CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Estimands, missing data, and sensitivity analysis Authors:  Geert Molenberghs - UHasselt (Belgium) [presenting]
Abstract: Estimands, a crucial topic in clinical trials, are considered. A connection is made with the much older use in survey sampling theory. Using an example from surrogate marker evaluation, it is discussed where information comes from data, design, and assumptions. The latter may be unverifiable, hence the need to perform sensitivity analysis. The setting is then broadened to various forms of enrichment; that is, every situation where the model contains more aspect than the data can provide information about. Subsets of the enrichment class are: (a) coarsening; (b) augmentation. The focus is then placed on incomplete data for the rest of the presentation. A general framework for missing data is given, starting from Rubins seminal work. The defining and transforming role of the National Academy of Sciences report from 2010 about the Prevention and Treatment of Missing Data in Clinical Trials is evocated. It is argued that the role of the patient should not be forgotten, next to academe, regulators, and industry. It is shown that for every MAR model, there is a family of MNAR models that exhibits the same fit to the data. Hence, one cannot show that MAR holds or not, solely depending on the data. The implications for standard and sensitivity analyses are discussed.