Keywords: Combination drug, dose escalation, Bayesian, CRM
It is believed that combination drugs have a greater efficacy by inhibiting/targeting two or more pathways at the same time instead of the individual pathways alone. Developing a combination drug however, comes with increased safety risk and operational complexity. Therefore, it needs an efficient dose-escalation scheme that fits increased design complexity to find the most efficacious and safe combination drug level(s) quickly. Most of cases, there is prior information about the dose-toxicity/efficacy relationship of each combination drug from previous single agent trials, publications, preclinical data, etc. Bayesian model-based dose escalation designs will utilize these prior information of each drug as well as the accumulated on-going clinical trial data, relevant external data. There are several methods to find a maximum tolerate dose (MTD) in combination drugs, e.g., mTPI (Ji et al., 2007), BOIN (Liu and Yuan, 2015), Keyboard design (Yan, Mandrekar and Yuan, 201), or the waterfall design (Yuan and Zhang, 2017). With cytotoxic drugs, the objective is to find the MTD under the assumption that toxicity is monotonically increasing as the dose increases. However, the dose–toxicity surface of the molecularly targeted agents (MTAs) or immunotherapy drugs is likely to plateau at higher dose levels, and the dose–efficacy relationship may exhibit a nonmonotonic pattern. Therefore, the objective for MTAs or immunotherapy drugs is to find the optimal biological dose (OBD), defined as the lowest dose combination having the highest efficacy and tolerable toxicity. Therefore, it is important to consider both efficacy and toxicity in finding the OBD of combination drugs. Recently, several phase I/II drug combination trial designs have been proposed to account for both toxicity and efficacy (Huang et al. (2007), Wages and Conaway (2014), Guo and Li (2015), Riviere et al. (2015)).