ISSUE: This issue panel will discuss the suitability of the Mixed Model Repeated Measures framework when analyzing change from baseline in Clinical Outcome Assessment (COA) data, as well as alternative methods dealing with treatment discontinuation and death.

OVERVIEW:

Clinical Outcome Assessment (COA) data are typically collected at several timepoints in clinical trials. Change from baseline in COA scores at a specific timepoint is a frequently proposed endpoint, sometimes supporting labeling claims.

Change from baseline historically has been analyzed under the Mixed Model Repeated Measures (MMRM) framework. MMRM assumes missing data are missing at random, i.e., that missing data can be inferred by previous values of the same patient, or values from similar patients on the same treatment. This assumption might be viewed as suitable when it is reasonable to assume that patients with missing data have similar outcomes as those who are still being observed on study. This may not be a reasonable assumption depending on the disease setting, therapy, and reasons for ‘missing’ data (e.g., death). Additionally, the plan may be to collect COA data after randomized treatment discontinuation but only some patients are actually observed.

Occurrences that may impact the interpretation of data or cause missing data are called Intercurrent Events (ICE) under the estimand framework (ICH E9 Addendum (2020)). It is important to address ICEs based on the clinical question of interest.

This issue panel seeks to discuss how suitable the MMRM framework is when analyzing and interpreting change from baseline in COA data, as well as propose alternative methodological approaches that may be more suitable when addressing certain ICEs, such as treatment discontinuation or death.