|
Activity Number:
|
295
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2003 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #300911 |
|
Title:
|
Nonparametric Simultaneous Estimation of Multiple Quantiles Following Nonparametric Phase I Clinical Trial Designs with Ordinal Response
|
|
Author(s):
|
Ranjan K. Paul*+ and Nancy Flournoy and William F. Rosenberger
|
|
Companies:
|
The Boeing Company and University of Missouri and University of Maryland
|
|
Address:
|
PO Box 3707, Seattle, WA, 98124-2207,
|
|
Keywords:
|
adaptive design ; isotonic regression ; proportional odds model ; random walk rule ; sequential design
|
|
Abstract:
|
There is no consensus among statisticians as to the appropriate approach to design phase I clinical trials, in part because trials with many different objectives are are labeled "phase I." In this paper, the goal is to estimate a set of target quantiles from an ordinal toxicity scale. We compare two designs in the literature for ordinal response trials with a new design based on a multistage random walk rule. Each design is nonparametric. We develop two multidimensional isotonic regression estimators to capture the ordinal data by simultaneously estimating quantiles of cumulative toxicity probabilities. These estimators are also nonparametric. We compare these estimators for the three designs to parametric maximum likelihood estimators from a proportional odds model. We find that a modified multidimensional isotonic regression estimator far exceeds others in terms of accuracy and efficiency. A rule by Simon et al. (1997) yields particularly efficient estimators, slightly more so than the random walk rule, but it has a higher frequency of dose-limiting toxicity.
|
