Modelling and Simulation based Techniques to support Trial Design and Submission of Daxas
Axel Facius (1), Laurent Claret (2), Rene Bruno (2), Gezim Lahu (1), Dirk Bredenbr÷ker (1)
(1) Department of PKPD and Biomarker Sciences, Nycomed GmbH, Konstanz, Germany; (2) Pharsight, A CertaraÖ Company, Marseilles, France
Roflumilast, an oral selective PDE4 inhibitor has been recently approved in EU, US and Canada as treatment that reduces the rate of exacerbations in patients with severe COPD associated with chronic bronchitis and a history of exacerbations. Model based techniques were used to describe the primary clinical endpoint (reduction in the number of exacerbations) and secondary endpoint (change from baseline (cfb) FEV1) in two pivotal phase III trials.
Objectives: The objectives of this analysis were to develop statistical models to describe the effect sizes in both clinical endpoints and to simulate these outcomes in two pivotal trials.
Methods: Data from six phase II/III trials (~5600 patients) were used to develop a nonlinear mixed effects model describing FEV1 with time. Data from two phase III trials (~2500 patients) were used to develop a generalized linear model (negative binomial model) to describe exacerbation rates per patient per year. The models were qualified using posterior predictive checks.
Results: The FEV1 model described the data from all six trials very well. Significant covariates were dose (PK information was not available), baseline FEV1 % pred., reversibility, and the symptom (cough and sputum) score. The predicted effect size for the two pivotal trials was 47.2 mL difference between placebo and treatment.
The initial exacerbation model did describe the data with relatively large variability. The exacerbation data contains comparatively little information because there is only one value per subject and the response variable is categorical. PK information was not available to develop a PK/PD model. We used the strong correlation between the effect sizes on predicted mean cfb FEV1 and exacerbations as an additional source of information by adding the predicted change from baseline FEV1 as covariate. The final exacerbation model included predicted mean cfb FEV1, baseline FEV1 % pred, symptom score, ICS pre-treatment, gender and dose. The model predicted effect size was 16.7%.
Conclusions: Mean cfb FEV1 could be used as a marker of exacerbation rate. Combining correlated endpoints might substantially increase model quality and precision of predictions when used as additional sources of information about individual effect sizes. The model predictions were very accurate: Observed effect sizes were 48 mL (FEV1) and 17% (exacerbations) in recently completed Phase III studies.
 Calverley PMA, et al. Lancet 2009; 374:685-94