|
Abstract:
|
Missing data in clinical trials has been a prominent topic over the last several years. Since the 2010 National Research Council's report on missing data, much focus has been on methods corresponding to more principled assumptions than single imputation (eg LOCF). Under the assumption of ignorable missingness, Multiple Imputation (MI) and Mixed Model Repeated Measures (MMRM) are both sensible approaches and clearly superior to single imputation. Both have been successfully applied in numerous submissions in many therapeutic areas. One universal approach is not likely to be recommended. There is some evidence however, that MMRM provides better control of some statistical properties than MI (Siddiqui, 2011), but it is still not clear when or whether MMRM should be recommended over MI for primary analyses. We will explore the relative merits, implications, and potential drawbacks of MI and MMRM with the goal of providing more clarity on best practices in our industry.
|
