Randomization Inference and Conformal Selective Borrowing in Hybrid Controlled Trials

Music City Center 

External controls from historical trials or observational data can enhance randomized controlled trials (RCTs) when large-scale randomization is impractical or unethical, such as in rare disease drug evaluations. However, non-randomized external controls may introduce biases, and existing Bayesian and frequentist methods can inflate type I error rates, especially in small-sample trials where borrowing is most needed. To address this, we propose a randomization inference framework that ensures finite-sample exact and model-free type I error control, adhering to the "analyze as you randomize" principle to mitigate hidden biases. Since biased external controls reduce randomization test power, we leverage conformal inference to develop an individualized test-then-pool procedure that selectively borrows comparable controls. Our approach accounts for selection uncertainty, providing valid post-selection inference. We also introduce an adaptive procedure to optimize selection by minimizing mean squared error. The methods are validated through theory, simulations, and a lung cancer trial with external controls.

causal inference

data fusion

randomization test

real-world data and evidence

small sample size 

Biopharmaceutical Section