Developing Algorithms forDeveloping Algorithms for Early ...
Post on 21-Apr-2022
13 Views
Preview:
Transcript
SIDARTHaSIDARTHaSIDARTHaSIDARTHa
European Emergency DataEuropean Emergency DataEuropean Emergency DataEuropean Emergency Data----based Syndromic Surveillance Systembased Syndromic Surveillance Systembased Syndromic Surveillance Systembased Syndromic Surveillance System
Grant Agreement No. 2007208Grant Agreement No. 2007208Grant Agreement No. 2007208Grant Agreement No. 2007208
Developing Algorithms forDeveloping Algorithms forDeveloping Algorithms forDeveloping Algorithms for Early Public Early Public Early Public Early Public
Health Threat DetectionHealth Threat DetectionHealth Threat DetectionHealth Threat Detection in in in in EuropeEuropeEuropeEurope ReReReResults from the SIDARTHa projectsults from the SIDARTHa projectsults from the SIDARTHa projectsults from the SIDARTHa project
Draft report (January 2010)Draft report (January 2010)Draft report (January 2010)Draft report (January 2010)
Developing Algorithms for Early Public Health Threat Detection in Europe ii
© SIDARTHa 2010 DRAFT January 2010
SIDARTHa - European Emergency Data-based Syndromic Surveillance System
The project ‘European Emergency Data-based System for Information on, Analysis and Detection of Risks and Threats to Health – SIDARTHa’ is co-funded by the European Commission under the Programme of Community Action in the Field of Public Health 2003-2008 (Grant Agreement-No.: 2007208).
SIDARTHa Steering Committee
Luis Garcia-Castrillo Riesgo (Project Leader), Thomas Krafft (Scientific-Technical Coordinator), Matthias Fischer, Alexander Krämer, Freddy Lippert, Gernot Vergeiner SIDARTHa Project Group
Dispatch Centre Tyrol (Austria), contact person: Gernot Vergeiner; Federal Government, Department of Public Health (Belgium), contact person: Agnes Meulemans; Emergency Medical Service Prague (Czech Republic), contact person: Milana Pokorna; Capital Region (Denmark), contact person: Freddy Lippert; University Hospital Kuopio (Finland), contact person: Jouni Kurola; Emergency Medical Service Province Hauts de Seine (France), contact person: Michel Baer; Hospitals of County of Goeppingen (Germany), contact person: Matthias Fischer; GEOMED Research Forschungsgesellschaft mbH (Germany), contact person: Thomas Krafft; University of Bielefeld, Department of Public Health Medicine (Germany), contact person: Alexander Krämer; National Emergency Medical Service (Hungary), contact person: Gabor Göbl; San Martino University Hospital Genoa (Italy), contact person: Francesco Bermano; Haukeland University Hospital Bergen (Norway), contact person: Guttorm Brattebo; University of Cantabria (Spain), contact person: Luis Garcia-Castrillo Riesgo; University Hospital Antalya (Turkey), contact person: Hakan Yaman Advisory Board
Helmut Brand (The Netherlands, Chair), Andrea Ammon (ECDC), Enrico Davoli (WHO-Euro), Per Kulling (EU Health Threat - Unit), Javier Llorca (Spain), Jerry Overton (USA), Santiago Rodriguez (Spain), Mark Rosenberg (Canada) Coordination Office
Alexandra Ziemann (Science Officer), Weyma Notel (Project Assistant), Juan-José San Miguel Roncero (Financial Officer)
Developing Algorithms for Early Public Health Threat DetectionDeveloping Algorithms for Early Public Health Threat DetectionDeveloping Algorithms for Early Public Health Threat DetectionDeveloping Algorithms for Early Public Health Threat Detection in Europein Europein Europein Europe
Results from the SIDARTHa project.
Draft report (January 2010)
This report describes the methodology of developing detection algorithms for the European syndromic surveillance system for early public health threat detection and risk communication SIDARTHa and forms deliverable D6 as defined in the Grant Agreement. Compiled by
Nicole Rosenkötter, Janneke Kraan, Alexandra Ziemann, Martina Schorbahn, Genc Burazeri, Helmut Brand
Editors
Nicole Rosenkötter, Janneke Kraan, Alexandra Ziemann, Martina Schorbahn, Genc Burazeri, Helmut Brand, Thomas Krafft, Tim Tenelsen, Luis Garcia-Castrillo Riesgo, Matthias Fischer, Alexander Krämer, Freddy Lippert, Gernot Vergeiner for the SIDARTHa project group
Please cite as:
Rosenkötter N, Kraan J, Ziemann A, Schorbahn M, Burazeri G, Brand H, Krafft T, Tenelsen T, Garcia-Castrillo Riesgo L, Fischer M, Krämer A, Lippert F,
Vergeiner G, for the SIDARTHa project group (ed.) (2010): Developing Algorithms for Early Public Health Threat Detection in Europe – Results from
the SIDARTHa project. Draft Report (January 2010). Bad Honnef.
Cover Figure
Cluster detection based on emergency data (own creation)
© SIDARTHa 20© SIDARTHa 20© SIDARTHa 20© SIDARTHa 2010101010
SIDARTHa Scientific/Technical Coordination Office, c/o GEOMED Research Forschungsgesellschaft mbH, Hauptstr. 68, D-53604 Bad Honnef, Tel. +49 2224 7799896, Fax. +49 2224 7799897, mail@sidartha.eu, www.sidartha.eu
Developing Algorithms for Early Public Health Threat Detection in Europe iii
© SIDARTHa 2010 DRAFT January 2010
ContentsContentsContentsContents
CCCCONTENTSONTENTSONTENTSONTENTS .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... IIIIIIIIIIII
FFFFIGURESIGURESIGURESIGURES ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ IIIIVVVV
AAAABBREVIATIONSBBREVIATIONSBBREVIATIONSBBREVIATIONS ........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ VVVV
AAAACKNOWLEDGEMENTSCKNOWLEDGEMENTSCKNOWLEDGEMENTSCKNOWLEDGEMENTS ........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ VVVVIIIIIIII
1111 INTRODUCTION: THE SIINTRODUCTION: THE SIINTRODUCTION: THE SIINTRODUCTION: THE SIDARTHA PROJECTDARTHA PROJECTDARTHA PROJECTDARTHA PROJECT.................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 1111
2222 BACKGROUND, OBJECTIVBACKGROUND, OBJECTIVBACKGROUND, OBJECTIVBACKGROUND, OBJECTIVES & METHODOLOGYES & METHODOLOGYES & METHODOLOGYES & METHODOLOGY .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 3333
2.12.12.12.1 BBBBACKGROUNDACKGROUNDACKGROUNDACKGROUND:::: AAAALGORITHMS FLGORITHMS FLGORITHMS FLGORITHMS FOR OR OR OR EEEEARLY ARLY ARLY ARLY HHHHEALTH EALTH EALTH EALTH TTTTHREAT HREAT HREAT HREAT DDDDETECTIONETECTIONETECTIONETECTION ................................................................................................................................................................................................................................................................................................................................................................................ 3333
2.22.22.22.2 OOOOBJECTIVESBJECTIVESBJECTIVESBJECTIVES .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 4444
2.32.32.32.3 MMMMETHODOLOGYETHODOLOGYETHODOLOGYETHODOLOGY:::: IIIINTRODUCTIONNTRODUCTIONNTRODUCTIONNTRODUCTION ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 5555
2.42.42.42.4 PPPPREPARATION OF THE REPARATION OF THE REPARATION OF THE REPARATION OF THE DDDDATA ATA ATA ATA SSSSETSETSETSETS .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 6666
2.52.52.52.5 SIDARTHSIDARTHSIDARTHSIDARTHA A A A SSSSTANDARD TANDARD TANDARD TANDARD DDDDATA ATA ATA ATA SSSSETETETET .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 6666
2.62.62.62.6 SSSSYNDROME YNDROME YNDROME YNDROME GGGGENERATIONENERATIONENERATIONENERATION .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 6666
2222.7.7.7.7 DDDDESCRIPTIVE ESCRIPTIVE ESCRIPTIVE ESCRIPTIVE AAAANALYSISNALYSISNALYSISNALYSIS ............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 6666
2.82.82.82.8 DDDDETECTION ETECTION ETECTION ETECTION AAAALGORITHMSLGORITHMSLGORITHMSLGORITHMS .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 6666
2.92.92.92.9 SSSSOFTWAREOFTWAREOFTWAREOFTWARE ............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 7777
3333 APPLYING DETECTION AAPPLYING DETECTION AAPPLYING DETECTION AAPPLYING DETECTION ALGORITHMS IN THE SIDLGORITHMS IN THE SIDLGORITHMS IN THE SIDLGORITHMS IN THE SIDARTHA IMPLEMENTATIONARTHA IMPLEMENTATIONARTHA IMPLEMENTATIONARTHA IMPLEMENTATION SITES: FIRST RESULTSSITES: FIRST RESULTSSITES: FIRST RESULTSSITES: FIRST RESULTS ............................................................................................................................................................................................................ 8888
