A Simulation-Based Evaluation of Statistical Methods for Hybrid Real-World Control Arms in Clinical Trials

Real-world (RW) data have been a source for creating external control arms to evaluate results from randomized controlled trials (RCTs) in rare diseases and scenarios where randomization to a control group is unethical or unfeasible. However, the validity of any decision making based on such comparative results depends heavily on the appropriateness and quality of the control arm data. FDA guidance lists multiple bias-generating concerns with the use of real-world controls arising from data quality and validity issues, which we frame as a data source ignorability assumption under the potential outcome framework. Hybrid control designs, RCTs with a full treatment group and a small underpowered control group supplemented with RW control data, have the potential to address some of these bias concerns. Statistical methods have been proposed for the analysis of hybrid designs and can adjust for potential violations of the data source ignorability assumption. A simulation study is presented to evaluate the operating characteristics of single and hybrid real-world control methods across the bias-generating scenarios mentioned in FDA guidance. Results suggest that certain methods can adjust for potential biases under these scenarios but may result in reduced efficiency through larger standard errors, or type I error inflation. Implications for the use of such methods and suggestions for additional work are discussed.