|
Abstract:
|
The cluster randomized trial (CRT) allocates treatment at the cluster level (e.g. school or hospital) and measures outcomes at the individual level. Often, the number of enrolled clusters is small (< 20), and there could be chance imbalance on baseline covariates across the two arms. Such imbalance is particularly problematic when the covariates are predictive of the outcome because it can threaten the internal validity of the CRT. Matching and stratification are two restricted randomization procedures commonly used to balance covariates at the design stage. An alternative, less commonly-used restricted randomization method is covariate-constrained randomization. This approach quantifies baseline imbalance of cluster-level covariates using a balance metric and is able to accommodate multiple covariates, both categorical and continuous. The data may then be analyzed using a clustered permutation test, accounting for the constrained design. We illustrate covariate-constrained randomization and permutation analysis using a real CRT for increasing child immunization rates, and implement them in our newly developed R and Stata packages, cvcrand.
|
