|
Activity Number:
|
207
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 7, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
IMS
|
| Abstract - #307068 |
|
Title:
|
Minimax Interval Estimation of Optimal Treatment Stategies
|
|
Author(s):
|
James Robins*+ and Eric Tchetgen and Lingling Li and Aad Van der Vaart
|
|
Companies:
|
Harvard School of Public Health and Harvard School of Public Health and Harvard School of Public Health and Amsterdam
|
|
Address:
|
Kresge, Boston, MA, 02115,
|
|
Keywords:
|
higher order u-statistics ; causal inference ; nonparametric statistics ; semiparametric statistics ; optimal treatment strategy
|
|
Abstract:
|
James Robins (joint with Aad van der Vaart, Lingling Li, Eric Tchetgen) We use our unified theory of inference in non-,semi-, and fully parametric problems based on higher order U-statistics to construct optimal confidence intervals for the conditional treatment effect at each level of a high dimensional vector X of pretreatment continuous covariates based on data from either randomized or observational study data. We then use this interval to construct optimal confidence intervals for the optimal treatment strategy function d*(X) that determines the optimal treatment as a function of a patient's covariates X. This methodology helps solve the vexing problem of how to determine which apparent qualitative interactions observed in a randomized experiment are real rather than do to data snooping and sampling variability.
|
