A critique of recent research on the two-treatment crossover design.

A critical review is provided of recent articles on two-treatment crossover trials that purport to provide solutions to the problems caused by the existence of differential carryover effects. Included in the review are papers suggesting that extra-period designs be employed in which treatments follow themselves, that the bias caused by a difference between the carryover effects may sometimes be ignored, that changes from baseline be analyzed, and that a series of preliminary tests using baseline values be performed. The validity of many of these suggestions is shown to depend on whether mathematical assumptions that are generally unverifiable are satisfied. The two-period crossover design is recommended only when prior experience suggests that its assumptions are likely to be satisfied.