|
Activity Number:
|
364
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 14, 2002 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Biopharmaceutical Section*
|
| Abstract - #300902 |
|
Title:
|
Covariate Measurement Error Estimation in a Mixed Model for Longitudinal Data
|
|
Author(s):
|
Ying Wan*+ and Vernon Chinchilli
|
|
Affiliation(s):
|
Merck Research Laboratories and Pennsylvania State University
|
|
Address:
|
518 Township Line Rd., Blue Bell, Pennsylvania, 19422, USA
|
|
Keywords:
|
measurement error ; covariate ; mixed model
|
|
Abstract:
|
Covariate measurement error is a common source of bias in the regression analysis of biostatistical data. We explore the effects of measurement error in intra-individual covariates on estimates of the mean and variance parameters for a linear mixed model for longitudinal data. We present a conditional measurement error model and apply it to a clinical study of "Nutrition Education for Hypercholesterolemic Children." In particular, we derived a longitudinal mixed-effects model to examine the plasma LDL-cholesterol level in response to two nutrition education programs (fixed effects), when dietary intakes are treated as covariates (random effects) subject to measurement error. We compared the resulting coefficient estimates and variances in the error-adjusted model with those in the unadjusted model. Although the final inferences for both analyses are similar, variance estimates for the coefficients in the measurement error model are higher. This magnitude of covariate measurement error in the mixed model could be evaluated by efficiency of adjustment, which is measured by var (adjusted model)/var (unadjusted model).
|
