Subgroup Debate Followup

A post-Phd topic:

Extend on `subgroup debate chapter’ by running the argument that the trialist’s view on the reliability of subgroup data amounts to a reductio for relying on frequentist methods to interpret results of clinical trials (especially in terms of applying the results to patients).

The trialists acknowledge that clinicians need reliable subgroup data. They then give many good arguments for why reliable subgroup data is not possible.

I agree with these arguments. Where I differ from the trialists, however, is what I think these arguments establish. For mine they show that frequentist methods are not up to the task of informing therapeutic decisions. (I am happy to accept frequentist methods have some good features for tests of efficacy).

Indeed—and this is the thesis of the paper—the trialists argument against the possibility of reliable subgroup data is a pretty strong argument for Bayesian approaches to interpreting clinical trials.