Revisiting Adaptive Methodology for Identifying Non-Toxic Immunotherapy Regimens
Thomas Braun, MS, PhD
Professor of Biostatistics, University of Michigan
ABSTRACT: There is a plethora of published methodology for adaptive Bayesian designs for Phase I trials in oncology when a single dose of an experimental agent is given to participants. However, in practice, cancer is not treated with a single administration, but instead through repeated administrations, of an agent, which we refer to as schedules or cycles. Thus, we seek to describe and contribute to the methodology for designing studies in which several potential treatment schedules are being compared for potential dose-limiting toxicities (DLTs). This talk will review two existing designs in which the probability of DLT is generated through a cumulative hazard function applied to each participant’s time-to-DLT, and also describe the strengths and limitations of those designs. We will then present an alternative approach in which participants instead provide a series of longitudinal binary outcomes of DLT for each administration, and we model each outcome’s conditional probability of DLT on the previous outcome using traditional models from single-dose Bayesian adaptive designs. The operating characteristics of this method and the two existing methods will be examined and compared via simulations motivated by a pharmaceutical clinical trial.
The meeting agenda includes a presentation, programmatic announcements, and updates from Cancer Center program leaders.
This group meets once a month from noon – 1 p.m.