Sensitivity Analyses for Clinical Trials with Missing Binary Outcomes

• Activity Number: 395
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Michael O'Kelly*+
• Companies: Quintiles
• Keywords: Clinical trials ; Missing data ; Binary outcomes ; Multiple imputation ; Tipping point analysis ; Delta adjustment

Abstract

Primary and sensitivity analyses for binary outcomes in clinical trials tend to select from the same small set of assumptions about missing data, when estimating treatment differences and when checking those estimates for sensitivity to missing data. Worst-case assumptions and last observation carried forward (LOCF) assumptions are common, and are often used to check one another, though both assumptions are often clinically implausible. While the direct likelihood or MMRM approach for continuous data implements the more plausible assumption that outcomes are missing at random (MAR), the counterpart generalized linear mixed model (GLMM) for binary outcomes involves compromises. Multiple imputation provides a flexible method of implementing both MAR and missing not at random (MNAR) assumptions, that is not yet widely used. This presentation describes how a mix of the above methods can provide clinically meaningful sensitivity analyses for a clinically plausible primary analysis of binary outcomes.