Phase I clinical trials in oncology: A review of the old and a look at the new.

Abstract

e14030 Background: In Phase I oncology trials the primary goal is to assess dose limiting toxicities (DLT) and estimate the maximum tolerated dose (MTD). The classical 3+3 design is still used in greater than 90% of studies. We review the classical 3+3 design and two new model-based designs: Bayesian Optimal Interval (BOIN) design and the recently updated modified toxicity probability interval (mTPI-2) design. These designs are easy to implement like the 3+3 using a simple table to guide dose escalation/de-escalation. As opposed to the 3+3 design, these designs can target a DLT rate well above or below the standard 33% target. In general, an expansion cohort is added to 3+3 designs to verify safety and get an early efficacy signal. In the 3+3 expansion cohort it is unclear how to proceed if excessive toxicity is observed; whereas, with BOIN and mTPI-2, expansion can be built in with toxicity monitoring and MTD updating. Methods: We carefully explain how computer simulations can be used to evaluate phase I designs and present results from simulations comparing the designs under several true dose toxicity curves. Results: We show that BOIN and mTPI-2 have better performance than the 3+3. These new designs select the true MTD at a much higher rate and treat a higher percentage of patients at the MTD. The new designs generally tend to allocate fewer patients to high toxicity doses and fewer patients to low toxicity doses, thus ensuring high ethical standards. Unlike older Bayesian designs (e.g., modified continual reassessment method), the newer designs do not require a statistician to update the model during the course of the trial. Readily available, free software make these designs simple to implement. Conclusions: We recommend the use of the new model-based designs over the classical 3+3 design.