All Times EDT
Keywords: prior-data conflict, hybrid designs, effective sample size, real word evidence
Hybrid designs are an important approach to leveraging historical control data to reduce the sample size of standard randomized controlled trial (RCT) designs by utilizing hybrid controls, which consist of concurrent controls and augmented controls “borrowed” from historical data. In practice, it is of great interest to quantify the effective sample size of the hybrid controls. We develop a pragmatic method to determine the effective sample size of the hybrid controls by benchmarking the RCT design that the hybrid design targets, referred to as the benchmark effective sample size (BESS). The BESS of the hybrid controls is determined by the number of RCT controls that would be needed to match the conditional power curve from the hybrid design as compared to the power achieved from the benchmark RCT design. The BESS has several desirable properties. First, it depends on the observed data (e.g., observed effect size), rather than an average over hypothetical datasets. Consequently, BESS inherently adjusts for prior-data conflict and takes into account the effect size of the observed data. Second, by benchmarking to a RCT design, BESS does not rely on the notion of a chosen non-informative/vague prior (e.g., a chosen baseline prior model) to quantify the effective sample size. Third, BESS can be computed with any selected prior (e.g., power prior, commensurate prior, robust meta-analytic predictive prior). Examples are shown using Normal and binary endpoints.