Crossover designs: issues in construction, use, and communication

Abstract

In a trial with a crossover design, participants receive a sequence of treatments over two or more periods, with the outcome measured at the end of each period. In order to estimate contrasts between direct treatment effects and between carryover treatment effects, we model each observation as a linear combination of the effects of period, participant, the direct effect of the current treatment, and, for all except the first period observations, the carryover effect of the treatment in the preceding period. In this thesis, we will consider some aspects of the design and use of crossover experiments. Our focus will be on methods for construction and comparison of designs which improve performance and are accessible to those researchers who need to use crossover designs but who are not specialists in statistical methodology or the design of experiments. In Chapter 2 we discuss the construction of balanced crossover designs. In Chapter 3 we consider visual methods for determining the connectedness of block, row- column, and a restricted class of crossover designs. In Chapter 4 we discuss participant dropout in crossover designs, and introduce a new criterion for selecting a design that is less likely to result in non-estimable treatment contrasts in the event of some participants not completing the trial. In Chapter 5 we present a review of the use of crossover designs in the scientific literature during a one-year period. In Chapter 6 we discuss the relationship between the theory of crossover designs as described in the earlier chapters, and the reality of the use of crossover designs as described in Chapter 5. We conclude by discussing potential practical approaches for making some experimental design methods more widely known and used by researchers who implement trials with crossover designs.