ABSTRACT
The role of cross-over trials in clinical research is explained and illustrated. Various types of design are presented and issues to do with analysis and planning are addressed. Designs illustrated by example include the simple AB/BA design, in which patients are treated with two treatments, each being given in one of two periods and the order of assignment being random, to more complex designs, including those based on Latin Squares and incomplete block designs in which the number of treatments being compared exceeds the number of periods. Sample size determination based on power analysis is covered.
The nature of periods, which has been misunderstood by many commentators, is explained. These should not be understood in terms of calendar time, since patients will nearly always be recruited sequentially. But in terms of order in which treatments are given.
The problem of carry-over is discussed and it is shown that conventional approaches to deal with this problem are unlikely to be successful, since they rely on mathematically convenient but biologically and pharmacologically implausible assumptions.
Brief mentions are made of the related topic of n-of-1 trials and also of the use of cross-over trials in infertility.
