Sensitivity Analyses for Handling Sample Heterogeneity and Missing Data in Clinical Trials in Alcohol Dependence

• Activity Number: 300
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Ralitza Gueorguieva* and Stephanie O'Malley and John H. Krystal
• Companies: Yale University and Yale University and Yale University
• Keywords: mixture models ; tree-based methods ; clinical trials ; alcohol dependence ; missing data ; sensitivity

Abstract

Heterogeneity of clinical findings in studies evaluating treatment efficacy for alcohol dependence can be partly attributed to the use of standard analyses of summary drinking measures that poorly reflect the distributions of drinking data. In addition, despite restrictive inclusion/exclusion criteria, subject populations may be quite heterogeneous with subgroups that show enhanced treatment effects. Mixture models allow for more appropriate handling of both heterogeneity in treatment response and drinking data distributions. Tree-based methods can be used to identify potential subgroups with enhanced treatment effect. We will present results from two clinical trials of alcohol dependence that demonstrate sensitivity of results to assumptions about population heterogeneity and to distributional assumptions about drinking data. Additionally, we will discuss issues of missing drinking data that are routinely imputed as heavy drinking in planned statistical analyses leading to bias in estimates of treatment effect. Multiple imputation methods based on different assumptions about the relationship between the missing data mechanism and the observed drinking outcomes will be presented.