A Bayesian Pattern-Mixture Model for Longitudinal Data with Informative Dropouts

• Activity Number: 30
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Niko Kaciroti*+
• Companies: University of Michigan
• Keywords: Missing data ; missing not at random ; shrinkage sensitivity analysis ; randomized clinical trial

Abstract

In randomized trials some subjects may drop out prematurely for various reasons. Many of the data analyses assume that missing data are missing at random (MAR), which may not be valid. We propose a Bayesian pattern-mixture model to analyzed longitudinal data where dropout mechanism is potentially not missing at random. The new method uses easy-to-understand identifying constrains that relate the unidentifying parameters to the identifying parameters at different follow-up times and across different dropout patterns. These identifying parameters are then used for sensitivity analysis. The large number of the sensitivity parameters is reduced by considering them as random and assumed to follow a log-normal distribution with pre-specified mean and variance. We vary the mean and the variance to explore sensitivity of inferences. The MAR is a special case within the proposed model, thus allowing exploration of the sensitivity to inferences as departures from the inferences under the MAR assumption. The proposed approach is applied to blood pressures data from a randomized trial.