|
Abstract:
|
Often when clinical trials are designed, there is a wealth of background information available for the treatments under investigation. Usually there are numerous studies with control/standard of care arm information and it is desirable to use, i.e. borrow from, these studies to maybe lower the number of controls that need to be enrolled or gain higher precision in effect estimates. In situations where numerous studies with control data are available, studies may be grouped into clusters based on similar results. When performing the analyses the amount of borrowing desired would depend on how similar the results are to a cluster, as well as the size of the cluster. A particular benefit to this method is the ability for the current study to become its own cluster if the results indicate that borrowing is not appropriate. This alleviates some of the concern related to type I error inflation in historical borrowing. We propose a clustered hierarchical models approach for the historical borrowing and provide a practical example.
|
