Hierarchical Bayesian space-time small area mapping of COVID-19 risk: The case of Netherlands
Students should be knowledgeable in the R software.
The current trend of COVID-19 epidemic threatens to decimate populations even in advance countries. The irony is that advance countries rather suffer the burden perhaps due to their complacency in implementing optimal disease surveillance techniques. Cases in the Netherlands generally keep rising within the different epidemic waves. Currently, vaccination has become the agreed effective alternative to reduce the burden of COVID-19. Mapping the risk can inform optimal vaccination process has. Most of the existing maps only show the spatial distribution of raw/crude rates (Figure 1). Maps of the raw rates have biases due to population variability. This is known as variance instability and can spuriously influence interpretation and impact.
The objective of this topic is to develop space-time risk maps that takes into account population variabilities, demographic, environmental and socioeconomic factors, spatial, temporal, and space-time variations. This will require the development of hierarchical Bayesian space-time models. The student is expected to develop a hierarchical based conditional autoregressive (CAR) model for mapping COVID-risk. Here, the risk is assumed to unknown parameter in a base/likelihood distribution (either Poisson, negative Binomial or generalized Poisson). The model should include deterministic components (effects of risk factors like demographic or environmental factors) purely temporal trend, spatial variation, and space-time variations. There are different structures of the CAR models. These should be compared be compared under the different likelihood distributions. There are also different space-time structures which ought to be developed. Knowledge of Bayesian statistics is a prerequisite here.