Shuying Yang, Misba Beerahee
CPMS, GSK
Objectives: The aims of this work are: 1) to describe the method of evaluating probability of success by applying exposure response modelling and simulation approach; 2) to investigate the impact of model uncertainty and sample size on the probability of success.
Methods: A model-based exposure-response (ER) relationship was derived post hoc from a Phase IIb study, to describe the relationship between COPD exacerbation rate and the steady state trough drug exposure. Several forms of relationship including linear, log linear and Emax models were applied. The final model was determined using Akaike information criterion (AIC). Clinical trial simulation (CTS) was conducted to assess the probability of success of a new trial in Phase III under various scenarios. The success of a trial was defined as mean ratio of exacerbation rate
Results: A negative binomial linear model best described the relationship between COPD exacerbation rate and steady state trough drug exposure. With every unit (1 ng/mL) increase in exposure, there was about 3% (95%CI: 0.5-4.7%) reduction in the mean exacerbation rate. CTS results showed that the POS was related to dose and sample size in a nonlinear manner. The uncertainty in model selection and model parameter estimate significantly impacted the POS. Although the POS was increased with an increase in the sample size, the magnitude of increase was small. For example, at the highest dose (15mg), for a particular scenario, the POS increased from 60% to 79% as the sample size increased from 70 to 150 per arm.
Conclusions: POS can be an effective tool at each stage of drug development to support risk-benefit decision making as shown in this example. This approach integrates the various components of study design including dose, sample size, historical data and model uncertainties, and allows balanced review of impact of each component.
References:
[1] SAS/STAT® 9.2 User’s Guide
[2] R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://www.R-project.org/
Reference: PAGE 23 (2014) Abstr 3156 [www.page-meeting.org/?abstract=3156]
Poster: Methodology - Study Design