Longitudinal Data with Informative Visit Times

• Activity Number: 210
• Type: Invited
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Charles Elliott McCulloch*+ and John M Neuhaus
• Companies: University of California at San Francisco and University of California at San Francisco
• Keywords: shared random effects ; missing at random ; missing not at random ; biased estimation

Abstract

Clinical databases are increasingly being used for research to address longitudinal questions. Because visit times in such data are often driven by patient characteristics and may be informative, the danger is that this will result in biased estimation compared to designed, prospective studies. Through a series of case studies we motivate a variety of potential determinants of the visit time process. Key differences are whether the process depends on underlying patient characteristics, current values (observed or unobserved) of the outcome process, only observed values of outcomes, or auxiliary variables. We use this to characterize the visit time process and survey previously suggested models. This characterization gives guidance as to which methods of analysis will tend to be biased under which visit process models. We investigate and quantify the bias using approximations and simulation studies.