基于时空聚集面板模型的肺结核病高危区域探测及影响因素研究

发布时间：2017-12-28 00:11

  本文关键词：基于时空聚集面板模型的肺结核病高危区域探测及影响因素研究　出处：《山西医科大学》2017年博士论文　论文类型：学位论文

  更多相关文章： 肺结核病 空间自相关 时空扫描统计 空间截面回归模型 空间面板数据模型

【摘要】：目的:针对疾病发病水平的监测数据存在时间、空间自相关性和空间异质性的特点,采用时空统计分析方法和空间计量经济模型,对青海省肺结核病监测数据和地区主要社会经济指标及气象因子数据,在生态学层面开展肺结核病系统研究,准确探测发病高危区域和定量分析影响发病率的相关社会环境因素,并借助气象因子的变化,对发病率进行合理预测。通过本研究探讨空间地理信息系统、时空聚集性分析方法和空间计量经济模型在具有时空属性的传染病监测数据挖掘中的应用价值,为类似研究提供分析思路和方法学参考,也为政府决策提供科学依据。方法:通过中国疾病预防控制信息系统收集青海省肺结核病发病资料,《青海省统计年鉴》收集青海省人口学资料和相关社会因素及气象因素资料,开展如下研究:1、采用流行病学“三间”分布描述、集中度法、圆形分布法、季节指数法和三维趋势分析等方法对2009-2013年青海省肺结核病流行病学特征进行分析。2、针对传染病监测数据时空不独立性,采用Moran’s I和Getis-Ord G空间自相关分析以及Sa TScan时空扫描分析对2009-2013年青海省县级水平肺结核病空间、时间以及时空聚集特征进行分析,对发病高危聚集区域及范围进行准确探测,并评价其风险强度。3、针对横截面数据空间分布非独立,采用Moran’s I空间自相关分析和空间截面回归模型,对2011年和2013年青海省各县级行政区域肺结核病年发病率与政府医疗卫生支出(千元/人)、教育支出(千元/人)、医疗机构床位数(张/千人)、医疗机构人员配备情况(人/千人)、农村居民人均纯收入(千元)以及人均GDP(万元)等6项社会指标关系进行双变量空间自相关分析和回归分析,研究社会因素的空间聚集特征,以及在扣除邻近地区发病率的影响后,定量探讨影响肺结核病年发病率的社会因素。4、由于肺结核病发病具有一定的季节特征,气象因素和前期发病率可能对本地区发病率的影响存在时间滞后性,以及邻近地区间发病率的相互影响,以2009-2013年青海省各市(州)肺结核病月发病率数据为应变量,同期到6个月滞后间隔的月平均气温(°C)、降水量(mm)、日照总时数(hours)、平均风速(m/s)和发病率为自变量,进行不同滞后间隔的面板数据模型回归分析,探讨气象因素对发病率影响的最佳滞后期;利用F检验、Hausman检验、误差项Moran’s I检验以及拉格朗日乘数(LM)检验确定最优分析模型,定量分析影响肺结核病月发病率的气象因素。5、利用上述气象因素与发病率的最佳滞后期空间面板数据模型进行发病率预测时,由于自变量中包含着邻近地区同期发病率,故尝试采用专家建模器对预测时段各地区发病率进行预估,再利用空间面板数据模型对各地区发病率进行预测,评价预测精度,比较专家建模器联合空间面板数据模型的预测效果;探讨联合模型进行发病率短期预测的可行性和可靠性。结果:1、青海省肺结核病年均报告发病率为98.26/10万,明显高于全国平均水平,且近年有略微上升趋势;中老年人群发病率最高,其次为青壮年;男性高于女性,新发病例以农牧民为主;具有明显的周期性和微弱的季节性,发病高峰主要集中在3-5月;南北方向呈北低南高的弧形变化趋势,东西方向呈明显的倒“U”型。2、全局空间自相关Moran’s I和General G值均大于期望值,取值范围分别为0.398-0.581和0.029-0.034,表明肺结核病年发病率地区分布存在明显的高发地区聚集倾向;Sa TScan时空扫描分析结果显示青海省肺结核病在时间、空间以及时空上存在明显的高发病风险聚集性,时空一类聚集区域位于青海省西南部,中心位置在囊谦县(东经96.12°,北纬32.17°),覆盖囊谦县、玉树市和杂多县等8个县(市),聚集半径为421.00Km,发病高峰期为2012年1月到2013年6月,相对危险度(RR)为4.58;叠加分析显示,以RR2作为判断标准比较合理,高危聚集区域主要集中在青海省西南部的玉树和果洛州所辖12个县(市)。3、双变量Moran’s I空间自相关分析显示2011年医疗机构床位数、医疗机构人员配备情况、农村居民人均纯收入和人均GDP共4项社会指标与肺结核病年发病率间均具有统计学意义(P0.05),提示以上社会因素可能影响地区发病水平;以发病率对数值建立的普通最小二乘回归显示回归残差不独立(Moran’s I=0.16,P0.05),而依据LM检验,空间滞后模型为最佳模型,该模型显示:空间自相关系数r=0.4041,说明相邻区域的发病率存在空间外溢现象(空间自相关性),即当其它影响因素固定不变时,相邻地区肺结核病年发病率每增加9倍,本地区年发病率将增加1.54倍;在扣除了发病率的空间自相关性后,农村居民人均纯收入是影响肺结核病年发病率的主要社会因素,b=-0.0657,即农村居民人均纯收入每增加1千元,本地区肺结核病年发病率将降低14%;空间截面回归模型与普通最小二乘回归模型相比,回归系数绝对值有所下降(-0.0657 vs-0.0883),说明充分考虑了发病率的空间自相关性后,估计结果更为合理,而传统回归模型没有考虑空间自相关性,夸大了社会因素的作用。2013年分析结果与2011年结果一致。4、面板数据模型分析结果显示气象因素对发病率的影响存在3个月的滞后期;发病率对数转换构建的固定效应模型(F=193.90,H=10.41,P0.05)显示回归残差不独立(Moran’s I=0.20,P0.05),而依据LM检验,空间滞后固定效应面板数据模型为最佳模型,该模型显示:不同地区截距项不同,体现了发病率的空间异质性;空间自相关系数r=0.3017,说明相邻区域的月发病率存在空间外溢现象,即相邻地区肺结核病月发病率每增加9倍,本地区月发病率将增加1倍;相比气象因素而言,当前发病率对滞后3个月的发病率影响更明显;在扣除了发病率的空间自相关性、空间异质性以及前期发病率的影响后,平均气温和降水量是影响滞后3个月发病率的主要气象因素;当前发病率每增加9倍,滞后3个月的发病率将增加36%,平均气温每升高10°C,滞后3个月的发病率将降低9%,降水量每增加2cm,滞后3个月的发病率将降低3%;与空间截面回归模型类似,空间面板数据模型与传统回归模型相比,估计结果也更为合理。5、2013年10-12月各地区发病率时间序列专家建模器预测相对误差为0.90%-136.14%,平均相对误差为28.99%;专家建模器联合空间面板数据模型预测相对误差为0.17%-94.20%,平均相对误差为21.09%;2014年1-3月平均相对误差分别为26.60%和19.79%;专家建模器联合空间面板数据模型的预测精度明显提高。结论:本研究首次采用时空统计分析方法和空间计量经济模型,从生态学角度对青海省肺结核病监测数据进行了详细探讨,得出如下结论:1、针对传染病监测数据时空非独立性特点,空间自相关分析和时空扫描分析是疾病时空聚集特征和高危区域探测的理想分析方法,准确探测出了青海省肺结核病高危聚集区域主要集中在该省西南部,最大危险区以玉树市为中心,r=259Km,覆盖玉树、囊谦、称多、杂多、玛多和曲麻莱等6县(市),RR=3.77。2、考虑到发病率的时空属性,空间截面回归模型和空间面板数据模型是生态学影响因素研究的理想分析模型,在公共卫生领域具有广泛应用价值。扣除发病率的时空影响后,农村居民人均纯收入、气温以及降水量是影响地区肺结核病发病率的主要社会环境因素。3、相比单纯时间序列专家建模器预测,专家建模器联合空间面板数据模型的预测策略,在考虑了邻近地区间发病率的相互影响以及气象因子的作用后,预测精度明显提高,可应用到实际工作中发挥预警作用。通过本研究既为青海省肺结核病防控措施的合理制定提供了理论依据,也为具有时空属性特征数据的研究提供了分析思路和方法学参考。
