气温，湿度对上海市流感样病例人次的影响

发布时间：2018-03-11 21:49

本文选题：气温　切入点：湿度　出处：《复旦大学》2013年硕士论文　论文类型：学位论文

【摘要】：研究目的：提出带自回归项的广义相加模型,并研究气温,湿度对上海市每日流感样病例的影响。研究意义：流感样病例人次是一个时间序列,而时间序列分析往往基于数据对发生在它之前的数据的依赖。但常用的GLM和GAM本质上是概率模型,没有使用因变量的自相关性的信息。另外,采用最大似然法估计GLM和GAM的参数值时,一般假定观测之间相互独立,但这不一定成立。文中提出的GAMAR可以解释因变量的自相关性,不仅仅对本研究的数据较好的拟合和解释提供了方法,而且为其他类似研究方法的选择提供一种新的可能性。流感是一个对人类健康危害极大的疾病。试验证实了气候因素与流感病毒的传播有着密切的关系。但气候因素与流感样病例关联性的流行病学研究还较少。上海是一个国际大都市,面临着流感季节性流行和暴发的风险。了解气象因素对流感的作用,有利于人们应用相应的措施来控制流感的流行,减少流感暴发的风险,有助于医院管理人员对医疗资源的调配,也有利于对流感的暴发作出预警。统计方法：提出带自回归项的广义相加模型,并用模拟研究来比较GAM与GAMAR在应用于因变量有自相关性的数据中的表现。建立模型：先采用基于GAM的Poisson回归,利用自然立方样条函数来代表长期趋势,以及日均气温,相对湿度,日均气压,大气污染物效应PM10,SO2,NO2的非线性效应,用哑变量描述星期效应,分析流感样病例与气候因素的关联性。发现其Pearson残差存在自相关的现象,继而采用GAMAR进行估计。预测模型：前文中模型包含一个长期时间样条,这使得它不可能被用于预测。若用GAM模型进行预测,未来流感样病例趋势的不确定性可能会影响预测的精度。而采用GAMAR进行预测,由于AR项本来就包含了以往时间趋势的信息,因此GAMAR本质上能够用以往的数据来预测以后的变化趋势。为证明这一点,对调整时间趋势前后的流感样病例采用GAM/GAMAR拟合结果进行交叉验证。结果和结论：本文提出了拟合环境流行病学中时间序列的GAMAR模型,模拟研究证明了在拟合有自相关性的数据时,GAM的参数估计结果有错误,而GAMAR可以给出正确的估计结果。接下来采用GAM构建了流感样病例关于气候与大气污染物的统计模型,但是发现其Pearson残差有严重的自相关性,违背了GAM的模型假设；因而采用GAMAR建模。GAMAR给出的Pearson残差基本上没有自相关性。通过对比发现,GAMAR给出的变量效应比GAM的更平缓一些。对GAMAR而言,23天前的日均气温在10℃C和30℃C时对应的当天流感样病例最高,在15～25℃时对应的流感样病例最低,效应曲线呈双峰状。4天前的相对湿度较低时,当天的流感样病例比较高,然后随着湿度的增加而降低,在65～80%时流感样病例有一些波动,在高于80%时继续降低。因为未来的总体变化趋势是未知的,所以预测模型中不能包括时间的趋势样条,自然能够包含过去信息的GAMAR是预测模型更好的选择。通过交叉验证发现,GAMAR果然比GAM在是否调整长期趋势的问题上更稳健。并且无论是否调整长期趋势,GAMAR均比GAM在交叉验证中表现得更好。
[Abstract]:Objective: to propose a generalized additive model with autoregressive term, and to study the influence of temperature and humidity on the daily influenza like cases in Shanghai.
Significance: Ili people is a time series and time series analysis are often based on data dependence on the data before it happened. But the GLM and GAM is essentially a probabilistic model, because without the use of the self correlation of information variables. In addition, using the maximum likelihood method to estimate the parameters of GLM and GAM value, generally assumed that observations are independent of each other, but this is not true. The proposed GAMAR can be explained by self correlation variables, provides a method for the data of this study not only better fitting and interpretation, but also for other similar research methods provide a new possibility for a flu is. Human health harmful diseases. Experiments have confirmed that there is a close relationship between the climate factors and the spread of influenza virus. But the epidemiological association between climate factors and the Ili Shanghai is less. Is an international metropolis, facing the seasonal flu epidemic and outbreak risk. To explore the effects of meteorological factors on influenza, help people apply the corresponding measures to control the flu, reduce the risk of the outbreak, to help hospital managers on medical resources allocation, but also conducive to the early warning of influenza the outbreak.
Statistical methods: a generalized additive model with autoregressive terms is proposed, and the performance of GAM and GAMAR in the data that are applied to dependent variables are compared by simulation.
Model: first using GAM Poisson regression based on the use of natural cubic spline function to represent the long-term trend, and the daily average temperature, relative humidity, average air pressure, air pollutants SO2, PM10 effect, nonlinear effect of NO2, describe the week effect with dummy variables, correlation analysis and climatic factors of influenza like cases are found. The Pearson residual autocorrelation phenomenon, then estimated by GAMAR.
Prediction models: previous model contains a long time spline, which makes it impossible to be used to predict if the use of GAM model to predict the future trend of influenza like illness uncertainty may affect the prediction accuracy. The use of GAMAR to predict, because AR already contains the time trend information therefore, GAMAR can essentially change trend after using past data to predict. To prove this point, of influenza like cases before and after adjustment of time trends using the fitting results of GAM/GAMAR cross validation.
Results and conclusion: This paper proposes a GAMAR model fitting in environmental epidemiology sequence, simulation study proved that fitting self correlation data, GAM parameter estimation results in error, while GAMAR can give the correct estimation results. Then GAM is used to build a statistical model on climate and atmospheric pollutants in Ili but the Pearson has serious residual autocorrelation, contrary to the assumption of GAM model; and using GAMAR modeling.GAMAR Pearson are basically no residual autocorrelation. By comparison, the effect of variables given by GAMAR than GAM. Some more slowly for GAMAR, the daily average temperature of 23 days at 10 degrees C and 30 C C C corresponds to the day of influenza like cases of influenza like cases corresponding to the highest, at 15~25 DEG C when the lowest effect curve of relative humidity in Shuangfeng.4 days ago was low, the day Influenza like cases are higher and then decrease with the increase of humidity. There are some fluctuations in influenza like cases at 65 to 80%, and continue to decrease at higher than 80%.
Because the overall trend of the future is unknown, so the prediction model not including the time trend of spline, nature can contain information in the past GAMAR prediction model is a better choice. The cross validation showed that GAMAR was more than GAM in the long-term trend of adjustment is more robust. And regardless of whether the long-term trend of adjustment, GAMAR better than GAM in cross validation.

【学位授予单位】：复旦大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：R511.7

【相似文献】