The goal of cancer therapy often is to induce remission of the disease using powerful chemotherapeutic agents, and then, if remission is achieved, to maintain remission as long as possible before relapse/recurrence, e.g., by administering additional agents that intensify or augment the effects of the initial therapy. When several induction and maintenance options are available, a natural question is which "adaptive treatment strategy," or "dynamic treatment regime," consisting of a particular induction therapy followed by a particular maintenance treatment if the former elicits a response, should be recommended to the population of patients. A common clinical trial design in this context is one in which patients are randomized at study entry to one of two induction therapies, and, if they achieve remission and consent to further participation, they are subsequently randomized to one of two maintenance regimens; otherwise, if remission is not achieved, patients are followed-up and treated at their physician's discretion. Such trials are capable of providing evidence to address the broad question of which strategy to recommend. However, standard practice in most cases is to conduct separate analyses of the effects of induction therapies on remission rates and of maintenance therapies on a longer-term time-to-event endpoint (e.g., survival) among those who achieve remission, which separately or together do not directly address this question. In the first part of the talk, we will discuss the general idea of a dynamic treatment regime and simple principles for making (frequentist) inference on quantities such as survival distribution and mean restricted survival time associated with the regimes embedded in such a trial.In the second part of the talk, we turn to the real, shameless goal, which is to stimulate interest among statisticians and others in the very important and open area of methodological development for studying and constructing dynamic treatment regimes. The talk is not technical and is meant to be broadly accessible.
This talk discusses joint work with Jared Lunceford, Butch Tsiatis, Abdus Wahed, Tom Banks, Eric Rosenberg, Susan Murphy, and many others.
Return to Biostatistics Working Group