Parametric Models for Optimal Treatment Schedule Finding in Adaptive Early-Phase Clinical Trials.

Cure Models;Bayesian Estimation Procedures;Public Health;Statistics and Numeric Data;Health Sciences;Science;Biostatistics

Liu, ChangyingTaylor, Jeremy M. ;

University of Michigan

关键词: Cure Models;    Bayesian Estimation Procedures;    Public Health;    Statistics and Numeric Data;    Health Sciences;    Science;    Biostatistics;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/57655/liuchang_1.pdf?sequence=2&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Recently, a Bayesian paradigm was constructed for Phase I trial designsthat allows for theevaluation and comparison of several nested treatment schedules, each consisting of a sequence ofadministration times. In contrast to traditional Phase I trial designs that seek to find a maximum tolerateddose (MTD), the goal of this new design was to determine a maximum tolerated schedule (MTS). Subject accrual, Bayesian estimation procedure and outcome adaptive decision-making are done in a sequential fashion as inclassical Phase I trial designs. As competing approaches to the additive triangular hazard model proposed with the Bayesian paradigm, we propose several classes of parametric models for optimal treatment schedulefinding by both maximum likelihood and Bayesian approaches.In part I of our research, we propose a mixture cure model to identify the MTS from a fixed number of nestedtreatment schedules.We model the cure rate with logistic regression and the conditional hazard function for the susceptible patients using a combination of two Weibull distributions to account for the non-monotonicnature of the hazard of toxicity. We use a modified likelihood approach to estimate parameters of interest.In part II of our research, we propose using maximum likelihood to estimate the parameters of the triangularhazard model in a single adminstration setting. We describe how to derive estimators for the change-point and boundary parameters of the triangular hazard model and discuss their large sample properties. In part III of our research, we propose a parametric non-mixture cure model to identify the optimal treatment schedule from a fixed number of nested treatment schedules. With such a model, we generate a continuous non-monotonic hazard function for the time to toxicity of each administration, as well as model the population probability of toxicity to increase with the number of administrations.Via simulation, we compare the performance of our proposed approaches to the existing method in a variety of settings motivated by an actual study in allogeneic bone marrow transplant patients. The parameters of interest are estimated by both maximum likelihood method (EM algorithm) and Bayesian approach (MCMC procedures).