Cluster Randomized Trials Designed to Support Generalizable Inferences.

Area of Science:

• Epidemiology
• Biostatistics
• Clinical Trials

Background:

• Cluster randomized trials (CRTs) often face challenges in generalizing findings due to practical sampling constraints.
• Oversampling clusters based on characteristics can improve trial economy but may compromise generalizability.

Purpose of the Study:

• To describe and evaluate a nested trial design for CRTs that allows for known, characteristic-dependent cluster sampling probabilities.
• To develop methods for analyzing data from this design to ensure generalizable causal inferences to the target population.

Main Methods:

• A nested trial design embedding randomized clusters within a larger cohort of eligible clusters.
• Development and evaluation of statistical methods for identifying and estimating average potential outcomes and average treatment effects.
• Simulation studies to assess bias and precision of proposed estimators.

Main Results:

• The proposed methods allow for precise quantification of treatment effects in the target population.
• Estimators demonstrated low bias across simulation studies.
• Different estimators exhibited varying levels of precision.

Conclusions:

• The nested trial design, combined with efficient estimation, can address practical trial conduct needs while maintaining generalizability.
• This approach enables precise causal inference in the target population, even with characteristic-based cluster selection.