A Seamless Phase I/II Design With Randomization for Biomarker-Guided Oncology Trials

The FDA's Project Optimus CALLS for randomized exploration of multiple doses to better refine selection of the optimal biologic dose (OBD). Project FrontRunner encourages the use of randomized controlled trials earlier in the drug development process. We propose a seamless three-stage trial design that addresses both goals by incorporating biomarker-subgroup stratified randomization in both the dose-ranging and efficacy stages. Stratification in the dose-ranging stage allows for possibly different OBDs within each biomarker subgroup. A Bayesian optimal interval design guides dose escalation, followed by a one-sample Bayesian optimal efficiency predictive probability design in the dose-ranging stage. A Bayesian optimal efficiency stratified control arm design is employed in the efficacy stage. The components are linked by two novel decisions: (1) selecting sufficiently safe doses to advance from escalation to dose-ranging, and (2) ranking doses based jointly on safety and efficacy to choose an OBD for the efficacy stage. In simulation studies of three potential dose levels, correct selection of the true OBD in the dose-ranging stage was challenging in some settings, with rates of correct selection ranging from 12% when the true OBD was the same as the true MRD to 80% when the true OBD was below the MTD. In many settings the proposed design outperformed BOIN12 and DROID in selection of the true OBD. Over the entire seamless study, adequate power of at least 80% was never achieved and the Type I error was controlled at too low rates, below the target of 10%. This novel design provides an important option for both settings with well-defined molecularly targeted subgroups and in settings where the molecular target is still being refined. By gathering multifaceted information across both dose levels and targeted subgroups, more efficient and better-informed decisions can be made about targeted dosing.