Keywords: Bayesian, Adaptive, Rare Disease, Clinical Trial Design
Bayesian methods were considered in a randomized study of a rare disease to expose the minimum number of children to control treatment while still having a conclusive study, i.e., with similar or better performance characteristics as a randomized classical group sequential design (GSD). Therefore we explored the option to borrow information from an external patient registry via Bayesian methods with the aim to mimic the operational characteristics of the classical GSD. The primary objective of this randomized event-driven study was to show delay of time to first clinical worsening with test treatment as compared to standard of care (SoC) over several months duration of treatment. Prior SoC data on time to first disease progression was available from a single prospective, contemporary worldwide patient registry. The Bayesian framework is a natural consideration to combine prior information with clinical trial data on SoC to lower the sample size required to achieve the same statistical power as available with a classical GSD. This talk discusses the pros and cons of the Bayesian approach. In particular, if the TRUE underlying hazard rates can be assumed the same for the prior SoC dataset and the clinical study dataset, then the Bayesian approach achieves the same power as a standard frequentist GSD, but with smaller required sample size. However, without that equality assumption, methods to adjust for non-negligible difference under type 1 error control in the presence of potential heterogeneity of prior and trial SoC rates required similar sample sizes as a traditional frequentist GSD. This talk will summarize the methods and simulation results supporting these findings.