Exact samples sizes for clinical trials subject to size and power constraints

Abstract

This paper first describes the difficulties in providing the required sample sizes for clinical trials that guarantee type 1 and type 2 error control. The required sample sizes obviously depend on the test employed, and in this study we use the so-called E-test, which is known to have extremely favourable size properties and higher power than alternatives. To compute exact powers for this test in real time is not currently feasible, so a corpus of pre-computed exact powers (and sizes) was created, covering sample sizes up to 500. When there are no solutions within the corpus, a novel extrapolation technique is used. Exact size can be computed after the sample sizes have been extracted; however, for the E-test the exact size is virtually always very close to the nominal target. All the code has been converted into an R-package, which is available on CRAN and illustrated.