Underestimation of treatment effect in crossover trials

In crossover trials each subject serves as his own control. For the study of cardiovascular diseases such as hypertension and angina pectoria, properly designed crossover studies are preferred to parallel studies. There is a considerable between-subject variability of symptoms in some of these conditions. Bias due to this is eliminated by the use of a crossover design. However, a problem is the so-called treatment-by-period interaction. The present study analyzes the potential influences of this on the outcome of the trial. Physical carryover effect, defined as a physical effect of the first treatment period carrying on into the second, tends to minimize differences between two consecutive treatment periods. So does the frustrating experience of an inactive agent in the first treatment period. Outside influences such as the change of the seasons may affect lengthy crossover trials in a similar way. The author concludes that the treatment effect in a crossover trial tends to be underestimated. The current concept that reports of clinical trials are generally biased toward an exaggeration of treatment effects does not seem to apply to crossover trials.