Modelling impact of dropout mechanisms in Chronic Obstructive Pulmonary Disease (COPD)
C. Ambery, M. Beerahee
Clinical Pharmacology Modelling and Simulation, GlaxoSmithKline
Objectives: The objective of this analysis was to investigate the impact of modelling jointly the drug-disease and dropout models to predict observed longitudinal response.
Methods: A population kinetic-pharmacodynamic (K-PD) model (Model 1) was developed describing the longitudinal trough FEV1 response in subjects with COPD after inhaled administration of anti-inflammatory and bronchodilator medicines as part of monotherapy and combination therapy including placebo over 24 weeks. One of the combination components included different doses of the anti-inflammatory drug. Model 1 included disease progression and PD parts. The disease progression part was described by a linear decline over time. Consistent with previous reports [Celli, 2008], Model 1 also showed the FEV1 baseline was influenced by gender, smoking status, severity of disease, previous medication, age factoring smoking status and height factoring gender. The slope of the disease progression was influenced by baseline and smoking status. The PD part was described by an indirect response model, which also incorporated a less than additive interaction term for describing the FEV1 response of combination therapy. Dropout was modelled jointly with the drug-disease progression model using a parametric hierarchical model for evaluating various dropout mechanisms namely, completely at random, random, and informative [Hu & Sale, 2003]. Parameters were estimated by maximising the approximate joint likelihood as implemented in the software NONMEM. This work was motivated by the recent EMEA guidance [EMEA, 2010] on missing data in confirmatory clinical trials.
Results: Although Model 1 was adequate in describing the longitudinal FEV1 observed data, inclusion of the joint dropout model showed improvement in the model diagnostics and also reflected in the observed versus predicted FEV1 response. The longitudinal FEV1 response profiles over 24 week treatment suggest patient dropout can be explained with the missing at random mechanism or with the informative dropout mechanism. Inclusion of the drop out model did not appear to influence parameters of the disease progression model and provided some insight in the dose response for the combination therapy.
Conclusions: Inclusion of a dropout model jointly with the drug-disease progression model in COPD has potential to improve the prediction of longitudinal FEV1 response specially for prospective simulation of future study designs in COPD.
 Hu & Sale, J Pharmacokin & Pharmacodyn, (2003)
 EMEA, Guidance on missing data in confirmatory clinical trials (2011)