Semiparametric Estimation for Longitudinal Binary Response Model Parameters Under Outcome-Dependent Sampling Designs

• Activity Number: 318
• Type: Invited
• Date/Time: Tuesday, August 2, 2011 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Jonathan Schildcrout*+
• Companies: Vanderbilt University School of Medicine
• Keywords: longitudinal data ; outcome dependent sampling ; epidemiological study designs

Abstract

We will discuss a general semi-parametric estimation strategy for analyses of longitudinal binary response data ascertained from outcome dependent sampling designs. The strategy permits time-varying and time-invariant estimation targets. It can be used for prospective or retrospective designs where sampling is based on an ancillary variable that is related but not equal to the response vector. The strategy involves two models. The first (sampling variable) model is estimated with a logistic regression of the sampling variable on the exposure and response variables. Estimates from this model are then used in an offset term for the second (target) model that regresses the response variable on exposure. The offset term is used to acknowledge or correct for the biased study design. The second model, in particular, is estimated with covariance weighted generalized estimating equations. We will discuss several designs and will explore optimal designs for various estimation targets.