All Times EDT
Keywords: multiple imputation, sensitivity analysis, missing data, Repeated Measurements
In clinical trials, the presence of missing data due to withdrawal or partial compliance is often inevitable and creates complexity. If a statistical analysis makes a wrong assumption about the missing, the obtained parameter estimates will be biased, resulting in misleading inferences. The importance of conducting sensitivity analysis has been increasingly emphasized by regulatory agencies. Systematically controlled multiple imputation (MI) enables accessible sensitivity analysis for missing data via combining patten mixture modeling, which includes the Delta-based MI methods and the reference-based MI approaches. This poster demonstrates how different assumptions for unobserved data may be made for different treatment arms of subjects in the same clinical trial with worked examples. The software implementation is based on the five macros, which fit a Bayesian Normal repeated measures model and then impute missing data under a series of post-withdrawal profiles including Jump to Reference, Copy Increment from Reference, and Copy Reference, as described by Carpenter et al. (2012). We have made modification and enhancement to the macros to allow more choices on the priors, provide flexible structures for covariance over time, and improve sampling convergence and efficiency.