3.13.13.13.1 PPPPREPARATIONREPARATIONREPARATIONREPARATION:::: IIIIMPLEMENTATION SITES MPLEMENTATION SITES MPLEMENTATION SITES MPLEMENTATION SITES &&&& DATA SOURCESDATA SOURCESDATA SOURCESDATA SOURCES ................................................................................................................................................................................................................................................................................................................................................................................................ 8888
3.23.23.23.2 SIDARTHSIDARTHSIDARTHSIDARTHA A A A SSSSTANDARD TANDARD TANDARD TANDARD DDDDATA ATA ATA ATA SSSSETETETET .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 10101010
3.33.33.33.3 SSSSYNDROME YNDROME YNDROME YNDROME GGGGENERATIONENERATIONENERATIONENERATION ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 11111111
3.43.43.43.4 DDDDESCRIPTIVE ANALYSISESCRIPTIVE ANALYSISESCRIPTIVE ANALYSISESCRIPTIVE ANALYSIS ........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 12121212
3.53.53.53.5 RRRRESULTS OF THE DETECTESULTS OF THE DETECTESULTS OF THE DETECTESULTS OF THE DETECTIIIION ALGORITHM ON ALGORITHM ON ALGORITHM ON ALGORITHM C1,C1,C1,C1, C2,C2,C2,C2, C3C3C3C3 ............................................................................................................................................................................................................................................................................................................................................................................................................ 14141414
4444 SUMMARY & NEXT STEPSSUMMARY & NEXT STEPSSUMMARY & NEXT STEPSSUMMARY & NEXT STEPS ........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 16161616
RRRREFERENCESEFERENCESEFERENCESEFERENCES ................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 17171717
Developing Algorithms for Early Public Health Threat Detection in Europe iv
© SIDARTHa 2010 DRAFT January 2010
FiguresFiguresFiguresFigures
FFFFIGURE IGURE IGURE IGURE 1:1:1:1: SIDARTHSIDARTHSIDARTHSIDARTHA A A A PPPPROJECT ROJECT ROJECT ROJECT MMMMETHODOLOGYETHODOLOGYETHODOLOGYETHODOLOGY .................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................... 2222
FFFFIGURE IGURE IGURE IGURE 2:2:2:2: SIDARTHSIDARTHSIDARTHSIDARTHA A A A AAAAPPROACHPPROACHPPROACHPPROACH ............................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................ 2222
Developing Algorithms for Early Public Health Threat Detection in Europe v
© SIDARTHa 2010 DRAFT January 2010
AbbreviationsAbbreviationsAbbreviationsAbbreviations
14 14 14 14 ---- AKFAM AKFAM AKFAM AKFAM –––– TRTRTRTR SIDARTHa associated partner abbreviation foSIDARTHa associated partner abbreviation foSIDARTHa associated partner abbreviation foSIDARTHa associated partner abbreviation for the University Hospital Antalya, Turkeyr the University Hospital Antalya, Turkeyr the University Hospital Antalya, Turkeyr the University Hospital Antalya, Turkey
AMPDSAMPDSAMPDSAMPDS Advanced Medical Priority Dispatch SystemAdvanced Medical Priority Dispatch SystemAdvanced Medical Priority Dispatch SystemAdvanced Medical Priority Dispatch System
ARIMAARIMAARIMAARIMA Autoregressive integrated moving averageAutoregressive integrated moving averageAutoregressive integrated moving averageAutoregressive integrated moving average
ATATATAT AustriaAustriaAustriaAustria
C 1, C2, C3C 1, C2, C3C 1, C2, C3C 1, C2, C3 Detection AlgorithmDetection AlgorithmDetection AlgorithmDetection Algorithm used in the Early Abused in the Early Abused in the Early Abused in the Early Abererererrrrration Reporting Systemation Reporting Systemation Reporting Systemation Reporting System
CDCCDCCDCCDC Centers for DiseCenters for DiseCenters for DiseCenters for Disease Control and Preventionase Control and Preventionase Control and Preventionase Control and Prevention
CICICICI Confidence IntervalConfidence IntervalConfidence IntervalConfidence Interval
CUSUMCUSUMCUSUMCUSUM Cumulative SumCumulative SumCumulative SumCumulative Sum
D6D6D6D6 Deliverable No. Deliverable No. Deliverable No. Deliverable No. 6666 of the SIDARTHa projectof the SIDARTHa projectof the SIDARTHa projectof the SIDARTHa project
DEDEDEDE GermanyGermanyGermanyGermany
EARSEARSEARSEARS Early Aberration Reporting SystemEarly Aberration Reporting SystemEarly Aberration Reporting SystemEarly Aberration Reporting System
ECDCECDCECDCECDC European Centre for Disease Prevention and ControlEuropean Centre for Disease Prevention and ControlEuropean Centre for Disease Prevention and ControlEuropean Centre for Disease Prevention and Control
EDEDEDED Emergency DepartmEmergency DepartmEmergency DepartmEmergency Departmentententent
EEDEEDEEDEED European Emergency DataEuropean Emergency DataEuropean Emergency DataEuropean Emergency Data
EMDEMDEMDEMD Emergency Medical DispatchEmergency Medical DispatchEmergency Medical DispatchEmergency Medical Dispatch
EMSEMSEMSEMS Emergency Medical Service Emergency Medical Service Emergency Medical Service Emergency Medical Service
EPEPEPEP Emergency PhysicianEmergency PhysicianEmergency PhysicianEmergency Physician
ESSENCEESSENCEESSENCEESSENCE Surveillance System for the Early Notification of CommunitySurveillance System for the Early Notification of CommunitySurveillance System for the Early Notification of CommunitySurveillance System for the Early Notification of Community----based Epidemicsbased Epidemicsbased Epidemicsbased Epidemics
EUEUEUEU European UnionEuropean UnionEuropean UnionEuropean Union
8 8 8 8 –––– FOD Health DG 1 FOD Health DG 1 FOD Health DG 1 FOD Health DG 1 –––– BE BE BE BE SIDARTHa associated partner abbreviation for the federal public service health, food chain safety and environment, SIDARTHa associated partner abbreviation for the federal public service health, food chain safety and environment, SIDARTHa associated partner abbreviation for the federal public service health, food chain safety and environment, SIDARTHa associated partner abbreviation for the federal public service health, food chain safety and environment,
BelgiumBelgiumBelgiumBelgium
2 2 2 2 ---- GEOMED GEOMED GEOMED GEOMED –––– DE DE DE DE SIDARTHa associated partner abbreviation for the GEOMED Research Forschungsgesellschaft mbH, GermanySIDARTHa associated partner abbreviation for the GEOMED Research Forschungsgesellschaft mbH, GermanySIDARTHa associated partner abbreviation for the GEOMED Research Forschungsgesellschaft mbH, GermanySIDARTHa associated partner abbreviation for the GEOMED Research Forschungsgesellschaft mbH, Germany
GISGISGISGIS GeograpGeograpGeograpGeographic Information Systemhic Information Systemhic Information Systemhic Information System
GLMGLMGLMGLM Generalized Linear ModelGeneralized Linear ModelGeneralized Linear ModelGeneralized Linear Model
GLMMGLMMGLMMGLMM Generalized Linerar Mixed ModelGeneralized Linerar Mixed ModelGeneralized Linerar Mixed ModelGeneralized Linerar Mixed Model
GPSGPSGPSGPS Global Positioning SystemGlobal Positioning SystemGlobal Positioning SystemGlobal Positioning System
HPRHPRHPRHPR Highest Priority ResponseHighest Priority ResponseHighest Priority ResponseHighest Priority Response
16 16 16 16 –––– HSanMartino HSanMartino HSanMartino HSanMartino –––– ITITITIT SIDARTHa associated partner abbreviation for the San Martino University HospSIDARTHa associated partner abbreviation for the San Martino University HospSIDARTHa associated partner abbreviation for the San Martino University HospSIDARTHa associated partner abbreviation for the San Martino University Hospital Genoa, Italyital Genoa, Italyital Genoa, Italyital Genoa, Italy
17 17 17 17 –––– HUS HUS HUS HUS –––– NONONONO SIDARTHa associated partner abbreviation for the Haukeland University Hospital Bergen, NorwaySIDARTHa associated partner abbreviation for the Haukeland University Hospital Bergen, NorwaySIDARTHa associated partner abbreviation for the Haukeland University Hospital Bergen, NorwaySIDARTHa associated partner abbreviation for the Haukeland University Hospital Bergen, Norway
ICDICDICDICD International Classification of DiseasesInternational Classification of DiseasesInternational Classification of DiseasesInternational Classification of Diseases
ILIILIILIILI InfluenzaInfluenzaInfluenzaInfluenza----LikeLikeLikeLike----IllnessIllnessIllnessIllness
