Tipping Point Sensitivity Analysis for Stress-Testing the Censored-at-Random Assumption in Survival Analysis

• Activity Number: 395
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Bohdana Ratitch*+ and Ilya Lipkovich and Michael O'Kelly
• Companies: InVentiv Health Clinical and Quintiles and Quintiles
• Keywords: tipping point analysis ; sensitivity analysis ; survival analysis ; censored at random ; missing data ; time-to-event data

Abstract

Over the past years, a significant progress was made in developing statistically valid and clinically justifiable methods for analysis of clinical trials with missing data for continuous and binary endpoints. Similarly, analyses of time-to-event data can be challenged with respect to the robustness and integrity of study conclusions when subjects leave the study prior to experiencing an event of interest and withdraw from treatment and/or follow-up prematurely. In this presentation, we will discuss a "tipping point" approach for stress-testing an assumption of Censored at Random (CAR), typically used in the survival-type analysis of clinical trials. The objective of the tipping point analysis is to introduce one or more sensitivity parameters with a clear clinical interpretation and to identify the regions of sensitivity parameters that nullify any conclusion of the analysis in favor of the experimental treatment. We discuss several approaches for conducting such analyses based on multiple imputation using parametric, semi-parametric, and non-parametric imputation models with a sensitivity parameter representing a hazard ratio for drop-out subjects compared to completers.