Abstract
In block-randomized clinical trials where the treatments can be identified by their appearance or effects, knowledge of the length and composition of the blocks enables prediction of the nature of one or more treatments at the end of a block and may introduce a selection bias. Matts and Lachin (1) have quantified “predictability with certainty” for two-arm trials with balanced randomization. In this paper, we provide a formula which can be used for measuring predictability in other situations: trials with three or more arms, and trials with unbalanced treatment allocation. In addition, we have provided tables of values for predictability according to common lengths and composition. Examples of results obtained in common situations are provided. The importance of bias associated with predictability and the methods to limit it are discussed, and practical recommendations are provided.
Get full access to this article
View all access options for this article.