|
Abstract:
|
While RCTs are the gold standard to assess the efficacy and safety of experimental treatments, external control arms, particularly from real-world (RW) data, are a viable alternative when RCTs are not feasible or ethical. FDA guidance on rare diseases raises concerns about data heterogeneity and validity when using external controls potentially resulting in bias, invalid inference, and incorrect decision making. The use of external control arms requires an assumption of strong data source ignorability in addition to unconfoundedness of treatment assignment, which are often non-verifiable in single arm trials. Hybrid control designs, RCTs with a full treatment group and a small control group supplemented with RW controls, can potentially detect and address violations of these assumptions. We present a simulation study to evaluate the operating characteristics of various statistical methods for single and hybrid RW control designs across bias-generating scenarios highlighted in the FDA guidance. Results suggest that while certain hybrid control methods can account for potential biases under these scenarios, they may lead to reduced efficiency or type I error inflation.
|
