Abstract:
|
When developing a new molecular entity through different phases, critical milestones are often established to review accumulated data and make Go/No Go decisions. Decision making always takes risk and it is thus desirable to create a decision rule that minimizes the chance of a false Go or a false No Go. Limitations associated with early phase clinical trials, e.g., small sample size, have increased the challenge of making the right decision. Several quantitative models have emerged in recently years to create Go/No Go decision rules and the core of these approaches is to walk away from using p-values and instead using models and simulations. In this presentation, we first compare the operating characteristics (OC) for various decision criteria and examine the dependency of OC on the study sample size and the true treatment effect through simulations. We then narrow our evaluations on two methods, one using Bayesian predictive probability of success at confirmatory phase and the other using credible interval boundaries to compare against the Low Reference Value or the Target Value, with the intent to understand the similarities and differences between these two approaches.
|