Sequential or fixed sample trial design? A case study by stochastic simulation.

ABSTRACT

The properties of Wilcoxon's rank sum test for fixed sample size and a Wilcoxon-type two-sample sequential test have been illustrated and compared by means of stochastic simulation. Data from a real fixed sample trial have been used, both for resampling from the original data, and for construction of an idealized theoretical distribution. The sequential and the fixed sample test obtain equal power, but the sequential test mostly includes considerably fewer patients to reach a conclusion, i.e. the mean and median number of patients included are both much lower than the fixed sample size. Under the hypotheses only a small fraction of the simulation runs exceed the fixed sample size. These findings exemplify results obtained in theoretical analyses and simulation studies covering a wide range of distributions. In our opinion sequential tests have obvious advantages and are in many cases better alternatives than fixed sample tests in clinical trials.