[Abstract]:Objective: the temporal and spatial characteristics of autocorrelation and spatial heterogeneity of the disease incidence rate of the monitoring data, the analysis method and the spatial econometric model of temporal statistics, Qinghai province tuberculosis monitoring data and area of main social and economic indicators and meteorological data, in the study of pulmonary tuberculosis ecology system to carry out accurate detection of high risk level. Regional and quantitative analysis of the influence of the social environment related morbidity factors, and with the change of meteorological factors, a reasonable forecast of incidence. Through this study, geographic information system, spatial aggregation value analysis method and spatial econometric model in infectious disease monitoring data with spatial attribute mining, provide analysis ideas and methodology reference for similar research, but also provide a scientific basis for government decision-making. Methods: the Chinese information system for Disease Control and prevention of tuberculosis incidence data collected in Qinghai Province, "Qinghai Province Statistical Yearbook" data collection factors and meteorological factors in Qinghai province and related social demographic data, carry out the research as follows: 1, "three" the epidemiological distribution description, degree method, the method of circular distribution, seasonal index method and focus on three-dimensional trend analysis method to analyze the epidemiological characteristics of pulmonary tuberculosis in Qinghai province 2009-2013. 2, according to the monitoring data of infectious diseases is not independent of time and space, using Moran s I and Getis-Ord G spatial autocorrelation analysis and Sa TScan space-time scan analysis to 2009-2013 at county level in Qinghai province tuberculosis space, time and space aggregation characteristics were analyzed, the accurate detection and scope of the gathering area of high risk, and to evaluate its risk strength. 3, according to the cross-sectional data of spatial distribution of non independent, using Moran s I spatial regression model and spatial autocorrelation analysis section, on 2011 and 2013 in Qinghai Province, county-level administrative region tuberculosis incidence and government health expenditure (1000 yuan / person), education expenditure (1000 yuan / person), medical institutions the number of beds (A / 1000), medical institutions (staffing / 1000), per capita net income of rural residents (thousand dollars), per capita GDP (million) 6 social indicators between the analysis and autocorrelation analysis and bivariate regression, accumulation characteristics of social factors in space, and net affected the incidence of adjacent areas, quantitative discussion on social factors influencing the annual incidence of pulmonary tuberculosis. 4, due to the occurrence of pulmonary tuberculosis has certain seasonal characteristics, meteorological factors and the incidence may be time effect on local incidence of lag, and adjacent areas of incidence between the mutual influence, in the cities of Qinghai Province during 2009-2013 (state) pulmonary tuberculosis incidence data for the month should be variable, the average monthly temperature over the same period to 6 months lag interval (C) and precipitation (mm), the total sunshine hours (hours), mean velocity (m/s) and the incidence rate as independent variables, regression analysis of panel data models with different lag intervals, to investigate the meteorological factors on the optimal lag influence the incidence rate; using F test Hausman test, the error Moran s I test and the Lagrange multiplier (LM) test to determine the optimal analysis model, quantitative analysis of the influence of meteorological factors on the incidence of pulmonary tuberculosis. 