Estimation of the disease prevalence when diagnostic tests are subject to classification error: Bayesian Approach

Abstract

The estimation of disease prevalence, which refers to the number of cases with a particular disease in a population divided by the total number of individuals, becomes highly precise when a 100% accurate test, known as the gold standard, is available. However, in many situations, due to the expensive nature of diagnostic tests or limited technological resources, it is not feasible to use a gold standard test. Instead, one or two less expensive tests with lower levels of accuracy, such as sensitivity or specificity, need to be employed. This study focuses on investigating two Bayesian approaches for estimating disease prevalence when it is not possible to obtain results from a 100% accurate diagnostic test. The first approach involves a model with two parameters that consider the association between test results. The second approach proposes the use of Bayesian Model Averaging, which combines four models, each making different assumptions regarding the association of test results. To assess the performance of both approaches, various scenarios will be examined using simulated datasets. Based on the results of this simulation study, the approaches can then be employed to estimate the prevalence of chronic kidney disease in Peru, utilizing data from the CRONICAS cohort study (Francis et al., 2015).