|
Activity Number:
|
163
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 3, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section for Statistical Programmers and Analysts
|
| Abstract - #305323 |
|
Title:
|
A Tree-Based Algorithm for Identifying Subgroups of Subjects with Treatment Effect
|
|
Author(s):
|
Ilya A. Lipkovich*+ and Alexei A. Dmitrienko and Jonathan Denne and Gregory Enas
|
|
Companies:
|
Eli Lilly and Company and Eli Lilly and Company and Eli Lilly and Company and Eli Lilly and Company
|
|
Address:
|
Lilly Corporate Center, Indianapolis, IN, 46285,
|
|
Keywords:
|
Recursive partitioning ; Tree-based search ; Data Mining ; Statistical computing
|
|
Abstract:
|
A tree-based algorithm for identifying subgroups of subjects with treatment effects has been proposed recently. The idea is to recursively partition data into two groups at each node so as to maximize treatment effect within selected subgroup until a predefined stopping condition has been satisfied. The procedure is similar to Classification and Regression Trees (CART) but allows generating many promising subgroups by choosing multiple splits at every node, while focusing only on subset(s) of data with desirable treatment effect. In this presentation we focus on implementation details of the algorithm: (i) fast updating of test statistics in tree nodes; (ii) accounting for correlation across multiple splits within a single predictor; (iii) accounting for multiplicity in potential covariates and levels of hierarchy; (iv) using cross-validation to select values of tuning parameters.
|
