Efficient interaction analysis in randomized controlled trials.

Area of Science:

• Biostatistics
• Clinical Trials
• Precision Medicine

Background:

• Identifying treatment-covariate interactions is crucial for understanding treatment effect heterogeneity.
• Continuous covariates pose challenges due to ambiguous definitions and model assumptions in interaction analysis.
• Existing methods may yield inaccurate uncertainty estimates in interaction analysis.

Purpose of the Study:

• To develop a model-free framework for interaction analysis in randomized controlled trials (RCTs).
• To address challenges with continuous covariates and covariate-adaptive randomization.
• To propose a semiparametric efficient method for interaction effect analysis.

Main Methods:

• Introduced a model-free framework defining a clear target parameter for interaction.
• Studied interaction analysis under covariate-adaptive randomization (simple, stratified, minimization).
• Developed a consistent variance estimator and a novel semiparametric efficient method using machine learning.

Main Results:

• The proposed framework avoids functional form assumptions of the data-generating mechanism.
• The new method corrects for exaggerated or understated uncertainty from usual methods.
• Demonstrated efficiency and wide applicability of the semiparametric efficient inference procedure.

Conclusions:

• The model-free framework offers a robust approach to interaction analysis in RCTs.
• The semiparametric efficient method enhances precision medicine by accurately assessing treatment effect heterogeneity.
• The approach is applicable across various covariate-adaptive randomization schemes.