Bayesian methods to estimate the accuracy of diagnostic tests in meta-analysis models.

Abstract

With the growing number of studies looking at the performance of diagnostic tests, combining the studies into a meta-analysis becomes an important and increasingly viable area of statistics, especially within the medical field. We begin by developing a hierarchical Bayesian prior structure to estimate prevalences and misclassi cation rates for a single diagnostic test. We provide the results from a simulation study which shows that this model has desirable operating characteristics. We then adapt the model to analyze a scenario in which the collected studies come from two populations, one of which having a known higher prevalence of the trait of interest. Next, we adapt the model from a previous article which constructs an estimate to the summary receiver operating characteristics curve for a diagnostic test. We develop a procedure to elicit prior distributions from an expert and to provide feedback once the priors are obtained. The model is demonstrated in detail and results are reported. We conclude by finding the necessary sample size to compare two diagnostic tests while using a meta-analysis to help power the study. Here we consider a brand new diagnostic test being compared to two established tests in a network meta-analysis. We present a model that provides a sample size needed to compare sensitivities and specificities in a reasonable computing time.