|
Abstract:
|
In pivotal phase 3 trials, it is important to demonstrate consistent treatment benefit across subgroups in oncology regulatory submission. Reversal observation in certain subgroups may lead to a narrow label indication or even failure of regulatory approval, thus having critical commercial and strategic sponsor impact. To address this widely-seen but challenging situation, we developed a Bayesian semi-parametric hierarchical model to accurately adjust the treatment effect estimation of subgroups by allowing information borrowing across subgroups. To avoid negative finding due to random variability from limited sample size, we use a Dirichlet Process prior to create a flexible shrinkage model that still allows for heterogeneity between individual subgroups and captures additional uncertainty in the prior distribution for the subgroup effect sizes. We used simulation studies to evaluate the model performance. Compared with traditional shrinkage models, our model gives more accurate estimation. We also conducted sensitivity analysis based on various assumptions to demonstrate model robustness. Furthermore, a case study is presented to illustrate the impact on regulatory approval.
|
