【作者】 王永；

【导师】 沈毅；

【作者基本信息】 浙江大学 ， 流行病与卫生统计学， 2008， 硕士

【摘要】 肺结核是由肺结核杆菌造成的一种传染性疾病,通过空气传播使人感染。1990年,世界卫生组织报告全球疾病负担排名时,肺结核位于第七位,预计到2020年,肺结核的全球疾病负担排名仍保持在第七位。在2001年,WHO估算,18.6亿的人口,即全球人口的32%,都感染了肺结核。每年全球874万的人口发生肺结核,将近200万人死亡。如果不能得到合适的治疗,肺结核病人一年能传染10-15的人。肺结核比其他传染病更容易导致成人死亡,每年大概会导致40万人死亡。它主要影响15-54周岁的人群的生产能力,如果一个成年人患了肺结核,每年平均会丧失3-4个月的劳动时间,相当于损失了每年20%-30%的家庭收入。由此可见,这对于一个国家或者地区会是巨大的社会和经济负担,因此探讨肺结核等传染病的时空分布及其防治研究对于研究者是一个严峻的挑战。众所周知,医学是微观和宏观系统的综合性学科,大量的数据具有空间分布的属性。而流行病学主要是研究疾病在空间、时间及人群中的分布,大约80%的流行病学资料具有空间属性,如人群或动物的患病总发生在一定的空间位置,而处在某一空间位置的地理环境或社会因素又影响着疾病的发生,因此只有准确分析疾病的空间分布特征,才能有效地研究疾病病因及其影响因素,进而确定高危人群和高发地区,制定预防措施。另外,传染病的发生与流行,地方病的分布,许多疾病的地方性高发性特点以及医药卫生机构的分布等都与空间信息密切相关。医学数据资料的这种空间相关特点成为空间统计学应用的前提。而总结前人的研究成果,也不难发现传染病的研究较多地运用传统的数量统计模型,在空间关系和数值的相关分析上顾此失彼。新兴的空间统计为在分析空间关系的基础上进行数值的相关分析提供了新的思路,其核心就是认识与地理位置相关的数据间的空间依赖。比如,针对肺结核等传染病的研究,多限于单一的对疾病发生率的描述,忽略了肺结核发病存在的地理相关性,不能从定量的水平上对疾病的空间分布做一定的探讨。而肺结核作为一种传染病,与当地的环境、人口、气候等都存在一定的联系,并且由于其传染性和普遍性,定性的研究不足以满足现有疾病控制工作的需要,因此,运用空间统计学的分析方法,从定量水平上探讨肺结核发病的热点区域,以及空间、时间和时空聚集性是非常必要的。研究目的本研究主要运用空间统计学中的空间自相关方法和时空扫描统计量方法,获得肺结核传染病发病的时空分布情况,旨在发现新发的传染病病例何时何地高发,即某地理区域内肺结核的发病的时空热点,从而探讨该地区肺结核发病的聚集性问题,为该地区公共卫生政策的制定和肺结核的防控提供理论依据,也可为同类研究提供参考。实例分析实例分析采用宁波市疾病控制中心提供的1997-2003年肺结核的汇编资料,对肺结核的发病数据进行整理,并获取1:25万的宁波地区矢量地图,在Excel中输入该地区十一个辖区的地区代码、发病数据、人口基本情况,每年的肺结核发病率数据,并把Excel格式的文件转变成可共使用SHP格式的地理属性文件。空间自相关分析方法采用的全局型和局域性Moran’s I、Moran’s I散点图、蒙特卡罗方法、空间经验贝叶斯光滑法均油空间统计软件包GeoDa095i实现;时空扫描统计方法分析由空间统计分析软件包时空分析软件SaTScan7.0来实现;空间聚类图,时空扫描窗口等疾病专题地图均由GIS公司的产品ArcView GIS软件来实现。主要结果一、Moran’s I结果运用空间统计分析软件包得出1997-2003年七年的全局型空间自相关系数,分别为0.1265、0.2628、0.2585、0.1386、0.1285、0.0012、0.0845,且均经过统计学检验,1997、1998、1999、2000、2001五年的Moran’s I指数均具有统计学意义,而2002和2003年均不具有统计学意义。局域型Moran’s I即LISA分析,1997年宁波地区象山县的肺结核发病处于高-高区域:1998年象山和宁海均处于高高区域,而慈溪处于低低区域;1999年象山、宁海处在高高区域,而慈溪、江北处于低低区域:2000年肺结核的发病分布的空间聚类结果,象山在高高区域,而慈溪,江北在低低区域;2001年象山在高高区域,而江北在低低区域;2002年象山在低-高区域,而江北在低低区域;2003年宁海在低.高区域。且以上阐述的高高、高低、低高、低低区域均代表经过统计学检验(P＜0.05),具有统计学意义的结果。二、空间扫描统计量方法(SaTScan)分析单纯空间扫描统计量分析结果1997高发病聚集区为象山所属人口区域(RR=1.620,P=0.01＜0.05),年平均发病率为45.5/10万,次级发病聚集区为奉化和鄞州(RR=1.342,P=0.01＜0.05),年平均发病率为37.2/10万,1998年高发病聚集区为奉化和鄞州所属人口区域(RR=1.493,P=0.001＜0.05),年平均发病率为35.7/10万,次级发病聚集区为象山和宁海(RR=1.324,P=0.001＜0.05),年平均发病率为32.9/10万,三级发病聚集区为北仑和镇海(RR=1.297,P=0.001＜0.05),年平均发病率为33.4/10万;1999年高发病聚集区为奉化和鄞州所属人口区域(RR=1.831,P=0.001＜0.05),年平均发病率为44.8/10万,次级发病聚集区为象山(RR=1.484,P=0.001＜0.05),年平均发病率为41.2/10万;2000年高发病聚集区为奉化和鄞州所属人口区域(RR=1.364,P=0.001＜0.05),年平均发病率为39.0/10万,次级发病聚集区为象山(RR=1.311,P=0.006＜0.05),年平均发病率为39.4/10万;2001年高发病聚集区为奉化和鄞州所属人口区域(RR=1.514,P=0.001＜0.05),年平均发病率为40.6/10万,次级发病聚集区为象山,宁海(RR=1.388,P=0.001＜0.05),年平均发病率为38.5/10万;2002年高发病聚集区为奉化和鄞州所属人口区域(RR=1.430,P=0.001＜0.05),年平均发病率为43.5/10万。2003年,高发病率聚集区域集中在象山、奉化和宁海(RR=1.670,P=0.001＜0.05),年平均发病率为54.7/10万,次级聚集区域为江北(RR=1.995,P=0.001＜0.05),年平均发病率为75.0/10万。时空分析的结果,为2001-2003年的肺结核发病的空间聚集区域主要为宁海和奉化(RR=1.477,P=0.001＜0.05),年平均发病率为46.3/10万;次级时空聚集区域为2003年的江北(RR=2.415,P=0.001＜0.05);三级聚集区域为1999-2001年的鄞州(RR=1.9364,P=0.001＜0.05),年平均发病率为41.8/10万。