Article
New methods for public health surveillance – a project summary
Search Medline for
Authors
Published: | September 13, 2012 |
---|
Outline
Text
Background: Detecting hints to public health threats as early as possible is crucial to prevent harm from the population. In indicator-based surveillance, data transmitted from hospitals, laboratories or physicians, drug subscriptions, emergency room visits etc. are monitored to become observe status changes in a population's health. Health organisations become more and more aware that other sources of information might provide such hints earlier. In event-driven surveillance, additional sources are monitored including online news or official reports.
Methods: In the M-Eco project (http://www.meco-project.eu/, [1]), we assessed the usefulness of social media and multi media data for supporting disease surveillance. For this purpose, methods have been developed that collect continously data from Twitter, blogs, and forums as well as from TV and radio channels. The data is annotated with linguistic information, disease names, persons and locations are identified. These features provide the input to machine learning algorithms that detect patterns in the data. By means of statistical methods such as the CUSUM method, the patterns are analysed and signals are generated automatically when unexpected behaviour is determined. A signal is a hint to some anomalous behaviour. Since the amount of signals generated can overwhelm a user, recommenation techniques are exploited to filter out those signals that are of potential interest to a user [2]. The information related to a signal is shown in charts and through personalized tag clouds to allow users to easily assess signals.
Results: Several health organizations were involved in developing and evaluating the M-Eco system, including the Robert Koch-Institut (Germany), WHO and ECDC, Institut de Veille Sanitaire and the organisation Mekong Basin Disease Surveillance. For some outbreaks located in Germany, such as the EHEC outbreak in 2011, it could be shown, that social media provides information earlier than it is available through official channels [3]. Another lesson learnt is that at least additional information can be found in social media that can support health officials in assessing the associated risk.
Conclusions: The methods are made available via web services for their easy integration into existing systems such as BioCaster, or HealthMap. Within the project, they were already integrated into the MedISys system. The current focus is on health events that are reported in German or English. However, the methods are adaptable to other languages as well. Future use of the system by health organizations will show the usefulness of social media in public health surveillance.
References
- 1.
- Denecke K, Kirchner G, Dolog P, Smrz P, Linge JP, Backfried G, Dreesman J. Event-driven Architecture for Health Event Detection from Multiple Sources. In: XXIII International Conference of the European Federation for Medical Informatics (MIE 2011); 2011 Aug 28-31; Oslo, Norway.
- 2.
- Lage R, Durao F, Dolog P. Towards Effective Group Recommendations for Microblogging Users. In: 27th Symposium on Applied Computing; 2012 March 26-30; Riva del Garda (Trento), Italy.
- 3.
- Diaz-Aviles E, Stewart A, Velasco E, Denecke K, Nejdl W. Towards Personalized Learning to Rank for Epidemic Intelligence Based on Social Media Streams. In: World Wide Web Conference; 2012; Lyon, France.