Published on in Vol 9, No 1 (2017):

Modeling spatial and temporal variability by Bayesian  multilevel model

Modeling spatial and temporal variability by Bayesian multilevel model

Modeling spatial and temporal variability by Bayesian multilevel model

Authors of this article:

Xiaoxiao Song1 ;   Yan Li1 ;   Wei Liu1 ;   Le Cai1 ;   Wenlong Cui1 ;   Mingyue Wang1
The full text of this article is available as a PDF download by clicking here.

ObjectiveThe purpose of this article was to quantitative analyses the spatialvariability and temporal variability of influenza like illness (ILI) bya three-level Poisson model, which means to explain the spatial andtemporal level effects by introducing the random effects.IntroductionThe early detection of outbreaks of diseases is one of the mostchallenging objectives of epidemiological surveillance systems. Inorder to achieve this goal, the primary foundation is using those bigsurveillance data for understanding and controlling the spatiotemporalvariability of disease through populations. Typically, publichealth’s surveillance system would generate data with the big datacharacteristics of high volume, velocity, and variety. One commonquestion of big data analysis is most of the data have the multilevel orhierarchy structure, in other word the big data are non-independent.Traditional multilevel or hierarchical model can only deal with 2 or3 hierarchical data structure, which bound health big data furtherresearch for modeling, forecast and early-warning in the public healthsurveillance, in particular involving complex spatial and temporalvariability of Infectious Diseases in the reality.MethodsAll the data based the ISSC project from April 1 2012 throughMarch 31 2014 in the China. We adopted Markov Chain MonteCarlo algorithm (MCMC) in Bayesian hierarchical (multilevel)model, which means to explain the spatial and temporal leveleffects by introducing the random effects. In order to calculate thegeographical variations and temporal variation of ILI cases duringtwo years surveillance, we constructed spatial and temporal modelof three levels, which was day-in-months → months-in-two-year→Monitoring Units (Fig-1). Level one was repeated measures withinevery month, which was referred as day-in-months and the maximumvalue was 31 days. Level two was the variation tendency of monthswhich was 24 months. Level three was the effect of spatial distributionof monitoring units, which took the spatial heterogeneity into accountrather than dependence. This model was then adopted to evaluate andimprove the early warning capacity of syndromic surveillance.ResultsWe adopted multilevel spatio-temporal model (day-in-months →months-in-two-year →Monitoring Units) to analyze the points datacollected from 2 counties in China, including two hospitals at countylevel, 15 central hospital at township level and 152 health care unitsin the villages. The analysis of totally 108163 pieces of point data onILI case indicated there are significant spatial and temporal variationamong these cases. Among two thirds of the variation attributes to thedifference of geographical locations of these monitoring sites. Theremaining one third of the variation attributes to the time dimensions,such as seasonal effect.ConclusionsThe variation of monitoring data collected from health careunits mainly attributes to the difference of geographic locations formonitoring sites, yet only one third of the change attribute to the timechange, such as seasons, holidays and festivals. Therefore, it is criticalto select the location of monitoring site, which is more rational toselect the hotspots with representative characters rather than try tocover the whole monitoring area.