Disease relative risk downscaling model to localize spatial epidemiologic indicators for mapping hand, foot, and mouth disease over China

Given the limitations of current approaches for disease relative risk mapping, it is necessary to develop a comprehensive mapping method not only to simultaneously downscale various epidemiologic indicators, but also to be suitable for different disease outcomes. We proposed a three-step progressive statistical method, named disease relative risk downscaling (DRRD) model, to localize different spatial epidemiologic relative risk indicators for disease mapping, and applied it to the real world hand, foot, and mouth disease (HFMD) occurrence data over Mainland China. First, to generate a spatially complete crude risk map for disease binary variable, we employed ordinary and spatial logistic regression models under Bayesian hierarchical modeling framework to estimate county-level HFMD occurrence probabilities. Cross-validation showed that spatial logistic regression (average prediction accuracy: 80.68%) outperformed ordinary logistic regression (69.75%), indicating the effectiveness of incorporating spatial autocorrelation effect in modeling. Second, for the sake of designing a suitable spatial case-control study, we took spatial stratified heterogeneity impact expressed as Chinese seven geographical divisions into consideration. Third, for generating different types of disease relative risk maps, we proposed local-scale formulas for calculating three spatial epidemiologic indicators, i.e., spatial odds ratio, spatial risk ratio, and spatial attributable risk. The immediate achievement of this study is constructing a series of national disease relative risk maps for China's county-level HFMD interventions. The new DRRD model provides a more convenient and easily extended way for assessing local-scale relative risks in spatial and environmental epidemiology, as well as broader risk assessment sciences.