|
Activity Number:
|
466
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 1, 2007 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
WNAR
|
| Abstract - #310083 |
|
Title:
|
Numeric EM for Linear and Nonlinear Mixed Effects Models with Censored Responses
|
|
Author(s):
|
Lin Liu*+ and Florin Vaida
|
|
Companies:
|
University of California, San Diego and University of California, San Diego
|
|
Address:
|
9500 Gilman Drive, La Jolla, CA, 92093-0717,
|
|
Keywords:
|
detection limit ; nonlinear models ; truncated normal
|
|
Abstract:
|
Laboratory-based data are often censored due to limitations of measuring technology. In AIDS research, HIV viral load is subject to both upper and lower detection limits. The usual EM estimation procedure for mixed effects model needs to be modified to deal with such data issue. We propose a numeric implementation of the EM algorithm for linear/nonlinear mixed effects models with censored response. It is shown that the E-step reduces to computing the first two moments of certain truncated multinormal distributions. We use a numerical implementation which greatly improves the speed and precision in computing the MLE compared to the Monte Carlo EM algorithm. The likelihood function can be calculated as a bi-product of E-step and is used for monitoring convergence. We apply this work in the analysis of the longitudinal viral load of a cohort of subjects with primary HIV infection.
|