4 4 4 4 –––– ILL GmbH ILL GmbH ILL GmbH ILL GmbH –––– AU AU AU AU SIDARTHa associated partner abbreSIDARTHa associated partner abbreSIDARTHa associated partner abbreSIDARTHa associated partner abbreviation for the Dispatch Centre Tyrol, Austriaviation for the Dispatch Centre Tyrol, Austriaviation for the Dispatch Centre Tyrol, Austriaviation for the Dispatch Centre Tyrol, Austria
7 7 7 7 –––– KAE KAE KAE KAE –––– DE DE DE DE SIDARTHa associated partner abbreviation for the Klinik am Eichert (Clinics of the County of Goeppingen), GermanySIDARTHa associated partner abbreviation for the Klinik am Eichert (Clinics of the County of Goeppingen), GermanySIDARTHa associated partner abbreviation for the Klinik am Eichert (Clinics of the County of Goeppingen), GermanySIDARTHa associated partner abbreviation for the Klinik am Eichert (Clinics of the County of Goeppingen), Germany
9 9 9 9 –––– KUH KUH KUH KUH –––– FIFIFIFI SIDARTHa associated partner abbreviation for the University HospitSIDARTHa associated partner abbreviation for the University HospitSIDARTHa associated partner abbreviation for the University HospitSIDARTHa associated partner abbreviation for the University Hospital Kuopio, Finlandal Kuopio, Finlandal Kuopio, Finlandal Kuopio, Finland
MMMM Month of the SIDARTHa projectMonth of the SIDARTHa projectMonth of the SIDARTHa projectMonth of the SIDARTHa project
MINDMINDMINDMIND Minimaler Notarztdatensatz (minimum emergency physician data set)Minimaler Notarztdatensatz (minimum emergency physician data set)Minimaler Notarztdatensatz (minimum emergency physician data set)Minimaler Notarztdatensatz (minimum emergency physician data set)
nHPRnHPRnHPRnHPR NonNonNonNon----Highest Priority ResponseHighest Priority ResponseHighest Priority ResponseHighest Priority Response
NYCDONYCDONYCDONYCDOHHHH New York City Department of HealthNew York City Department of HealthNew York City Department of HealthNew York City Department of Health
5 5 5 5 –––– OMSZ OMSZ OMSZ OMSZ –––– HU HU HU HU SIDARTHa associated partner abbreSIDARTHa associated partner abbreSIDARTHa associated partner abbreSIDARTHa associated partner abbreviation for the viation for the viation for the viation for the National Emergency Medical Service HungaryNational Emergency Medical Service HungaryNational Emergency Medical Service HungaryNational Emergency Medical Service Hungary
3 3 3 3 –––– RegH RegH RegH RegH –––– DK DK DK DK SIDARTHa associated partner abbreviation for the Capital Region DenmarkSIDARTHa associated partner abbreviation for the Capital Region DenmarkSIDARTHa associated partner abbreviation for the Capital Region DenmarkSIDARTHa associated partner abbreviation for the Capital Region Denmark
Developing Algorithms for Early Public Health Threat Detection in Europe vi
© SIDARTHa 2010 DRAFT January 2010
RLSRLSRLSRLS Recursive Least SquareRecursive Least SquareRecursive Least SquareRecursive Least Square
RODSRODSRODSRODS RealRealRealReal----time time time time OutbreakOutbreakOutbreakOutbreak and Disease Surveillance and Disease Surveillance and Disease Surveillance and Disease Surveillance
6 6 6 6 –––– SAMU SAMU SAMU SAMU –––– FR FR FR FR SIDARTHa assSIDARTHa assSIDARTHa assSIDARTHa associated partner abbreviation for the System of Emergency Medical Assistance Garches, Franceociated partner abbreviation for the System of Emergency Medical Assistance Garches, Franceociated partner abbreviation for the System of Emergency Medical Assistance Garches, Franceociated partner abbreviation for the System of Emergency Medical Assistance Garches, France
SIDARTHaSIDARTHaSIDARTHaSIDARTHa European Emergency DataEuropean Emergency DataEuropean Emergency DataEuropean Emergency Data----based System for Information on, Detection and Analysis of Risks and Threats to Healthbased System for Information on, Detection and Analysis of Risks and Threats to Healthbased System for Information on, Detection and Analysis of Risks and Threats to Healthbased System for Information on, Detection and Analysis of Risks and Threats to Health
SMARTSMARTSMARTSMART Small Area Regression and TestingSmall Area Regression and TestingSmall Area Regression and TestingSmall Area Regression and Testing
SPO2SPO2SPO2SPO2 Pulse oximetryPulse oximetryPulse oximetryPulse oximetry
15 15 15 15 –––– UNIBI UNIBI UNIBI UNIBI –––– DE DE DE DE SIDARTHa associated partner abbreviation for the University of Bielefeld, GermanySIDARTHa associated partner abbreviation for the University of Bielefeld, GermanySIDARTHa associated partner abbreviation for the University of Bielefeld, GermanySIDARTHa associated partner abbreviation for the University of Bielefeld, Germany
1 1 1 1 –––– UNICAN UNICAN UNICAN UNICAN –––– ES ES ES ES SIDARTHa associated partner abbreviation for the University of Cantabria, SpainSIDARTHa associated partner abbreviation for the University of Cantabria, SpainSIDARTHa associated partner abbreviation for the University of Cantabria, SpainSIDARTHa associated partner abbreviation for the University of Cantabria, Spain
USAUSAUSAUSA United States of AmericaUnited States of AmericaUnited States of AmericaUnited States of America
WPWPWPWP Work Package of the SIDARTHa projectWork Package of the SIDARTHa projectWork Package of the SIDARTHa projectWork Package of the SIDARTHa project
WHOWHOWHOWHO World Health OrganizationWorld Health OrganizationWorld Health OrganizationWorld Health Organization
13 13 13 13 –––– ZZSHMP ZZSHMP ZZSHMP ZZSHMP –––– USZS USZS USZS USZS –––– CZ CZ CZ CZ SIDARTHa associated partner abbreviation for the Emergency Medical Service Prague, Czech RepublicSIDARTHa associated partner abbreviation for the Emergency Medical Service Prague, Czech RepublicSIDARTHa associated partner abbreviation for the Emergency Medical Service Prague, Czech RepublicSIDARTHa associated partner abbreviation for the Emergency Medical Service Prague, Czech Republic
Developing Algorithms for Early Public Health Threat Detection in Europe vii
© SIDARTHa 2010 DRAFT January 2010
AcknowledgementsAcknowledgementsAcknowledgementsAcknowledgements
This report presents the results of the sixth Work Package
(WP) of the European Commission co-funded project
“SIDARTHa – European Emergency Data-based Syndromic
Surveillance System” and forms deliverable D6 as defined in
the Grant Agreement (No. 2007208). The results presented
in this report were compiled by the leader of the Dutch
Country Consortium and the Scientific/Technical Coordination
Office. The authors would like to thank all project group
members for their contributions:
Country Consortia (leading organisations/Associated Country Consortia (leading organisations/Associated Country Consortia (leading organisations/Associated Country Consortia (leading organisations/Associated
ParParParParttttners)ners)ners)ners)
� AustriAustriAustriAustria:a:a:a: Dispatch Centre Tyrol, Ing. Gernot Vergeiner,
Andreas Maurer (4 – ILL GmbH – AU)
� Belgium:Belgium:Belgium:Belgium: Federal Government, Department of Public
Health, Dr. Agnes Meulemans, Dr. Jean Bernard Gillet
(8 – FOD Health DG 1 – BE)
� Czech Republic:Czech Republic:Czech Republic:Czech Republic: Emergency Medical Service Prague,
Dr. Milana Pokorná, Dr. Petr Zajíĉek
(13 – ZZSHMP-USZS – CZ)
� Denmark:Denmark:Denmark:Denmark: Capital Region, Prof. Freddy Lippert
(3 – RegH – DK)
� Finland:Finland:Finland:Finland: University Hospital Kuopio, Dr. Jouni Kurola,
Dr. Tapio Kettunen (9 – KUH – FI)
� France:France:France:France: Emergency Medical Service Province Hauts de
Seine, Dr. Michel Baer, Dr. Anna Ozguler
(6 – SAMU – FR)
� Germany:Germany:Germany:Germany: Hospitals of the County of Goeppingen, Prof.
Matthias Fischer, Dr. Martin Messelken (7 – KAE – DE)
� Hungary:Hungary:Hungary:Hungary: National Emergency Medical Service, Dr. Gábor
Gőbl (5 – OMSZ – HU)
� Italy:Italy:Italy:Italy: San Martino University Hospital Genoa,
Prof. Francesco Bermano, Dr. Lorenzo Borgo
(16 – HSanMartino – IT)
� Norway:Norway:Norway:Norway: Haukeland University Hospital Bergen,
Dr. Guttorm Brattebø, Lars Myrmel (17 – HUS – NO)
� Spain:Spain:Spain:Spain: University of Cantabria, Prof. Luis Garcia-Castrillo
Riesgo, Weyma Notel, Juan José San Miguel Roncero, Prof.
Francisco Javier Llorca Diaz (1 – UNICAN – ES)
� Turkey:Turkey:Turkey:Turkey: University Hospital Antalya, Dr. Hakan Yaman,
Sercan Bulut (14 - AKFAM – TR)
The country consortia consist of emergency medical care
institutions and local/regional public health authorities.
