A Varying Coefficient Method for Analyzing Longitudinal Data with Nonignorable Dropout

• Activity Number: 311
• Type: Contributed
• Date/Time: Tuesday, August 5, 2008 : 2:00 PM to 3:50 PM
• Sponsor: WNAR
• Author(s): Jeri E.F. Harwood*+ and Samantha MaWhinney+ and Diane Fairclough
• Companies: University of Colorado Denver and Health Sciences Center University of Colorado Denver and University of Colorado Denver
• Keywords: B-splines ; Clinical trials ; HIV/AIDS ; Longitudinal data ; Nonignorable dropout ; Varying-coefficient models

Abstract

Dropout is common in longitudinal studies and is nonignorable when the probability of dropout depends on unobserved outcomes even after conditioning on available data. In this case, mixture models can be used to account for the relationship between a longitudinal outcome and dropout. We propose a varying-coefficient mixture model using natural cubic B-spline basis functions (VCM_ns) to semiparametrically model the outcome-dropout relationship. This method is computationally stable, highly flexible and relatively simple to implement. We apply the VCM_ns and comparable methods to an HIV/AIDS clinical trial that shows evidence of nonignorable dropout. In addition, we conduct simulation studies to evaluate performance and compare methodologies. The simulation studies suggest that the VCM_ns is an improvement over existing methods when the dropout mechanism is nonlinear.