|
Activity Number:
|
405
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2008 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Nonparametric Statistics
|
| Abstract - #302361 |
|
Title:
|
GEE Performance in Analyzing Longitudinal Binary Data Collected Over Unequal Time Intervals
|
|
Author(s):
|
Soe Soe Thwin*+ and David Gagnon and Howard Cabral and Adrienne Cupples
|
|
Companies:
|
Boston University Medical Center and Boston University School of Public Health and Boston University School of Public Health and Boston University School of Public Health
|
|
Address:
|
Geriatrics Section, Robinson-2, Boston, MA, 02118,
|
|
Keywords:
|
Longitudinal Data ; Unequal Intervals ; Binary Response ; Estimation ; GEE ; ALR
|
|
Abstract:
|
Binary response data observed over unequal time intervals were generated with a modified AR(1) correlation structure, whereby, pair-wise correlations were adjusted based on their interval length relative to the interval between the first 2 observations. This structure simplifies to the AR1 structure when time intervals are equal. Bias in parameter estimation and relative efficiency were evaluated for scenarios when working correlation structures were misspecified or when alternating logistic regression was invoked. We observed that performance of GEE depends on characteristics of the study design, namely the sample size (Large vs. Small), covariate (Time Stationary vs. Time Varying), and cluster Size (# observations per subject). Adjustment of interval length in correlation structure provided better estimate for highly correlated data, especially for time-stationary covariates.
|
