Study investigators enrolled eligible patients who were 40–80 years of age, had a clinical history of COPD as defined by the American Thoracic Society (ATS)/European Respiratory society , were current or former cigarette smokers with a history of cigarette smoking of ≥10 pack-years, and had a post-salbutamol FEV1/forced vital capacity (FVC) ratio of <0.70 and a post-salbutamol FEV1 of ≥35% and ≤70% of predicted normal values [12, 13]. Patients were excluded if they had a current diagnosis of asthma, known α1-antitrypsin deficiency, active lung infections, lung cancer, any clinically significant uncontrolled disease, or an abnormal and significant electrocardiogram (ECG) or significantly abnormal clinical laboratory finding. Concomitant use of inhaled salbutamol as a rescue medication was allowed. Inhaled corticosteroids (ICS) at a dose up to 1000 μg/day of fluticasone propionate or equivalent were permitted, provided the dose remained stable throughout the treatment period. Initiation or discontinuation of ICS or long-acting β2-agonist/ICS combinations within 30 days prior to screening was prohibited; however, patients were allowed to discontinue long-acting β2-agonist/ICS up to 2 days prior to screening if ICS alone was continued.
Written consent was obtained prior to the start of study-specific procedures. The study (clinicaltrials.gov NCT01372410; GSK study number AC4115321) was approved by the local ethics review committee (Chesapeake IRB, Columbia, MD) and was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines .
Study design and treatment
This randomised, incomplete block, three-period cross-over, placebo-controlled study was conducted in 15 centres in the United States from 25 July 2011 to 27 October 2011. In accordance with the randomisation schedule, generated using SAS and RandAll version 2.5, patients were assigned to receive a sequence of three of eight potential treatments for a total of three treatment periods per patient. UMEC 15.6 μg QD, 31.25 μg QD, 62.5 μg QD, 125 μg QD, 15.6 μg BID, 31.25 μg BID, open-label tiotropium 18 μg QD, and placebo were evaluated. UMEC and matching placebo study medication were administered via the ELLIPTA™ dry powder inhaler a in a double-blind fashion where neither patients nor the study investigators knew which study medication was administered. Tiotropium was an open-label comparator administered via the Handihaler® b. To maintain blinding, patients taking QD treatments also took placebo in the evening. Treatment consisted of three 7-day treatment periods, with two intervening 10–14 day washout periods. A 7–9 day washout period followed the third treatment period, before a follow-up phone call.
Study withdrawal criteria included COPD exacerbation (defined as acute worsening of COPD symptoms requiring treatment beyond study medication or rescue salbutamol, including antibiotics or systemic corticosteroids, and/or hospitalisation or emergency treatment), clinically significant change in laboratory parameters, or elevated liver chemistry.
Treatment adherence was assessed on Day 7 of each treatment period using dose counters on the inhaler or by counting blister doses remaining (tiotropium).
Outcomes and assessments
The primary efficacy endpoint was trough FEV1 on Day 8, defined as the mean of the FEV1 values obtained 23 and 24 hours after morning dosing on Day 7 of each treatment period.
The secondary efficacy endpoints were weighted mean FEV1 over 0–24 hours after morning dosing on Day 7, and serial FEV1 over 24 hours after morning dosing on Day 7. Serial spirometry was measured 30 and 5 minutes pre-morning dose, 1, 3, 6, 9, 12 (pre-evening dose), 13, 15, 23 and 24 hours after morning dosing.
Additional efficacy endpoints included trough FEV1 at other time points, weighted mean FEV1 over other time periods, serial FEV1, trough FVC, weighted mean FVC, and rescue salbutamol use (mean number of puffs per day and percentage of rescue-free days). Safety assessments included incidence of adverse events (AEs), haematology and clinical chemistry evaluations, incidence of COPD exacerbations, and vital signs.
Plasma and urine samples were collected for pharmacokinetic (PK) analysis. Assessments included area under concentration-time curve from time 0 to time t (AUC(0-t)), maximum observed plasma concentration (Cmax), time of maximum observed plasma concentration (tmax), amount of drug excreted unchanged in urine, and fraction of dose excreted unchanged in urine. Accumulation was calculated using plasma Cmax, plasma AUC using a common sampling time and amount excreted in urine over the same time interval on Day 7 versus Day 1.
