Testing Crossover Studies for Carryover Effects

For the study of cardiovascular disease, for instance, hypertension and angina pectoris, properly designed crossover studies are to be preferred over parallel-design studies. Many statisticians now propose that these studies should always be tested for carryover effects. In the present paper the author analyzes the crossover studies published after November, 1985, in two general and two specialist journals (New England Journal of Medicine, Lancet, Angiology, British Heart Journal). Of 73 papers only 23 (32%) consider the possibility of carryover effects. Of these latter papers 12 (16%) perform statistical tests for these effects. The results of these tests were shown in only 7 papers (10%). It is concluded that there is a discrepancy between the statistical recommendations and the data presented in the literature.