手足口病流行规律的贝叶斯多水平时空建模关键技术研究

Hand, foot, and mouth disease (HFMD) was associated with a lot of complicated natural and social factors. Using multi-discipline and cross-discipline knowledge, we suppose to build Bayesian multilevel spatio-temporal models to investigate epidemiological characteristics, spatial-temporal patterns and the potential risk factors of HFMD. The data included in our research will be collected from the national web-based reporting system of communicable diseases，GIS, Meteorological station, and remote sensor, etc. Based on our previous studies, we aim to investigate some key points in the process of model building: ①The HFMD cases and related data have the properties of multi-scale, complicated spatio-temporal effects. In order to explore the spatial pattern of lag and non-linear effects, we intend to apply the theory of space varying coefficient model and Gaussian Markov random field to the multilevel spatio-temporal modelling. ②One of the key challenges when building a prior for a Bayesian spatio-temporal model is controlling against over-fitting. We intend to adopt the recently proposed framework of penalised complexity (PC) priors to construction priors. We suppose to optimize the algorithm, realize the procedure and evaluate model. ③The standard solutions to obtain estimates of the posterior marginals of the unknown parameters are the Markov Chain Monte Carlo (MCMC) methods, which are very time-consuming. We intend to adopt the recently proposed approach that combines integrated nested Laplace approximations (INLA) and MCMC to improve the Bayesian inference. We suppose to optimize the algorithm, realize the procedure and evaluate the accuracy and efficiency of the algorithm. Our research will contribute to provide scientific information for public health decision making, responses and interventions of HFMD. As an essential aspect of statistical surveillance and spatial-temporal analysis, Bayesian multilevel spatio-temporal modelling will undoubtedly be the important theoretical and technological foundations of spatio-temporal epidemiological characteristics investigation of HFMD and similar infectious diseases, which possess brilliant prospect both in scientific research and application domain.

手足口病传播和流行是一系列自然和社会多源复杂因子综合作用的结果，反映手足口病时间和空间传播过程的规律就蕴含在手足口病时空数据中。本项目拟在前期研究基础之上，基于手足口病监测数据、GIS及气象、遥感等自然和社会因素数据，对贝叶斯多水平时空模型进行深入研究，着力解决以下关键技术问题：①针对手足口病及相关因素数据的复杂时空效应，将空间变系数模型的思想引入贝叶斯多水平时空建模，基于高斯马尔科夫随机场的图式建模理论，对非线性滞后效应的空间模式进行研究；②针对贝叶斯时空模型的过度拟合问题，基于国际上新近提出的PC先验框架改进模型先验分布的构建并评价其性能；③针对经典MCMC方法运算效率低下的问题，基于国际上新近提出的INLA与MCMC相结合的方法，改进模型的后验推断并评价其效率和准确性。在此基础上同时从时空两个维度深入研究手足口病在不同空间尺度和时间长度上的分布特征、传播过程及与之关联的自然和社会因素

手足口病传播和流行是一系列自然和社会多源复杂因子综合作用的结果，反映手足口病时间和空间传播过程的规律就蕴含在手足口病时空数据中。本项目在前期研究基础之上，基于手足口病监测数据、GIS及气象、遥感等自然和社会因素数据，对贝叶斯多水平时空模型进行深入研究，着力解决以下关键技术问题：①针对手足口病及相关因素数据的复杂时空效应，将空间变系数模型的思想引入贝叶斯多水平时空建模，基于高斯马尔科夫随机场的图式建模理论，对非线性滞后效应的空间模式进行研究；②针对贝叶斯时空模型的过度拟合问题，基于国际上新近提出的PC先验框架改进模型先验分布的构建并评价其性能；③针对经典MCMC方法运算效率低下的问题，基于国际上新近提出的INLA与MCMC相结合的方法，改进模型的后验推断并评价其效率和准确性。.对上述问题进行深入研究后发现，非线性滞后效应的空间模式的引入，可以更好地在时空模型中拟合气象、空气污染等变量的非线性滞后效应。PC先验能避免过拟合，并增强参数的可解释性。在INLA+MCMC中，仅少数参数估计采用MCMC法实现，绝大多数的参数估计采用INLA法实现，因此在保证估计结果准确的同时，可大幅度提高运算速度。且这些方法的应用具有普适性，能够方便的迁移到其它类似疾病与类似研究领域中，具有重要的学术与应用价值。.对上述关键技术进行程序化实现之后，本项目获得了四川省手足口病在不同空间尺度和时间长度上时空变化规律及影响手足口病发生和传播的重要自然和社会因素，可以为确定手足口病防控重点、调整或完善防控策略和政策制定提供参考。此外，本项目还将相关方法用于其他类似疾病，拓宽了相关方法的应用范畴，具有重要的应用价值。.

内容获取失败，请点击重试