Abstract:
|
In clinical trials, it is common to compare multiple candidate treatments in an early phase, pick the most promising treatment based on observed response data, and carry it forward into a later phase. A major issue with this approach is that the effect size of the "most promising treatment" is overestimated, especially with small sample sizes. Furthermore, such overestimation is amplified when multiple effective treatments have similar effect sizes. This can lead to serious consequences: if the sample size calculation of the later phase depends on the estimated treatment effect size, then the overestimation will cause power loss and decrease the probability of study success. To address this issue, we propose a bias reduction procedure by weighting all effect sizes in estimating the best treatment effect size. Through an example with continuous endpoints, we demonstrate that the proposed method reduces the bias and eventually improves the probability of study success. We also demonstrate that our approach can be extended to other areas such as estimating the best subgroup effect or examining multiple indications.
|