Abstract
Population selection is a crucial subject in clinical development nowadays as personalized medicine is growing in interest. Evolution in biomarker scanning techniques allows for the composition and detection of sub-populations of interest when analyzing new drug responses in a disease. Seamless adaptive trials could allow for subgroup analysis with the selection of the most promising population at interim analysis. We propose a hybrid Bayesian design for seamless Phase II/III trials with binary and time-to-event outcomes for the first and second phases, respectively. In this work, at interim analysis, several prior distributions, including shrinkage prior, are compared to possibly select/discard a population, and a final test using a conditional error function as a combination method testing procedure to control the frequentist type I error is used. Simulation studies showed that the logistic regression model performs better than frequentist testing for the population selection problem when the subgroup should be selected. Shrinkage prior distributions tend to be more conservative than simpler normal distributions as studies that would have ended positively are stopped at interim analysis.