Spatial Autocorrelation Analysis of Infectious Disease Incidence Rates at State and District Level Using Supra-Adjacency Weights Matrix

The spatiotemporal correlation in disease incidence rates resulting from the spatial arrangement of neighboring geographical units is often conceptualized through constructing contiguity-based spatial weights. However, these weights specifications are not meant for capturing the spatial relationships across multiple spatial scales and disjoint spatial units. Modifications to existing spatial weights specifications are highly required. Hence, this study used supra-adjacency matrix in network science to analyze the spatial autocorrelation of COVID-19 incidence rates at Sarawak’s district and Malaysia’s state levels. Flight routes between these regions were embedded as spatial interaction submatrix to represent their inter-layer adjacency. Segmentation of data based on respective Sarawak’s and Malaysia’s daily cases was conducted to investigate the consistency in the spatial autocorrelation and the type of local clustering. When global spatial autocorrelations at state level were high, both the Sarawak districts’ and Malaysia states’ incidence rates became more spatially related with the inclusion of spatial interaction. Several districts, including Sibu, in Sarawak were now classified as high-high cluster with supra-adjacency weights. These high-high clusters can only be discovered with second-order contiguity weights in previous literature. The numbers of significant spatial clusters and outliers in district level were substantially greater than its state-level counterpart. This research provides evidence on how spatial dependencies of disease incidence rates between two spatial aggregation levels and geographically disjoint regions can be quantified using supra-adjacency matrix for disease surveillance. Capturing inter-layer spatial dependencies allows for more targeted interventions such as optimizing vaccine distribution and planning mobility restrictions during pandemics