Sample size estimation for pilot animal experiments by using a Markov Chain Monte Carlo approach

Area of Science:

• Biostatistics
• Animal Research Methodology

Background:

• Statistical sample size determination is crucial for animal experiments but often hindered by a lack of prior data.
• Pilot studies are particularly affected, making reliable assumption verification difficult.

Purpose of the Study:

• To evaluate the utility of statistical simulation, specifically a Markov Chain Monte Carlo (MCMC) approach, for sample size calculations in animal studies.
• To verify pragmatic assumptions for power and sample size calculations using common distributions (binomial and normal).

Main Methods:

• Simulated binomial and normal distributions for categorical and continuous endpoints, respectively, using a Markov Chain Monte Carlo (MCMC) approach.
• Assessed statistical power based on varying group sizes and effect sizes.

Main Results:

• For continuous endpoints, group sizes of 5-6 animals per group demonstrated sufficient statistical power (≥ 80%), even for small effect sizes.
• For categorical outcomes, a minimum of 8 animals per group is necessary to guarantee adequate statistical power, irrespective of effect size.

Conclusions:

• The MCMC approach is a valuable tool for sample size calculation in animal studies, especially when prior data is scarce.
• Simulation results underscore the importance of assumptions regarding distributional properties and effect sizes, applicable whether or not prior data is available.
• MCMC offers a promising method for more informed planning of pilot animal research.