ABSTRACT
For evaluation of the efficacy of a test treatment for progressive disease such as cancer or HIV, a parallel-group active control randomized clinical trial is often conducted. Under the study design, qualified patients are randomly assigned to receive either an active control (a standard therapy or a treatment currently available in the marketplace) or the test treatment under investigation. Patients are allowed to switch from one treatment to another due to ethical consideration such as lack of response or evidence of disease progression. In practice, it is not uncommon that up to 80% of patients may switch from one treatment to another. This certainly has an impact on the evaluation of the efficacy of the test treatment. Despite allowing a switch between two treatments, many clinical studies are to compare the test treatment with the active control agent as if no patients had ever switched. Sommer and Zeger (1991) referred to the treatment effect among patients who complied with treatment as biological efficacy. In practice, the survival time of a patient who switched from the active control to the test treatment might be on the average longer than his/her survival time that would have been if he/she had adhered to the original treatment (either the active control or the test treatment), if switching is based on prognosis to optimally assign patients’ treatments over time. We refer to the difference caused by treatment switch as switching effect. The purpose of this chapter is to discuss some models for treatment switch with switching effect and methods for statistical inference under these models.
