Abstract
ObjectiveThe presentation describes the design and the main functionalitiesof two user-friendly applications developed using R-shiny to supportthe statistical analysis of morbidity and mortality data from the Frenchsyndromic surveillance system SurSaUD.IntroductionThe French syndromic surveillance system SursaUD® has beenset up by Santé publique France, the national public health agency(formerly French institute for public health - InVS) in 2004. In 2016,the system is based on three main data sources: the attendancesin about 650 emergency departments (ED), the consultations to62 emergency general practitioners’ (GPs) associations SOSMédecins and the mortality data from 3,000 civil status offices [1].Daily, about 60,000 attendances in ED (88% of the nationalattendances), 8,000 visits in SOS Médecins associations (95% ofthe national visits) and 1,200 deaths (80% of the national mortality)are recorded all over the territory and transmitted to Santé publiqueFrance.About 100 syndromic groupings of interest are constructed fromthe reported diagnostic codes, and monitored daily or weekly, fordifferent age groups and geographical scales, to characterize trends,detect expected or unexpected events (outbreaks) and assess potentialimpact of both environmental and infectious events. All-causesmortality is also monitored in similar objectives.Two user-friendly interactive web applications have beendeveloped using the R shiny package [2] to provide a homogeneousframework for all the epidemiologists involved in the syndromicsurveillance at the national and the regional levels.MethodsThe first application, named MASS-SurSaUD, is dedicated to theanalysis of the two morbidity data sources in Sursaud, along with dataprovided by a network of Sentinel GPs [3]. Based on pre-aggregateddata availaible daily at 10:30 am, R programs create daily, weeklyand monthly time series of the proportion of each syndromic groupingamong all visits/attendances with a valid code at the national andregional levels. Twelve syndromic groupings (mainly infectious andrespiratory groups, like ILI, gastroenteritis, bronchiolitis, pulmonarydiseases) and 13 age groups have been chosen for this application.For ILI, 3 statistical methods (periodic regression, robust periodicregression and Hidden Markov model) have been implementedto identify outbreaks. The results of the 3 methods applied to the3 data sources are combined with a voting algorithm to compilethe influenza alarm level for each region each week: non-epidemic,pre/post epidemic or epidemic.The second application, named MASS-Euromomo, allowsconsulting results provided by the model developed by the Europeanproject EuroMomo for the common analysis of mortality in theEuropean countries (www.euromomo.eu). The Euromomo model,initially developed using Stata software, has been transcripted inR. The model has been adapted to run in France both at a national,regional and other geographical administrative levels, and for 7 agegroups.ResultsThe two applications, accessible on a web-portal, are similarlydesigned, with:- a dropdown menu and radio buttons on the left hand side to selectthe data to display (e.g. filter by data source, age group, geographicallevels, syndromic grouping and/or time period),- several tab panels allowing to consult data and statistical resultsthrough tables, static and dynamic charts, statistical alarm matrix,geographical maps,... (Figure 1),- a “help” tab panel, including documentations and guidelines,links, contact details.The MASS-SurSaUD application has been deployed in December2015 and used during the 2015-2016 influenza season. MASS-Euromomo application has been deployed in July 2016 for the heat-wave surveillance period. Positive feedbacks from several users havebeen reported.ConclusionsBusiness Intelligence tools are generally focused on datavisualisation and are not generally tailored for providing advancedstatistical analysis. Web applications built with the R-shiny packagecombining user-friendly visualisations and advanced statistics can berapidly built to support timely epidemiological analyses and outbreakdetection.Figure 1: screen-shots of a page of the two applications