The All Randomized/Full Analysis Set (ICH E9)—May Patients Be Excluded from the Analysis? *

The intention-to-treat (ITT) principle has become an important component for a fair comparison of two or more treatments. Deviations from the principle are frequently found, because in practice the ideal of perfect protocol adherence and complete follow-up is difficult to achieve. The question of whether an exclusion of patients from the statistical analysis compromises the ITT principle must be oriented at the preservation of (1) the comparability initially achieved by randomization, (2) the appropriateness of the answer to the underlying clinical question, and (3) the need for an unbiased estimate of the treatment effect. While some situations in which the exclusion of patients will usually not impair the comparability can be specified, there are only few exceptions in which the deviation from the principle does not lead to an inadequate answer to the clinical question. Equivalence and noninferiority studies, safety analyses, and special trial designs require specific considerations.