10 General Recommendations
- Sponsor launches a study for internal use: prior up to sponsor but should be in SAP before study unblinded
- Trial for regulatory submission: prior specification is iterative process between sponsor and medical and statistical reviewers, before study begins
- If have preexisting trustworthy, relevant information from similar disease, treatment, dosing
- often reasonable to use such information in prior
- Seldom appropriate for expert opinion alone to form the prior
- Strong pharmacology and existing drug class information could be used for form weakly optimistic prior
- Strong Phase 2 design that is similar to Phase 3 design: may be OK to use Phase 2 posterior as Phase 3 prior
- may discount using mix of skeptical prior and previous study posterior
- See Weber et al. (2018) for very useful information about using prior knowledge and giving new data a chance
- Dallow et al. (2018) is the best available paper about how to elicit priors from experts
- No applicable prior info: somehwat skeptical prior
- in exchange: sponsor can take unlimited looks at the data (if study blinding/integrity can be preserved)
- recall that this skepticism is equivalent to ignoring small number of subjects
- Reporting:
- give details of prior, how developed, when
- report P(efficacy) and P(efficacy > ε)
- no hard cutoff on a “win”; instead use above 2 PPs with less need for large 2nd PP
- quantify totality of evidence using PP(compound condition)
- 0.95 credible interval for efficacy parameter but emphasize directional PPs
- For tx not found to be superior, report PP(similarity)
- for non-inferiority studies report PP(similarity) and PP(non-inferiority)