A causal inference framework for leveraging external controls in hybrid trials.

Area of Science:

• Biostatistics
• Causal Inference
• Clinical Trial Design

Background:

• Estimating average treatment effect (ATE) can be inefficient with limited randomized trial data.
• Augmenting internal trial data with external control data offers potential efficiency gains.

Purpose of the Study:

• To develop methods for causal inference using augmented data from randomized trials and external controls.
• To assess the efficiency gains from incorporating external controls in treatment effect estimation.

Main Methods:

• Formal causal inference framework to address lack of full randomization.
• Development of estimators and efficiency bounds for augmented data.
• Doubly robust estimation using machine learning for nuisance models.
• Graphical criteria for exchangeability assumptions.

Main Results:

• External controls can increase the efficiency of treatment effect estimation.
• Proposed methods demonstrated robust performance in simulation studies.
• Application to the SUNFISH trial confirmed efficiency gains.

Conclusions:

• Augmenting randomized trial data with external controls is a viable strategy to improve ATE estimation efficiency.
• The proposed methods provide a robust framework for causal inference in such settings.
• This approach offers practical benefits for clinical trial analysis, as shown in the SUNFISH trial.