An application of minimization for ensuring balanced study arms in a group-randomized COVID-19 educational intervention trial

Abstract

Background

Our institution carried out a multi-center, group-randomized controlled trial to evaluate the effectiveness of two community-based interventions on promoting the uptake of COVID-19 testing and vaccination in three regions with high levels of health disparities in Texas. We selected Census Block Groups (CBGs) with high disparity for randomization. In each study region, selected CBGs were randomized into two intervention groups and one control group. An important goal was to ensure balanced distributions of two continuous covariates, the disparity index and population size, across study arms. In this paper, we describe a novel minimization method used to ensure balanced study arms.

Methods

We employed a minimization method to balance distributions of disparity index and population size among the selected CBGs across three study groups. First, we used the means and standard deviations at the baseline to standardize covariates. Second, we used the maximum of pairwise Manhattan distances as the imbalance score. When randomizing a set of CBGs, we computed the imbalance scores for all possible assignments and used unequal allocation probabilities to implement randomization. We conducted a simulation study to evaluate the performance of the imbalance score.

Results

In both the simulation study and the actual randomization results of the trial, minimization yielded balanced groups on the marginal distributions of the disparity index and population size. In the trial, the study groups were highly homogenous regarding the joint distribution of disparity index and population size (p = 0.91).

Conclusion

The results indicate that the maximum of pairwise Manhattan distances is a practically useful imbalance score. Using this imbalance score, the minimization procedure satisfactorily balances the distribution of continuous covariates among three study groups.