Abstract:
|
Introduction: Schoenfeld et al. (2009) proposed Bayesian design and analysis for a pediatric randomize controlled trial (RCT), to be informed by adult data, using a pre-specified borrowing fraction parameter (BFP). This incorrectly assumes that the BFP is intuitive and corresponds to the degree of similarity between the populations. Methods: We generalize the results of Schoenfeld et. al to K>1 historical datasets, derive simple expressions and show that it leads to Empirical Bayes meta-analysis (EBMA, Raudenbush & Byrk, 1985). We also show that the powers of the induced power-prior approach (Ibrahim & Chen 2000) are driven by uncertainty, rather than affinity between datasets. We proceed to offer power and sample size calculations for a future RCT, to be informed by K historical RCTs. The method is applied to simulated datasets with a dichotomous outcome. Results: We demonstrate how designing RCTs with EBMA in mind, can save resources and improve estimation accuracy. Conclusion: Bridging Schoenfeld’s design and EBMA can help inform inference on treatment effects for independent trials involving either different patient populations or various therapies from a common class.
|