Understanding Outbreak Data Dissemination In Event Based Surveillance Systems. Application On Avian Influenza Using PADI-web

Abstract

Purpose: Epidemic intelligence (EI) has been adopted by several countries to reach fast detection of new and emerging infectious diseases. EI collects information from two types of sources: official sources (i.e. health reports from OIE or FAO) and unofficial sources (i.e. online media outlets, scientific publications, etc.). In France, the EI system PADI-web (Platform for Automated extraction of Disease Information from the Web) is used since 2014 to detect signals of animal health events with risk of introduction to France.The objective of this work was to understand how health information (signal) is disseminated from a primary source (transmitter) to a final source (EI system) through quantitative and qualitative network analysis methods. Methods & Materials: We analysed all English reports related to avian influenza detected by PADI-web between August 2018 and June 2019. We used the sources cited in the detected reports to trace the path of each signal. Signals were categorized as official and unofficial according to the source. We have built a directed network where the nodes represented the sources (characterized by type, location and geographical focus) and the edges represented the signal flow. To describe the network, we used network centrality measurements (degree, betweenness and eigenvector) to determine which nodes were important in the data dissemination. We also included the reactivity, calculated as the difference (in days) between the detection of an outbreak by PADI-web and its official notification by Empres-i (gold-standard) with a distinction between wild and domestic birds. Results: PADI-web detected 202 official signals and 26 unofficial signals. The OIE occupies a central position in the PADI-web information network. National veterinary authorities were the major primary sources. Online news outlets followed by press agencies were the main secondary sources. A significant portion of the signals was detected early in wild birds (41%) and in domestic birds (13%). Conclusion: This work showed PADI-web's capacity to detect early signals and can be used to define priority sources to improve this tool in terms of reactivity and data quality. In the future, similar work will be conducted on other diseases and EI systems to improve these systems.