New Country Consortium (leading organisation, from New Country Consortium (leading organisation, from New Country Consortium (leading organisation, from New Country Consortium (leading organisation, from
SepteSepteSepteSeptemmmmber 2009)ber 2009)ber 2009)ber 2009)
� Netherlands: Maastricht University, Department of
International Health, Prof. Helmut Brand,
Dr. Genc Burazeri, Nicole Rosenkötter, Janneke Kraan
ScientificScientificScientificScientific----Technical partners (Associated Partners/Technical Technical partners (Associated Partners/Technical Technical partners (Associated Partners/Technical Technical partners (Associated Partners/Technical
Unit) Unit) Unit) Unit)
� GEOMED Research Forschungsgesellschaft mbH
(Germany): Dr. Thomas Krafft, Alexandra Ziemann,
Tim Tenelsen, Martina Schorbahn, Nico Reinke,
Dr. Axel Kortevoß (2 – GEOMED – DE)
� University of Bielefeld, Department of Public Health
Medicine (Germany): Prof. Alexander Krämer,
Dr. Paulo Pinheiro (15 – UniBi – DE)
External Scientific Advisory BoardExternal Scientific Advisory BoardExternal Scientific Advisory BoardExternal Scientific Advisory Board
� Prof. Helmut Brand, Maastricht University (Netherlands,
Chair)
� Dr. Andrea Ammon, European Centre for Disease
Prevention and Control (Sweden)
� Dr. Enrico Davoli, World Health Organization Regional
Office for Europe, Division of Country Health Systems
(Spain)
� Dr. Per Kulling, European Commission, Health Threat Unit
(Luxembourg)
� Prof. Francisco Javier Llorca Diaz, University of Cantabria
(Spain)
� Jerry Overton, MPA, Road Safety International (USA)
� Dr. Santiago Rodriguez, Health Service Cantabria (Spain)
� Prof. Mark Rosenberg, Queen’s University, Department of
Geography and Department of Community Health and
Epidemiology (Canada)
Developing Algorithms for Early Public Health Threat Detection in Europe 1
© SIDARTHa 2010 DRAFT January 2010
1111 Introduction: The SIDARTHa ProjectIntroduction: The SIDARTHa ProjectIntroduction: The SIDARTHa ProjectIntroduction: The SIDARTHa Project
Syndromic surveillance can detect public health threats earlier
than traditional surveillance and reporting systems. Pre-
hospital emergency medical services (EMS) and emergency
medical dispatch centres (EMD), and in-hospital emergency
departments (ED) across Europe routinely collect electronic
data that provides the opportunity to be used for near real
time syndromic surveillance of communicable and non-
communicable health threats such as heat-related diseases or
Influenza-Like-Illness (ILI). The European Commission co-
funded project SIDARTHa (Grant Agreement No. 2007208) for
the first time systematically explores the use of emergency
data to provide a basis for syndromic surveillance in Europe.
The project started in June 2008 and will run until December
2010. It is an initiative of emergency medical professionals
organised in the European Emergency Data (EED) – Research
Network1.
ObjectivesObjectivesObjectivesObjectives
The objective of the European project SIDARTHa is to
conceptualise, develop, implement/test and evaluate the
European Emergency Data-based System for Information on,
Detection and Analysis of Risks and Threats to Health
(SIDARTHa).
Methodology Methodology Methodology Methodology
During the conceptualisation phase, information on
international state-of-the-art in the early detection of health
threats and on the current practice of health surveillance and
alert systems in Europe are brought together with the
possibilities of emergency data for detection of health threats
and specific public health authority and emergency
professional desires for SIDARTHa’s system features. On this
basis the Geographic Information-System (GIS)-based
surveillance system SIDARTHa will be tested and evaluated
during the implementation phase in four regions2
(cf. Figure 1).
The project group constitutes a high-level expert panel of
emergency professionals, public health experts and health
1 www.eed-network.eu 2 SIDARTHa Implementation sites: District of Kufstein, Austria; Capital Region,
Denmark, County of Goeppingen, Germany, Autonomous Region Cantabria,
Spain
authority representatives under guidance of an
interdisciplinary steering committee. A sequence of focused
methods such as group discussions, Strengths - Weaknesses -
Opportunities - Threats analysis of existing procedures, half-
standardised surveys to seek input from potential futures
users, statistical analyses and modelling, and geo-processing
methods will be applied.
Expected Results & ProductsExpected Results & ProductsExpected Results & ProductsExpected Results & Products
The SIDARTHa project provides a methodology and software
application for syndromic surveillance at the regional level3 in
Europe based on routinely collected emergency data. The
SIDARTHa syndromic surveillance system automatically
analyses the actual demand for emergency services and
detects temporal and spatial aberrations from the expected
demand. The system will automatically alert decision makers in
the emergency medical institution and the regional public
health authority. Via the established reporting ways the
regional public health authority can inform national or
supranational authorities on an event (cf. Figure 2).
It is expected that SIDARTHa improves the timeliness and
cost-effectiveness of European and national health
surveillance by providing a basis for systematic syndromic
surveillance that supplements the existing surveillance
structures.
The main outputs of the project will be a syndromic
surveillance application (software) publicly available free-of-
charge and guidelines for future users on how to use the
application and how to transform emergency data into
syndromes and into the common SIDARTHa data set that the
application can analyse, including recommendations on
technical infrastructure, reporting procedures and
interpretation of the results. Furthermore, the guidelines will
cover the utilisation of the interactive user display and risk
communication platform.
3 In the SIDARTHa project the term regional is used referring to the smallest
administrative level at which a health authority responsible for surveillance
and reporting is established in a European country depending on the national
definition and rules. This level can be a community, city, county, district or
state. The implementation of the SIDARTHa syndromic surveillance system can
be based on data collected for the same administrative level or also for a part
of this area or based on the catchment areas of one or more participating
emergency institutions.
Developing Algorithms for Early Public Health Threat Detection in Europe 2
© SIDARTHa 2010 DRAFT January 2010
Evaluation
Implemen-
tationInformation
Possibilities
Needs
PHASE I - Conceptualisation PHASE II - Implementation
Project Coordination
Dissemination of Project Results
Project Evaluation
Figure Figure Figure Figure 1111: SIDARTHa Project Methodology: SIDARTHa Project Methodology: SIDARTHa Project Methodology: SIDARTHa Project Methodology
M = Month of the project time
ROUTINE DATA
ROUTINE DATA
ROUTINE DATA
REPORT/ALERT
REPORT/ALERT
REPORT/ALERT
REPORT/ALERT
REPORT/ALERT
REPORT/ALERT
REPORT/ALERT
Routine data from (i) emergency
medical dispatch centres, (ii)
ambulance patient documentations
and (iii) emergency department
information systemsis analysed for spatial and
temporal abberations at the
regional level.
SIDARTHa alerts emergency professionals and regional public
health authorities if a threshold is exceeded;
Via national authorities the European Commission, ECDC and
WHO can be informed about regional and cross-border alerts;
SIDARTHa can be used for risk communication about the event;
SIDARTHa only complements
but does not replace any existing system.
Figure Figure Figure Figure 2222: SIDARTHa Approach: SIDARTHa Approach: SIDARTHa Approach: SIDARTHa Approach
ECDC = European Centre for Disease Prevention and Control, WHO = World Health Organization
Developing Algorithms for Early Public Health Threat Detection in Europe 3
© SIDARTHa 2010 DRAFT January 2010
2222 Background, Background, Background, Background, ObjectivesObjectivesObjectivesObjectives & Methodology& Methodology& Methodology& Methodology
2.12.12.12.1 Background: Background: Background: Background:
Algorithms for Early Health Algorithms for Early Health Algorithms for Early Health Algorithms for Early Health
Threat DetectionThreat DetectionThreat DetectionThreat Detection
Early detection of public health threats relies on two major
components: timely and reliable data and the sensitivity,
specificity, and timeliness of detection algorithms. Detection
algorithms should be assessed considering costs of false
alerts versus the delay for a confirmed true alert. There are
three major groups of detection algorithms: control charts,
temporal modelling approaches and spatial-temporal
algorithms (Siegrist and Pavlin 2004 (1), Mandl et al. 2004
(2)).
Cumulative Sum (CUSUM)Cumulative Sum (CUSUM)Cumulative Sum (CUSUM)Cumulative Sum (CUSUM)
CUSUM is a short-term surveillance algorithm to indicate
recent data changes by comparing moving averages. Major
syndromic surveillance systems applying CUSUM are CDC’s
BioSense and EARS, the syndromic surveillance system of the
New York City Department of Health (NYCDOH) and FirstWatch.
Variations from the average of more than two standard
deviations issue an alert. Because of high variations in the
data individual CUSUM values are calculated for each data
source – syndrome combination at a regional level. CUSUM
can be applied to data from population and hospital-based
reporting systems. It controls for fixed confounders, and
hence allows for more robust evaluation of potential
associations than methods based on spontaneous adverse
event reporting. CUSUM detects aberrations very quickly and
is able to detect small shifts from the mean (Hutwanger et al.
2003 (3)).
C1, C2, C3 C1, C2, C3 C1, C2, C3 C1, C2, C3
The C1, C2, C3 detection algorithm which is applied in the CDC
syndromic surveillance system EARS standardises each
observation by using a moving sample average and sample
standard deviation Fricker et al. 2008 (9). The C1 algorithm
uses seven previous days to calculate the sample average
and sample standard deviation. For the C2 algorithm the same
threshold as for the C1 algorithm applies. The C3 method
uses the C2 statistic from the current day and two days prior
to the current observation. C1 is better for detection of point-
source distribution while C2 is more sensitive than C1 in
signalling a continued outbreak. With emphasis on timely
detection of outbreaks within the first few days of onset C2 is
suggested. It appears to be also least effected by serial
correlation. C3 is not more sensitive than C2 once the false
alert rate is held constant (Jackson et al. 2007 (4),
Hutwagner et al. 2003 (3), Zhu et. al. 2005 (5)).
