Analysing Multiple Epidemic Data Sources

Abstract: Evidence-based knowledge of infectious disease burden, including prevalence, incidence, severity and transmission, in different population strata and locations, and possibly in real time, is crucial to the planning and evaluation of public health policies. Direct observation of a disease process is rarely possible. However, latent characteristics of an epidemic and its evolution can often be inferred from the synthesis of indirect information from various routine data sources, as well as expert opinion. The simultaneous synthesis of multiple data sources, often conveniently carried out in a Bayesian framework, poses a number of statistical and computational challenges: the heterogeneity in type, relevance and granularity of the data, together with selection and informative observation biases, lead to complex probabilistic models that are difficult to build and fit, and challenging to criticize. Using motivating case studies of influenza, this chapter illustrates the cycle of model development and criticism in the context of Bayesian evidence synthesis, highlighting the challenges of complex model building, computationally efficient inference, and conflicting evidence.