Introduction

Previous chapters focused on estimands, methods of and models for estimation, and means for dealing with missing data. These topics are now considered jointly in order to appreciate the interactions between them. For example, different estimands applied to the same data may yield greater or fewer missing values, which can influence the method of estimation and the validity of assumptions.

In the following section, the six estimands introduced in Chapter 3 are discussed with regard to data choices, methods of estimation, and assumptions. To refocus the discussion, key attributes of the 6 estimands are summarized in Table 11.1.

Estimands, Estimators, and Choice of Data

Estimand 1

Difference in Outcome Improvement at the Planned Endpoint for all Randomized Subjects

Estimand 1 tests a de facto (effectiveness) hypothesis at the planned endpoint of the trial, based on all randomized patients. Data after discontinuation of the initially randomized study medication and/or initiation of rescue medication are included in the primary analysis. Given the confounding influences of the follow-up data, inferences are based on the treatment regimen, not the initially randomized treatments.