Kulessa, Moritz Alexander Claus (2022)
Data-driven Disease Surveillance.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00020415
Dissertation, Erstveröffentlichung, Verlagsversion
Kurzbeschreibung (Abstract)
The recent and still ongoing pandemic of SARS-CoV-2 has shown that an infectious disease outbreak can have serious consequences on public health and economy. In this situation, public health officials constantly aim to control and reduce the number of infections in order to avoid overburdening health care system. Besides minimizing personal contact through political measures, a fundamental approach to contain the spread of diseases is to isolate infected individuals. The effectiveness of the latter approach strongly depends on a timely detection of the outbreak as the tracking of individuals can quickly become infeasible when the number of cases increases. Hence, a key factor in the containment of an infectious disease is the early detection of a potential larger outbreak, commonly known as outbreak detection.
For this purpose, epidemiologists rely on a variety of statistical surveillance methods in order to maintain an overview of the current situation of infections by either monitoring confirmed cases or cases with early symptoms. Mainly based on statistical hypothesis testing, these methods automatically raise an alarm if an unexpected increase in the number of infections is observed. The practical usefulness of such methods highly depends on the trade-off between the ability to detect outbreaks and the chances of raising a false alarm. However, this hypothesis-based approach to disease surveillance has several limitations. On the one hand, it is a hand-crafted approach which requires domain knowledge to set up the statistical methods, especially if early symptoms are monitored. On the other hand, outbreaks of emerging infectious diseases with different symptom patterns are likely to be missed by such a surveillance system.
In this thesis, we focus on data-driven disease surveillance and address these challenges in the following ways. To support epidemiologists in the process of defining reliable disease patterns for monitoring cases with early symptoms, we present a novel approach to discover such patterns in historic data. With respect to supervised learning, we propose a fusion classifier which can combine the output of multiple statistical methods using the univariate time series of infection counts as the only source of information. In addition, we develop algorithms based on unsupervised learning which frame the task of outbreak detection as a general anomaly detection task. This even includes the surveillance of emerging infectious diseases. Therefore, we contribute a novel framework and propose a new approach based on sum-product networks to monitor multiple disease patterns simultaneously. Our results show that data-driven approaches are ideal to assist epidemiologists by processing large amounts of data that cannot fully be understood and analyzed by humans. Most significantly, the incorporation of additional information into the surveillance through machine learning techniques shows reliable and promising results.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2022 | ||||
Autor(en): | Kulessa, Moritz Alexander Claus | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Data-driven Disease Surveillance | ||||
Sprache: | Englisch | ||||
Referenten: | Binnig, Prof. Dr. Carsten ; Fürnkranz, Prof. Dr. Johannes ; Spiliopoulou, Prof. Dr. Myra | ||||
Publikationsjahr: | 2022 | ||||
Ort: | Darmstadt | ||||
Kollation: | xi, 137 Seiten | ||||
Datum der mündlichen Prüfung: | 17 Dezember 2021 | ||||
DOI: | 10.26083/tuprints-00020415 | ||||
URL / URN: | https://tuprints.ulb.tu-darmstadt.de/20415 | ||||
Kurzbeschreibung (Abstract): | The recent and still ongoing pandemic of SARS-CoV-2 has shown that an infectious disease outbreak can have serious consequences on public health and economy. In this situation, public health officials constantly aim to control and reduce the number of infections in order to avoid overburdening health care system. Besides minimizing personal contact through political measures, a fundamental approach to contain the spread of diseases is to isolate infected individuals. The effectiveness of the latter approach strongly depends on a timely detection of the outbreak as the tracking of individuals can quickly become infeasible when the number of cases increases. Hence, a key factor in the containment of an infectious disease is the early detection of a potential larger outbreak, commonly known as outbreak detection. For this purpose, epidemiologists rely on a variety of statistical surveillance methods in order to maintain an overview of the current situation of infections by either monitoring confirmed cases or cases with early symptoms. Mainly based on statistical hypothesis testing, these methods automatically raise an alarm if an unexpected increase in the number of infections is observed. The practical usefulness of such methods highly depends on the trade-off between the ability to detect outbreaks and the chances of raising a false alarm. However, this hypothesis-based approach to disease surveillance has several limitations. On the one hand, it is a hand-crafted approach which requires domain knowledge to set up the statistical methods, especially if early symptoms are monitored. On the other hand, outbreaks of emerging infectious diseases with different symptom patterns are likely to be missed by such a surveillance system. In this thesis, we focus on data-driven disease surveillance and address these challenges in the following ways. To support epidemiologists in the process of defining reliable disease patterns for monitoring cases with early symptoms, we present a novel approach to discover such patterns in historic data. With respect to supervised learning, we propose a fusion classifier which can combine the output of multiple statistical methods using the univariate time series of infection counts as the only source of information. In addition, we develop algorithms based on unsupervised learning which frame the task of outbreak detection as a general anomaly detection task. This even includes the surveillance of emerging infectious diseases. Therefore, we contribute a novel framework and propose a new approach based on sum-product networks to monitor multiple disease patterns simultaneously. Our results show that data-driven approaches are ideal to assist epidemiologists by processing large amounts of data that cannot fully be understood and analyzed by humans. Most significantly, the incorporation of additional information into the surveillance through machine learning techniques shows reliable and promising results. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
Status: | Verlagsversion | ||||
URN: | urn:nbn:de:tuda-tuprints-204152 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 000 Allgemeines, Informatik, Informationswissenschaft > 004 Informatik | ||||
Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Data Management (2022 umbenannt in Data and AI Systems) |
||||
TU-Projekte: | G-BA|01VSF17034|ESEG | ||||
Hinterlegungsdatum: | 14 Mär 2022 13:18 | ||||
Letzte Änderung: | 15 Mär 2022 09:16 | ||||
PPN: | |||||
Referenten: | Binnig, Prof. Dr. Carsten ; Fürnkranz, Prof. Dr. Johannes ; Spiliopoulou, Prof. Dr. Myra | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 17 Dezember 2021 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |