New approach to compare treatments in adaptive seamless designs while maintaining Type I error and ensuring adequate power
In superiority "exploratory" Phase II clinical trials, we often compare the efficacy of several doses of an experimental product versus a control group (often a placebo). We then use the results from the Phase II to design the subsequent confirmatory superiority Phase III trial, and we use statistical methods in the Phase III trial to demonstrate that the "best" (most efficacious) dose selected from the Phase II study is superior to the control. The two phases are usually separate and independent: the Phase III trial does not incorporate patient data from Phase II except, again, in designing the Phase III trial. If we can combine data from the two phases into one design and use data from both phases to assess efficacy of the most efficacious dose of the experimental treatment versus control, we can potentially shorten the overall time of clinical development by reducing the overall sample size across Phase II and Phase III combined. This kind of design is the so-called Adaptive Seamless Designs (ASD). In this dissertation, we first review two commonly used combination approaches for the adaptive seamless designs. These approaches combine the stagewise p-values and apply the closed testing procedure to control the familywise error rate at the nominal level. Due to their complexity in both understanding and implementation, we propose an approach that uses a standard statistical test to compare treatments on the endpoint at the final analysis; we derive the distribution of the final test statistic and the critical value required to maintain Type I error rate at the nominal level Our simulation studies show our approach is comparable to the combination approaches in terms of Type I error rate and power. An extension to Denne's sample size re-estimation method is applied in order to estimate the final Phase III sample size required to maintain desired power, conditioned on Phase II results, when using our proposed adaptive seamless design statistical test. Simulation results demonstrate that the type I error rate and power are maintained at the desired level.