Dose finding in cancer clinical trials is dominated by algorithmic designs based on the principle that the highest tolerated dose is also the most effective dose. New treatments do not necessarily follow this principle: higher doses may not be more efficacious. As such, novel trial designs identifying the optimal dose based on safety and efficacy are needed. We propose a phase I/II design with two stages focusing on immunotherapy trials. In the 1st stage, doses are escalated in a standard algorithmic fashion, with dose escalation decisions based on likelihood principles. Continuous immunotherapy outcomes are used to evaluate the relative efficacy of the doses. Adaptive randomization is used to assign patients in the 2nd stage. The randomization probabilities are based on the T-cell persistence (patients are assigned to doses showing higher persistence with higher probability). Safety data is also collected: doses may be 'closed' in the 2nd stage based on likelihood ratios as described for the first stage. We will compare the proposed design and competitor designs using percent allocation per dose and estimation of outcomes under different dose-response scenarios.
Dept of Biostatistics
biostats [at] columbia [dot] edu