Constructing Relative Effect Priors for Research Prioritization and Trial Design: A Meta-epidemiological Analysis

Area of Science:

• Biostatistics
• Clinical Trial Design
• Health Economics

Background:

• Bayesian methods can improve randomized clinical trial (RCT) design by integrating prior evidence.
• Value of Information (VOI) methods assess the value of reducing uncertainty for research prioritization.
• Both methods require prior distributions for key parameters, like relative treatment effect (RTE), but data for these priors may be lacking at the research design stage.

Purpose of the Study:

• To present methods for constructing prior distributions for relative treatment effects (RTEs) using existing collections of previous RCTs.
• To demonstrate the application of these empirically derived priors in VOI analyses for research prioritization and RCT design.

Main Methods:

• Developed two Bayesian hierarchical models to capture variability in RTEs across studies, accounting for study characteristics.
• Applied these models to a dataset of 743 published RCTs across 9 disease areas to generate predictive distributions for RTEs.
• Illustrated the use of these priors in a VOI analysis for an RCT in bladder cancer, comparing results to those obtained with uninformative priors.

Main Results:

• For most disease areas, predicted RTEs favored new interventions over comparators.
• Significant differences in predicted effects and uncertainty were observed across the 9 disease areas.
• VOI analysis indicated a substantially lower expected value of research when using empirically derived priors compared to uninformative priors.

Conclusions:

• The study demonstrates a novel approach for generating informative priors to support research prioritization and clinical trial design.
• The methodology can be extended to integrate RCT evidence with expert opinion.
• Further development of a comprehensive database of RCT evidence is needed to create readily available 'off-the-shelf' priors.