Influences of using different spatial weight matrices in analyzing spatial autocorrelation of cardiovascular diseases mortality in China

Objective: To explore the potential influences and applicability of different spatial weight matrices used in analyzing spatial autocorrelation of cardiovascular disease (CVD) mortality in China. Methods: Using data from the National Cause-of-death Reporting System, we used adjacency-based Rook and Queen contiguity and distance-based K nearest neighbors/distance threshold. We then conducted global and local spatial autocorrelation analysis of CVD mortality at the county level in China, 2018. Results: All four categories and 26 types of spatial weight matrices had detected significant global and local spatial autocorrelation of CVD mortality in China. Global Moran's I statistics reached its peak when using first-order Rook (0.406), first-order Queen (0.406), K nearest neighbors including five spatial units (0.409), and distance threshold with 100 kilometers (0.358). Meanwhile, apparent local spatial autocorrelation was found in CVD mortality. Substantial disparities were observed when detecting "High-High clusters", "Low-Low clusters", "High-Low clusters" and "Low-High clusters" of CVD mortality spatial distribution by using different weight matrices. Conclusions: Using different spatial weight matrices in analyzing the spatial autocorrelation of CVD mortality, we could understand the spatial distribution characteristics of CVD mortality in-depth at the county level in China. In this way, adequate supports could also be provided on CVD premature death control and rational medical resource allocation regionally.