Correcting the Variance of Effect Sizes Based on Binary Outcomes for Clustering.

Area of Science:

• Biostatistics
• Epidemiology
• Health Research Methods

Background:

• Systematic reviews and meta-analyses frequently include cluster randomized trials.
• A common issue is statistical analysis that fails to account for clustered data.
• This can lead to inaccurate results, especially when analyzing secondary outcomes with summary data.

Purpose of the Study:

• To provide accurate variance calculations for risk differences, log risk ratios, and log odds ratios from clustered binary data.
• To offer a method for meta-analysts to handle studies where clustering is not accounted for in the analysis.

Main Methods:

• Derivation of approximate variance expressions for key effect measures.
• Utilization of intraclass correlations (ICCs) to adjust for clustering.
• Illustrative example demonstrating the calculation process.

Main Results:

• Formulas are provided for approximate variances of risk differences, log risk ratios, and log odds ratios.
• The method allows for the analysis of clustered binary data even when standard analyses do not account for clustering.
• Empirical estimates of intraclass correlations are referenced.

Conclusions:

• The proposed methods enable more accurate meta-analyses of studies with clustered binary outcomes.
• Researchers can now better address analytical challenges in systematic reviews involving cluster randomized trials.
• Accurate variance estimation is crucial for reliable synthesis of evidence from clustered data.