Model averaging for robust extrapolation in evidence synthesis.

Area of Science:

• Biostatistics
• Statistical Modeling
• Evidence Synthesis

Background:

• Meta-analyses often face sparse data from a limited number of studies.
• External data sources can be valuable but challenging to integrate effectively.
• Extrapolation methods offer a way to leverage auxiliary information.

Purpose of the Study:

• To develop a simple and robust extrapolation strategy for meta-analysis.
• To address the challenge of sparse data by utilizing external information.
• To provide an easy-to-implement solution with R code.

Main Methods:

• Utilized heavy-tailed mixture priors for effect estimation in meta-analysis.
• Implemented a model-averaging technique.
• Designed the method to anticipate and manage prior-data conflicts.

Main Results:

• The proposed method provides a robust approach to effect estimation.
• Demonstrated the ease of implementation through provided R code.
• Simulations confirmed the general robustness of the extrapolation strategy.

Conclusions:

• The heavy-tailed mixture prior approach is effective for meta-analysis with sparse data.
• The method offers a robust way to integrate external information, handling potential discrepancies.
• This strategy enhances evidence synthesis by enabling effective use of auxiliary data.