|
Activity Number:
|
426
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics in Epidemiology
|
| Abstract - #310003 |
|
Title:
|
Modeling Correlated Longitudinal Processes with Some Processes Partially Observed
|
|
Author(s):
|
Qilu Yu*+ and Laurel Beckett and David Bennett and Robert Wilson
|
|
Companies:
|
Johns Hopkins University and University of California, Davis and Rush University and Rush University
|
|
Address:
|
2024 E Monument St, Baltimore, MD, 21205,
|
|
Keywords:
|
longitudinal data ; aging ; partially observed process ; profile likelihood ; EM algorithm
|
|
Abstract:
|
Complex processes in growth and aging may have multiple markers of change, and result from a complex interplay between risk factors. Our goal is to model correlated processes and to understand their relationship over time. Such studies encounter particular challenges when some processes are only partially observed. The motivating example is pathology-mediated clinical change in aging, where the pathology development can only be observed with a single post-mortem measurement. We propose a family of mixed effects models for investigating the pathologic mechanisms linking risk factors to clinical decline. According to different model assumptions, two sets of approaches are developed. In Method I we propose a profile likelihood method, while in Method II a generalized EM algorithm is adapted for parameter estimation. We illustrate the analysis with an example from the Religious Orders Study.
|
