Bayesian sample-size determination for inference on two binomial populations with no gold standard classifier

Abstract

We consider the impact of test properties on the required sample size for the Bayesian design problem for comparing two proportions with error-prone data. Specifically, we examine four cases: a single diagnostic test and two independent diagnostic tests, both when the test properties are identical across populations and when they differ. Interval-based and moment-based sample-size determination criteria are contrasted using Monte Carlo simulation methods. We consider an application in which Strongyloides infections are compared in two populations.