Discrete Survival Analysis Incorporating both Two-Phase Sampling and Annual-Biennial Track Designs

• Activity Number: 448
• Type: Contributed
• Date/Time: Thursday, August 7, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Yougui Wu*+ and Amy Graves and James Mortimer and Eric Larson
• Companies: University of South Florida and University of South Florida and University of South Florida and Group Health Cooperative of Puget Sound
• Keywords: two-phase sample ; annual-biennial track ; nonignorable missing data

Abstract

Two-phase sampling designs have been widely used in the field of psychiatry where the disease status is ascertained only for the individuals who are believed to be at higher risk based on information measured at baseline as well as each follow-up wave. In addition to this sampling design, the KAME project, a prospective study of 1,985 individuals of Japanese descent in Seattle-King County, Washington, also adopted an annual-biennial track design, whereby individuals who have been clinically evaluated at each wave and found to be disease-free were divided into high- and low-risk groups; the individuals in the high-risk group were then followed annually and those in the low-risk biennially. The statistical implication of this annual-biennial track design is that this design-based missing data are nonignorable. We propose to use an enriched model (polytomous logistic regression model) for three-levels of disease status to model the missing data. We adopt a selection model approach to incorporate the enriched model into the likelihood to obtain the MLE estimates for the parameters in a main model (discrete survival model). The proposed approaches are applied to data from the KAME project.