|
Abstract:
|
Phase I studies in oncology involve the first clinical application of a novel treatment and are designed to assess safety by escalating the dose until severe toxicities are seen. This talk will point out several situations in which information is wasted in current phase I designs and give simple solutions. Bayesian and other model-based methods provide a rich source of extensions to traditional designs and can incorporate a variety of sources of information. However, they are complex and can appear to be unintuitive "black box" methods to clinical investigators who may thus be reluctant to use them. As simple, if not uniformly superior alternatives, I will present generalizations of the traditional algorithmic designs to the following situations: incomplete followup of long-term toxicities, ordered risk group cohorts, and information about the severity of toxicities.
|
