统计与数据科学学院知识库

  传染病不仅影响人类健康，而且对地区经济发展产生约束。国家疾病预防控制局报告显示，乙肝发病数在传染病例总数中占有很高比重。由于当前发病人数较多，能接受规范性抗病毒治疗的比例较低，防治乙肝任重道远，了解我国乙肝的流行特征并进行合理预测是防治乙肝的关键。

  论文以我国乙肝发病率为研究对象，选取2004-2017年乙肝发病率的时空多维度数据，就乙肝发病率基本特征及预测方法进行挖掘与探讨。首先利用传统流行病学中的“三间分布”对不同维度的乙肝发病情况进行描述性分析，探讨乙肝的分布差异；其次结合空间统计方法和GIS软件挖掘其空间特征，一方面，从可视化角度制作我国乙肝疾病专题地图，以直观展示乙肝发病率的空间分布特征。另一方面，借助Moran’s I指数和LISA集聚图，进一步捕捉其空间集聚特征；最后探讨乙肝发病率的科学预测方法，通过比较单一模型和组合模型，最终选取合理模型对其未来趋势进行预测。

  根据上述分析得出以下结果：                

  （1）乙肝在不同时间、不同地区和不同人群间的分布都存在明显差异。时间分布中，不同年份的发病率波动范围在68.57/10万到89.00/10万之间，2009年发病数最多达到1179607例，3月份是乙肝的高发病期，12月份乙肝发病数最少。地区分布中，不同地区的乙肝年发病率在7.79/10万-251.05/10万之间，2004-2017年广东省发病数最多，西藏发病数最少。性别层面，男性发病率明显高于女性发病率。年龄层面，0-19岁间的低年龄段发病数逐渐减少，40岁以上人群的发病数逐渐增多。职业层面，农民乙肝发病数最高，其次是家政、家务及待业和工人。

  （2）空间相关性分析结果表明，2004-2017年各省市区乙肝发病率存在较强的空间正相关性。发病率具有明显的空间聚集效应，主要集聚特征为高-高发病率地区相邻、低-低发病率地区相邻，其中高发病率地区主要位于西部的青海、新疆和南部的广东，低发病率地区主要包含上海和江苏。

  （3）使用三种模型预测乙肝月发病率，结果显示ARIMA-BP神经网络组合模型预测效果最优，该模型各项评价指标得出的误差都较小，模型拟合精度高，最终使用该组合模型预测乙肝发病率的未来趋势。

Infectious diseases affect human health, it can also hinder regional economic development. According to the report of Chinese Centre for Disease Control and prevention, the number of cases of Hepatitis-B accounted for a high proportion of the total number of infectious cases. Because of the large number of patients, the proportion of patients who can receive standardized antiviral treatment is low. It is an urgent issue to prevent, diagnosis and treatment of Hepatitis-B. An important issue for prevention and treatment of Hepatitis-B is to understand the epidemic characteristics and introduce a reasonable prediction.

Taking the incidence of Hepatitis-B in China as the research object, this paper selected the spatial and temporal multi-dimensional data of the incidence of Hepatitis-B from 2004 to 2017 to dig and explore the basic characteristics and prediction methods of the incidence of Hepatitis-B. Firstly, a descriptive analysis of the incidence of Hepatitis-B in different dimensions was carried out by using the "Distribution of three" in traditional epidemiology to explore the distribution differences of Hepatitis-B. Secondly, spatial statistical methods and GIS software are combined to mine its spatial characteristics. On the one hand, a thematic map of Hepatitis-B disease in China is made from the perspective of visualization to visually display the spatial distribution characteristics of the incidence of Hepatitis-B. On the other hand, Moran's index and LISA cluster map were used to further capture the spatial clustering characteristics. Finally, the scientific prediction method of the incidence of Hepatitis-B was discussed. By comparing single model and combination model, a reasonable model was selected to predict the future trend of Hepatitis-B.

The following results are obtained.

(1) There are significant differences in the distribution of Hepatitis-B at different times, different population and different regions. In the time distribution, the incidence of Hepatitis-B in different years fluctuated from 68.57/100,000 to 89.00/100,000. In 2009, the number of cases reached the maximum of 1179,607 cases. March was the high incidence period of Hepatitis-B, and December was the lowest. In terms of regional distribution, the annual incidence of Hepatitis-B in different regions ranged from 7.79/100,000 to 251.05/100,000. From 2004 to 2017, Guangdong Province had the highest incidence, while Xizang Province had the lowest incidence. The incidence rate of male is significantly higher than that of female. The incidence of low age among 0-19 is gradually decreasing, and the incidence rate of people aged 40 or above is increasing. The highest incidence was found in farmers, followed by housekeeping, housework, unemployed and workers.

(2) Spatial correlation analysis showed that there was a strong spatial positive correlation between the incidence rates of Hepatitis-B in each province from 2004 to 2017. The incidence rate is obviously spatial aggregation effect, and the main clustering characteristics are high- high incidence rate adjacent area and low-low incidence area. High incidence rate is mainly located in Qinghai, Xinjiang and Guangdong in the west, and the low incidence rate mainly includes Shanghai and Jiangsu.

(3) Fitting the prediction of the incidence rate of Hepatitis-B through three models, the results show that the ARIMA-BPNN combined model has the best prediction effect, and the error of each evaluation index of this model is small. The model has high precision and can be used for the prediction of Hepatitis-B incidence rate.