Title: Spatio-temporal cluster detection and disease risk estimation using clustering-based adjacency modelling
Authors: Xueqing Yin - University of Glasgow (United Kingdom) [presenting]
Gary Napier - University of Glasgow (United Kingdom)
Craig Anderson - University of Glasgow (United Kingdom)
Duncan Lee - University of Glasgow (United Kingdom)
Abstract: Globally spatially smooth conditional autoregressive (CAR) models are typically used to capture the spatial autocorrelation in areal unit disease count data when estimating the spatio-temporal trends in disease risk. In these models, the spatial autocorrelation structure is typically induced by a binary neighbourhood matrix based on a border sharing specification, such that spatial correlation is always enforced between geographically neighbouring areas. However, enforcing such correlation in the model will mask any discontinuities in the disease risk surface, thus impeding the detection of clusters of areas that exhibit higher or lower risks than their neighbours. Therefore, we propose a novel methodology to account for these discontinuities via a two-stage modelling approach, which either forces the spatial clusters to be the same for all time periods or allows them to evolve dynamically over time. Stage one produces a set of candidate neighbourhood matrices using a variety of common clustering methods that allow for these risk surface discontinuities. Then in stage two, an appropriate spatial autocorrelation structure(s) is selected by estimating the neighbourhood matrix from the candidate set as part of a hierarchical Bayesian spatio-temporal model. The novel methodology is applied to the motivating study of respiratory disease risk in Greater Glasgow, Scotland, from 2011 to 2017.