ABSTRACT
Issues involving missing data are often linked with issues involving the choice of the proper analysis set. However, it is also true that missing data issues are often confused with issues involving the proper choice of analysis set. Consider a randomized, double-blind, two-arm clinical trial in which some subjects drop out at randomization before undergoing study therapy or any other therapy, and have no follow-up for the study endpoint. Because these subjects should be fairly distributed between arms, they may be excluded from the analysis without compromising the integrity of the randomization. Whether to include such subjects in the analysis is an analysis set issue. If these subjects were included in the analysis [i.e., as in an intent-to-treat (ITT) analysis], the imputation or representation of their values for the endpoint should not depend on treatment arms (since such subjects should have been fairly distributed between arms) and should consider the actual adherence or nonadherence to therapy. A variation of this “imputation under the null” can be used for non-inferiority trials and will be discussed later.
