Multi-Arm Non-Inferiority Tests in Cluster-Randomized Design

This technical report details the methodology for conducting power and sample size calculations for multi-arm non-inferiority tests comparing treatment means to a control mean in a cluster-randomized design. The document outlines the significance of using cluster-randomized designs, particularly in scenarios where individual randomization may lead to contamination among subjects within clusters. It explains the adjustment of sample size calculations based on existing literature, specifically referencing the work of Campbell and Walters (2014) and others. The report describes the formulation of hypotheses for non-inferiority tests, including cases where higher or lower values are deemed better. Additionally, it covers power calculations, the Bonferroni adjustment for type I error rates, and recommendations for control group sizes. The report emphasizes the importance of proper statistical methods in ensuring valid results in clinical trials, particularly when multiple treatment groups are involved.