Measurements for FEV1 and FVC were obtained using standard spirometry equipment that met or exceeded the minimal ATS performance recommendations . Spirometry was performed at screening, during a 6-hour interval on the first day of each treatment period, and during a 24-hour interval on the last day of each treatment period. A minimum of three acceptable spirometry efforts were obtained for FEV1 and FVC, and the highest measurement was recorded. Pre- and post-salbutamol spirometry measures at screening determined patient eligibility.
Sample size determination
The sample size for the population model-based dose response analysis was determined using the Monte-Carlo Mapped Power approach for mixed effects . The dose response from a UMEC dose-ranging study was used as reference . Using a model-based approach for a cross-over design study, 16 patients would provide at least 90% power to show a significant dose response with 25 patients showing ~95% power. The sample size was increased to reduce the risk of a false positive result for the lower UMEC doses – a sample size of 40 patients per arm would provide <10% chance that lower doses of UMEC would falsely show a trough FEV1 response of >100 mL. This number of patients would also provide approximately 85% power for the comparison of active treatments with placebo for the primary efficacy endpoint on Day 8 (ANCOVA analysis). This calculation assumes a two-sided 5% significance level, a within-patient standard deviation of 0.170 L (based on Donohue et al. ) and a treatment difference from placebo of 0.130 L. Therefore, approximately 160 patients were recruited to compensate for a possible 30% dropout rate.
The primary population for all efficacy and safety analyses was the modified intent-to-treat (mITT) population, comprising all patients who were randomised and received at least one dose of study medication. The population was modified in that outcomes were analysed based on the actual treatment received rather than the randomised treatment. The PK population comprised all patients in the mITT population for whom a PK sample was obtained and analysed.
Model-based and statistical analysis
Two approaches were used to characterise the relationship between dose and trough FEV1. The first approach (i.e., the primary analysis) was a model-based analysis whereby an Emax model was selected from a suite of dose response shape models to describe the observed trough FEV1 data as a function of dose. Two key parameters were estimated from this model - Emax which is an estimate of the maximum response predicted by the model given the observed trough FEV1 and ED50 which represents the dose that achieves 50% of Emax. The modelling approach investigated the impact of the inter-patient variability in trough FEV1 by examining the influence of patient demographic and physiologic factors (Additional file 1), and the effect of QD and BID regimen on the model parameters. Established model diagnostics were derived [17, 18] to demonstrate the suitability of the chosen dose–response model. Using the population Emax model, the predictive distribution of trough FEV1 across treatments was derived by simulating 1000 sets of individual model parameters using the covariance matrix (model uncertainty and random effects) of parameter estimates from the model. Key outputs included median trough FEV1 (95th percentiles) for QD and BID regimens, the probability of achieving a certain target trough FEV1 with each dose (adjusted for baseline and placebo), and median estimates of trough FEV1 (adjusted for baseline and placebo) across the dose range and by dose regimen. Both mean baseline FEV1 and period were included as covariates.
A Day 8 dataset and a pooled dataset for Days 7 and 8 (post-hoc analysis) were analysed and reported for the primary efficacy analysis. The rationale for pooling Day 7 and Day 8 was to ensure informative interpretation of FEV1 response as function of dose, given the repeated measures for trough FEV1 response within each patient on different days.
The second approach (also referred to as the secondary analysis) involved comparison of trough FEV1 at each dose versus placebo using Analysis of Covariance (ANCOVA). The change from baseline FEV1 (defined as the mean of the two pre-morning dose assessments at Day 1) to trough FEV1 at Day 8 was analysed using a mixed model which included period baseline FEV1, mean baseline FEV1, treatment and period as fixed effects, and patient as a random effect. A similar methodology was used to analyse weighted mean FEV1 and trough and weighted mean FVC endpoints. Serial FEV1 was analysed using a similar mixed model. Sensitivity analyses were conducted to assess the effect of any interaction of treatment with mean baseline, period baseline or period by repeating the analysis of trough FEV1 on Day 8 and adding a variable to indicate the previous treatment received, removing baselines from the model or both.
Due to issues with Good Clinical Practice at one investigator site, a decision was made after unblinding to re-evaluate the dose–response model and ANCOVA analysis of trough FEV1 on Day 8, excluding all patients enrolled at that site.
A Bayesian analysis of the primary endpoint (using the same covariates as the original mixed- model analysis) provided the posterior probability distribution of the treatment difference of each treatment against placebo, i.e. the distribution of the true treatment difference given the data observed in the study.