Abstract
ObjectiveHFMD incidence varies between geographic regions at thetownship in Beijing. The objective of this study was to examinespatial heterogeneity for the association between HFMD incidenceand demographic and socioeconomic factors.IntroductionHand-foot-mouth disease (HFMD) is a common childhood illnessand the drivers of HFMD incidence are still not clear [1]. In mainlandChina, continuing and increasing HFMD epidemics have beenrecorded since 2008, causing millions of infections and hundreds ofdeaths annually. In Beijing, 28,667 cases were reported in 2015 andthe incidence was 133.28/100,000. The variations in Beijing HFMDepidemics over population, space, and time that have been revealed[2] emphasize the need for further research about risk factors ofHFMD occurrence. This study aims to explore local effects on HFMDincidence led by potential factors.MethodsHFMD Data. Beijing HFMD data during 2008–2012 period wereprovided by the Beijing Center for Disease Prevention and Control.HFMD incidence adopted in this study was the annual average valueduring the five years.Predictor variables. Potential risk factors obtained from the caserecords (demographic, occupation, health-seeking behavior) andspatial POIs (points of interest) consisted of 22 variables involvingresidence, restaurant, education, medical facilities, business facilities,infrastructure. The scale of different kinds of POIs (1/100,000) wasnoted by calculating the ratio of the number of POIs to the populationat certain township or street committee.Model Specification. Some initial associations between HFMDincidence and 8 predictor variables (population density, shoppingmall, supermarket, pharmacy, kindergarten, middle school, parkinglot, health seeking behavior) were revealed using Pearson correlationanalysis and the exploratory regression. An ordinary least squares(OLS) model was fitted to diagnose the residual normality anddependence. Geographically weighted regression (GWR) was chosento model the relationship, compare the difference from OLS regressionand measure how much improvement the local model gained.ResultsGWR model with residual independence (Moran’s I = 0.0214,p = 0.3405) and lower AICc, performing much better than OLSmodel with residual dependence (Moran’s I=0.1271, p = 0.0000)and higher AICc. Prediction accuracy by GWR (local R2rangingfrom 0.42 to 0.90, R2=0.88) was higher than that by OLS (R2=0.57).The higher local R2values clustered in the east of Fangshan andUrban-Rural Transition Area. Higher coefficient for intercept mainlyoccurred in north-western and south-eastern portion of Beijing.The coefficients for predictors showed shifting patterns from positiveto negative at different township. The local effects led by supermarketand shopping mall showed similar spatial pattern, as well as thoseled by kindergarten and middle school. The scale of pharmacy waspositively related to HFMD incidence in the west of Daxing and thejunction part of Chaoyang and Tongzhou.ConclusionsThis study quantitatively assessed local risk factors of BeijingHFMD occurred in China using GWR model which outperformedOLS regression. The findings could provide valuable information foradequate disease intervention measures and regional policy.