【作者】 赵飞；

【导师】 姜庆五；

【作者基本信息】 复旦大学 ， 流行病与卫生统计学， 2011， 博士

【摘要】 我国的血吸虫病防治工作经过60多年的努力取得了举世瞩目的成就,但目前也面临着巨大挑战。血吸虫病人数明显下降,但急性感染者人数增多；血吸虫病治疗药物疗效显著,但高风险地区居民化疗依从性降低；钉螺孳生地广泛存在,但灭螺比例相对较低；加之全球气候变暖、流动人口增多、血吸虫病防治工作投入不足等诸多原因都对我国的血吸虫病防治工作提出了新的难题。湖沼型地区患者多、钉螺孳生地广泛,仍然是今后血吸虫病防控工作的重点。为了更好地为血吸虫病防治工作提供参考与指导,本研究以湖沼型血吸虫病流行区——安徽省为现场,以现有血吸虫病防治工作为基础,在乡镇尺度采用空间信息分析技术和多水平模型、Bayesian时空模型等先进的模型理论,对钉螺分布进行空间分析,以解决血吸虫病防治工作中的关键问题,同时建立的研究方法也为相关疾病的聚集性分析和时空分析提供参考。第一部分钉螺分布和血吸虫病的空间聚集性分析本部分旨在探索安徽省乡镇级别的钉螺和血吸虫病的空间聚集性并加以可视化展示,找出血吸虫病高风险区域的同时,揭示现象之间的时空相互作用机制,为今后的血吸虫病防控提出建议。主要通过回顾性调查方法,收集整理2000-2008年安徽省乡镇血吸虫病患病率和有螺面积百分比,并将该数据与安徽省乡镇级电子地图进行匹配,以构建起本次研究所用的时空分析数据库。通过地统计学中全局空间自相关与局部空间自相关分析方法和空间动态窗口扫描统计方法,探索血吸虫病、钉螺的空间聚集性,并通过ArcGIS软件将结果进行可视化表达。2000-2008年钉螺分布全局空间自相关结果均具有统计学意义(Moran’s I>0,P<1.05),采用局部空间自相关和SaTScan探测的具体聚集位置较为吻合。整体来看有螺面积百分比不随时间变化而发生更为聚集的趋势(t=1.85,P>1.05)。SaTScan探测结果提示该指标聚集区域主要集中分布在沿长江周围62千米的带状区域内。2005年之前,钉螺聚集分布主要集中在长江安徽段的中下游；从2005年开始,钉螺分布的聚集区域转移到了长江上游。对于血吸虫病患病率指标,2000～2001年和2005-2007年间全局空间自相关结果显示其分布无空间聚集效应(P>1.05)。局部空间自相关结果虽有少量高值聚集区域,但是聚集乡镇数量少且位置较为分散,其与SaTScan探测的聚集中心又有所差异,整体上说明其分布呈空间随机分布。2002-2004年和2008年间全局空间自相关结果显示血吸虫病患病率分布呈现空间聚集性(Moran’s I>0,P<1.05),但是聚集程度较弱。局部空间自相关结果与SaTScan探测结果较为一致。SaTScan探测的聚集半径下包括了绝大部分局部空间自相关探测出的高值聚集乡镇,整体上说明这些年份该指标分布呈现空间聚集性。全局空间自相关可以从整体上把握空间数据是否具有空间聚集性；局部空间自相关可以探测出各个单元格之间的关系,从而判断其具体聚集区域；SaTScan可以计算出其具体的中心位置和范围。三种方法联合使用,逐步深入,使得结果呈现更加全面、系统。第二部分钉螺分布的多水平模型构建与分析本部分旨在进一步控制影响钉螺分布的混杂因素,探讨乡镇级别钉螺分布变化的时间趋势,同时为建立Bayesian时空模型筛选出主要的危险因素。研究采用多水平模型中的发展模型排除组内相关对研究结果的影响,对影响钉螺分布的环境因素(归一化植被指数、地表温度)和气候因素(年平均降水量、年极端最低温度)以及灭螺措施、地理位置等相关因素综合分析。空模型结果显示,组内相关系数ICC为84.17%,随机截距的方差具有统计学意义(t=-8.19,P<1.01),因此推断ICC具有统计学意义且乡镇各测量值之间相关性较大,采用多水平模型分析该纵向数据较传统模型更为合适。纳入随时间变化协变量所建立的最终模型最适合拟合该数据。结果显示,随机截距的方差存在统计学意义(σ2u0=33.57,t=5.81,P<1.01),即各乡镇是否具有钉螺的概率初始值不同；时间变化率的方差存在统计学意义(σ2u1=0.74,t=5.87,P<1.01),即各个乡镇有螺概率随时间的变化率存在不同；协方差存在统计学差异(σ2u01=-3.02,t=-4.71,P<1.01),说明研究对象初始有螺概率越高,其有螺概率随时间变化的变化率越小。乡镇离长江的距离对乡镇有螺概率有影响(t=-8.03,P<1.01),有螺概率随乡镇离长江的距离增加而降低(P<1.01)。灭螺情况和地表温度均具有统计学意义(P<1.05),即前一年对有螺乡镇采取灭螺措施后,其有螺概率仍然高于无螺乡镇(95%CI：2.9,6.7)；地表温度高于平均温度27℃的乡镇其有螺风险低于平均温度低于27℃的乡镇(95%CI：0.5,0.9)。第三部分钉螺分布的Bayesian时空模型构建与分析本部分在排除其它危险因素和空间自相关性的基础上,旨在定量刻画乡镇有螺风险的时间变化趋势及空间聚集性。以第二部分研究中筛选出的危险因素(乡镇离长江的距离、灭螺情况和地表温度)的基础上,构建起Bayesian非时空模型、时空独立模型和时空交互模型。根据DIC值越小越好的准则,时空独立模型拟合数据最合适。每年各研究乡镇之间存在空间随机效应,且每年的空间效应一致,不随时间变化而变化。每年乡镇级别的有螺概率呈现逐年递减趋势(OR=0.87,95%CI：0.84,0.90)；地表温度高于27℃的乡镇其有螺概率为低于27℃乡镇的0.7倍(OR=0.73,95%CI：0.59,0.90)；乡镇距长江的距离每增加10千米,其有螺风险降低约20%(OR=0.83,95%CI：0.76,0.91)：前一年灭螺后的乡镇其有螺风险仍是无螺且未灭螺乡镇的10倍(OR=9.97,95%CI：7.85,12.81)。在校正其它危险因素和空间随机效应的情况下,距长江超过62千米的乡镇有螺概率是在此范围内的0.43倍(OR=0.43,95%CI：0.22,0.84)。空间随机效应夸大了时间和乡镇离长江的距离对乡镇有螺概率的影响,高估了灭螺措施对有螺概率的影响。整体上在长江安徽段上游高风险有螺乡镇仍然较多,且分布较为集中,主要位于池州市东至县和贵池区。随时间的推移,其它地区高风险有螺乡镇数呈现减少的趋势。建议血防工作者加强对历史有螺区的监测和灭螺力度,特别是距离长江62千米范围内,地表温度低于27℃的地区,因为这些地区更适于钉螺生存和繁殖,有螺风险也要比其它地区更高,血吸虫病传播风险更大,对血吸虫病防治工作的意义更为重大。同时也应加强对以往无螺地区的监测,因为这些地区随时间变化的变化率更大。更多还原

【Abstract】 In spite of the great achievements in the national schistosomiasis control during the past half century, the new century will indeed be confronted by the great challenges. The number of the schistosomiasis decreased sharply, but the acute schistosomiasis increased. The drugs for the schistosomiasis were effective, but the long-term chemotherapy decreased compliance of the patients in the high risk areas. Snail habitats existed wildly, but the percent of snail control areas was relatively small. In addition, there were some new problems including global warming, floating population increase, limited financial supports and so on. The focus of schistosomiasis control was in the lake and marshland areas. In our study, modern spatial analysis technology, multilevel model and Bayesian spatial-temporal model were complemented and analyzed based on the data collection of schistosomiasis control program in Anhui province, in order to solve the key problems from the practical work and provide suggestions for the development of integrated schistosomiasis control program. This study can be referenced by the other related diseases for clustering and spatial-temporal analysis.Part I Clustering analysis