Abstract
ObjectiveLANL has built a software program that automatically collectsglobal notifiable disease data—particularly data stored in files—andmakes it available and shareable within the Biosurveillance Ecosystem(BSVE) as a new data source. This will improve the prediction andearly warning of disease events and other applications.IntroductionMost countries do not report national notifiable disease data in amachine-readable format. Data are often in the form of a file thatcontains text, tables and graphs summarizing weekly or monthlydisease counts. This presents a problem when information is neededfor more data intensive approaches to epidemiology, biosurveillanceand public health as exemplified by the Biosurveillance Ecosystem(BSVE).While most nations do likely store their data in a machine-readableformat, the governments are often hesitant to share data openly fora variety of reasons that include technical, political, economic, andmotivational issues [1]. For example, an attempt by LANL to obtaina weekly version of openly available monthly data, reported by theAustralian government, resulted in an onerous bureaucratic reply. Theobstacles to obtaining data included: paperwork to request data fromeach of the Australian states and territories, a long delay to obtaindata (up to 3 months) and extensive limitations on the data’s use thatprohibit collaboration and sharing. This type of experience whenattempting to contact public health departments or ministries of healthfor data is not uncommon.A survey conducted by LANL of notifiable disease data reportingin 52 countries identified only 10 as being machine-readable and42 being reported in pdf files on a regular basis. Within the 42 nationsthat report in pdf files, 32 report in a structured, tabular format and10 in a non-structured way.As a result, LANL has developed a tool-Epi Archive (formerlyknown as EPIC)-to automatically and continuously collect globalnotifiable disease data and make it readily accesible.MethodsWe conducted a survey of the national notifiable disease reportingsystems notating how the data is reported in two important dimensions:date standards and case definitions.The development of software to regularly ingests notifiabledisease data frand makes this data available involved four main stepsscraping, extracting, parsing and persisting.For scraping: we would examine website designs and determinereporting mechanisms for each country/website as well as what variesacross the reporting mechanisms. We then designed and wrote codeto automate the downloading of report pdf files, for each country.We stored report pdfs along with appropriate metadata for extractingand parsing.For extracting: we developed software that can extract notifiabledisease data presented in tabular form from a pdf file. We combinedthe methodology of figure placement detection with the in-housedeveloped table extraction and annotation heuristics.For parsing: we determined what to extract from each pdf dataset from the survey conducted. We then parsed the extracted datainto uniform data structures correctly accommodating the dimensionssurveyed and the various human languages. This task involvedingesting notifiable disease data in many disparate formats extractedfrom pdf files and coalescing the data into a standardized format.For persisting: We then store the data in the Epi ArchivePostgreSQL database and make it available through the BSVE.ResultsThe EpiArchive tool currently contains subnational notifiabledisease data from 10 nations. When a user accesses the EpiArchivesite, they are prompted with four fields: country, region, disease,and date duration. These fields allow the user to specify the location(down to the state level), the disease of interest, and the durationof interest. Upon form submission, a time series is generated fromthe users’ specifications. The generated time series can then bedownloaded into a csv file if a user is interested in performingpersonal analysis. Additionally, the data from EpiArchive can bereached through an API.ConclusionsLANL as part of a currently funded DTRA effort so that it willautomatically and continuously collect global notifiable diseasedata—particularly data stored in pdf files—and make it available andshareable within the Biosurveillance Ecosystem (BSVE) as a newdata source. This will provide data to analytics and users that willimprove the prediction and early warning of disease events and otherapplications.