A novel predictor for micro‐scale covid‐19 risk modeling: An empirical study from a spatiotemporal perspective

Abstract

Risk assessments for COVID‐19 are the basis for formulating prevention and control strat-egies, especially at the micro scale. In a previous risk assessment model, various “densities” were regarded as the decisive driving factors of COVID‐19 in the spatial dimension (population density, facility density, trajectory density, etc.). However, this conclusion ignored the fact that the “densi-ties” were actually an abstract reflection of the “contact” frequency, which is a more essential deter-minant of epidemic transmission and lacked any means of corresponding quantitative correction. In this study, based on the facility density (FD), which has often been used in traditional research, a novel micro‐scale COVID‐19 risk predictor, facility attractiveness (FA, which has a better ability to reflect “contact” frequency), was proposed for improving the gravity model in combination with the differences in regional population density and mobility levels of an age‐hierarchical population. An empirical analysis based on spatiotemporal modeling was carried out using geographically and temporally weighted regression (GTWR) in the Qingdao metropolitan area during the first wave of the pandemic. The spatiotemporally nonstationary relationships between facility density (attrac-tiveness) and micro‐risk of COVID‐19 were revealed in the modeling results. The new predictors showed that residential areas and health‐care facilities had more reasonable impacts than traditional “densities”. Compared with the model constructed using FDs (0.5159), the global prediction ability (adjusted R2) of the FA model (0.5694) was increased by 10.4%. The improvement in the local‐scale prediction ability was more significant, especially in high‐risk areas (rate: 107.2%) and densely pop-ulated areas (rate in Shinan District: 64.4%; rate in Shibei District: 57.8%) during the outset period. It was proven that the optimized predictors were more suitable for use in spatiotemporal infection risk modeling in the initial stage of regional epidemics than traditional predictors. These findings can provide methodological references and model‐optimized ideas for future micro‐scale spatio-temporal infection modeling.