Outside Help: Can Bayesian Borrowing Help Supplement Limited Sample Sizes in Pediatric and Rare Disease Trials While Mitigating Risk of Bias?

PURPOSE: Bayesian borrowing methods are seeing growing uptake in regulatory settings for supplementing limited sample sizes in rare indications–particularly in pediatric populations. These methods can allow for the amount of information borrowing to depend on the degree of heterogeneity across data sources, with greater borrowing from the external data when outcomes are compatible across data sources and down-weighting of the external data when outcomes are incompatible. This dynamic borrowing approach has also been used to partially borrow information across different subpopulations in basket trials, to supplement limited sample sizes in pediatric trials by borrowing from adult trial populations, and to augment small concurrent control arms in randomized controlled trials. The goal of the workshop will be to introduce these methods to a health economics and outcome research (HEOR) audience and outline some recent applications. Lastly, potential for application of these methods to support future health technology assessments (HTA) will be discussed.

DESCRIPTION: Paul Arora will provide a brief introduction, including challenges faced when evaluating novel therapies for rare diseases and key considerations for selecting data sources for borrowing (10 minutes). Emma Mackay will introduce Bayesian borrowing methods, static vs. dynamic borrowing and prior-based modelling approaches that have seen recent application in regulatory and HTA settings. She will conclude with an example application (15-20 minutes). Aaron Springford will provide a focused discussion on the use of meta-analytic predictive (MAP) prior approaches, outlining recent examples of borrowing to augment a small concurrent control arm in a randomized controlled trial (RCT), and supplementing of limited sample sizes in pediatric trials using adult data (15-20 minutes). The session will conclude with a Q&A / panel discussion on the potential for Bayesian borrowing methods in future HTA applications. An interactive smartphone-based polling feature will be used to encourage audience participation.