Abstract 3594: Adaptively randomized seamless-phase multiarm platform trial: Glioblastoma Multiforme Adaptive Global Innovative Learning Environment (GBM AGILE)

Abstract

Abstract Traditional phase 3 clinical trials compare an experimental arm with control. They inefficiently use patients, time, and finances. Dramatic and rapid changes in biology makes such trials untenable. We describe an alternative drug development strategy that we are using in a particular setting, the trial GBM AGILE (Glioblastoma Multiforme Adaptive Global Innovative Learning Environment). The trial’s design employs many innovations. Some aspects are similar to those of I-SPY 2 (see 4 articles in July 7, 2016 NEJM) but GBM AGILE extends I-SPY 2 in many ways. (1) It is a Bayesian platform trial that simultaneously evaluates many treatment arms (including combinations) from many companies. (2) Arms are added to the trial at any time and leave when they have been evaluated, whether positively or negatively. (3) An arm’s sample size is adaptive and based on frequent analyses of the trial results. (4) Every arm has an initial stage in which it is randomized adaptively: arms performing better in disease subtypes are assigned with higher probability to such patients. (5) An arm that performs sufficiently well in a disease subset moves seamlessly into a small (50-patient) confirmatory, registration stage in the same subset, with equal randomization against control. (6) All experimental arms are compared against a common control arm that is assigned to 20% of patients in every subtype; a bridging model takes advantage of having many arms in the trial and many comparisons among arms, and enables indirect randomization comparisons of all arms with all controls. (7) Patient subtypes are defined by line of therapy, MGMT methylation status for newly diagnosed patients, and biomarkers associated with targeted therapies, although adaptive randomization enables us to draw conclusions about off-target effects. The many possible subtypes means that there are many possible drug indications. So there are many possible “error types” and no single definition of statistical power. For example, the trial may conclude that a drug’s indication is “recurrent, biomarker-positive” disease when in truth it is “all recurrent” disease. We show how the design addresses this issue and we define “pure type I error.” GBM AGILE’s primary endpoint is overall survival (OS). To make the design more efficient we incorporate evaluations of patients’ statuses over time using a longitudinal model based on periodic MRI assessments and performance status. The longitudinal model and its components are not end points but rather provide auxiliary information that enables multiply imputing OS for surviving patients. We represent the trial’s coordinating committees that are made up of more than 150 enthusiastic and devoted disease experts and advocates from around the globe, including from Australia and China. The U.S. FDA has been enormously helpful in designing GBM AGILE, especially as regards its potential for drug and biomarker registration. Our approach provides a model for other diseases, including those outside of cancer. Citation Format: Donald A. Berry, Todd Graves, Jason Connor, Brian Alexander, Timothy Cloughesy, Anna Barker, Scott M. Berry, for the GBM AGILE Global Alliance. Adaptively randomized seamless-phase multiarm platform trial: Glioblastoma Multiforme Adaptive Global Innovative Learning Environment (GBM AGILE) [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 3594. doi:10.1158/1538-7445.AM2017-3594