Sample size calculations for studies designed to evaluate diagnostic test accuracy

Abstract

We developed a Bayesian approach to sample size calculations for cross-sectional studies designed to estimate sensitivity and specificity of one or more diagnostic tests. Sample size calculations can be made for common study designs such as one test in one population, two conditionally independent or dependent tests in ≤2 populations, and three tests in ≤2 populations. We determine a sample size combination that yields high predictive probability, with respect to the future study data, of accurate and precise estimates of sensitivity and specificity. We also consider hypothesis testing for demonstrating the superiority or equivalence of one diagnostic test relative to another. The predictive probability can also be computed when the sample size combination is fixed in advance, thereby providing a “power-like” measure for the future study. The method is straightforward to implement using the S-Plus/R library emBedBUGS together with WinBUGS.