Background
The Multiphase Optimization Strategy (MOST) is designed to maximize the impact of clinical healthcare interventions, which are typically multicomponent and increasingly complex. The MOST framework often relies on factorial experiments to identify which components of an intervention are most effective, efficient, and scalable. When assigning participants to conditions in factorial experiments, researchers must be careful to select the assignment method that will result in balanced sample sizes and equivalence of covariates across conditions without being predictable. Historically, procedures most often include simple randomization and stratification with blocking; minimization is an increasingly utilized method that assigns participants to the condition that minimizes differences in covariates and sample size across study conditions.
Methods
In the context of a MOST optimization trial with a 2x2x2x2 factorial design (4 components, 16 cells), we used computer simulation to empirically test three subject assignment methods: simple randomization, stratification with blocking, and minimization. We compared these methods with respect to: sample size balance across condition, equivalence across conditions on key covariates, and unpredictability of assignments. Leveraging an existing dataset to compare three different allocation methods, we conducted 250 computerized simulations using bootstrap samples of 304 participants, which was the planned sample size for the proposed study.
Result
Simple randomization, the most unpredictable subject allocation method, generated the least balance of sample and equivalence of covariates across the 16 study cells. Stratification with blocking performed well on stratified variables, and resulted in similar sample balance and predictability as minimization. In contrast, minimization, which had a higher degree of complexity and cost, was most successful in achieving balanced sample sizes and equivalence across a large number of covariates.
Conclusions
Unlike simple randomization, minimization procedures and stratification with blocking are both methodologically sound options for factorial designs. Based on the computer simulation results and priorities within context of this MOST optimization trial, minimization was selected as the optimal subject allocation method. Minimization is utilized infrequently in randomized experiments but represents an important technical advance that should be considered by researchers implementing multi-arm and factorial studies.