Bayesian Estimation of Cluster-Level Test Accuracy Based on Different Sampling Schemes

摘要

We develop Bayesian models to estimate cluster-level test characteristics, sensitivity, specificity, prevalence, and predictive values, based on four different sampling schemes: a single test case and three sequential test cases. The corresponding cluster-level characteristics are calculated and compared for different sample sizes, sampling schemes, individual-level sensitivities, specificities, and cut-off values. We compared posterior estimates of individual-level and cluster-level characteristicsfor these four sampling schemes with simulated data. Two illustrations, one for Johne's disease in cattle and another for Salmonella in pig herds, are used to demonstrate application of the methods.