Crossover Designs for Two-treatment Clinical Trials

Crossover designs in clinical trials are very popular for comparing several non-curative treatments for their eecacy, while having each patient as his/her own control. Many investigators have constructed universally optimal designs for certain repeated measurement experiments. However, most of these statistically optimal designs have certain undesirable features from a clinical perspective. In this work, we compare savings in cost and thus the gain in power attainable from employing crossover designs against completely randomized designs. We note that the assumption of a positive within-subject correlation between measurements has been implicit in earlier comparisons and in most models for repeated measures data. We explore other models for repeated measures data that allow for a negative correlation among repeated measurements. Under such models, we will consider some useful crossover designs and compare their cost eeectiveness with those of completely randomized designs and parallel group designs.