Model-based adaptive randomization procedures for heteroscedasticity of treatment responses

Area of Science:

• Clinical Trials
• Biostatistics
• Statistical Methods

Background:

• Patient responses in clinical trials often depend on treatment and covariates, leading to heteroscedasticity.
• Standard clinical trial designs may lack sufficient power to detect treatment-covariate interactions.
• Heteroscedasticity in treatment responses can complicate efficacy assessments.

Purpose of the Study:

• To develop model-based adaptive randomization procedures to address heteroscedasticity in clinical trials.
• To enhance the power of tests for treatment-covariate interactions.
• To provide methods for hypothesis testing and sample size estimation in the context of adaptive randomization.

Main Methods:

• Development of two model-based adaptive randomization procedures.
• Derivation of limiting allocation proportions generalizing Neyman allocation.
• Investigation of hypothesis testing and sample size estimation.

Main Results:

• The proposed adaptive randomization procedures demonstrated greater power compared to complete randomization.
• These methods effectively detect differences in systematic effects, main treatment effects, and treatment-covariate interactions.
• Simulation studies validated the effectiveness and the limiting allocation proportions.

Conclusions:

• Model-based adaptive randomization offers improved power for detecting interactions in heteroscedastic clinical trial data.
• The derived allocation proportions provide efficient treatment assignments.
• These methods enhance the statistical rigor of clinical trial analysis.