|
Abstract:
|
Imagine a study with two parts. Part 1 of the study contains a randomization of patients to treatment arms A and B, followed by Part 2 with treatment arms C and D. The intent of the study design is to have a balanced number of patients at the end of the study who were randomized to receive the four possible combinations of treatments in Parts 1 and 2. Operationally, this can be done in two ways. Patients could be randomized to one of four arms defining both treatments assigned in Parts 1 and 2 at the initial randomization (patients randomized 1:1:1:1 to receive treatments A then B, A then C, B then C, and B then D in Parts 1 and 2, respectively). Alternatively, patients could be randomized to arms A or B and then re-randomized to arms C or D in Part 2 of the study, stratified by the treatment assigned in Part 1. While this corrects an imbalance in Part 1 due to dropouts, it also potentially creates new imbalances based on other factors. We will investigate the probability of treatment arm imbalance in these two scenarios under different assumptions. Simulations will be performed to examine the impact of sample size and dropout rate on the probability of imbalanced arms.
|
