Determining the sample size for a cluster-randomised trial using knowledge elicitation: Bayesian hierarchical...

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Health Services Research

Background:

• Sample size determination in cluster-randomised trials critically depends on the intracluster correlation coefficient (ICC).
• Inaccurate or imprecise ICC estimates can lead to underpowered or inefficiently large trials.
• Existing ICC estimates are often problematic due to limited data or lack of relevance.

Purpose of the Study:

• To develop a robust method for estimating the ICC using a Bayesian approach.
• To determine an appropriate sample size for a cluster-randomised trial on post-stroke incontinence management.
• To leverage existing ICC data from published research for improved estimation.

Main Methods:

• A Bayesian hierarchical model was employed to combine ICC estimates from 16 relevant published trials.
• Expert knowledge was elicited to assign relevance weights to external ICC estimates.
• A posterior distribution for the ICC was constructed to inform sample size calculations.

Main Results:

• The Bayesian approach combined 34 ICC estimates using expert-defined relevance weights.
• The estimated ICC resulted in a sample size of 450-480 participants, compared to 500-600 using classical methods.
• A posterior median and 95% credible interval provided a range of plausible sample sizes.

Conclusions:

• The Bayesian method offers a more robust approach to sample size calculation by accounting for ICC uncertainty.
• Incorporating external ICC data and expert knowledge enhances efficiency and reduces sample size compared to conservative classical methods.
• This approach can lead to significant savings in trial resources and time.