A Two-part Model for Longitudinal Data with Missing Values

• Activity Number: 273
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Yeonjoo Yi*+ and Leann Myers
• Companies: Tulane University and Tulane University
• Keywords: semicontinuous variable ; longitudinal data ; two-part model ; generalized estimating equations (GEE) ; missing values ; simulation study

Abstract

Semicontinuous variables have a proportion of responses which are a single value (often zero) and the remaining responses which follow a continuous, often skewed, distribution. In a two-part model, the first level models the probability that the semicontinuous variable takes on its point mass value, and the second level models the distribution of the variable given that it is not at its point mass. The two parts are then combined into a single model. We extend this two-part model approach to longitudinal settings with missing observations at both levels. We use generalized estimating equations regression methods (GEE) with a logit link to predict who will respond at different time points. Using only the nonzero responses, GEE methods using a normal link are used to predict the mean level of response at different time points. The two parts are then combined to obtain a single prediction model. We will present epidemiological examples of this two-part model for longitudinal studies by GEE with missing values. Limitations are also discussed.