Abstract:
|
The idea of supplementing the control arm of a current study with data from a historically similar control arm seems to offer several advantages including devoting more resources to the new treatment arm, reduced total sample size, ethical and cost saving. Previous authors have shown that such historical data should be down-weighted or discounted during analysis. A major question then is what should be the amount of discounting that will be appropriate and what are the consequences of a wrong choice. The inclusion of similar historical data may allow for reduce mean squared error, increased power and reduced Type 1 error rate. However, if the historical control data are inconsistent with current trial control data, there is potential for bias and inflated Type 1 error rate. We investigate the impact of borrowing historical active control data in a non-inferiority clinical data with binary outcome using both frequentist and Bayesian methods. Historical data are incorporated via hierarchical modeling and meta-analytic approaches.
|