A tutorial in estimating the prevalence of disease in humans and animals in the absence of a gold standard diagnostic

Accurate estimation of the prevalence of disease is an essential part of both human and veterinary public health. For many pathogens this estimation is complicated by the lack of an appropriate reference test, that is, a diagnostic test which when applied to samples taken from a given target population has known accuracy (e.g. a gold standard/error free, or where the misclassification error is reliably known and understood). An important fact which is often overlooked is that the accuracy of a diagnostic test is a population specific parameter [1], as opposed to some intrinsic constant, as it depends upon the specific biological characteristics of the study population.Despite the longstanding availability of approaches for estimating disease prevalence in the presence of diagnostic uncertainty, the use of such methods is still far from common. For example, a recent review of 69 prevalence studies [2] found that despite the lack of an available reference test, none of the studies provided either estimates of true prevalence or indications as to the accuracy of the diagnostics used.When estimating disease status it is crucially important to distinguish between analytical and diagnostic accuracy of a test. Analytical accuracy is concerned with repeatability and robustness of the assay under laboratory conditions, when applied to samples usually with a known disease status [3]. In contrast, diagnostic accuracy is the ability of the assay to correctly identify a truly diseased subject from a non-diseased subject when applied to a sample from a randomly chosen individual from a given population of interest. This population may be defined in terms of biological characteristics, or else by geography or any other relevant commonality. High analytical accuracy does not imply high diagnostic accuracy. For example, a diagnostic test may reasonably be considered a gold standard test when applied to one study population but not another if these are epidemiologically different (e.