[Skip to Content]
[Skip to Content Landing]
Views 284
Citations 0
Comment & Response
April 23/30, 2019

Post Hoc Bayesian Analyses—Reply

Author Affiliations
  • 1Interdepartmental Division of Critical Care Medicine, University of Toronto, Toronto, Ontario, Canada
  • 2Department of Medicine, University Health Network, Toronto, Ontario, Canada
JAMA. 2019;321(16):1632-1633. doi:10.1001/jama.2019.1202

In Reply Dr Aberegg raises several important issues about our Bayesian reanalysis of the EOLIA trial.1

First, in raising the specter of confirmation bias, the author describes exactly the kind of intuitive Bayesian analysis that clinicians perform regularly, combining a hypothetical prior consistent with treatment harm and a hypothetical marginally “significant” trial result to imagine a posterior consistent with treatment harm. He may or may not be correct in speculating that such a result would not have been met by a Bayesian reanalysis; a prior for harm might have motivated this. Nevertheless, his concern about confirmation bias is not an argument against Bayesian analysis as such, but rather that these analyses should be performed consistently regardless of the anticipated result and specified as part of the design, a contention that we fully support.