This Chapter continues to develop the arguments related to conservative sample size estimation revealed in the previous one. In fact, the problems which arise when the effect size of phase II is different from that of phase III are considered.
Indeed, broader and more heterogeneous patient populations are often pursued in phase III clinical trials as compared to those in phase II studies. This might induce larger variability and/or lower differences between means of the variables related to primary endpoints of effectiveness of different drugs. Both phenomena imply that phase III effect size δt is lower than the phase II one, namely δII. It should be noted that the effect the drug has is not a function of the development phase of the sponsor. Instead, there are different treatment effects for the population studied for particular phase II and phase III protocols.
In practice, when δt ≠ δII a structural bias arises within SSE. This difference between the effect size of the two phases in question implies that the estimation of the parameters of interest (e.g. the true effect size, the ideal sample size, the overall power) is structurally biased. Nevertheless, phase II data can be useful in estimating the sample size of phase III. The aim here is to show this usefulness.
In this Chapter, the numerical bias of CSSE strategies is evaluated first, and their robustness is compared in different scenarios. Some techniques for ...