The temporal and spatial modeling of children's lead poisoning in Syracuse, New York.

摘要

This dissertation investigated the temporal trends and spatial patterns of children's lead poisoning in the early 1990s through the early 2010s in the city of Syracuse, NY. During the time period the continuous efforts on lead exposure reduction shifted from lead sources control to residential areas lead mitigation. This research used the 20-years surveillance data of children's blood lead levels (BLLs), and focused on selecting effective statistical modeling techniques to detect the temporal changes and spatial patterns. These methods included interrupted time series analysis, generalized linear mixed models, and geographically weighted generalized linear regression. The results of time series analysis showed that the children's BLLs reduced 50% from 8.77microg/dl to 3.94microg/dl in Syracuse, NY over the past two decades. After a decade of lead hazard control treatment program, the average children's BLLs reduced 2.1microg/dl, and the seasonal variation of the children's BLLs also decreased. Further, this research explored the statistical techniques to model the number of children's lead poisoning cases in each census block across the geographical areas of study and used the building year, town taxable value, and soil lead concentration as the predictors. The modeling results showed that the spatial negative binomial Hurdle model was the optimal model to deal with the overdispersion, excessive zeros, and spatially correlation in the spatial count data. The localized modeling method, geographical weighted logistic model, yielded an 11% improvement on predicting the true positives, and 4% improvement on the overall accuracy of model prediction. The geographical weighted Poisson model showed that the spatially varying relationships outperformed the global relationships because it incorporated the spatial autocorrelation and heterogeneity in the spatial count data.