Related Experiment Videos
Properties of permuted-block randomization in clinical trials
1Department of Surgery, University of Minnesota, Minneapolis 55414.
Controlled Clinical Trials
|December 1, 1988
Summary
Statistical analysis of permuted-block designs requires incorporating blocking for accurate results. Ignoring blocking leads to conservative tests, especially with positive intrablock correlation, impacting clinical trial validity.
Area of Science:
- Biostatistics
- Clinical Trial Design
- Statistical Inference
Background:
- Permuted-block designs (blocked-randomization) are common in clinical trials.
- Analysis often ignores the blocking, potentially affecting statistical test validity.
- Intrablock correlation can influence the conservatism of statistical tests.
Purpose of the Study:
- To describe statistical properties of permuted-block designs.
- To analyze the impact of ignoring blocking in statistical tests.
- To assess selection and covariate imbalance bias in permuted-block designs.
Main Methods:
- Permutation models for statistical tests (e.g., Mantel-Haenszel, ANOVA F-test, linear rank test).
- Analysis of test size (T vs. TI) related to intrablock correlation (R) using TI = T(1-R).
- Application of Blackwell-Hodges and Efron models to assess bias.
Main Results:
- Ignoring blocking leads to unduly conservative tests when intrablock correlation is positive (TI = T(1-R)).
- Permuted-block designs can introduce selection bias due to predictable assignments.
- Covariate imbalances can cause bias in treatment effect estimation, reduced by larger block sizes.
Conclusions:
- Proper analysis of permuted-block designs must incorporate blocking.
- Selection bias is a concern, particularly in unmasked trials, and is not fully mitigated by random block sizes.
- Larger block sizes in permuted-block designs reduce bias from covariate imbalances.