of snail distribution and schistosomiasisThis part aimed to analyze the spatial clustering of schistosomiasis and snail for the high risk areas based on data collection in Anhui province and indicate the spatial-temporal association and mechanism for the development of integrated schistosomiasis control program. Two variables of the prevalence rate and the percentage of snail areas were computed according to the data collection of schstosomiasis prevalence in Anhui province from 2000 to 2008 and the spatial analysis database was constituted by the two variables matched with the spatial database of polygon. Global Autocorrelation Analysis, Local Autocorrelation Analysis and spatial scan statistics approach through moving windows were complemented to detect the clusters of schstosomiasis and snails. The results were visualized through the software of ArcGIS.For the variable of the percentage of snail areas, the spatial clustering of Global autocorrelation analysis were statistically significant from 2000 to 2008(Moran’s I>0, P<0.05). And the clusters by Local autocorrelation analysis and SaTScan analysis were almost consistent. The clusters of different radius by SaTScan analysis covered most of the towns with spatial clustering of high values of Local autocorrelation analysis and it didn’t appear to cluster more towns with the positive spatial clustering(T=1.85,P>0.05). The result of exploration by SaTScan suggested that the cluster area of the index mainly gathered in the band area 62 kilometers far away from both sides of the Yangtze River. The main clusters located near the downstream of Yangtze River through Anhui province before 2005. However, the clusters moved to the upstream since 2005.For the variable of prevalence rate of schistosomiasis, the results from Global Autocorrelation Analysis were not statistically significant (P>0.05) during the periods of 2000 to 2001 and 2005 to 2007. A few towns covered by small clusters were detected by Local Autocorrelation Analysis, however, the positions of the clusters were different from the result of SaTScan anslysis and distributed randomly. On the other hand, it was significant that some clusters were detected by Global Autocorrelation Analysis from 2002 to 2004 and in 2008(Moran’s I>0,P<0.05), but the clustering effect was weak. The results were consistent between the Local autocorrelation analysis and SaTScan analysis. The clusters of different radiuses by SaTScan analysis covered most of the towns with spatial clustering of high values of Local Autocorrelation Analysis. According to the analysis of the three methods mentioned above, it was proved that the spatial clustering was significant to the prevalence rate of schistosomiasis from 2002 to 2004 and in 2008.Global autocorrelation analysis can identify the clustering from the global view. Local Autocorrelation Analysis can detect the spatial association and the locations with positive spatial clustering with high values. The cluster centers and radius can be identified by SaTScan analysis. The three methods can be combined and implemented gradually and the integrated results are more systematic and comprehensive. PartⅡEstablishment and analysis of Multilevel Model of snail distributionThis part aimed to study the time trend of snail distribution based on town data collection adjusted by the confounding factors and identified the significant factors influencing the snail distribution in order to control the spatial autocorrelation by Bayesian spatial-temporal model. Growth Model of Multilevel model was used to adjust the within-group autocorrelation and study the effect of environmental factors(Normalized Difference Vegetation Index, Surface Temperature), climatic factors(Average Annual Precipitation, Annual Extreme Minimum Temperature), snail elimination and local position.The results of Empty model showed that ICC was 84.17% and the variance of the random intercept was statistically significant(t=8.19, P<0.01), so that ICC was significant and there was a