利用空间聚集的贝叶斯网络评估手足口病发病风险

手足口病是一种常见的传染病,以往的研究表明该疾病与气象、环境和社会经济等因素相关联,其影响关系复杂,而疾病本身体现出较强的区域聚集性,采用普通的线性风险建模方法无法捕捉影响因素的复杂性及空间聚集性。因此,本文以山东省为例,在前人研究的基础上,提出了采用贝叶斯网络综合风险建模方法研究手足口病的发病风险与气象、土地利用、社会经济及空气污染等要素间的关系,并通过引入空间扫描统计聚集结果,将空间聚集引入到贝叶斯网络模型加强其空间推理功能,减少模型的偏差,提高评估的精度。结果表明,本文建立的手足口病空间贝叶斯网络风险模型具有较高的估计效果,引入的空间聚集性较好地融入到贝叶斯概率推理模型中,合理建立预测因子同手足口病发病风险之间的关系。通过对建模结果的解译,分析了手足口病的发病风险影响因素,特别是气候、社会经济及空气污染的影响。本文的空间贝叶斯建模方法及研究结果对手足口病暴发的防控预警具有重要的意义。

A Bayesian Network Method Considering Spatial Cluster to Evaluate Health Risk of Hand,Foot and Mouth Disease

Hand, foot and mouth disease (HFMD) is a common infectious disease. Previous studies showed that multiple factors, such as meteorological, geographical, environmental and socio-economic factors were associated with HFMD. The associations between these risk factors and disease are complex. HFMD incidences present strong spatial clustering and auto-correlation. It is difficult to capture such complex non-linear associations and spatial auto-correlation using ordinary linear regression. Based on the previous studies, we proposed a Bayesian network based integrated risk approach to explore the relationship between HFMD incidence risk and the influential factors, such as meteorological parameters, land-use pattern, socio-economic status and air pollution. HFMD is a typical disease of children in Shandong Province of China and it was taken as our study case. Our approach incorporated the output of spatial clusters obtained by scanning statistics to enhance spatial reasoning of the proposed Bayesian network. This could also reduce the bias and improved the performance of the prediction. The results showed that the integrated Bayesian network model proposed achieved higher accuracy than the other methods. Also, spatial hot spots incorporated well in our model. By interpreting the marginal probability of every influential factor in the model, we analyzed the effect of these risk factors, in particular meteorological parameters, socio-economic factors and air pollution on the HFMD incidence. Our spatial Bayesian network approach is useful and the results provided important information for early-warning, prevention and control of HFMD.