The Slope-Up Pattern Mixture Model with Multiple Imputation

• Activity Number: 88
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Kenneth Liu* and Gregory Golm and James Mancuso
• Companies: Merck Research Laboratories and Merck and Pfizer Inc.
• Keywords: missing data ; Missing Not at Random ; sensitivity analysis ; LOCF ; Jump to Control ; Missing at Random

Abstract

Regulatory agencies are questioning the Missing at Random (MAR) mechanism assumed by mixed models. For example, The Prevention and Treatment of Missing Data in Clinical Trials (2010) written by the National Academy of Sciences (endorsed by FDA) and the Guideline on Missing Data in Confirmatory Clinical Trials (2010) written by the EMA (endorsed by CHMP) recommend performing sensitivity analyses that assume alternative missing data mechanisms specifically, Missing Not at Random (MNAR). Unlike MAR that assumes that missingness is due to observed data, MNAR assumes that missingness is due to unobserved data so observed data cannot provide information about missing data.

Data following a MNAR mechanism can be imputed using Pattern Mixture Models (PMM). We present data from a 2-arm placebo-controlled clinical trial and perform sensitivity analyses including: a mixed model, LOCF, and 2 PMMs, the Jump to Control (J2C) PMM, and a novel approach, the "slope-up" PMM with a "slope-up tipping point" extension.