Exact group sequential designs for two-arm experiments with Poisson distributed outcome variables.

Area of Science:

• Biostatistics
• Statistical Methodology
• Experimental Design

Background:

• Two-arm experiments often involve count data that follows a Poisson distribution.
• Traditional fixed sample size designs may not be sample-efficient for accumulating data over time.
• Group sequential methods allow for interim analyses and early stopping, potentially improving efficiency.

Purpose of the Study:

• To describe and compare two group sequential design methods for Poisson distributed data.
• To present a framework for determining near-optimal stopping boundaries in such designs.
• To evaluate the sample size reduction achievable with group sequential designs versus fixed sample approaches.

Main Methods:

• Development and comparison of two group sequential design methodologies.
• One method utilizes a normal approximation for Poisson data.
• The second method employs exact calculations for increased precision.

Main Results:

• A framework for identifying near-optimal stopping rules was established.
• For a specific example, group sequential designs reduced the expected sample size by up to 44% under the null hypothesis compared to fixed sample designs.
• The study highlights the potential for substantial sample size savings.

Conclusions:

• Group sequential designs offer a more sample-efficient alternative to fixed sample size designs for Poisson data.
• Both normal approximation and exact calculation methods provide viable approaches, each with distinct advantages and disadvantages.
• The presented framework aids in optimizing stopping boundaries for improved experimental efficiency.