|
Activity Number:
|
324
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Biopharmaceutical Section
|
| Abstract - #305483 |
|
Title:
|
Handling Missing Outcome Data When Estimating and Testing the Average Causal Effect of Treatment for a Subset Selected by a Post-Randomization Event
|
|
Author(s):
|
Robin Mogg*+ and Devan V. Mehrotra and Peter Gilbert and Thomas Ten Have and Marshall Joffe
|
|
Companies:
|
Merck Research Laboratories and Merck Research Laboratories and University of Washington/Fred Hutchinson Cancer Research Center and University of Pennsylvania and University of Pennsylvania
|
|
Address:
|
, , ,
|
|
Keywords:
|
Causal inference ; Missing at random ; Principal stratification ; Selection bias ; Sensitivity analysis
|
|
Abstract:
|
Principal stratification based methods to estimate and test the average causal effect (ACE) of treatment on an outcome measured after a post-randomization event have been described in considerable detail. Selection models that identify the ACE can be applied in a sensitivity analysis to quantify how the ACE varies over a range of presumed bias. The validity of these methods relies on an often untenable MCAR missing data assumption. Inverse weighting and robust likelihood approaches have been described under a more plausible MAR assumption. We use simulation to compare these methods with a MI-based approach to accommodate MAR missing outcome data under various conditions. We illustrate the approaches using data from an HIV vaccine clinical trial.
|