结论1.Moran散点图具有较强的局域型相关分析功能,即对应于Moran散点图的不同象限,可识别出空间分布中存在着哪几种不同的实体,一定程度上也表达了局部区域之间的一些联系。结合1997年的Moran’s I散点图,可初步判断,象山在第一象限,余姚,奉化,鄞州,镇海在第二象限,慈溪、海曙,江东,江北在第三象限,而宁海,北仑在第四象限。但是,散点图不提供各区域点的具体的Moran’s I值,所以需要借助于局域型自相关系数及其检验。2.1997-2001年LISA(Local Indication of Spatial Autocorrelation)的综合分析,因此一定程度上可以推断,宁波的两个县级辖区东南部的象山和南部的宁海,为发病的热点区域。另外由于肺结核的传染病特点,其发病具有相应的扩散效应,宁波地区肺结核的发病可能存在着由北向南逐渐增高的趋势3.单纯空间扫描统计量分析显示,1997年集中在象山、1998-2002年高发病区域均集中在奉化和鄞州,2003年象山、宁海、奉化为高发区域,从纯空间分析角度来比较,与空间自相关的结果显然有所出入,其不同是由于不同方法基于的不同标准和指标不同所造成的。但是1998-2001年的纯空间扫描统计分析,奉化和鄞州为空间聚集区域,象山、宁海为次级高发聚集区域,和1998-2001年空间自相关分析时宁海和奉化处于高高值区域是一致的,仍能看出宁波南部和中部区域为发病的热点区域。根据对1997-2003年的时空扫描统计量结果显示2001-2003年奉化、宁海为高发病聚集区域,而2003年的江北为次级发病聚集区域;1999-2001年鄞州为三级发病聚集区域,从而探讨了宁波地区肺结核发病的时空聚集性。4.空间自相关Moran’s I法相比,时空扫描统计量方法借助于似然比的方法,用RR值判断窗口内外发病风险的大小,进行时空联合的交互分析,显然时空扫描统计量更适合对肺结核发病时空聚集性的分析。而空间自相关方法采用关联系数,运用空间经验贝叶斯光滑处理后的发病率,Moran’s I指数偏大,也容易受限于空间尺度大小的问题,显然更适合单纯空间模式的探讨。5.1997-2001年五年肺结核发病的全局型自相关,说明了该地区肺结核发病存在一定的聚集性,但并不能准确判断存在的聚集区域是低发区域,或是高发区域,还是两者兼有。2002年-2003年宁波地区肺结核的发病,在全局型Moran’s I检验后,提示不具有统计学差异,因此这两年不存在全局型的空间自相关。对2002年-2003年的两年的空间自相关分析,却提示不存在空间聚集性,这样的结果可能导致忽略宁波大部分辖区发病率水平在整体上处于较高水平的事实,而象山和宁海随时间变化的幅度小于同期的其他辖区,因此出现了不存在肺结核发病热点的结果。而且各辖区的发病率之间差距不大,该区域内的发病率的高低均是相对的,所以容易形成假阴性的结果,这可能是导致不存在全局型空间自相关的原因。更多还原

【Abstract】 Tuberculosis (TB) is an infectious disease caused by the bacillus Mycobacterium tuberculosis and spreads through air by a person suffering from TB. The 1990 World Health Organization (WHO) report on the Global Burden of Disease ranked TB as the seventh most morbidity-causing disease in the world, and expected it to continue in the same position up to 2020. In 2001, the WHO estimated that 1.86 billion persons (equal to 32% the world’s total population ) were infected with TB. 8.74 million in the world develop TB and nearly 2 million of them die every year. Unless properly treated, an infectious pulmonary TB patient can infect on average between 10 and 15 people in a year. TB kills more adults than any other infectious disease worldwide, accounting for almost 400,000 deaths annualy. It mainly afflicts people who are in the economically productive years of their life (equal to 15-54 years). An adult suffering from TB loses on average between three to four months of working time. This equals to a loss of 20%-30% of household’s annual income. This shows that it will cause large social and economic burden on a country. So TB is the biggest public heath challenge for experts and researchers.As we all know, medicine is a system of microcosmic and macroscopical comprehensive disciplines, a large number of data has characteristics of spatial distribution. Topic of Epidemiology is the distribution of space, time and population about the diseases, about 80 percent of the epidemiological data has space attributes, such as people or animals are always in the