Randomization-based inference in the presence of selection bias

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Nonparametric Statistics

Background:

• Clinical trial participant samples often lack representativeness of the target population.
• Small sample sizes and covariate imbalances can compromise trial validity.
• Traditional analyses may rely on population-based assumptions not met in practice.

Purpose of the Study:

• To propose a nonparametric statistical model providing a formal basis for randomization tests.
• To adapt the model for covariate imbalance and selection bias in clinical trials.
• To investigate the impact of bias on randomization test rejection probabilities and control for it.

Main Methods:

• Development of a novel nonparametric statistical model for randomization tests.
• Adaptation of the model to account for selection bias due to covariate imbalance.
• Monte Carlo simulations to assess the effects of bias on test performance.
• Application of ancillary statistics to control for bias.

Main Results:

• Covariate imbalance was shown to inflate the type I error probability in randomization tests.
• The proposed nonparametric model effectively controls for the influence of bias.
• Ancillary statistics can be used to achieve an unbiased, adjusted randomization test.

Conclusions:

• Nonparametric randomization tests offer a robust alternative when population assumptions are violated.
• The developed model and use of ancillary statistics provide a method to correct for bias in clinical trial analysis.
• This approach enhances the reliability and accuracy of treatment effect estimations in the presence of covariate imbalance.