Item Details

Print View

The CRM (Continual Reassessment Method) Design in the Presence of Population Heterogeneity

Shu, Jianfen
Thesis/Dissertation; Online
Shu, Jianfen
O'Quigley, John
This thesis first studies the parameterization issue in the CRM. The nature of the CRM design causes patients to be accumulated on a single dose [1], which leads to insufficient data to estimate two parameters. The two-parameter model is too flexible and has potentially erratic behaviors, which cause more early patients receiving toxic and untested doses. The performance of the two-parameter CRM is highly influenced by prior specification which is often expressed as pseudodata. Our study shows the two-parameter CRM with the commonly used pseudodata performs poorly for very safe treatments, compared with the two-stage oneparameter CRM. It is more reassuring to use one stage of the trial to collect the prior instead of specifying the prior or using imaginary patients. The main focus of this thesis is to investigate the extensions of CRM to twogroup or multiple-group studies where patients are from two or more different populations. Bridging studies, in which there are two distinct groups, such as adults and children, and where the adult study precedes the study in children, are a special case of the heterogeneity problem. In many two-group studies, it is known that one group of patients is less sensitive to treatment and the true MTD is higher compared to the other group. In this research, we focus on the CRM shift model. In this model, the MTD for one group is shifted either up or down by a certain number of dose levels depending on the known information about the group difference. The operating characteristic of this CRM model is demonstrated by simulated examples and performance is evaluated together with other CRM approaches and the optimal design. The shift model can be applied to situations with more than 2 groups. Stopping rules and method robustness are explored. Proof of asymptotical consistency with realistic conditions is provided as well. Note: Abstract extracted from PDF text
University of Virginia, Department of Statistics, PHD, 2012
Published Date
All rights reserved (no additional license for public reuse)
Libra ETD Repository


Read Online