Exact Unconditional Tests for Dichotomous Data When Comparing Multiple Treatments With a Single Control

Abstract

In contemporary clinical trials, often evaluated simultaneously are multiple new treatments or the same treatment at multiple dose levels. These treatments are first compared with a control, and the best candidate with sufficient activity is then picked for the following trial for further investigation. When the primary outcome is binary, several testing procedures including Dunnett's test, have been proposed for the assessment of hypotheses. The sample size of each group is predetermined; thus, an unconditional exact approach is aligned with the study design. The exact unconditional approach based on maximization has been studied for comparing multiple treatments with a control. The newly developed exact unconditional approach based on estimation and maximization could possibly increase the effectiveness of exact approaches by smoothing the tail probability surface. We compare these 2 exact unconditional approaches based on 3 commonly used test statistics under various design settings. Based on results from numerical studies, we provide recommendations on the usage of these exact approaches. A real clinical trial to treat psoriasis is used to illustrate the application of the considered exact approaches.