5, the incidence of the weather prediction using best lag factors and the incidence of spatial panel data model, the independent variables included in the adjacent regions in the same period of incidence, so try to use to predict the onset of each expert modeling area prediction rate with time, spatial panel data model to forecast the incidence of various regions. To evaluate the prediction accuracy, the prediction effect of modeling device combined with spatial panel data model comparison expert; to study the model of the incidence of the reliability and feasibility of the short-term forecast. Results: 1, Qinghai Province, the annual report of tuberculosis incidence rate of 98.26/10 million, significantly higher than the national average, and in recent years, there is a slight upward trend; the highest incidence in the elderly population, followed by young adults; men and women, new cases to farmers and herdsmen; with obvious periodicity and weak seasonal the peak incidence, mainly concentrated in 3-5 months; the north-south direction curved trend of North South High low, east-west direction is inverted "U" type obviously. 2, the global spatial autocorrelation of Moran 's I and General G values are greater than the expected value, range were 0.398-0.581 and 0.029-0.034, showed that the annual incidence of tuberculosis area distribution is significant in areas of high Sa TScan aggregation tendency; space-time scan analysis showed that Qinghai province tuberculosis in time, space and time are high incidence the risk of significant aggregation, a kind of space gathering area in Qinghai province is located in the southwest, central location in Nangqian county (96.12 degrees east longitude, latitude 32.17 degrees), 8 counties covered in Nangqian, Zaduo County, and Yushu city (city), gathering radius is 421.00Km, the peak incidence from January 2012 to June 2013, the relative risk degree (RR) is 4.58; superposition analysis showed that using RR2 as the judgment standard is reasonable, high accumulation area mainly concentrated in the southwest of Qinghai Province, Yushu and Guoluo Prefecture under the jurisdiction of 12 counties (city). 3, double variable Moran 's I spatial autocorrelation analysis showed that in 2011 the medical institutions beds, personnel, per capita net income of rural residents and the per capita GDP total of 4 social indicators and the incidence of pulmonary tuberculosis among years were statistically significant (P0.05), suggesting that social factors may influence the incidence level in area; the incidence of ordinary least squares regression to establish numerical display regression residuals are not independent (Moran' s I=0.16, P0.05), according to the LM test, the spatial lag model is the best model, the model shows that the spatial autocorrelation coefficient R =0.4041, shows that the incidence rate of adjacent area spatial spillover phenomenon (spatial autocorrelation), i.e. when other factors are fixed, the adjacent areas of TB incidence per year
【学位授予单位】：山西医科大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：R181.3;R521
，

本文编号：1343768