Missing Data Sensitivity Analyses with Small Sample Sizes

• Activity Number: 242
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Susan Huyck*+
• Companies: Merck
• Keywords: missing data ; sensitivity analysis ; small sample size

Abstract

Clinical trials often have incomplete data due to dropouts or intermittent missed measurements which can make analysis and interpretation of the study results difficult. Recent attention has focused on methods such as pattern mixture and selection models used as sensitivity analyses to support a primary analysis. In the case of continuous data measured longitudinally, a repeated measures model is often used as the primary analysis; however the method choice for sensitivity analysis is not always clear. Difficulties with implementation of the more complex pattern mixture and selection model methods can arise in the setting of small clinical trials where the sample size may not support these approaches. This talk will compare several methods and present a non-parametric approach for analyzing incomplete longitudinal data using ranks known as ETRank. This approach allows for incorporation of reasons for drop out such as lack of efficacy or adverse events and may offer an alternative sensitivity analysis approach in the case of small sample sizes. Application to a clinical trial data example will be presented.