ABSTRACT
The idea that each patient should be asked to assign their own numerical utilities to the possible outcomes and treated with the resulting optimal regimen may seem ethically appealing. Defending the subjectivity of utility-based decision making quickly leads to the simple fact that all statistical methods have highly subjective components, although they often are not recognized as such. In practice, once the trial's entry criteria, treatment regimes, and outcomes all have been established, the utilities may be elicited. More formal methods to obtain consensus utilities from the start certainly can be used, and this is especially appropriate for multi-institution studies. In the Bayesian framework, there are two different but closely related expressions that may be called the "mean utility of a treatment regime ρ." Simulating the design using alternative numerical utilities may provide the physicians with additional insights into how their utilities translate into design performance, which may motivate them to modify their utilities.
