Comments on 'A dose-finding approach based on shrunken predictive probability for combinations of two agents in phase I trials' by Akihiro Hirakawa, Chikuma Hamada, and Shigeyuki Matsui

摘要

It is not always easy to compare rival dose-finding methods, because they may not always have exactly the same objectives. In cases where the objectives are comparable, then a relative comparison is much easier to interpret when certain design parameters are fixed and common to both approaches. In the case of simulated comparisons, it is important to not only keep the same pattern of dose-response relationships, but it is necessary for the sample sizes to be comparable. For some sequential methods, the sample size is not known in advance, whereas others use a fixed sample size, making fair comparison awkward. Nonetheless, by using, say, average sample size or the same fixed sample size, we are able to make fair comparisons. In a recent paper, Hirakawa, Hamada, and Matsui (HHM; [1]) proposed a new design based on shrunken predictive probability, for the dose finding problem of multi-drug combinations. The authors compared their approach with the partial order continual reassessment method proposed by Wages, Conaway, and O'Quigley (WCO; [2]), coming to the conclusion that the new approach exhibited superior performance to WCO in some scenarios, and performed no worse than WCO in others.