ABSTRACT

This chapter provides an overview of simple and ad hoc methods of dealing with missing data in longitudinal clinical trials. These methods are generally not recommended for use but are of historic interest and provide a useful starting point from which to differentiate the more principled methods described in subsequent chapters. It is assumed throughout this chapter that the estimand of interest is the treatment effect that would have been estimated in an infinitely large trial with no missing values; that is, estimand 3 as introduced in Table 2.1, the de jure (efficacy) estimand. Focus is on the difference between treatments in mean change from baseline to endpoint.