Eliminating bias in randomized cluster trials with correlated binomial outcomes

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Clustered or correlated binary data are common in biomedical research, arising from repeated measures or subsampling.
• Intracluster correlation within clusters reduces the information available for intervention effect estimation.
• Standard Pearson's chi2 analysis may not adequately account for this correlation.

Purpose of the Study:

• To review and illustrate alternative statistical methods for analyzing randomized cluster trials with binary data.
• To outline an alternative analysis algorithm that adjusts for intracluster correlation.
• To provide a tested FORTRAN program for implementing these methods.

Main Methods:

• Review of alternative statistical methods beyond Pearson's chi2 analysis.
• Illustration of these alternative approaches with examples.
• Development and testing of a FORTRAN program for statistical analysis.

Main Results:

• Alternative methods effectively adjust variance estimators for intracluster correlation.
• The developed FORTRAN program accurately produces the outlined statistics.
• The program is available in an executable format.

Conclusions:

• Adjusting for intracluster correlation is crucial for accurate analysis of clustered binary data in biomedical studies.
• Alternative methods and the provided FORTRAN program offer robust solutions for randomized cluster trials.
• The availability of the program facilitates the application of these advanced statistical techniques.