Statistical Inference for Covariate-Adaptive Randomization Procedures With Missing Covariates

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Inference

Background:

• Covariate-adaptive randomization (CAR) balances baseline covariates in clinical trials.
• Missing covariates are a practical challenge often overlooked in CAR studies.
• Existing research lacks theoretical properties for hypothesis testing with imputed covariates in CAR.

Purpose of the Study:

• To establish theoretical properties of hypothesis testing for treatment and covariate effects under single imputation of missing covariates in CAR.
• To develop an adjusted test to correct conservativeness and improve statistical power.
• To extend the framework to multiple imputation and compare its performance with single imputation.

Main Methods:

• Theoretical analysis of hypothesis testing under single imputation for missing covariates in CAR.
• Development and evaluation of an adjusted statistical test.
• Simulation studies comparing single and multiple imputation methods.

Main Results:

• The traditional hypothesis test in CAR is conservative when covariates are missing and imputed using single imputation.
• An adjusted test effectively corrects this conservativeness, leading to increased statistical power.
• Multiple imputation shows comparable or improved performance over single imputation in simulations.

Conclusions:

• Provides a theoretical foundation for CAR procedures with missing covariates.
• The adjusted test offers a more powerful and reliable approach for hypothesis testing in CAR with imputed data.
• Highlights the importance of addressing missing data in CAR for accurate statistical inference.