Marginal Regression Analysis of Longitudinal Data with Irregular, Biased Sampling

• Activity Number: 490
• Type: Contributed
• Date/Time: Thursday, August 11, 2005 : 8:30 AM to 10:20 AM
• Sponsor: General Methodology
• Author(s): Petra Buzkova*+ and Thomas Lumley
• Companies: University of North Carolina, Chapel Hill and University of Washington
• Keywords: Longitudinal Data ; Biased Sampling ; Semiparametric Regression ; Estimating Equations ; Sampling-times Process

Abstract

In longitudinal studies, observations often are obtained at continuous subject-specific sampling times. Frequently, the availability of outcome data may be related to the outcome measure or other covariates related to the outcome measure. Under such biased sampling designs, unadjusted regression analysis yields biased estimates. Of our major interest is a mean-response model where we examine the marginal effect of the covariates X at time t on the mean of response Y at time t. Building on the work of Lin and Ying (2001) that integrates counting processes techniques with longitudinal data settings, we propose classes of estimators in generalized linear regression models that can handle biased sampling under continuous time. In linear regression and log-link models, we additionally allow for an unspecified baseline function of time. We call the proposed estimators ``inverse-intensity rate-ratio-weighted" (IIRR) estimators. They are $\sqrt n$-consistent and asymptotically normal. They do not require estimating any infinite-dimensional parameters. The estimators and estimators of their variance are relatively simple and computationally feasible.