Asymptotic Biases and Misspecified Models for Missing Data

• Activity Number: 242
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Melanie Poulin-Costello and Michael McIsaac*+ and Richard Cook
• Companies: Amgen and Department of Statistics and Actuarial Science University of Waterloo and University of Waterloo
• Keywords: inverse probability weights ; last observation carried forward ; missing data ; multiple imputation ; non-responder imputation

Abstract

In this talk we critically discuss commonly used ad hoc methods for dealing with incomplete data in clinical trials and evaluate these methods in terms of the limiting values of the resultant estimators. We contrast these methods with more generally appropriate methods of analysis including multiple imputation, inverse probability weighted estimating equations (IPWEE), and augmented IPWEE, and we explore the relative efficiency and robustness of resulting estimators. In particular, we explore the sensitivity of these methods to the misspecification of the models needed to adjust for the missingness. These methods and considerations are then illustrated through application to recent dermatology data.