Sensitivity Analyses in Clinical Trials via Pattern-Mixture Models Using Standard SAS Procedures for Multiple Imputations: How Much We Improve Over LOCF?

• Activity Number: 308
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Anjela Tzontcheva*+
• Companies: Merck
• Keywords: Sensitivity analysis ; pattern-mixture models ; multiple imputations

Abstract

The recent EMA guidance on missing data and the FDA-requested report by the National Research Council emphasize the importance of stress-testing trial results via appropriate sensitivity analyses. Single imputation methods like LOCF rely on unrealistic assumptions and may underestimate standard errors. Alternative sensitivity analysis approaches to handle the missing data exist such as pattern-mixture, selection and shared-parameter models but are complex and lack easy implementation. This talk is motivated by a need for an alternative to a planned LOCF sensitivity analyses in clinical trials with longitudinal data analyses on repeated measures. Pattern-mixture models are chosen for exploration for their clinically interpretable assumptions as well as their possible implementation using standard SAS procedures. Simulations based on observed trial dropout patterns and discontinuation reasons were performed using MI-based pattern-mixture models to assess the tipping point in the range of missing mechanism assumptions where estimated treatment effect conclusion might change.