Linear ModelsLinear ModelsLinear ModelsLinear Models
The RecursivRecursivRecursivRecursive Least Square (RLS)e Least Square (RLS)e Least Square (RLS)e Least Square (RLS) algorithm is an
autoregressive linear model applied for example in the Real-
time Outbreak and Disease Surveillance (RODS) syndromic
surveillance system. It predicts the current amount of
syndrome cases within a region based on historical data and
adjusts its model coefficients based on prediction errors. An
alert is triggered above the 95% CI of the estimated number
of cases (Najmi and Magruder 2005 (6)).
The Generalized Linear Model (GLM)Generalized Linear Model (GLM)Generalized Linear Model (GLM)Generalized Linear Model (GLM) uses a three-year
baseline and Poisson errors adjusting for day of week,
holiday, monthly and linear time trends. The model was found
to be more sensitive than C1, C2, C3. A weakness is the
detection of only large, rapidly increasing case numbers
(Jackson et al. 2007 (4)). The Generalised Linear MixedThe Generalised Linear MixedThe Generalised Linear MixedThe Generalised Linear Mixed
Model (GLMM) Model (GLMM) Model (GLMM) Model (GLMM) estimates the probability that a subject under
surveillance is a case, per spatial and temporal unit. Such a
model can be used especially for varying population sizes. The
model does not detect clusters extended across two
neighbouring spatial units (Kleinmann et al. 2004 (7)). The
Small Area Regression and TestingSmall Area Regression and TestingSmall Area Regression and TestingSmall Area Regression and Testing (SMART) model is an
adoption of the GLMM taking into account multiple
comparisons controlling also for day of week, holiday and day
after holiday, and seasonal trends. Predictions are based on a
Poisson distribution of the events (Kleinmann et al. 2005
(8)). SMART is used in the CDC’s BioSense syndromic
surveillance system.
Developing Algorithms for Early Public Health Threat Detection in Europe 4
© SIDARTHa 2010 DRAFT January 2010
Spatial Scan StatisticSpatial Scan StatisticSpatial Scan StatisticSpatial Scan Statisticssss (SaTScan)(SaTScan)(SaTScan)(SaTScan)
SaTScan is a free software that analyses spatial, temporal,
and space-time data using the spatial, temporal, or space-
time scan statistics. It uses either a Poisson-based model, a
Bernoulli model, a space-time permutation model, an ordinal
model, an exponential model, or a normal model. The software
is used in the syndromic surveillance systems of NYCDOH and
in the United States Department of Defense's Electronic
Surveillance System for the Early Notification of Community-
based Epidemics (ESSENCE) syndromic surveillance system.
SaTScan imposes a circular window on the map and lets the
circle centroid move across the region under study. For any
given position in the centroid, the radius of the window is
changed continuously to take any value between zero and the
upper limit. SaTScan does not require specification of the
location or size of a cluster. It uses a circular window to
determine potential cluster boundaries which may not
represent the population at risk (Kleinmann 2005 (8)).
2.2.2.2.2222 ObjectivesObjectivesObjectivesObjectives
The main objective of this task of Work Package 6 (Task 8) is
to assess the utility of different detection algorithms for the
SIDARTHa syndromic surveillance system by applying different
algorithms on historical emergency data from the four
implementation sites, simulating events and comparing the
detection results with actual public health department reports.
Another objective of the historical data analysis was to test
the coding manual developed as part of WP 5 with real
emergency data to suggest adjustments to the Coding
Manual.
Developing Algorithms for Early Public Health Threat Detection in Europe 5
© SIDARTHa 2010 DRAFT January 2010
2.2.2.2.3333 MethodologyMethodologyMethodologyMethodology: Introduction: Introduction: Introduction: Introduction
During the first Technical Workshop the implementation site
representatives4 and the technical unit came together with
additional experts for statistical modelling and software
programming to discuss the different established approaches
towards detection methods and a methodology for testing the
utility of detection algorithms for the system. The algorithms
will be adjusted and assessed during simulated events based
on factitious or actual event data provided by regional health
authorities. A methodology for historical data analysis
foreseeing the following steps that are applied to every data
set of the different emergency institutions in every
implementation site was developed:
Preparation of the dataPreparation of the dataPreparation of the dataPreparation of the data: : : :
� Quantity structure
� Data fields/contents
� Data completeness & quality
SIDARTHa SIDARTHa SIDARTHa SIDARTHa Standard Standard Standard Standard Data SetData SetData SetData Set::::
� Possibility to generate the SIDARTHa Standard Data Set
SyndromeSyndromeSyndromeSyndromes:s:s:s:
� Possibility to generate syndromes
DetectionDetectionDetectionDetection algorithms:algorithms:algorithms:algorithms:
� Descriptive analysis of regular patterns in time and
space
� Familiarisation with algorithms and identification of
software for testing algorithms and for automatic
programming of algorithms for SIDARTHa syndromic
surveillance system;
� Test of algorithms for different spatial and temporal
levels, different syndromes and different detection
algorithms, including simulations;
4 SIDARTHa Implementation sites: District of Kufstein, Austria; Capital Region,
Denmark, County of Goeppingen, Germany, Autonomous Region Cantabria,
Spain
� Calculation of initial baselines and thresholds per
algorithm, per implementation site/data set for actual
test runs during test/evaluation phase.
Next to the task leader the implementation sites will be
important partners not only in providing historical data but
also in testing algorithms with their own modelling and
statistical analysis expertise. The Dutch country consortium
leader, Maastricht University, accomplished the historical data
analysis and the tests/simulations.
During the Technical Workshop I it was agreed that the tests
start with the temporal algorithms C1, C2, C3, Holt-Winter-
Smoothing and CUSUM, time series modelling using ARIMA
and then will go on to spatial scan statistics. Software used for
historical data analysis and test of the algorithms will be SPSS,
Microsoft Excel, and SaTScan.
Data from the four implementation sites from previous years
(referred to as historical data) are sent to Maastricht
University.
The historical data analysis can be seen as test phase before
the real implementation of the surveillance system. Deeper
insights in the implementation site specific data are gained.
Variables which are available are identified, and a selection of
variables useful for syndrome generation can be made.
It is explored if regular differences in the occurrence of
emergency cases, like seasonal or daily patterns exist.
Furthermore, knowledge is gained about the overall
occurrence of events. Do cases occur frequently or do they
occur seldom? This information is necessary and prerequisite
for the correct application of detection methods.
In this first draft report historical EMD and ED data from
Austria (State of Tyrol) as well as Emergency Physician (EP)
data from Germany (County of Goeppingen) are analysed. The
historical data analysis is performed first on the overall
amount of cases starting with C1, C2, C3. In the following
months the analysis will be extended to analyse syndrome-
specific events and apply the other algorithms. Data from the
other implementation sites and other emergency institutions
will be included.
The tests and simulations are currently ongoing and will be an
integral part of the implementation phase until mid 2010.
For better readability, all tables and figures are included in a
separate appendix to this report.
Developing Algorithms for Early Public Health Threat Detection in Europe 6
© SIDARTHa 2010 DRAFT January 2010
2.2.2.2.4444 Preparation of the Data SetsPreparation of the Data SetsPreparation of the Data SetsPreparation of the Data Sets
A general preparation and cleansing of the raw data must take
place in the first step. Useful data variables and data set parts
are selected for syndromic surveillance from the raw data and
the quality and quantity of data are described to get an
overview on the availability of data useful for syndromic
surveillance (e.g., missing data, implausible data entries). In
this step adjustments to the data might be necessary (e.g.,
recoding of text into numeric data fields) to allow an analysis.
2.2.2.2.5555 SIDARTHa Standard Data SetSIDARTHa Standard Data SetSIDARTHa Standard Data SetSIDARTHa Standard Data Set
Seven data fields must be generated from individual
emergency data sets in order to allow the automated
SIDARTHa system to analyse the data:
1. Anonymous case identifying number (necessary)
2. Date/time (necessary)
3. Geographic reference (spatial surveillance)
4. Syndrome (syndromic surveillance)
5. Age (modifier)
6. Gender (modifier)
7. Severity (modifier)
Syndromes in their complexity are analysed in a separate
step. For the other data fields the quantity and quality of all
potential relevant data fields in the original emergency data
set is evaluated which allows for the selection of the most
suitable data variables for each implementation site or
emergency data set.
2.2.2.2.6666 Syndrome GenerationSyndrome GenerationSyndrome GenerationSyndrome Generation
The SIDARTHa syndromes Influenza-Like-Illness,
Gastrointestinal Syndrome, Respiratory Syndrome, Intoxication
Syndrome, and Environment-related Illness can be defined by
using specific variables of the original emergency data sets.
The quantity and quality of all potential relevant data fields in
the original emergency data set is evaluated which allows for
the selection of the most suitable data variables for each
implementation site or emergency data set. The total amount
of cases can be used to identify aberrations in the number of
cases for a defined time period and area, i.e., the Unspecific
Syndrome.
