Abstract:
|
In the last few years, various model-based designs have been developed to identify the maximum tolerated dose (MTD) in phase 1 trials. Encouragingly, an increasing number of trials have started to implement some of these novel methods, amongst which the Bayesian Continual Reassessment Method (CRM) is the most widely used. The CRM proposes two main dose-finding strategies: one-stage design and two-stage design. The latter requires the specification of an initial design (e.g., '3+3') that stays in effect until the first observed toxicity; at that point the trial turns to the model-based CRM. The two-stage CRM represents a more conservative escalation alternative by starting the trial at the lowest dose and potentially reducing the risk of overdosing. In order to facilitate the planning stage of a trial and provide a comparison of the two designs, we extended the sample size formulae for the one-stage CRM (Cheung, 2013) to the two-stage design. Simulation results are presented for several scenarios by varying parameters such as: targeted toxicity, accuracy probability, or number of pre-specified dose levels. Practical implementation is illustrated via a Shiny App.
|