On the Implementation of Robust Meta-Analytical-Predictive Prior

Abstract

Prospectively leveraging historical control information has become increasingly popular in biometrics research. For placebo or drugs that are already being used widely, such historical controls may provide useful prior information. However, robustness of analysis in the presence of prior-data conflicts is arguably an important topic. Meta-analytical-predictive (MAP) prior is a well-known dynamic borrowing method for this purpose. A robust MAP (rMAP) prior has been shown to improve the robustness by adding a weakly informative component to the original MAP prior. Specific implementations of rMAP can vary depending on where the weakly informative prior is integrated to the relevant historical data. In this article, we outline the differences among three approaches and design simulation studies in which their performances are compared. Our results suggest that their performances can differ considerably, contrary to the perception that they are equivalent in the robustification setting.

Keywords:

• Historical data borrowing
• Meta-analytical-predictive prior
• Robustness

Supplementary Material

R programs used in this article are available on GitHub: https://github.com/squallteo.

Acknowledgments

The authors would like to thank the editor, the associate editor and two referees for their constructive comments, which lead to significant improvement of this article.