The Two-Period Crossover Design in Medical Research

Abstract

The crossover design has enjoyed popularity with many clinical researchers, but has been criticized by biostatisticians. The central problem is the inability to derive an unbiased estimate of the treatment effect when differences occur because of the different sequences in which treatments are applied. This problem can be traced to a deficiency of the logic of the crossover arrangement itself. Factors that can invalidate the findings of a crossover trial include nonuniform pharmacologic and psychologic carry-over effects, failure to return patients to their baseline state before the crossover, nonuniform changes in the patients over time, and the use of time-dependent response measures. When these problems can be anticipated, a parallel-groups design should be used instead of a crossover trial.