At the request of the U.S. Food and Drug Administration, the National Academy of Sciences convened the Panel on the Handling of Missing Data in Clinical Trials to prepare a report that would make recommendations that could be used to aid in the FDA’s eventual development of a Guidance for Industry on that topic.
Chief among the findings and recommendations of the resultant report was the need to make all efforts to prevent the occurrence of missing data. However, when missing data are unavoidable, the report also stressed the need for the careful specification of data analysis models that would explicitly account for any missingness.
A major first step in that process is the identification of both the scientific and statistical estimands that are hidden by the incomplete data. Furthermore, because all such analysis methods are based on untestable assumptions, the report also stressed the need for carefully conducted sensitivity analyses that would quantify the robustness of trial results to departures from any assumed mechanism of missingness.
In this module we explore an approach to defining estimands and thinking about sensitivity analyses that are most relevant to the underlying scientific questions.