Digital Epidemiology and Surveillance  (DES) Session 2

Schedule Top Page

Time and Date: 14:15 - 18:00 on 20th Sep 2016

Room: Z - Zij foyer

Chair: Daniela Paolotti

36005 Monitoring vaccine confidence (with deep learning): the VCMP platform Marco Cristoforetti
36006 Flu-Now : Nowcasting Flu based on Product Sales Time Series [abstract]
Abstract: Big Data offer nowadays the capability of creating a digital nervous system of oursociety, enabling the measurement, monitoring and prediction of various phenomena inquasi real time. But with that, comes the need of more timely forecast, in other wordsnowcast of changes and events in nearly real-time as well. The goal of nowcasting is toestimate up-to-date values for a time series whose actual observations are available onlywith a delay. Choi and Varian introduced the term nowcasting to advocate the tendencyof web searches to correlate with various indicators, which may reveal helpful for shortterm prediction. In the field of epidemiology, it was showed in various works that searchdata from Google Flu Trends, could help predict the incidence of influenza-like illnesses(ILI). But as Lazer and al. notice, in February 2013, Google Flu Trends predicted morethan double the proportion of doctor visits for ILI than the Center for Disease Control.In this work we are studying the flu time series, of cases from 2004/05 to2014/2015 flu season, from physicians and pediatricians from all over Italy. We are interestedto examine whether is possible to use retail market data as a proxy for flu prediction.Our dataset consists of economic transactions collected by COOP, a system of Italian consumers?cooperatives which operates the largest supermarket chain in Italy. The wholedataset contains retail market data in a time window that goes from January 1st, 2007 toApril, 27th 2014. First, we identified the products that have adoption trend similar to theflu trend with the help of an 1-nearest neighbor classifier that uses dynamic time warpingas the distance measure between time series. Based on these products, we identified thecustomers that buy them during the flu-peak, since those individuals would have higherpossibility to be either infected or close to an infected individual. We extracted their mostfrequent baskets during the peak using the Apriori algorithm, an algorithm for frequentitem set mining and association rule learning over transactional databases, and we usethose baskets-sentinels as control set for the following year flu peak. Monitoring the behaviorof these baskets-sentinels we are able to detect patterns similar to the ones of theprevious year?s flu peak, and as a result obtain an alarm for the appearance of the flu.Many lines of research remain open for future work, such as studying whether theretail market data can manage to predict the flu peak even in particular cases such as theyear 2009 non-seasonal H1N1 influenza (flu) pandemic that peaked in October and thendeclined quickly to below baseline levels by January. Close
Ioanna Miliou, Salvatore Rinzivillo, Giulio Rossetti, Dino Pedreschi and Fosca Giannotti.
36007 TOWARDS A BRIGHT FUTURE OF DATA-DRIVEN SIMULATION FOR POLICY DECISION SUPPORT Modeling and Simulation of the Zika outbreak [abstract]
Abstract: Background: The potential health threats associated to Zika virus infections (ZIKV) have alarmed the global community. Unusual incidences of microcephaly and Guillain-Barr? syndrome have attracted attention towards a disease that was for decades considered relatively harmless. At present, health authorities worldwide are urged to implement policies to prevent further spreading, in spite of only having scarce and uncertain knowledge on the epidemiology and potential solutions. Complex problems such as infectious diseases are often addressed by simulation techniques. However, vector-borne diseases such as ZIKV present three major challenges that these approaches find difficult to tackle. First, the worldwide distribution of tropical and sub-tropical regions with suitable conditions for ZIKV?s vectors call for a global scope of the study and the use of global data. Second, human travel is the main dissemination mode. Therefore, policy decisions support needs to cope with the high connectivity of regions. Third, the emergence of ZIKV is driven by geographically specific environmental and socio-economic factors. Thus, an appropriate representation requires georeferenced information. Objective: The aim of this research was to explore ways to overcome the limitations of simulation methods conventionally used to study epidemics. More specifically, we aimed at designing robust response strategies to prevent the potential spread of Zika virus infections to regions worldwide. Approach: To achieve this goal, we applied a data-driven simulation approach that integrates different modelling and simulation methods and adopts concepts from other disciplines. We divided the world according to their global administrative divisions and by adopting an object-oriented approach replicated our core model structure describing the epidemiological process in each region. The spread of the disease across regions occurs along two distinct network structures. First, an undirected and symmetric adjacency matrix. Second, model-based predictions of global air passenger flows represented in a directed matrix . Data: The input data was collected from diverse publicly available sources including geospatial data from maps, conventional databases provided by government agencies and open-access modelling outcomes of reported research. We aggregated multiple layers of high resolution raster datasets to administrative divisions by geo-processing zonal statistics and combined this with data on air passenger flows from 3416 airports across the world. After pre-processing steps for data alignment, the data was used as input for our simulation model. Since viewing data on maps enhances users? understanding, we projected back our simulation outcomes on maps. In this way users can interactively plot different plausible scenarios over time and explore their consequences. Sensitivity analysis and policy exploration were conducted on the higher levels of aggregation. Simulation methods: The model components of each region combine two modeling methods taking advantage of their respective strength. Our approach, loosely based on Bobashev et al. (2007), starts with an agent-based model. After, reaching a threshold number of infectious individuals, it switches to equation-based. Next steps: By collecting and managing data at global scale, coping with regions increasingly connected by air travel, and integrating georeferenced environmental and social information, our approach promises to deliver breakthroughs in model based policy analysis. The validity of the presented approach and results of policy exploration will be shown in future work. Close
Philipp Schwarz, Erik Pruyt.
36008 Digital Epidemiology Through the Ages Rumi Chunara