Poster Presented at ACOP 2020

Two typical pioneering Bayesian adaptive designs for the phase I cancer clinical trials are (i) continual reassessment method (CRM) and (ii) escalation with overdose control (EWOC). Drawbacks of CRM and EWOC are as follows: (i) adverse event is recorded with a binary response: DLT (dose limiting toxicity) or non DLT and (ii) they cannot accommodate grade information for multiple adverse events (AE). To accommodate grade information from multiple AEs, recently, EWOC using normalized equivalent toxicity score (EWOC-NETS) was developed. Because EWOC-NETS is based on a pseudo-Bernoulli likelihood, its drawback is that the results cannot be reproduced by simulation. In this study, we propose a two-parameter linear dose-finder (2PLD) to address a continuous toxicity response in the phase I cancer clinical trials. The 2PLD assumes a linear relationship between a toxic reaction and a dosage (or possibly an exposure). 2PLD treats the toxicity response as a continuous variable, aimed at utilizing Common Toxicity Criteria for Adverse Events provided by the National Cancer Institute. The proposed search procedure suggests an optimal exposure to each patient by using accrued patients' information while controlling the posterior probability of overdose.

The simulations show that this novel 2PLD method has good operating characteristics and improved accuracy in achieving MTD. Simulations showed that the 2PLD requires a fewer step to converge to the targeted MTD than using adaptive clinical trial design methods that are based on binary responses such as EWOC or CRM. This novel 2PLD can be an attractive tool for clinical scientists because of its parsimonious description of a toxicity-dose curve, medically interpretable hyperparameters, and an automated posterior computation.