prevalence of certain space location, but the space geographical or social factors of a certain location also affect the incidence of the disease, so only the precise analysis of the spatial distribution of the disease can effectively study the cause of the disease, then we can identify the high-risk population and areas and develope measures of prevention. For instance, the incidence and prevalence of infectious diseases and endemic diseases of the distribution, many local high incidence of the disease , the distribution of medicine and public health agencies are all closely related to spatial information. Space-related characteristics of medical data is the premise of the space application of spatial statistics.According to the results of the previous research, the study of infectious diseases we can easily find always use the method of traditional statistical model, failed in the anlalysis of the correlation of space and numerical analysis. The emerging spatial statistics supports a new way for the spatial autocorrelation on the basis of spatial analysis, its core is the understanding and location-related data between the space-dependent. For example, tuberculosis and other infectious diseases researches are more limited to simple description of incidence of the disease, ignoring the geographic correlation, not doing a study on the basis of the quantitative level of the spatial distribution of diseases. And tuberculosis as an infectious disease are related to the local environment, population, climate and so on, because of its infectivity and universality, qualitative study is insufficient to meet the needs of prevention and control of diseases. Therefore, we explore the hot spot of tuberculosis incidence from the quantitative level with the use of spatial statistics of the analysis, as well as space, time and space-time cluster is very necessaryObjectivesThe main objectives of this research: exploring the spatial distribution and spatio-temporal distribution of TB incidence with spatial autocorralation and spatial scan statistics method; exploring where and when the new cases happen, that is to say ,exploring the spatio-temporal hot spot to know the spatio-temporal cluster of TB, aim to supply with the theroy evidence for the policy of disease prevention and control and the reference for the same study.Case analysisWe accumulate the TB incidence data about 1997-2003 year from Ningbo Center of Disease prvention and Control center (CDC), and get the Ningbo Map of scale with 1:250000, input the region code, incidence data, population and TB incidence data each year of the 11 regions in Ningbo; translate the excel data into SHP format data.Moran’s I, Moran’s I scatter plot, Monter-Carlo and spatial empirical Bayes smooth method of spatial autocorrelation can be realized with software GeoDa0985i; spatial and spatio-temporal