| Summary: | Infectious diseases are a major threat to the wellbeing of humans, livestock, and wildlife. However, there is often a paucity of information for responding to these threats, and thus a need to make efficient use of existing data. This thesis shows how to use Bayesian analysis to maximise the information gained from already collected diagnostic test data. First, the commonly used latent class analysis of multiple binary diagnostic tests is ex- tended to account for vaccinated individuals, and used to estimate the effect of study size on sensitivity and specificity estimates of DIVA (”Distinguishing Infected and Vaccinated Animals”) tests for bovine Tuberculosis. It is then shown how quantitative test responses can be used as clocks indicating the time since infection to “hindcast” historic trends of disease incidence using cross-sectional data. This is used to determine whether an endemic disease is increasing or decreasing up to the time of sampling, enabling the tracking of trends in populations where routine surveillance data is not available. It is further demonstrated how to hindcast the rise and fall of disease outbreaks. Using the 2007 UK Bluetongue virus outbreak and a whooping cough outbreak as examples, it is shown that hindcasting can be used to determine whether an outbreak is increasing or past its peak at the time of sampling, thus informing potential outbreak responses. In the light of these methods for analysing quantitative test data, the challenges of generating data on test kinetics are discussed. Suggestions are given for how to improve on current methods by modelling the development of paired diagnostic tests as a dynamic host-pathogen system. This thesis demonstrates that multiple quantitative tests can be used to recover disease trends in a population. These methods have far-reaching consequences for the design and practice of disease surveillance in all contexts.
|
|---|
