BOIN-UT:A New Bayesian Optimal Interval Clinical Design For Dose Finding Based On Utility And Toxicity Outcomes

Background:The purpose of phase I trials for oncology drug development is to determine the optimal dose(OD).For cytotoxic agents,the maximum tolerated dose(MTD),defined as the highest dose level that can be administered to patients with clinically acceptable toxicity,was usually selected as the OD,by assuming that the dose-toxicity and dose-efficacy relationships monotonically increase with the dose level.But the landscape of oncology drug development has recently changed with the emergence of molecularly targeted agents and immunotherapies.For those molecular-targeted,cytostatic,biological agents and immune-oncology therapy,they may exhibit non-monotonic patterns in their dose-efficacy relationships.Therefore,dose-finding methods should take account of efficacy,toxicity at the same time for the clinical development of OD.For immune-oncology therapy,the immune response may also be considered.Many methods that based on both efficacy and toxicity outcomes have been proposed,but all the methods are model-based,which highly rely on the model assumptions and need to fit the models during the conduction of clinical trials continuously.The nonparametric Bayesian optimal interval(BOIN)design has been proposed to identify the MTD which is robust and does not require the assumption used in model-based designs.In our study,we aim to extend the BOIN design to consider both efficacy,immune response and toxicity outcomes at the same time.Objective:In this study,we aim to propose a new BOIN design that takes account of the efficacy,immune response and toxicity outcomes in determinating the OD.The operating characteristics of the proposed BOIN new design will be assessed by comparing with those model-based approaches,incorporating both ① efficacy and toxicity ② efficacy,immune response and toxicity for a phrase Ⅰ/Ⅱ dose-finding trial in oncology.Methods:In this study,by first minimize the posterior probability of inappropriate classification of the dose efficacy/utility and toxicity and then choose the best dose assignment under the classification,we proposed new dose-finding design BOIN-UT.We call our extended BOIN design the "BOIN-UT" design,where"U" is for short of "utility" and "T" is for short of "Toxicity".The utility here can be the efficacy outcome or the user-defined utility.We compared our new design with the EffTox design and the SHH design for taking an accounting of efficacy and toxicity outcomes by using a simulation study.We also compared our new design with the Bayesian method proposed by Liu et.al for taking an accounting of efficacy,immune response and toxicity outcomes by a simulation study.Results:The simulation showed that BOIN-UT has advantages in both the percentages of correct OD selected and the average number of patients allocated to the OD across a variety of realistic settings when comparing with the EffTox and SHH designs.Our proposed method also showed a comparable performance when compared with the Bayesian method for immunotherapy,such as LGY.Conclusion:We have proposed a new approach,BOIN-UT,to identify an OD based on both efficacy,immune response and toxicity in oncology drug development.The proposed method is nonparametric,and is thus robust,and is easy to implement simply.