great autocorrelation between the adjacent towns. It was indicated that the multilevel model was preferable to the classical statistical model for the longitudinal study.The model including the covariants with time was judged as the best model to fit our data. The variances of both the random intercept and slope were significant (σu02 =33.57,t=5.81, P<0.01;σu12=0.74, t=5.87, P<0.01). The results indicated that both the initial probability of snail and the rate of change of the probability were different and changing with the time moving for different towns. The significant covariance (σu012=-3.02,t=-4.71, P<0.01) suggested that if the initial probability of snail was great, the rate of change of snail probability would be small.The index of distance from the Yangtze River to the towns had the effect to the probability of snail for different towns (t=-8.03, P<0.01). The probability of snail would decreased with the distance increase (P<0.01). The effect of snail elimination and Land Surface Temperature were significant(P<0.05). Although the snail elimination was implemented in some towns, the probability of snail in these towns would be greater than that without snail elimination (95%CI:2.9,6.7). The probability of towns with Land Surface Temperature above 27℃was smaller than that of Land Surface Temperature below 27℃(95%CI:0.5,0.9).Part III Establishment and analysis of Bayesian spatial-temporal model for snail distributionThis part aimed to study the time trend and spatial clustering for quantitative measurement adjusted by the significant factors from Multilevel Model and spatial autocorrelation. Non-spatial model, separate spatial-temporal model and spatial-temporal interaction model were modeled including risk factors adjusted from Multilevel Model.According to the DIC principle, the separate spatial-temporal model was selected as the best model to fit the longitudinal data. It indicated that the random spatial effect was significant every year respectively and had no interaction as the time went by. The probability of snail decreased at town level year by year(OR=0.87,95%CI:0.84, 0.90). The probability of snail of towns with Land Surface Temperature above 27℃was 0.7 time of those with the temperature below 27℃(OR=0.73,95%CI:0.59,0.90). The risk of snail for different towns decreased by 20% with the distance increase away from the Yangtze River every 10 km (OR=0.83,95%CI:0.76,0.91). Though some towns implemented the snail elimination, the risk of snail was 9 times than that of the towns without snail and any control measures(OR=9.97,95%CI:7.85,12.81). In general, when the towns were over 62 km far away from the Yangtze River, the risk of this area was 0.43 times that of the towns inside this zone adjusted by other risk factors and spatial correlation(OR=0.43,95%CI:0.22,0.84). The effect of the distance from town to the Yangtze River on the probability of snail was overestimated due to the spatial correlation. The snail elimination had same effect on the probability of snail as well. More generally, the towns with high risk of snail clustered in the upstream of the Yangtze River though Anhui province, especially near the boundary of Dongzhi County and Guichi District in Chizhou City. The probability of other towns decreased with the time going.It was suggested that the surveillance and snail elimination should be reinforced greatly to the former snail habitats, especially in the zone of 62km away from the both sides of the Yangtze River and with the Land Surface Temperature below 27℃. Because the environment in these areas was suitable for the survival and reproduction of snail and people had more risks than other places to schistosomiasis infection. The snail control to this zone played an important role in controlling schistosomiasis. At the same time, it was necessary to strengthen the surveillance to the areas without detection of snail, because these areas had the greater rate of change of probability than other areas with snail as the time went on.更多还原