|
Activity Number:
|
489
|
|
Type:
|
Invited
|
|
Date/Time:
|
Thursday, August 2, 2007 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
ENAR
|
| Abstract - #307909 |
|
Title:
|
Analyzing the Effects of Compliance Measured with Error in Randomized Trials
|
|
Author(s):
|
Els Goetghebeur*+ and Stijn Vansteelandt and Manoochehr Babanezhad
|
|
Companies:
|
Gent University and Ghent University and Ghent University
|
|
Address:
|
Krijgslaan 281 S9, Gent, International, 9000, Belgium
|
|
Keywords:
|
causal inference ; compliance ; measurement error ; structural models
|
|
Abstract:
|
For active drugs, exposure patterns are on the causal pathway from prescription to response, and tend to vary. Structural (generalized) mean models allow for randomization based estimates of their causal effects in terms of potential outcomes. Their estimating equations remain unbiased with mean zero errors on compliance measures, but systematic errors are common. With known (covariate specific) mean (and variance) for the error on compliance, we can build (better) adjusted unbiased structural linear and approximate logistic mean estimating equations. We also propose a new estimator relying on an additional instrumental variable to estimate the effect of true exposure on the response of interest. The unbiased estimating equations are obtained in the structural linear mean context and modified to obtain better mean squared errors. We explore their value in several randomized clinicals.
|
