Estimation of the odds ratio from multi-stage randomized trials

Area of Science:

• Biostatistics
• Clinical Trial Design
• Medical Research Methodology

Background:

• Multi-stage randomized trials allow early study termination based on efficacy.
• Early stopping rules can introduce bias into treatment effect estimations, particularly the maximum likelihood estimator.
• Estimating odds ratios in small sample sizes, common in phase II cancer trials, presents challenges.

Purpose of the Study:

• To investigate bias in odds ratio estimation within multi-stage randomized trials.
• To evaluate existing odds ratio estimation methods in this context.
• To propose novel bias-corrected estimators for multi-stage randomized trials.

Main Methods:

• Considered multi-stage randomized trials with dichotomous outcomes.
• Evaluated current odds ratio estimation techniques.
• Developed and tested new bias-corrected estimators through numerical simulations.

Main Results:

• Existing estimation methods exhibit bias in multi-stage randomized trials.
• The proposed bias-corrected estimators demonstrate reduced bias.
• The new estimators also show a smaller mean squared error compared to existing methods.

Conclusions:

• Bias-corrected estimators are essential for accurate odds ratio estimation in multi-stage randomized trials.
• The proposed methods offer improved precision, especially for challenging designs like randomized phase II cancer trials.
• These findings enhance the reliability of treatment effect evaluation in adaptive clinical trial settings.