Confidence intervals and P-values for meta-analysis with publication bias

Area of Science:

• Biostatistics
• Medical Research Methodology

Background:

• Publication bias is a significant concern in meta-analysis, potentially distorting overall treatment effect estimates.
• Previous work analyzed maximum bias under specific selection functions but did not address uncertainty.
• Selection functions determine study inclusion probability, influencing meta-analysis outcomes.

Purpose of the Study:

• To extend previous research on publication bias by incorporating the impact of study selection on estimation uncertainty.
• To develop a novel sensitivity analysis for publication bias that considers confidence intervals and P-values.

Main Methods:

• The study models treatment effect estimates (y) from a population of studies as approximately normally distributed N(theta, sigma^2).
• A selection function is used to describe the probability of study inclusion in a meta-analysis.
• The analysis extends previous worst-case sensitivity analysis by deriving corresponding confidence intervals and P-values.

Main Results:

• The new sensitivity analysis quantifies the impact of publication bias on the uncertainty of meta-analysis estimates.
• The approach provides a more comprehensive assessment of potential bias compared to methods focusing solely on point estimates.
• Two illustrative examples are presented to demonstrate the application of the developed methods.

Conclusions:

• The extended sensitivity analysis offers a robust method for evaluating publication bias in meta-analysis.
• Accounting for selection effects on uncertainty is crucial for reliable interpretation of meta-analysis results.
• This research provides valuable tools for researchers aiming to mitigate the impact of publication bias.