Statistical inference on the relative risk following covariate-adaptive randomization.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Statistical Inference

Background:

• Covariate-adaptive randomization (CAR) is crucial for balancing treatment groups in clinical trials based on baseline covariates.
• While average treatment effects are well-studied, inference for relative risk under CAR remains less explored.
• Existing methods for relative risk analysis in CAR may exhibit conservative properties.

Purpose of the Study:

• To examine a covariate-adjusted estimate of relative risk under CAR.
• To investigate the properties of hypothesis tests for relative risk in CAR experiments.
• To develop and validate improved methods for relative risk inference in CAR.

Main Methods:

• Derivation of theoretical properties for covariate-adjusted relative risk across various CAR procedures.
• Introduction of model-based and model-robust methods for enhanced standard error estimation.
• Conducting extensive numerical studies to validate theoretical findings and proposed methods.

Main Results:

• Conventional hypothesis tests for relative risk under CAR were found to be conservative, resulting in reduced Type I error rates.
• The proposed model-based and model-robust methods effectively enhance standard error estimation.
• Demonstrated validity and favorable properties of the adjusted tests through numerical simulations.

Conclusions:

• The study provides a theoretical framework for covariate-adjusted relative risk inference in CAR.
• New statistical methods are proposed to address the conservativeness of conventional tests.
• The developed methods offer improved accuracy and reliability for relative risk analysis in CAR clinical trials.