate inhalers Any combination therapy of ICS + LABA + LAMA, in combined or separate inhalers Phosphodiesterase type 4 (PDE4) inhibitors Kleijnen Systematic Reviews Ltd 53

272 Appendix 2 provides a detailed list of interventions included in the scope of this review. Outcomes The review will include at least the following outcomes of interest (non exhaustive list, further outcomes can be added up to the moment of finalisation of the protocol): COPD exacerbations (based on resource utilisation, symptoms, or any other definition) o Moderate COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing a moderate COPD exacerbation; Total number of moderate COPD exacerbations experienced over the duration of the study; Mean rate of moderate COPD exacerbations per patient per year; Time to first moderate COPD exacerbation) o Severe COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing a severe COPD exacerbation; Total number of severe COPD exacerbations experienced over the duration of the study; Mean rate of severe COPD exacerbations per patient per year; Time to first severe COPD exacerbation o Composite of moderate/severe COPD exacerbations (Definition, including timepoint; Percentage of patients experiencing moderate/severe COPD exacerbations; Total number of moderate/severe COPD exacerbations experienced over the duration of the study; Mean rate of moderate/severe COPD exacerbations per patient per year; Time to first moderate/severe COPD exacerbation) Change from baseline in FEV 1 o Prebronchodilator FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial); Percentage of patients achieving an increase in postbronchodilator FEV 1 of 12% and 200 ml above baseline at any time during 0 6 hours post dose on treatment day 1 o Postbronchodilator FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial)) o Peak FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value)) o Trough FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value, serial); Percentage of patients achieving an increase in trough FEV1 of 100mL above baseline, Time to onset (increase 12%+ from baseline FEV1)) o AUC FEV 1 (Definition, including timepoint; As reported (e.g., mean, percentage of predicted value)) Exercise capacity e.g. six minute walk test,12 minutes walk test, incremental shuttle walk test, cycle ergometry test, treadmill test Inspiratory capacity, Residual volume (measure of lung hyperinflation), Functional residual capacity, Forced expiratory vital capacity (FVC), Exertional dyspnoea intensity (10 point Borg scale) during exercise, etc Inhaled rescue medication (Type of inhaled rescue medication used; Percentage of patients using inhaled rescue medication) Change from baseline and Clinically relevant improvement in Patient reported Outcomes (PRO) measures 2 : o St. George s Respiratory Questionnaire (SGRQ) (MCID SGRQ 4 point change) o St. George s Respiratory Questionnaire for COPD patients (SGRQ C) 2 Assessing disease-related symptoms and/or health-related quality of life Kleijnen Systematic Reviews Ltd 54

273 o Baseline Dyspnea Index Transition Dyspnea Index (BDI TDI) (MCID TDI 1 point change) o Chronic Respiratory Questionnaire (CRQ) o Self administered Chronic Respiratory Questionnaire (CRQ SAS) o Shortness of Breath Questionnaire (SOBQ) o Shortness of Breath with Daily activities (SOBDA) o Exacerbations of Chronic Pulmonary Disease Tool Patient Reported Outcome (EXACT PRO) o Modified Medical Research Council (mmrc) questionnaire o COPD Assessment Test (CAT) o EuroQol 5D (EQ 5D) o SF 36 o Rescue medication: Number of doses taken in a specified time period, % rescue medication use free days, Decrease in number of rescue med puffs per day, % symptom free days o Healthcare resource utilisation o Any other relevant patient reported outcome in COPD, to be specified in study protocol Safety and tolerability endpoints, including: Adverse events (any reported, any severity, by type, i.e. MI, arrhythmia, congestive heart failure, pneumonia, bone fractures, bone mineral density, specifying incidence, severity / grade, aortic pulse velocity, etc..) Withdrawals (all, due to AEs and any other specified, time to discontinuation) Study design Randomised controlled trials only (we suggest to include all types of RCTs, including phase 2 RCTs) The following study/publication types will be excluded: Studies with < 10 participants Pre clinical and animal Studies including Short acting beta agonist (SABAs) and Short acting muscarinic antagonist (SAMAs) aimed at symptoms control Assessment of methodological quality A checklist for randomised clinical trials, we propose the one based on the Centre for Reviews and Dissemination Guidance for undertaking reviews in health care 4 will be used for quality assessment. An alternative checklist can be decided on up to finalisation of the protocol. Quality assessment for the domains specified in the table will be carried out independently by two reviewers. Any disagreements will be resolved by consensus. The results of the quality assessment will be used for descriptive purposes to provide an evaluation of the overall quality of the included studies and to provide a transparent method of recommendation for design of any future studies. We plan to investigate the possible modifying effects of the methodological quality of primary studies using meta regression. This is described in the analysis section below. Table 1: Checklist for randomised clinical trials Kleijnen Systematic Reviews Ltd 55

