|
Activity Number:
|
58
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Sunday, July 29, 2007 : 4:00 PM to 5:50 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308894 |
|
Title:
|
Latent Variable Model for Multiple Outcomes with Nonignorable Missing Data
|
|
Author(s):
|
Xiaohong Yan*+ and W. John Boscardin
|
|
Companies:
|
University of California, Los Angeles and University of California, Los Angeles
|
|
Address:
|
Department of Biostatistics, Los Angeles, CA, 90095,
|
|
Keywords:
|
latent variable ; multiple outcomes ; non-ignorable missing data ; multivariate longitudinal data
|
|
Abstract:
|
In clinical trials, multiple outcomes are usually collected over time to measure the same quantity of interest, such as treatment effect. These measures could be in different data type (discrete or continuous), and tend to be incompletely recorded due to drop out, death or other reasons. Ignoring these missing data might induce bias in inference analysis. We propose a latent variable shared parameter model for multiple outcomes in different data type (binary and continuous) with non-ignorable missing data. The latent variable accommodates the correlation structure among different response outcomes. The mechanism for those non-ignorable missing data is modeled by a shared parameter model. We fit the model using Markov chain Monte Carlo method. Our method is illustrated using real data from scleroderma lung study (SLS).
|
