Inclusive designs that allow the inclusion of a broader population in randomized controlled trials: An evaluation of randomization ratio and analysis strategy

In most multi-arm randomized controlled trials, all participants need to be eligible to be randomized to all the arms. This means some participants are excluded from participating in multi-arm studies. We propose to relax this by considering a 'differential randomization' approach that allows the inclusion of participants who are eligible for some but not all arms, as in some innovative trial approaches. We refer to a multi-arm design that employs differential randomization as an 'inclusive design' because it maximizes participant inclusion. Considering superiority comparisons for a normal endpoint, we evaluate the performance of some analysis methods for a three-arm inclusive design by simulation studies. Methods include pairwise or overall regression analysis, an averaging approach that pools summary statistics using prevalence rate, and a meta-analysis type framework. We compare the statistical power when the inclusive design employs different treatment allocation schemes. We find that using equal allocation ratio within each subpopulation leads to a higher disjunctive power than using equal allocation across arms, when the many-to-one comparisons are analysed using the corresponding pairwise data separately as opposed to using all data in a single regression analysis. We find that differential covariate-outcome relations among patients can affect the property of the inference.