A comparison of allocation strategies for optimising clinical trial designs under variance heterogeneity

Abstract

Balanced allocation of patients across treatments is a widely adopted and practical strategy in clinical trials. Under the assumption of variance homogeneity, a balanced allocation is known to possess desirable properties in terms of commonly considered operating characteristics such as power, type I error rate, and estimation accuracy. When this assumption is violated, the balanced allocation strategy can perform suboptimally in terms of these (and other) metrics of interest, compared to alternative allocation rules. Such allocation rules are examined under the assumption of response variance heterogeneity across treatments and within an adaptive framework. A blocked design is proposed in order to account for additional sources of variability which are incorporated into the proposed design through a mixed effects model. For this setting, two allocation strategies are derived: an efficiency-oriented and an outcome-oriented strategy. These target two different and potentially conflicting objectives (estimation accuracy and within-trial patient response, respectively) that may be optimised in the presence of variance heterogeneity. A comparison of the resulting allocation strategies is provided in the context of a clinical trial studying the effect of different treatment protocols on the level of inflammation caused by Rheumatoid Arthritis. The interrelation of common clinical trial objectives under the examined allocation strategies is explored, demonstrating the benefits but also the costs in terms of objectives that are not formally incorporated into the optimisation criterion in each case.