Multiple Imputation Using Control Quantile Statistics

• Activity Number: 607 - Recent Advances in Missing Data Methods: From Estimands to Assumptions for Primary and Sensitivity Analyses
• Type: Topic Contributed
• Date/Time: Thursday, August 3, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): David Li*
• Companies: Pfizer
• Keywords: Multiple imputation ; MNAR ; Jumping to Control ; quantile statistics

Abstract

It has been shown that jumping-to-control (JC) is not conservative compared to copying-difference-from-control (CDC) when used to impute data missing not at random (MNAR), and CDC works well when the proportion of MNAR data is not off balance too much between two comparison groups. The research for approaches to imputing missing data under a more general setting is still of need. This presentation considers imputation approaches for the following setting: 1) Missing data, though not observed, have the worst efficacy outcome; (2) The proportion of missing data in the control group is less than 50%. We will discuss approaches based on quantile statistics from the control group and assess the properties of proposed approaches.