Navigation and service

Use of cookies

Cookies help us to provide our services. By using our website you agree that we can use cookies. Read more about our Privacy Policy and visit the following link: Privacy Policy

OK

Signale - Projects

The Signale project is supported by the Federal Ministry of Health. Its mission and objectives are defined in the application "DEMIS Signale 2" linked below at the end of this page (in German). We are also partner in the "ESEG" project.

Focus Group AI for Health, Topic Group Outbreaks

The FG-AI4H is an initiative of the International Telecommunication Union (ITU) and the World Health Organization (WHO). It aims at developing standards for the evaluation of artificial intelligence (AI) approaches in health. We coordinate the subgroup TG Outbreaks, which addresses the benchmarking of outbreak detection algorithms. Further information under https://www.itu.int/en/ITU-T/focusgroups/ai4h/Pages/default.aspx

Signal reports

The core mission of the Signale team is the development of the early-warning system of the RKI. A detailed description (as of May 2015) has been published in Eurosurveillance and is attached at the end of this page.

The early-warning system aims at detecting notable events automatically and quickly from surveillance data. It is mainly about possible outbreaks, but other "anomalies" can be of interest. The strength of the system is to indicate quickly, through the analysis of a large quantity of data, where a closer epidemiological investigation might be useful.

Concretely a signal is defined as a "statistically unlikely accumulation of cases within a week". This is an indication for a possible outbreak, not a proof.

The signal detection works as follows:

1. a "filter combination" is chosen, i.e. a combination of properties like pathogen, sex or reporting place;
2. all reported cases that correspond to the filter combination are aggregated in a time series;
3. the number of cases that can be expected today for this time series under normal conditions is estimated;
4. an upper bound is determined, above which the case count of today can be considered abnormal;
5. if the observed case is indeed above the upper bound, a signal is generated for the given filter combination.

These steps are done for a large number of different filter combinations.

The signal reports are provided to our colleagues in public-health agencies. Please contact us should you be interested in receiving them!

The signal reports give an overview of the signals of the last 6 weeks for a selection of diseases. Source: RKI The signal reports give an overview of the signals of the last 6 weeks for a selection of diseases. They consist in an HTML file containing two tables, a map, and two bar charts for age and sex. The reports are interactive: One can sort and filter the tables; zoom in the map, show either case count or incidence, as well as county or state names; and much more.

Machine learning

We apply and adapt methods of machine learning to improve outbreak detection, evaluate and optimise existing algorithms, and gain new insights about how infectious diseases spread, e.g. through the integration of external data.

Interactive dashboards

We develop custom automatic analyses and interactive visualisations, made available within the RKI as dashboards. They allow the dynamic exploration of the epidemic situation and the uncovering of correlations.

BeSSA: Berliner Syndromische Surveillance in Notunterkünften für Asylsuchende

For the BeSSA project we have developed tools for outbreak detection and automatic reporting and advised domain experts on statistical questions. A description of the project and the tool for outbreak detection can be found at http://www.rki.de/synsurvrefugees

Date: 23.01.2020

Publications

  • Stojanović O, Leugering J, Pipa G, Ghozzi S, Ullrich A (2019): A Bayesian Monte Carlo approach for predicting the spread of infectious diseases
    PLoS ONE 14 (12): e0225838. Epub Dec 18. doi: 10.1371/journal.pone.0225838. more

  • Abbood A (2019): Automatic Information Extraction and Relevance Evaluation of Epidemiological Texts Using Natural Language Processing
    Berlin/Osnabrück: Robert Koch Institute/Osnabrück (master thesis).
    (PDF, 2 MB)

  • Abbood A, Ullrich A, Busche R, Ghozzi S (2019): EventEpi — A Natural Language Processing Framework for Event-Based Surveillance.
    medRxiv: 19006395. Epub Sep 26. doi: 10.1101/19006395. more

  • Busche R (2019): Systematic Evaluation and Optimization of Outbreak-Detection Algorithms Based on Labeled Epidemiological Surveillance Data
    Osnabrück: Universität Osnabrück, Fachbereich Humanwissenschaften, Institute of Cognitive Science, master thesis.
    (PDF, 8 MB)

  • Sarma N, Ullrich A, Wilking H, Ghozzi S, Lindner AK, Weber C, Holzer A, Jansen A, Stark K, Vygen-Bonnet S (2018): Surveillance on speed: Being aware of infectious diseases in migrants mass accommodations – an easy and flexible toolkit for field application of syndromic surveillance, Germany, 2016 to 2017
    Euro Surveill. 23 (40): pii=1700430. doi: 10.2807/1560-7917.ES.2018.23.40.1700430. more

  • Salmon M, Schumacher D, Burmann H, Frank C, Claus H, Höhle M (2016): A system for automated outbreak detection of communicable diseases in Germany
    Euro Surveill. 21 (13): pii=30180. doi: 10.2807/1560-7917.ES.2016.21.13.30180. more