SAM-HC: a Bayesian nonparametric construction of hybrid control for randomized clinical trials using external data.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Data Science

Background:

• Augmenting control arms with external data is beneficial when participant recruitment is difficult, such as in rare or pediatric diseases.
• A key challenge is addressing heterogeneity in subpopulations between external and clinical trial datasets.

Purpose of the Study:

• To introduce a novel Bayesian nonparametric approach for borrowing information from external data.
• To develop a method that accounts for subpopulation heterogeneity across datasets to improve treatment effect estimation.

Main Methods:

• Application of the Shared Atoms Model (SAM), a Bayesian nonparametric model, to identify overlapping and unique subpopulations.
• Restricting information borrowing to common subpopulations to form a hybrid control (HC).

Main Results:

• The Shared Atoms Model effectively identifies shared and distinct subpopulations across datasets.
• The proposed hybrid control approach leads to more precise estimation of treatment effects compared to traditional methods.
• Simulation studies confirmed the robustness of the method.

Conclusions:

• The Shared Atoms Model provides a robust framework for integrating external data in clinical trials by managing subpopulation heterogeneity.
• This approach enhances the precision of treatment effect estimation, particularly valuable in scenarios with limited control arm recruitment.
• The method demonstrated improved treatment effect estimation in an Atopic Dermatitis dataset application.