On the sample size for one‐sided equivalence of sensitivities based upon McNemar's test

Abstract

In this paper we formalize the problem of testing the one‐sided equivalence in the sensitivities of two medical diagnostic tests under a matched‐pair study design. We derive conditional and unconditional sample size formulae which are decreasing functions of the probability of being diagnosed by both tests. We calculate upper boundary and midpoint sample sizes. Results of a Monte Carlo simulation study that compares the proposed sample size formulae with that of Lachenbruch suggest that our midpoint conditional sample size is the best choice to obtain the desired power for the type of equivalence studies discussed in the paper.