REDUCING THE PATIENTS-AT-RISK (PaR) IN A RESPONSE-ADAPTIVE TRIAL: A NUMERICAL STUDY

For a response adaptive allocation of subjects in a clinical trial, optimality with respect to the power function benefits future patients, whereas increasing the number of patients allotted to the superior treatment is helpful for the current patients in the study. This is similar to the fundamental explore-exploit trade-off that is discussed in the multi-armed bandit literature. Both the above aspects must be considered while designing a clinical trial. Now which one is more important? One way to approach this problem is to put a minimum proportion of patients allocated to the superior treatment as a constraint in the optimisation problem. However, an additional constraint to an optimisation problem typically leads to some loss in optimality. What is the social cost of this loss? Can this be quantified? The article begins with answering these questions and builds upon them to propose a new robust criterion for evaluating response-adaptive designs.