Methods for calculating confidence and credible intervals for the residual between-study variance in random effects...

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Meta-regression is increasingly utilized for modeling study-level covariate effects.
• Challenges exist in accurately estimating residual between-study variance in these models.
• Accurate quantification of variance is crucial for reliable meta-regression interpretation.

Purpose of the Study:

• To develop and present methods for calculating confidence intervals for residual between-study variance in random effects meta-regression.
• To extend existing meta-analysis techniques to the meta-regression context.
• To propose a Bayesian approach using informative priors for credible intervals.

Main Methods:

• Extension of exact confidence interval methods for between-study variance to meta-regression.
• Adaptation of Generalized Cochran heterogeneity statistics for meta-regression.
• Implementation of a Newton-Raphson procedure for the Q profile method.
• Utilization of WinBUGS for Bayesian meta-regression with informative priors.

Main Results:

• Developed and validated methods for calculating confidence and credible intervals for residual between-study variance.
• Demonstrated application using examples with binary and continuous covariates.
• Observed wide intervals for residual between-study variance in both examples, highlighting the need for uncertainty reporting.

Conclusions:

• Exact frequentist methods and Bayesian approaches are available for computing confidence/credible intervals for residual between-study variance in meta-regression.
• These methods are comparable in ease of implementation to those for meta-analysis.
• Routine reporting of residual between-study variance estimates with uncertainty measures (confidence/credible intervals) is recommended.