Efficient design of partially nested randomized trials: A maximin approach.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Health Economics

Background:

• Randomized trials with clustered arms and continuous outcomes present unique design challenges.
• Optimizing sample size and resource allocation is crucial for efficient clinical research.
• Existing optimal designs require precise parameter knowledge, which is often unavailable.

Purpose of the Study:

• To develop and present efficient study designs for two-treatment randomized trials with clustering in one arm.
• To minimize the number of subjects and research budget while achieving a desired statistical power level.
• To address the challenge of unknown parameters during the design phase by introducing maximin designs.

Main Methods:

• The study presents designs that optimize treatment-to-control allocation ratios and the balance between the number of clusters and cluster size.
• Maximin designs are introduced to ensure a pre-specified power level for plausible parameter ranges, maximizing power for worst-case scenarios.
• Designs are also derived for fixed numbers of clusters or fixed cluster sizes, accommodating practical constraints.

Main Results:

• Maximin designs provide robust power guarantees across a range of unknown parameters.
• These designs optimize the allocation of subjects and the structure of clusters (number vs. size).
• Empirical examples demonstrate significant reductions in research budgets compared to equal allocation designs.

Conclusions:

• Maximin designs offer a practical and efficient approach to sample size and budget optimization in clustered randomized trials.
• The proposed methods provide a reliable way to achieve desired statistical power despite parameter uncertainty.
• An R Shiny app is available to facilitate the calculation of sample sizes for these practical designs.