2222....7777 Descriptive AnalysisDescriptive AnalysisDescriptive AnalysisDescriptive Analysis
For the description of the data sets the frequencies of
relevant variables and the mean are calculated. The focus lies
on describing daily or seasonal patterns like cases per year,
month, week and day of the week. Furthermore, the
occurrence according to gender and age of the emergency
cases is examined. This information is a prerequisite and
should be taken into account when applying detection
algorithms for surveillance activities.
2.2.2.2.8888 Detection AlgorithmsDetection AlgorithmsDetection AlgorithmsDetection Algorithms
The SIDARTHa consortium decided to apply the detection
algorithms on a daily basis, which means that each day the
observed and expected frequencies are compared and
assessed.
C1, C2, C3 C1, C2, C3 C1, C2, C3 C1, C2, C3
As a first algorithm the C1, C2, C3 detection algorithm which is
implemented in the early aberration reporting system (EARS)
of the Centers for Disease Control and Prevention (CDC) is
applied.
The C1, C2, C3 detection methods standardise each
observation by using a moving sample average and sample
standard deviation (Fricker et al. 2008 (9)). The C1 algorithm
uses seven previous days to calculate the sample average
and sample standard deviation.
)(
)()()(
1
1
1tS
tYtYtC
−=
∑−
−=
=7
1
1 )(7
1)(
t
ti
iYtY and
[ ]∑−
−=
−=7
1
2
)()(6
1)(
1
2 t
ti
iYiYtS
The threshold of the C1 algorithm is fixed at the point when
the C1 statistic exceeds a value of three. This is
correspondent to a value being higher than three sample
standard deviations above the sample mean (Fricker et al.
2008 (9)).
Developing Algorithms for Early Public Health Threat Detection in Europe 7
© SIDARTHa 2010 DRAFT January 2010
The C2 algorithm includes also seven days for calculation but
inserts a 2-day lag to avoid influences of an upswing of a
probable outbreak. Therefore, the observations nine to three
days before the day of interest are included.
)(
)()()(
3
32
tS
tYtYtC
−=
∑−
−=
=9
3
3 )(7
1)(
t
ti
iYtY and
[ ]∑−
−=
−=9
3
2
3 )()(6
1)(
3
2 t
ti
iYiYtS
For the C2 algorithm the same threshold as for the C1
algorithm applies. C2 statistics >3 exceed the expected
values (Fricker et al. 2008 (9)).
The C3 method uses the C2 statistic from the current day and
two days prior to the current observation.
[ ]∑−
=
−=2
23 1)(,0max)(t
ti
iCtC
Values > 2 should be judged as alerting signals (Fricker et al.
2008 (9)).
In medical emergency data different patterns of case
occurrence are known. There are more cases during
weekdays than on weekends and also public holidays
influence the amount of emergency cases. Since in C1, C2
and C3 the baseline is build by the amount of cases which
occurred in the same season (7 to 9 days before the current
observation) seasonal variations are taken indirectly into
account (Hutwagner et al. 2005 (10)). Tokars and colleagues
(2009 (11)) modified the algorithms to be able to take also
the day-of-week into account which might increases sensitivity.
Therefore the detection methodology will be used in two
different ways:
1. Taking the last seven to nine days into account as it is
described by Fricker et al. 2008 (9) (referred to as
unstratified baselineunstratified baselineunstratified baselineunstratified baseline)
2. Stratifying baseline data in weekdays and weekend days.
Depending on the day of the current observation,
weekdays or weekend days are used to calculate the
sample mean and standard deviation. This stratification is
known as W2 algorithm (1, 3) (referred to as stratified stratified stratified stratified
baselinebaselinebaselinebaseline).
In the following sections the two ways of applying C1, C2, C3
are used for the overall, daily amount of cases/events. The
predefined thresholds are at a value of three for C1 and C2
and a value of two for the C3 algorithm.
2.2.2.2.9999 SoftwareSoftwareSoftwareSoftware
Data are stored in Microsoft Office Access databases (Version
2003). For descriptive analyses data are exported in SPSS
(Version 15.01). In SPSS frequency tables are generated, the
distribution of missing values is analysed and cleansing of the
data sets has been performed as well as generation of
variables like day of the week, number of the week, etc.
For developing the graphs frequency tables of SPSS are
exported to Microsoft Office Excel (Version 2003).
Furthermore the C1, C2, C3 algorithm has been applied in
Microsoft Office Excel (Version 2003).
Developing Algorithms for Early Public Health Threat Detection in Europe 8
© SIDARTHa 2010 DRAFT January 2010
3333 Applying Applying Applying Applying Detection Detection Detection Detection Algorithms in Algorithms in Algorithms in Algorithms in the SIDARTHa the SIDARTHa the SIDARTHa the SIDARTHa
Implementation SImplementation SImplementation SImplementation Sitesitesitesites: First Results: First Results: First Results: First Results
3333.1.1.1.1 Preparation: Preparation: Preparation: Preparation: ImplementationImplementationImplementationImplementation
sites sites sites sites &&&& data sourcesdata sourcesdata sourcesdata sources
AustriaAustriaAustriaAustria: State of Tyrol: State of Tyrol: State of Tyrol: State of Tyrol
ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein –––– ED(AT)ED(AT)ED(AT)ED(AT)
The ED(AT) data set contains admissions from the General
Districht Hospital Kufstein. The hospital is settled in the
Kufstein district of the State of Tyrol in Austria. The district
has a geographical area of 969.9 km2. 99,394 inhabitants are
living in the district on 31st December 20095 which results into
a population density of 102 inhabitants per km2.
In this data set, all inpatients that accessed the ED in 2008
(1/1/2008 - 31/12/2008) are included (n = 30,669).
The origin counties/regions of patients are mainly the Tyrol
region and other regions in Austria (93%). 1,571 patients are
from Germany and 558 from other countries.
As a geographic reference, the place of residence, the zip
code and the country code are given in the data set. Patients
can be described by age and gender. Information on
syndromes or diagnosis of the patients is not part of the data
set.
The data set contains no missing values or coding errors
(Table 1). The variables which are currently used are marked
grey in Table 1. For the historical data analysis it was
necessary to add or recode some variables.
Added variables are
� The day of the week (Monday, Tuesday, …Sunday);
� The week number of the year (from 1 to 52).
5 Amt der Tiroler Landesregierung, Raumordnung Statistik. Demografische
Daten Tirol. Innsbruck 2009
(http://www.tirol.gv.at/fileadmin/www.tirol.gv.at/themen/zahlen-und-
fakten/statistik/downloads/BEV2008.pdf, accessed January 2010)
Recoded variables are
� Gender was recoded from text into a numeric, nominal
(dichotomous) variable;
� The zip code and the zip category (region) of the patient
were added and recoded from the free text
(alphanumerical) variable Residence.
EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of
Kufstein Kufstein Kufstein Kufstein –––– EMD(AT)EMD(AT)EMD(AT)EMD(AT)
The EMD Tyrol currently is responsible for the area of the City
of Innsbruck (118,035 inhabitants), the County of Innsbruck
(164,027 inhabitants) and the District of Kufstein (99,394
inhabitants).
In this data set, all the events of the EMD from 1/1/2003 until
31/12/2008 (six years) are documented (n = 937,604).
The origin countries/regions of reported patients were mainly
the Tyrol region and other regions in Austria (99.75%).
2,112 patients were from Germany and 247 patients were
from other countries. As a geographic reference of the event,
the country, region, zip code, city, street, house-number and
GPS coordinates are given.
No further personal information like gender or age is
represented in the data set. Diagnostic information is given by
AMPDS codes and sub-codes. These variables are used for
syndrome generation.
Since the EMD(AT) data set contains all activities dispatched
by the EMD it was necessary to select the events to be used in
SIDARTHa.
These events are emergency medical events that require
immediate attention and that occur in an unplanned manner.
Planned activities like transportation activities (transporting
patients to regular dialysis appointments, etc.) or stand-by
activities during public events are not selected for syndromic
Developing Algorithms for Early Public Health Threat Detection in Europe 9
© SIDARTHa 2010 DRAFT January 2010
surveillance. These are therefore excluded when applying the
detection algorithms in historical data analysis.
It turned out that the EMD(AT) data contained 46.5% events
that did not meet the inclusion criteria. After exclusion of
these cases the data set consisted of 500,977 cases. The
documentation of case selection can be found in Table 2.
A description of the available variables is given in Table 3;
rows marked in grey indicate that the variables have been
used for analysis. The dataset contains missing values and
coding errors (Table 3). For historical data analysis it was
necessary to add or recode some variables.
Added variables are
� The day of the week (Monday, Tuesday, …Sunday);
� The week number of the year (from 1 to 52):
� The month of the year (January, February, … December).
Furthermore, a filter variable has been defined to extract the
relevant events. The selection criteria are described in Table
2. The basis for the filter generation was a recoded variable
giving the event type (übergeord_Einsatzstichwort) and the
variable giving the complete event/AMPDS code
(Einsatzcode).
GermanyGermanyGermanyGermany: County of: County of: County of: County of GöppingenGöppingenGöppingenGöppingen
EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)
The County of Goeppingen belongs to the federal State of
Baden-Wuerttemberg in Germany. The county has a
geographical area of 642.4 km2. 255,807 inhabitants were
living in the county on 31st December 20096 which results into
a population density of 398.2 inhabitants per km2.
