A selection model for accounting for publication bias in a full network meta-analysis

Area of Science:

• Biostatistics
• Medical Research Methodology
• Evidence Synthesis

Background:

• Publication bias significantly impacts the reliability of meta-analysis results.
• Existing selection models have limitations in handling complex network structures and consistency assumptions.
• Frequentist approaches to fitting these models often encounter numerical and identification challenges.

Purpose of the Study:

• To develop a generalizable Bayesian selection model for complex network meta-analysis.
• To account for publication bias while maintaining the consistency assumption in network meta-analysis.
• To provide a flexible framework for sensitivity analysis in the presence of publication bias.

Main Methods:

• Developed a design-by-treatment selection model within a Bayesian framework.
• The model describes the probability of study publication based on design and precision.
• Implemented the model for complex network meta-analysis, including full networks and multiple treatments.

Main Results:

• The proposed Bayesian model successfully accounts for publication bias and consistency in complex networks.
• It overcomes the numerical and generalization limitations of previous frequentist models.
• Demonstrated improved handling of additional uncertainty arising from publication bias compared to standard models.

Conclusions:

• The new Bayesian design-by-treatment selection model offers a robust approach to adjusting for publication bias in network meta-analysis.
• This methodology is generalizable and overcomes limitations of prior selection models.
• The model provides a more accurate and flexible tool for evidence synthesis, particularly in complex scenarios.