More powerful randomization-based p-values in double-blind trials with non-compliance

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Inference

Background:

• Standard intent-to-treat (ITT) analyses in randomized clinical trials (RCTs) are valid but often lack power with alternative hypotheses.
• Non-compliance in RCTs necessitates alternative analytical approaches beyond standard ITT.

Purpose of the Study:

• To develop a more powerful randomization-based procedure for RCTs with non-compliance.
• To introduce a Bayesian approach for estimating treatment effects under an 'exclusion' hypothesis.

Main Methods:

• Utilized Bayesian analysis under an 'exclusion' hypothesis, where treatment assignment only matters if it alters actual treatment received.
• Developed a novel randomization-based procedure using posterior predictive checks with a non-compliance model.
• Ensured the new procedure maintains frequentist validity.

Main Results:

• The new procedure yields improved estimates for the effect of treatment receipt.
• Demonstrated substantially greater power compared to standard intent-to-treat procedures.
• Highlighted the distinction from biased 'as treated' and 'per protocol' analyses.

Conclusions:

• The proposed Bayesian-informed randomization-based method offers a more powerful and valid approach for RCTs with non-compliance.
• This method provides a robust alternative for estimating treatment effects when adherence varies.
• The findings suggest a significant advancement in statistical methodologies for clinical trial analysis.