Despite incorporating design-based control for confounding, such as stratification, pair matching, or constrained randomization, it is sometimes advisable to also include in the analysis covariates that might still be unbalanced across the arms of the study. Depending on the goals of the study, these covariates might be at the cluster level or even at the individual level. However, depending on the sample size, the number of covariates might be limited.
When the number of clusters is small, permutation tests might be recommended. Again, the study statistician, in collaboration with the investigators, will determine the appropriate design and analysis methods.