|
Activity Number:
|
332
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biometrics Section
|
| Abstract - #309591 |
|
Title:
|
Estimating the Survivor Average Causal Effect in Nonrandomized Studies Where Treatment Assignment Is Missing for Some Patients
|
|
Author(s):
|
Abbie Stokes-Riner*+ and Jason Roy and Sally W. Thurston
|
|
Companies:
|
University of Rochester and Geisinger Health System and University of Rochester
|
|
Address:
|
Department of Biostatistics, 601 Elmwood Avenue, Rochester, NY, 14642,
|
|
Keywords:
|
causal effect ; truncation due to death ; missing data ; Bayesian inference ; posttreatment variable
|
|
Abstract:
|
Evaluating the causal effect of treatment can be complicated when the outcome measurement is truncated due to a post treatment event. For example, obstetricians are interested in the effect of early epidural use on the duration of second stage labor. However, this measurement will not be available if the patient had a cesarean section delivery (C-section). The situation is further complicated when the treatment for some patients is unknown. We take a fully Bayesian approach to (1) adjust for confounding due to non-randomization, and (2) impute missing treatment assignments using auxiliary information. A sensitivity analysis is proposed to account for assumptions whose validity cannot be assessed from the data. The method is applied to obstetric data, to identify the effect of early epidural use on duration of second stage labor.
|
