Analysis
of Clustered Binary Outcomes Measured with Uncertainty
Na Li
Department of Biostatistics, University of Minnesota
Friday, February 15, 2008, 12:30–1:30 pm
GEMS classroom, 3rd Floor in
Shriner's Building
Coffee, tea, and cookies will be provided
Abstract
In genetic epidemiological studies, we may be interested in clustered binary
outcomes that are not directly observed. Instead we either observe one or
more surrogate outcomes which tend to be error-prone, or the true outcome is
a latent variable not directly observable but is obtained from a statistical
model. Through simulations, we compare several methods dealing with
misclassification in binary responses. In the analysis, we use a flavor of
generalized estimating equation (GEE) called alternative logistic regression
(ALR) because we are interested in not only estimating the mean parameters
(regression coefficients) but also the correlation parameters in the form of
odds ratios. We compared imputation based methods (in particular, multiple
imputation (MI)), simulation and extrapolation (SIMEX), and a GEE model with
a modified link function to accommodate the measurement error. We evaluated
their performances on estimating both the regression coefficients and the
correlation parameter. We will also present some results from a study of
delayed graft function in kidney transplant patients through data available
in the United States Renal Data System (USRDS). As a work in progress, we
propose a modified ALR procedure to allow measurement uncertainty in the
correlation model in order to correct the bias.