A1174
Title: A new similarity-based spatiotemporal model for Covid-19 infection prediction and forecasting
Authors: Helena Baptista - Universidade Nova de Lisboa (Portugal) [presenting]
Jorge M Mendes - NOVA Information Management School, NOVA University Lisbon (Portugal)
Ying C MacNab - University of British Columbia (Canada)
Abstract: A conditionally specified Gaussian random field (CS GRF) model with a similarity-based non-spatial weight matrix to facilitate non-spatial smoothing in Bayesian disease mapping (BDM) has been proposed. The model, named similarity-based GRF, is motivated for modelling disease mapping data in situations where the underlying small area relative risks and the associated determinant factors do not vary systematically in space, and the similarity is defined by similarity with respect to the associated disease determinant factors. The method, designed to handle cases when there is no evidence of positive spatial correlation, was also used when the appropriate mix between local and global smoothing is not constant across the region. It showed again that results consistent with the published knowledge were produced and that accuracy was increased to clearly determine areas of high- or low-risk. Now, the proposed method was used when the underlying small area relative risks vary systematically in space and time. This spatiotemporal approach of the method employs a dynamic CS GRF model with a novel approach to characterize infection risk dependencies through the similarity of areal-level covariates. Furthermore, the method produces the best-balanced forecast. Once again, BDM models gain significantly in explanatory and forecasting power by including extra information, according to the specific knowledge of the epidemiologists, to fit the right suitable model to the problem at hand.
