|
Activity Number:
|
138
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, July 30, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #309380 |
|
Title:
|
A Semiparametric Joint Modeling Approach for Nonignorable Missing Data
|
|
Author(s):
|
Liansheng Zhu*+ and Sujit Ghosh
|
|
Companies:
|
Pharmaceutical Product Development, Inc and North Carolina State University
|
|
Address:
|
311 Castle Hayne Dr, Cary, NC, 27519,
|
|
Keywords:
|
Bayesian inference ; GAM ; Joint modeling ; Missing data
|
|
Abstract:
|
It is generally difficult to analyze incomplete data when the missingness is informative. Pattern-mixture models are commonly used in practice. Concerns include that not all parameters are identifiable and marginalization over patterns can be tricky for discrete outcomes. Zhu et al. (2007) proposed a parametric joint-modeling approach within a Bayesian framework in which patterns are defined by surrogate variables and treated as random effects. A parametric generalized linear model may not be appropriate especially when the missingness is non-ignorable. An extension to accommodate nonlinear models is proposed within the framework of generalized additive models (GAMs). Our small simulation study indicates that the joint model by using GAMs performs well when the underlying model is non-linear. We reanalyzed the CPCRA trial data and confirmed the significance of the treatment effect.
|
