A two‐stage trial design for testing treatment, self‐selection and treatment preference effects

Abstract We propose a two‐stage randomized clinical trial design for separating treatment effects from those resulting from choosing treatment. At the first stage all patients are randomly allocated to one of two groups, the random group and the option group. At the second stage, patients in the random group are randomized a second time to treatment A or B, whereas patients in the option group are given a free choice between the two treatments. If there are differences in treatment response between the random group and the option group, there are two potential sources of bias: self‐selection by choosing treatment, and effects of suggestion by receiving the preferred treatment. A linear model is presented to estimate these effects separately along with test statistics which are approximately normally distributed.