274 1. Was the assignment to the treatment groups really random? 2. Was the treatment allocation concealed? 3. Were the groups similar at baseline in terms of prognostic factors? 4. Were the eligibility criteria specified? 5. Were outcome assessors blinded to the treatment allocation? 6. Was the care provider blinded? 7. Was the patient blinded? 8. Were the point estimates and measure of variability presented for the primary outcome measure? 9. Did the analyses include an intention to treat analysis? Below an example how the input screen might look in an Access database. Data collection and presentation We will create a database in Microsoft Access, with categorisations (drop down menus, including options for free text), based on the specific questions identified above. This database can serve as a knowledge platform that can be interrogated in various ways. Using the database, data extraction tables will be produced in Microsoft Word for each study. In addition, the database can be used to easily export data for analyses. Possible outputs include forest plots showing the relative effects of different drugs. An example of such a forest plot is shown in Figure 2 below (figure taken from our earlier work in this area, Puhan et al ). Kleijnen Systematic Reviews Ltd 56

275 Figure 2 For each study, data will be extracted independently by two reviewers. Any disagreements will be resolved by consensus. Data extraction will be discussed and decisions documented. Trials will be identified by the name of the first author and year of the main trial publication. The following are examples of general data which will be extracted from each included study: Name of first study author Date study publication Study location and setting Study group investigators (if known) Source of study funding (Government/Industry/Non profit organisation/ not reported) Study design characteristics (and quality assessment) Study aim Period of participant recruitment Study duration and Length of follow up Patient baseline characteristics (including GOLD classification of COPD severity) Kleijnen Systematic Reviews Ltd 57

276 Summary of participants inclusion/exclusion criteria Types of interventions and comparators Total number of patients included (numbers per group) Outcomes of interest Total number of patients lost to follow up/withdrawals/exclusions/refusals (numbers per study group) Reasons for withdrawals and dropouts per study group Study conclusions Any additional reviewer s comments Bibliographic details of all reports (full papers, abstracts, trial reports etc) relating to the study Dichotomous data will be extracted as the number of individuals with the outcome of interest and the total numbers of individuals in the intervention and control group. Continuous data will be extracted as the mean and standard deviation (SD) for the intervention and control group, and where appropriate as survival data (hazard ratios). We will also provide a summary statement of the overall quality of each study. We will use the database in Microsoft Access for data extraction and also as a functional deliverable. The figure below gives an example of how GSK might be able to use the database. In the left window will be all included studies, and in the right window there will be a range of selection options to filter the studies with an aim of e.g. to get all studies in unstable COPD with drug family LABA that report the outcome exacerbations. Figure 1: Example of COPD database in Access Studies can be filtered by Author, year, treatment (Relvar, Zephyr, ICS, LAMA, LABA, etc), population (severity of disease), outcome reported, etc. Kleijnen Systematic Reviews Ltd 58

