|
Activity Number:
|
104
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #300798 |
|
Title:
|
Model Selection for Individualized Treatment Rules
|
|
Author(s):
|
Min Qian*+ and Susan A. Murphy
|
|
Companies:
|
The University of Michigan and The University of Michigan
|
|
Address:
|
439 West Hall, Ann Arbor, MI, 48104,
|
|
Keywords:
|
decision making ; machine learning ; oracle inequality
|
|
Abstract:
|
In medicine, no single treatment has been shown to be effective for all. This has motivated researchers to incorporate the heterogeneity in need for treatment across individuals. A decision rule that assigns treatment according to patient characteristics is called an individualized treatment rule. Our goal is to estimate the treatment rule that maximizes the mean reward. Penalized regression is used to estimate the optimal individualized treatment rule. We first give nontrivial upper bounds on the excess mean reward in terms of the excess prediction error. Fast rate of convergence of the excess mean reward is obtained in the case of low noise. Then we provide a non-asymptotic upper bound of the excess prediction error. Finally we present an oracle inequality of the excess mean reward.
|
