Abstract: Master protocols address multiple questions. Some master protocols consider separate clinical trials within a single framework. Others integrate the many questions within the same protocol to improve efficiency and accuracy with some extent of “borrowing” across the answers to the various questions. The latter trials usually have many adaptive aspects. I will address the latter type of master protocols for platform trials, those with many treatment arms for the same disease or diseases. These trials frequently have what the FDA calls “Complex Innovative Designs.” The FDA has been instrumental in encouraging and promoting master protocols and innovations in clinical trial design, for example, “Interacting with the FDA on Complex Innovative Trial Designs for Drugs and Biological Products.” link I will describe platform trials in oncology and COVID-19 that qualify for the label “CID.” These include I-SPY 2, an ongoing phase II trial in neoadjuvant breast cancer, and GBM AGILE (GlioBlastoma Multiforme Adaptive Global Innovative Learning Environment) and Precision Promise, seamless phase II/III trials in glioblastoma and pancreatic cancer. The latter two trials take a Bayesian approach and include many statistical innovations never before used in drug registration trials. These include utilizing nonconcurrently randomized controls (what we call the “time machine”) and adaptive randomization. GBM AGILE and Precision Promise are sponsored by nonprofit organizations and are currently accruing patients. I will also describe aspects of an ongoing international platform trial called REMAP-COVID.
Abstract: In recent years, single arm basket trials have been used for accelerated approval of oncology drugs in patients selected by a putative predictive biomarker who have an unmet medical need. The approval decision is often based on a heuristic clinical review of trial data. While being much more efficient than the conventional approach of conducting one trial for each tumor indication, a randomized basket trial can supplement the current paradigm by expanding the scope of approval (accelerated or full) to any line of therapy. The randomized basket design is achieved with pruning-and-pooling. The associated Type I error control and various other issues will be discussed.
An umbrella trial attempts to test multiple new drugs in the same tumor indication. In its simplest form, it is analogous to the traditional multiple-arm trials except that the experimental arms may enter/exit the trial at a different time. While it is well known statistically that sharing a control arm within an umbrella trial leads to smaller Type I error than for independent trials, regulatory requirement on Type I error correction remains uncertain. In this talk, Type I error distribution will be provided to further the discussion. In addition, we will discuss about optimal randomization ratio between control and experimental arms.