Application of a model based longitudinal network meta-analysis of FEV1 in COPD trials in clinical drug development
Julia Korell (1), Steven W. Martin (2), Mats O. Karlsson (1), Jakob Ribbing (1,3)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Sweden, (2) Pfizer Inc, Cambridge, MA, USA, (3) Pfizer AB, Sollentuna, Sweden
Objectives: Efficacy benchmarking is an important decision making component in clinical drug development. Comparative effectiveness determines ranking and pricing among alternative treatments in many countries. In chronic obstructive pulmonary disease (COPD), the forced expiratory volume in 1 second (FEV1) is used to assess lung function [1], and serves as biomarker for dose selection [2]. A model based meta-analysis of literature data on FEV1 in COPD can facilitate this process.
Methods: Randomized COPD maintenance trials on long-acting bronchodilators (BD) and anti-inflammatory (AI) treatments published until July 2013 were identified from literature. Suitable summary level FEV1 data, treatment information and covariates were extracted. The literature data was analysed using NONMEM 7.2 by expanding and revising a previously developed database and FEV1 model [3].
Based on the final meta-model, comparative effectiveness among the treatments was assessed using log-likelihood profiling (LLP). The LLPs were set up to assess uncertainty in the relative efficacies, i.e. the efficacy of compound A over the efficacy of compound B for all contrasts.
Results: In total, 142 studies were included, comprising 106,422 subjects who received 19 compounds (11 BD, 8 AI) in 105 treatment combinations across 419 study arms. 1982 FEV1 observations were available for analysis, each representing the mean FEV1 for a treatment arm at a specific time point.
The final model included baseline, disease progression, placebo effect, and drug effects for all 19 compounds. Dose-response was identifiable for 10 compounds. Time course for on/offset of drug effects were included where information was available. Drug-drug interactions and the effect of concomitant background COPD treatment were also accounted for.
As the NONMEM covariance matrix was unreliable and non-parametric bootstrapping was unsuitable, sampling-importance resampling [4] was used to obtain parameter uncertainty, which allowed for asymmetric confidence intervals.
Comparative effectiveness among the BD treatments showed superiority of once-daily and novel BD over the established twice-daily BD, salmeterol and formoterol. Among the AI treatments, roflumilast showed the highest efficacy.
Conclusions: The FEV1 meta-model can serve as a tool for assessing comparative effectiveness across different compounds. It also provides FEV1 predictions for the development of a meta-model linking FEV1 with exacerbation rate in COPD.
References:
[1] Vestbo et al. Am. J. Respir. Crit. Care Med. 2013;187(4):347-365. [2] Cazzola et al. European Respiratory Journal. 2008;31(2):416-469. [3] Ribbing et al. Population Analysis Group Europe (PAGE) 21, Abstract 2530. 2012. [4] Dosne et al. Population Approach Group Europe (PAGE) 22, Abstract 2907. 2013.