In this data set, all EP responses from 1/7/2005 until
31/12/2008 (two and a half years) (n = 14,869).
The origin countries/regions of patients are the district of
Göppingen and other regions in Germany. As a geographic
reference of the emergency event, the community code, the
zip code, and place of the event (e.g. home, work place) are
given.
6 Statistisches Landesamt Baden-Wuerttemberg. Statistische Berichte Baden-
Wuerttemberg. Bevoelkerung und Erwerbstaetigkeit. Stuttgart 2009
(http://www.statistik.baden-
wuerttemberg.de/Veroeffentl/Statistische_Berichte/3126_08001.pdf,
accessed January 2010)
Information on age and gender is represented in the data set.
Furthermore, information on diagnosis and severity are
available. The medical information is given as ICD-10 codes
and codes (KRANK 1-8) which are part of a standardised
system used by the physicians in Göppingen to describe the
disability or illness (MIND 2). Additionally, information on
breathing status, oxygen saturation, severity (Glasgow Coma
Scale), pain and body temperature is available. This
information can be used to generate syndromes.
A description of selected variables and their availability is
shown in Table 4. The dataset contains missing values and
coding errors (Table 4). Unfortunately, the field which gives
information on the body temperature contained no valid data.
The only values available do not correspond to a realistic
body temperature (values were -1 or -01). For historical data
analysis it was necessary to add or recode some variables.
Added variables are
� The day of the week (Monday, Tuesday, …Sunday);
� The week number of the year (from 1 to 52).
Due to coding errors it was necessary to recode the variables
� Age: This variable was present (PATALTER) but not
usable in SPSS. Therefore, the age has been newly
calculated on the basis of date of the event (DATUM) and
date of birth of the patient (GEBDAT);
� Gender: Due to some double coding for the same variable
values the existing gender variable (GESCHL) has been
cleansed and the new variable gender has been
generated;
� ATM1, KRANK1, KRANK2, KRANK3, KRANK4, KRANK5,
KRANK6, KRANK7, KRANK8: All of these variables included
two different codes for the same variable value. The
variables have been cleansed and new variables have
been generated.
Developing Algorithms for Early Public Health Threat Detection in Europe 10
© SIDARTHa 2010 DRAFT January 2010
3333....2222 SIDARTHa Standard Data SetSIDARTHa Standard Data SetSIDARTHa Standard Data SetSIDARTHa Standard Data Set
AustriaAustriaAustriaAustria: State of Tyrol: State of Tyrol: State of Tyrol: State of Tyrol
ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein –––– ED(AT)ED(AT)ED(AT)ED(AT)
Data quality in the ED(AT) data set is good; there were no
missing values or coding errors. Five items of the SIDARTHa
Standard Data Set can be generated by information available
in the data set. Out of these items other information relevant
for a surveillance system can be generated (e.g. day of the
week). There is no information available which can be used for
syndrome generation or assessment of severity of the cases.
VariablesVariablesVariablesVariables ED(AT)ED(AT)ED(AT)ED(AT) VariablesVariablesVariablesVariables
Identifier �
Date/time � Date and time
Geographic reference � Place of residence, ZIP-code
Age �
Gender �
Severity �
Syndrome 1-n �
EMEMEMEMD: City of Innsbruck, County of Innsbruck, District of D: City of Innsbruck, County of Innsbruck, District of D: City of Innsbruck, County of Innsbruck, District of D: City of Innsbruck, County of Innsbruck, District of
Kufstein Kufstein Kufstein Kufstein –––– EMD(AT)EMD(AT)EMD(AT)EMD(AT)
For five out of seven SIDARTHa Standard Data Set items
information is available in the data set. Out of these items
other information relevant for a surveillance system can be
generated (e.g. day of the week). There is no information
available on age and gender of the events. The geographic
information is very specific including besides address
information also GPS coordinates. Sufficient information for
syndrome generation is represented in the data set (0.01%
missing values regarding AMPDS codes (variable:
Einsatzcode). Specific information on the severity (NACA
score) is only available for approximately 10% of the events.
So, this information cannot be used in a surveillance system
but the general differentiation into HPR and nHPR events
provides information on severity.
VariablesVariablesVariablesVariables EMD(AT)EMD(AT)EMD(AT)EMD(AT) VariableVariableVariableVariablessss
Identifier �
Date/time � Date and time
Geographic reference
�
County, region, zip code, city, street, house number, GPS coordinates
Age �
Gender �
Severity � NACA Score
Syndrome 1-n � AMPDS codes and subcodes
GermanyGermanyGermanyGermany: County of: County of: County of: County of GöppingenGöppingenGöppingenGöppingen
EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)
In the EP(DE) data set information on all items of the
SIDARTHa standard data set is available. After data cleansing
the number of coding errors could be decreased remarkably.
Out of these items other information relevant for a
surveillance system can be generated (e.g. day of the week).
Sufficient information for syndrome generation and on severity
is available. Except the body temperature variables useful for
syndrome generation are available. However, the quality of
the variables varies. The first diagnosis coded by ICD-10 is
collected in 70% of the cases. The Goeppingen specific
diagnostic categories (KRANK 1-8) are only collected for one
third of the events, so the amount of cases with missing
values is 68.1%. The amount of missing values for the
following variables give an overview on the availability of
symptomatic information:
� 19% for first respiratory status (AF1);
� 21% for first oxygen saturation status (SAOZ1);
� 12% for first breathing status (ATM1);
� 13% for pain (SCHMERZ1);
� 100% for body core temperature (KTM1).
This leads to the decision that syndromes will be generated
mainly on the basis of the main ICD-10 diagnoses.
Developing Algorithms for Early Public Health Threat Detection in Europe 11
© SIDARTHa 2010 DRAFT January 2010
VariablesVariablesVariablesVariables EPEPEPEP(DE)(DE)(DE)(DE) VariablesVariablesVariablesVariables
Identifier �
Date/time � Date and time
Geographic reference
� community code, zip code, place (e.g., home, work place)
Age �
Gender �
Severity � NACA Score, Glasgow Coma Scale
Syndrome 1-n �
ICD-10 codes, Goeppingen specific codes (KRANK 1-8) and information on different symptoms
3333....3333 Syndrome GenerationSyndrome GenerationSyndrome GenerationSyndrome Generation
AustriaAustriaAustriaAustria: State of Tyrol: State of Tyrol: State of Tyrol: State of Tyrol
In the EMD(AT) data set syndromes can be generated from
the variable Einsatzcode containing AMPDS codes. This
variable describes the event and the chief complaint of the
patient. The exact use of AMPDS codes for syndrome
description can be found in the SIDARTHa Coding Manual
(Garcia-Castrillo Riesgo et al. 2009 (12)).
For the Unspecific Syndrome the total amount of all events
selected for SIDARTHA are used.
In the ED(AT) data set syndrome generation is not possible
since information on symptoms or diagnoses are not
available.
GermanyGermanyGermanyGermany: County of: County of: County of: County of GöppingenGöppingenGöppingenGöppingen
In the EP(DE) data set syndromes can be generated based on
ICD-10 codes. Additional information can be derived out of
eight specific diagnostic variables (KRANK 1-8) as well as
symptomatic variables (first respiratory status (AF1), first
oxygen saturation status (SAOZ1), first breathing status
(ATM1), pain and body core temperature (KTM1)).
ICD-10 codes are subdivided into first to third diagnosis. The
main diagnosis (ICD1) will be used for syndrome generation.
The exact use of ICD-10 codes for syndrome generation can
be found in the SIDARTHa Coding Manual (Garcia-Castrillo
Riesgo et al. 2009 (12)). The application of the variables
KRANK 1-8 and the symptomatic variables for syndrome
generation is described in the following paragraph and is
visualized in Table 5.
The symptomatic variables (first respiratory status (AF1), first
oxygen saturation status (SAOZ1), first breathing status
(ATM1), pain and body core temperature (KTM1)) provide
relevant information for generation of the SIDARTHa
syndromes:
Influenza-Like-
Illness
� Breathing: dyspnoematic, cyanotic, spastic or
rattling
� Respiratory rate: > 20 breaths per minute
� Pulse oximetry oxygen saturation: Sp02 <
95%
� Body core temperature: >38.5° C
Respiratory
Syndrome
� Breathing: dyspnoematic, cyanotic, spastic or
rattling
� Pulse oximetry oxygen saturation: Sp02 <
95%
Gastrointestinal
Syndrome VAS pain score >3
Developing Algorithms for Early Public Health Threat Detection in Europe 12
© SIDARTHa 2010 DRAFT January 2010
Regarding the variables KRANK1-8 the following selection was
chosen for syndrome generation:
Influenza-Like-
Illness
Airway disorder (KRANK3): Pneumonia/Bronchitis
or other respiratory disease
Respiratory
Syndrome
Airway disorder (KRANK3): Asthma or COPD
exacerbations or aspiration or
Pneumonia/Bronchitis or hyperventilation
or/Tetany or Croup/Epiglottitis or other respiratory
disease
Gastrointestinal
Syndrome
� Abdominal disorders (KRANK4): acute
abdomen or gastrointestinal bleeding or colic
or other disease abdomen
� Metabolic disease (KRANK6): Dehydrated
Intoxication
Syndrome
� CNS disorders (KRANK1): seizure or other
CNS disorders
� Psychiatric disorders (KRANK5): alcohol
intoxication or drug intoxication or intoxication
medical drugs
� Other diseases (KRANK8): other intoxication
For the generation of the Unspecific Syndrome the total
amount of EP(DE) cases is used.
