Design and Statistical Methods for Handling Covariates Imbalance in Randomized Controlled Clinical Trials: Dilemmas Resolved

Abstract

Introduction
In practice, between groups baseline imbalance following randomization not only opens effect estimate to bias in controlled trials,
it also has certain ethical consequences. Both design and statistical approaches to ensure balanced treatment groups in prognostic
factors are not without their drawbacks. This article identified potential limitations associated with design and statistical approaches
for handling covariate imbalance in randomized controlled clinical trials (RCTs) and proffered solutions to them.
Methods
A careful review of literatures coupled with a robust appraisal of statistical models of methods involved in a way that compared
their strength and weaknesses in trial environments, was adopted.
Results
Stratification breaks down in small sample size trials and may not accommodate more than two stratification factors in practice.
On the other hand, minimization that balances for multiple prognostic factors even in small trials is not a pure random procedure
and in addition, could present with complexities in computations. Overall, either minimization or stratification factors should be
included in the model for statistical adjustment. Statistically, estimate of effect by change score analysis (CSA) is susceptible to
direction and magnitude of imbalance. Only analysis of covariance (ANCOVA) yields unbiased effect estimate in all trial scenarios
including situations with baseline imbalance in known and unknown prognostic covariates.
Conclusion
Design methods for balancing covariates between groups are not without their limitations. Both direction and size of baseline
imbalance also have profound consequence on effect estimate by CSA. Only ANCOVA yields unbiased treatment effect and is
recommended at all trial scenarios, whether or not between groups covariate imbalance matters.