|
Abstract:
|
Two-stage randomization designs employ randomization in two stages. Subjects in the first stage are randomized to one of several primary treatments and then depending on the intermediate outcome of the exposure to the primary treatment, are randomized to one of several maintenance treatments in the second stage. We consider such a two-stage randomization design where the goal is to compare the mean survival time under various treatment combinations. Previous work includes IPW estimation, semi-parametric efficient estimation and PH models. The estimators proposed in these methods, especially in the first two are complicated. Considering the popularity of likelihood methods, we propose maximum likelihood estimators and compare them with the available estimators through simulation. The proposed estimators are easy to apply and have efficiency comparable to the existing estimators.
|