scan statistics can be realized with software SaTScan7.0; spatial cluster sketch map, spatio-temporal scan windows can be realized using Arc View GIS.Results1. Moran’s I methodWe get the global spatial autocorrelation indication from 1997 to 2003 year: 0.1265, 0.2628, 0.2585, 0.1386, 0.1285, 0.0012, 0.0845, the Moran’s I of 1997,1998, 1999, 2000, 2001 year have the significance, but 2002,2003 year’s not, with statistical test.According to LISA analysis, TB incidence of Xiangshan locates in high-high region in 1997; Xiangshan and Ninghai locate in high-high region in 1998,but Cixi locates in low-low region; Xiangshan and Ninghai locate in high-high region in 1999, but Jiangbei and Cixi locates in low-low region; Xiangshan locates in high-high region in 2000,but Jiangbei and Cixi locate in low-low region; Xiangshan locates in high-high region, but Jiangbei in low-low region in 2001; Xiangshan locates in low-high region, but Jiangbei in low-low region in 2002; Ninghai locates in low-high region, but Jiangbei in low-low region in 2003; the outcomes have the significance ( P<0.05 ) with statistical test.2. Spatial Scan Statistics(SaTScan)The outcome of purely spatial scan statistics is including that most likely cluster includes Xiangshan (RR=1.620, P=0.01<0.05 ), the annual cases/100,000 is 45.5/100,000, secondary cluster includes Fenghua and Yinzhou (RR=1342, P=0.01<0.05 ), the annual cases/100,000 is 37.2/100,000 in 1997; most likely cluster includes Fenghua and Yinzhou (RR=1.493, P=0.001<0.05 ), the annual cases/100,000 is 35.7/100,000, secondary cluster includes Xiangshan and Ninghai (RR=1.342, P=0.01<0.05 ), the annual cases/100,000 is 37.2/100,000 in 1998; most likely cluster includes Fenghua and Yinzhou (RR=1.831, P=0.001<0.05 ), the annual cases/100,000 is 44.8/100,000, secondary cluster includes Xiangshan and Ninghai (RR=1.484, P=0.001<0.05 ), the annual cases/100,000 is 41.2/100,000 in 1999; most likely cluster includes Fenghua and Yinzhou (RR=1.364, P=0.001<0.05 ), the annual cases/100,000 is 39.0/100,000, secondary cluster includes Xiangshan and Ninghai (RR=1.311, P=0.006<0.05 ), the annual cases/100,000 is 39.4/100,000 in 2000; most likely cluster includes Fenghua and Yinzhou (RR=1.514, P=0.001<0.05 ), the annual cases/100,000 is 40.6/100,000, secondary cluster includes Xiangshan and Ninghai (RR=1.388, P=0.001<0.05 ), the annual cases/100,000 is 38.5/100,000 in 2001; most likely cluster includes Fenghua and Yinzhou (RR=1.430, P=0.001<0.05), the annual cases/100,000 is 43.5/100,000 in 2002; most likely cluster includes Fenghua and Yinzhou (RR=1.670, P=0.001<0.05 ), the annual cases/100,000 is 54.7/100,000, secondary cluster includes Xiangshan and Ninghai (RR=1.995, P=0.001<0.05 ), the annual cases/100,000 is 75.0/100,000 in 2003.The outcome of spatio-temporal scan statistics is that the time of the most likely cluster