Informative priors are one of the major attractions of Bayesian and meta-analytic approaches for me (I’d much rather express my prior information directly than sweep it under the rug and play games of modus ponens/tollens with everyone), so I found this paper particularly interesting—all about how to draw upon previous clinical trials’ results for running new experiments. It’s a little dismaying how much effort seems to be going into approaches which strike me as completely unprincipled hacks like the ‘power priors’ rather than focusing on models which match the underlying facts, like hierarchical modeling or using covariates to directly estimate similarity of past & present data points, but at least people are working along those lines!
I enjoyed
“Bayesian methods for design and analysis of safety trials”, Price et al 2013.
Good overview, and I see many old friends like multi-level models & meta-analyses mentioned.
“Use of historical control data for assessing treatment effects in clinical trials”, Viele et al 2013
Informative priors are one of the major attractions of Bayesian and meta-analytic approaches for me (I’d much rather express my prior information directly than sweep it under the rug and play games of modus ponens/tollens with everyone), so I found this paper particularly interesting—all about how to draw upon previous clinical trials’ results for running new experiments. It’s a little dismaying how much effort seems to be going into approaches which strike me as completely unprincipled hacks like the ‘power priors’ rather than focusing on models which match the underlying facts, like hierarchical modeling or using covariates to directly estimate similarity of past & present data points, but at least people are working along those lines!