|
Activity Number:
|
157
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 9, 2004 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
ENAR
|
| Abstract - #301306 |
|
Title:
|
Longitudinal Modeling of Incomplete Data: Example of Plasma Viral Load in HIV Infection
|
|
Author(s):
|
Rodolphe Thiebaut*+ and Hélène Jacqmin-Gadda
|
|
Companies:
|
Université Bordeaux 2 and Université Bordeaux 2
|
|
Address:
|
INSERM E0338 Bordeaux France, Bordeaux, International, 33076, France
|
|
Keywords:
|
longitudinal data ; joint modeling ; informative dropout ; left-censoring ; HIV infection
|
|
Abstract:
|
Longitudinal data may be incomplete because of left-censoring of the response variable (e.g., in case of a lack of sensibility of the assay used to measure the outcome) or may be missing due to informative dropout. An example of such data is the plasma viral load in HIV infection that is a good prognostic factor of disease progression but may be left-censored (undetectable viral load). Moreover, follow-up of patients may also be censored due to clinical progression (i.e., opportunistic infection occurrence) leading to informative dropout. To study longitudinal evolution of such marker, we proposed a joint model including a piecewise linear mixed model for the marker evolution and an accelerated failure time model for the occurrence of informative drop-out. Parameters were estimated by a direct maximization of the likelihood (Marquardt algorithm) handling for left-censored viral load measurements (based on Gaussian or mixture distributions). We presented some checks of model assumptions. The model was applied to a dataset of HIV-1 infected patients followed in an observational cohort (Aquitaine Cohort).
|
