Featured Study: Flu Spread in France

The value of spatiotemporal analysis is well acknowledged among environmental health and epidemiology scientists. Space-time studies administer invaluable help by identifying and describing composite space-time heterogeneity of key attributes. The stochastic nature of analysis addresses the issue of inherent uncertainty. Visualization maps provide means for a deeper and comprehensive understanding of evolving events of interest.

To the above, our analysis in the specific study charted crucial paths to adding scientific support to commonly used space-time prediction techniques. To establish correlations among the observed data, we introduced new, model-based covariance functions driven by epidemic propagation models; then we yielded predictions for the attribute of interest on the basis of these models.

About the Analysis

The following animation displays the predicted ratio of new infectives (RNI) across France over a period of 15 weeks in the winter of 1998-1999. This quantity is defined for a specific region and time period as the number of newly infected individuals within this region and time over the whole population of the region. On a weekly time period in France, RNI ranges between 0 (white areas) and 0.005 (dark brown areas) in the animated map scale. At the peak of the flu season a few scattered higher values were predicted, among which RNI reached a maximum value 0.019.

Our focus in the French flu study was to research new avenues, in light of known conceptual and technical limitations of mainstream spatial and spatiotemporal environmental health analysis. The goal is to look for and employ knowledge bases such as health models, survey data, and empirical relationships to deal more efficiently with the problem uncertainty. Driven by the Knowledge Synthesis framework, we integrated the information of epidemic proagation models in the analysis. We achieved this in the form of new space-time covariance functions. That is, we composed mathematical constructs based on these models, and then fitted them appropriately to express the underlying correlation among the flu observations in the problem data set. 

Assembling these new functions led us to investigate in more depth the composite space-time RNI dependency. It consequently enabled us to generate informative, sufficiently detailed maps of the RNI patterns across space and time.

Reference

Kolovos A., Angulo J.M., Modis K., Papantonopoulos G., Wang J.-F., and G. Christakos. 2012. Model-Driven Development of Covariances for Spatiotemporal Environmental Health Assessment. Envir Monit and Assessm, doi: http://dx.doi.org/10.1007/s10661-012-2593-1