An online sample size calculator for designing partially clustered trials

Area of Science:

• Biostatistics
• Clinical Trial Design

Background:

• Partially clustered trials incorporate both independent and clustered observations, common in neonatal studies (e.g., twins).
• Accurate sample size determination is vital to avoid under or over-powering treatment comparisons in these trials.
• Existing tools for sample size calculation in partially clustered trials, especially with >2 cluster sizes, are limited.

Purpose of the Study:

• Introduce a novel online application for calculating target sample sizes in partially clustered trials.
• Address a broad range of partially clustered trial scenarios, including those with maximum cluster sizes greater than 2.

Main Methods:

• Utilizes recently derived design effects for two-arm partially clustered trials with pre-randomization clusters.
• Considers both cluster and individual randomization for clustered observations (nested and crossed designs).
• Incorporates key parameters: effect size, significance level, power, cluster size range, intracluster correlation, randomization method, and analysis model.

Main Results:

• Developed a freely accessible R Shiny web application implementing the sample size calculation methods.
• The calculator provides step-by-step guidance for designing partially clustered trials.
• Sample size calculations accounting for partial clustering can significantly differ from those ignoring clustering.

Conclusions:

• Partial clustering significantly impacts clinical trial power and sample size requirements.
• The presented calculator enables researchers to accurately account for partial clustering in two-arm trials.
• Ensures appropriate trial power for continuous and binary outcomes in partially clustered study designs.