Title: A Bayesian dose-finding design in oncology using pharmacokinetic/pharmacodynamic modeling
Authors: Xiao Su - The University of Texas MD Anderson Cancer Center (United States)
Yisheng Li - The University of Texas MD Anderson Cancer Center (United States) [presenting]
Peter Mueller - UT Austin (United States)
Kim-Anh Do - The University of Texas MD Anderson Cancer Center (United States)
Abstract: While a number of phase I dose-finding designs in oncology exist, the commonly used ones are either algorithmic or empirical model based. We propose a new framework for modeling the dose-response relationship via dynamic PK/PD modeling and modeling of the relationship between the pharmacologic effect and a binary toxicity outcome. Inference is implemented in one joint model that encompasses PK, PD and clinical outcome. This modeling framework naturally incorporates the information on dose, schedule and method of administration (e.g., drug formulation and route of administration) in their relationship with toxicity. Simulations show that the performance of the proposed DISCO design on average improves upon those of currently used designs, including the CRM, BOIN and mTPI designs, and a hypothetically optimal non-parametric design in some scenarios. A sensitivity analysis suggests that the performance of the DISCO design is robust with respect to assumptions related to the interindividual variability in the PK. The DISCO design is less expensive and more ethical than existing designs since it makes efficient use of the information from the enrolled patients, and it does not require PK data from all patients or real-time PK analysis. We illustrate the proposed design by applying it to the setting of a phase I trial of a gamma-secretase inhibitor in metastatic or locally advanced solid tumors.