A simple approach to test for interaction between intervention and an individual-level variable in community...

Area of Science:

• Biostatistics
• Epidemiology
• Public Health Research

Background:

• Community randomized trials (CRTs) often involve few communities but many individuals.
• Testing for interactions between community-level interventions and individual-level variables in CRTs presents statistical challenges.
• Existing methods may not be suitable for typical CRT scenarios with limited community numbers.

Purpose of the Study:

• To develop a simple, robust method for testing interactions in community randomized trials.
• To provide a statistically sound approach suitable for situations with a small number of communities and a large number of subjects per community.
• To derive approximate sample size formulas for the proposed method.

Main Methods:

• A novel method is proposed involving calculating within-community differences in summary statistics based on an individual-level attribute.
• These differences are then compared between trial arms using a two-sample t-test or Wilcoxon test.
• The method's performance is assessed through simulations and demonstrated with a health education intervention CRT.

Main Results:

• The t-test based approach demonstrated power close to expected levels across various scenarios, including non-symmetric distributions and varying numbers of communities (4-20 per arm).
• Type I error rates consistently remained close to the 5% nominal level, irrespective of distributional assumptions.
• The Wilcoxon test was found to be overly conservative in this context.

Conclusions:

• The proposed method for testing interactions in CRTs is statistically valid and user-friendly.
• The application of the t-test is robust and reliable for analyzing intervention effects in community settings with limited clusters.