is from 2001 to 2003, and at the same time ,the most likely cluster includes Ninghai and Fenghua (RR=1.477, P=0.001<0.05 ), the annual cases/100,000 is 46.3/100,000, secondary likely cluster time is from 1999 to 2001, and the secondary cluster is Yinzhou (RR=1.9364, P=0.001<0.05 ), the annual cases/100,000 is 41.8/100,000.Conclusions1. Moran scatter plot has local autocorrelation function, that is to say, the quadrant of Moran scatter plot can indicate different cluster, and evaluate the relation of local areas. According to the Moran scatter in 1997, we can decide Xiangshan is in the first quadrant, Yuyao, Fenghua, Yinzhou and Zhenhai in the second quadrant, Cixi, Haishu, Jiangdong and Jiangbei in the third quadrant, Ninghai and Beilun in the fourth quadrant. But Moran scatter can’t support the Moran’s I of local areas, if we get this Moran’s I, we must compute the LISA by statistical test.2. The LISA of 1997, 1998, 1999, 2000, 2001 indicate that Xiangshan in the south east and Ninghai in the south are hot spots of TB incidence in Ningbo. TB disease has diffusibility for the infectivity, and it maybe exsits the uprising trend from south to north..3. The result of Purely spatial scan statistics indicates that the most likely cluster is Xiangshan in 1997; Fenghua and Yinzhou are the most likely cluster; Xiangshan, Ninghai and Fenghua are the most likely clusters in 2003. This result is not consistant with the spatial autocorrelation, this difference maybe is based on different standard and different index. But the purely spatial scan statistics result is consistant with the result that Ninghai and Fenghua locates in high-high area and the hot spot in the south and middle of Ningbo regions.According to the spatio-temporal scan statistics, the most likely spatio-time clusters are Fenghua and Ninghai between 2001and 2003; Jiangbei is the secondary cluster; Yinzhou and Xiangshan between 1999 and 2001 are the tertiary clusters. So this method can evaluate the spatio-temporal cluster of TB incidence in Ningbo regions.4. Comparing to the Moran’s I method of Spatial autocorrelation, spatial scan statistics estimates the risk between the inside and outside of circle window with likelyhood ratio method and spatio-temporal mutual interation, so this method is suitable for evaluate the spatio-temporal cluster of TB incidence in Ningbo. Moran’s I tends to be consistently larger with Spatial empirical Bayes smooth method and often be limited for small spatial scale, so Moran’s I method is suitable for exploring the spatial model.5. The global spatial autocorrelation of the five years from 1997 to 2001 indicates the spatial cluster in Ningbo region, but we don’t decide whether this cluster is the area of high accurance; The statistical test of Moran’s I of 2002 and 2003 year indicates that there is not global spatial autocorrelation.更多还原