Sensitivity Analysis for Nonignorable Drop Out in Longitudinal Studies Using Semiparametric Selection Models

• Activity Number: 170
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Emin Tahirovic*+ and Andrea Beth Troxel
• Companies: University of Pennsylvania and University of Pennsylvania
• Keywords: drop-out ; longitudinal data ; semiparametric ; identifiability ; independent censoring ; sensitivity analysis

Abstract

A great number of techniques for adjusting for informative drop out in longitudinal studies have been developed over the years. A longstanding concern in this area has been that all of these methods make untestable assumptions in order to make inference on the target population. Relatively easy and comprehensive determination of sensitivity to these untestable identifiability assumptions is an important component of any new proposal. Fully parametric approaches often conflate modeling and identifiability assumptions. We define a general algorithmic approach for sensitivity analysis facilitated by semi-parametric selection models. Some new semi-parametric techniques that adjust for selection bias without specifying any model for probability of drop-out present a challenge for a global sensitivity analysis. In this setting we propose to adapt an existing semi-parametric paradigm to define a global sensitivity analysis for such methods. We discuss how to survey and transform experts' input on sensitivity parameters so they can be used and interpreted in the analysis.