Multi-period crossover trials

This paper presents a review of crossover designs for use in medical applications which have three or more treatment periods. Only outcomes which can be analysed as continuous variables are considered. Designs which purport to allow for carryover effects are reviewed in detail, as are methods for analysing data collected in such trials. In practice, it is often possible to eliminate carryover by interposing sufficiently long 'washout' periods between successive treatments, and suitable designs for this case are also mentioned. Much current practice revolves around a model which has been widely criticized: the shortcomings of this model and the implications of possible remedies, for design as well as analysis, are discussed.