Abstract by Matt Drewes
Bayesian Variable Selection for Evaluating Rapid Diagnostic Tests for Malaria
Diagnostic studies constitute an active area of current medical research. However, most such studies are univariate in nature, focusing only on the performance of individual tests in isolation. Some researchers have proposed an alternative multivariate approach to diagnostic research, where the added predictive value of an individual test is assessed within the framework of the diagnostic workup. This approach seeks to account for the patient information immediately available to the medical practitioner, as well as information obtained through previous diagnostic testing. In a previous paper, I provided a proof-of-concept regarding what this might look like in practice, using simulated tests based on fixed sensitivities and specificities. In this project, I expand upon this work, using real-world malaria data from Ghana in evaluating two candidate rapid diagnostic tests for the disease. Additionally, I incorporate Bayesian variable selection into the procedure for choosing a model as an alternative form of assessment for the added predictive value of these tests.