277 Examples of an Access database we created recently for treatment of varicose veins is presented below. The first page shows the baseline characteristics for a specific study and the second page shows the results for that study. Kleijnen Systematic Reviews Ltd 59

278 Kleijnen Systematic Reviews Ltd 60

279 ANALYSIS Studies will be grouped by research question. Narrative synthesis methods For each research question a narrative summary of all of the included studies will be presented. This will include a summary of the characteristics (e.g. study design, population size, geographical location, year, baseline population characteristics, intervention and outcome definitions etc.) and quality of the studies. This will include the identification of any risks which may introduce bias into the data or any factors which may limit the generalisability of the findings. The data will be summarised using text and where relevant, accompanying tables and figures (graphs, bar charts, etc.) will be used. An overall summary of the main findings will be presented at the end of results section for each of the research questions. Kleijnen Systematic Reviews Ltd 61

280 TIMETABLE Project start date 4th December 2012 Week 1 Protocol development Week 2 Kick off meeting with GSK and discussion of protocol, includes deliverables 1 and 2 Week 3 Protocol agreement and start of searching (Endnote database with all titles and abstracts to be screened deliverable 3a) Week 4 GSK review and comments on Endnote database Week 5 8 Checking titles and abstracts Week 8 GSK review of Endnote database with decisions about orders for full papers (deliverable 3b) Weeks 6 10 Inclusion assessment of full papers Week 9 Agreement on data extraction format Week 16 Update searches Weeks 9 18 Data extraction and creation of structured tables Weeks 9 18 Drafting of report Week Submission of first draft report to GSK Week Update searches and incorporate new studies and comments from sponsor Week Submission of second draft report to GSK Week Incorporate comments from sponsor Week 24 Submit final report (w/c 6 May 2013) Approx June 2013 Update searches and updating report Kleijnen Systematic Reviews Ltd 62

281 references [1] National Institute for Health and Clinical Excellence (NICE). Chronic obstructive pulmonary disease. Clinical guideline CG101 [Internet]. Manchester: NICE, 2010 [cited ] Available from: [2] Healthcare Commission. Clearing the air: a national study of chronic obstructive pulmonary disease [Internet]. London: Commission for Healthcare Audit and Inspection, 2006 [cited ] Available from: [3] Shahab L, Jarvis MJ, Britton J, West R. Prevalence, diagnosis and relation to tobacco dependence of chronic obstructive pulmonary disease in a nationally representative population sample. Thorax 2006;61(12): [4] Centre for Reviews and Dissemination. Systematic Reviews: CRD s guidance for undertaking reviews in health care [Internet]. York: University of York, 2009 [cited ] Available from: [5] Higgins JPT, Green S, eds. Cochrane handbook for systematic reviews of interventions [Internet]. Version [updated March 2011]: The Cochrane Collaboration, 2011 [cited ]. Available from: handbook.org/ [6] McGowan J, Sampson M, Lefebvre C. An evidence based checklist for the peer review of electronic search strategies (PRESS EBC). Evidence Based Library and Information Practice 2010;5(1):1 6. [7] Sampson M, McGowan J, Cogo E, Grimshaw J, Moher D, Lefebvre C. An evidence based practice guideline for the peer review of electronic search strategies. J Clin Epidemiol 2009;62(9): [8] Puhan MA, Bachmann LM, Kleijnen J, Ter Riet G, Kessels AG. Inhaled drugs to reduce exacerbations in patients with chronic obstructive pulmonary disease: a network meta analysis. BMC Med 2009;7:2. [9] DerSimonian R, Laird N. Meta analysis in clinical trials. Control Clin Trials 1986;7(3): [10] Lu G, Ades AE. Assessing evidence inconsistency in mixed treatment comparisons. J Am Stat Assoc 2006;101(474): [11] Spiegelhalter DJ, Best NG, Carlin BP, van der Linde A. Bayesian measures of model complexity and fit. J R Stat Soc Ser B 2002;64(4):