3333....4444 Descriptive analysis Descriptive analysis Descriptive analysis Descriptive analysis
AustriaAustriaAustriaAustria: State of Tyrol: State of Tyrol: State of Tyrol: State of Tyrol
ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein –––– ED(AT)ED(AT)ED(AT)ED(AT)
The ED(AT) data set contains information on hospital
admissions through the ED in the District of Kufstein in 2008.
In this period, 30,669 hospital admissions were documented.
On a monthly basis, approximately 2,400 to 2,800 hospital
admissions were counted, with 2,556 admissions per month
on average. A seasonal variation can be identified: more
hospital admissions were documented from January until April
and in July (Figure 1). Since this region is a touristic area with
a lot of skiing activities in the winter periods and hiking
activities during summer, these seasonal increases might be
caused by this circumstance. However, there is no increase in
admissions in August which is also a typical month for summer
holidays.
On a weekly basis, a stable amount of admissions was
documented (approximately 500 to 650), with an average of
579 admissions. As an exception, there were fewer
admissions in the last week of the year (Figure 2).
On a daily basis the higher amount of hospital admissions
occurred on working days (Figure 3.) with approximately 70
to 110 admissions ( x = 97). During the weekend the
amount of admissions decreased to 52 cases on average
(Figure 3).
Females were more often admitted to the hospital than males
(Figure 4). The mean age of admitted patients in 2008 was
52.3 years (Figure 5). Besides the age group of 25-64 years,
patients aged 65 or older were admitted most frequently
(Figure 6). Remarkable was the decrease in the admission of
elderly in week 32 and 52. At the same time, the admission of
children and adolescents increased (Figure 5, Figure 6).
SummarySummarySummarySummary
Daily amount of admissions
� Weekdays: x = 97
� Weekend days: x =52
Seasonal pattern
Higher amount of cases from January to April, and in July
Developing Algorithms for Early Public Health Threat Detection in Europe 13
© SIDARTHa 2010 DRAFT January 2010
EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of
Kufstein Kufstein Kufstein Kufstein –––– EMD(AT)EMD(AT)EMD(AT)EMD(AT)
The EMD(AT) data set contains information on dispatch
events in the Kufstein district during 2003 to 2008. In this
period 937,604 dispatch events were documented. The data
set of selected events for SIDARTHa contains 500,977 cases.
On average 156,267 events occurred per year. The amount
of dispatch events increased steadily during 2003 to 2008. A
clear linear trend can been observed (R2=0.95, p<0.05).
The number of events in 2008 (n=175,316) is 18% higher
than in 2003 (n=143,861) (Figure 7A). As can be seen in
Figure 7B, on average 83,496 events occurred per year that
were selected for SIDARTHa. The increasing trend of this
amount of relevant events per year is comparable to the
increasing trend of the total amount of events (R2=0.94,
p<0.05). The number of relevant events in 2008 (n =
95,483) is 22% higher than in 2003 (n = 74,338). The
increase of dispatch events over time is a known trend all over
Europe.
On a monthly basis, depending on the analysed year, on
average 12,000 to 14,500 events for the dispatch centre
occurred. Within the selected data set on average 6,000 to
9,000 events occurred. A seasonal variation can be identified:
higher amounts of events in December and January, in March,
and in July in each year. For 2008 a peek in October is visible
in addition (Figure 8A and 8B).
On a daily basis the higher amount of events for the dispatch
centre occurred during weekdays with an average of 496
events. During the weekend the average amount of events
decreased to 256 (Figure 9A). The same trend can be
observed in the selected data set with an average of 245
cases per day. During weekends, the average amounts of
events decrease to 186 cases per day.
GermanyGermanyGermanyGermany: County of: County of: County of: County of GöppingenGöppingenGöppingenGöppingen
EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)
The EP(DE) data set documented 14,869 cases during three
and a half years (01/07/2005 to 31/12/2008).
The yearly amount of cases can be seen in Figure 10.
Approximately 4,500 cases occurred in 2006 and 2007. In
2005 data were only available for the second half of the year.
The proportional amount of cases in 2005 and the total
amount in 2008 were less than in 2006 and 2007. The
reduced amount of cases in 2005 and 2008 becomes also
visible on a monthly and weekly basis (see Figure 11, 12).
On average 354 cases occurred per month. In 2006, 2007,
and 2008 a seasonal variation can be seen during summer.
Increases during autumn and/or winter are particularly visible
in 2007 (Figure 11). The average amount of cases per week
in Goeppingen was 80 (Figure 12) There were no daily
differences in the amount of cases in the EP(DE) data. On a
daily average, irrespective of a working day or during
weekends, 11 to 12 cases occurred. In the investigated
period at least two cases and at maximum 30 cases occurred
per day.
The proportion of male and female cases was mostly equally
distributed. In 2007 a slightly higher proportion of male
patients were treated (Figure 13). The mean age of the
patients was 53.8 years. During summer in 2006 to 2008 the
average age seems to decrease (Figure 14). There is no clear
pattern visible when analysing the data by age category for
2006 to 2008 (Figure 15).
SummarySummarySummarySummary
Yearly increase of events
22% more events in 2008 compared to 200322% more events in 2008 compared to 200322% more events in 2008 compared to 200322% more events in 2008 compared to 2003
Daily amount of events
� Weekdays: Weekdays: Weekdays: Weekdays: x =245=245=245=245
� Weekend days: Weekend days: Weekend days: Weekend days: x =186=186=186=186
Seasonal pattern
Higher amount of cases in December and January,
March, and in July
SummarySummarySummarySummary
� No differences in the daily amount of cases has been
observed
� Number of cases per day on average 11 to 12
Seasonal pattern
Higher amount of cases in June and July
Developing Algorithms for Early Public Health Threat Detection in Europe 14
© SIDARTHa 2010 DRAFT January 2010
3333....5555 Results of the detection Results of the detection Results of the detection Results of the detection
alalalalgorithm C1, C2, C3gorithm C1, C2, C3gorithm C1, C2, C3gorithm C1, C2, C3
AustriaAustriaAustriaAustria: State of Tyrol: State of Tyrol: State of Tyrol: State of Tyrol
ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein ED: District of Kufstein –––– ED(AT)ED(AT)ED(AT)ED(AT)
Unstratified baselineUnstratified baselineUnstratified baselineUnstratified baseline
Data are available for 2008. No aberration from the expected
values could be identified by C1, C2, C3 (Figure 16).
Stratified baselineStratified baselineStratified baselineStratified baseline
When stratifying baseline in weekdays and weekend days two
signals have been generated by C1. The first occurred on 3rd
March 2008 (Monday); the second one occurred on 1st June
2008 (Sunday).
EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of EMD: City of Innsbruck, County of Innsbruck, District of
Kufstein Kufstein Kufstein Kufstein –––– EMD(AT)EMD(AT)EMD(AT)EMD(AT)
UnstraUnstraUnstraUnstratified baselinetified baselinetified baselinetified baseline
Data are available for 2003 to 2008. For signal detection only
the relevant events (n=500,977) were included. On average
C1 signals 1 time per year, C2 signals 1.8 and C3 4.5 times
per year. For most of the years signals occurred mainly in the
beginning of the year, during summer, or in the end of
December.
Table 6 shows the date and the day of the week of the signals
which could be detected from 2003 to 2008. The results will
be summarized later on together with the results of the
stratified analysis.
In 2003 one C2 signal and three C3 signals have been
detected. These signals occurred in the beginning of August
(Table 6, Figure 17).
In 2004 one C1 and one C2 signal have been observed as
well as three C3 signals. All signals occurred in the beginning
of February (Table 6, Figure 18).
In 2005 one C2 signal and three consecutive C3 signals
occurred. They occurred in the end of June (Table 6, Figure
19).
In 2006 two C1 and C2 signals and six C3 signals have been
identified. There are two major time periods: First, at the end
of July, second, in the middle and at the end of December
(Table 6, Figure 20).
In 2007, two C1 signals occurred in mid-May and at the end
of December. There were also two C2 and four C3 signals in
mid-May. At the end of December three C2 and five C3 signals
have occurred (Table 6, Figure 21).
In 2008 one C1 signal occurred in the beginning of March.
One C1 and one C2 signal as well as three C3 signals occurred
in the end of December (Table 6, Figure 22).
C2 and C3 signaled often on the same day which was mostly
followed by C3 signals on two consecutive days.
Stratified baselineStratified baselineStratified baselineStratified baseline
After stratifying weekdays and weekend days for baseline
determination the overall number of C1, C2, C3 signals
increased approximately five-fold. The unstratified analysis
resulted into 44 signals, when using a stratified baseline 206
signals were detected. On average C1 signals six times per
year, C2 signals eight and C3 signals 20 times per year.
Figure 23 shows the occurrence of signals in a calendar
format. Signals of the unstratified analysis are also indicated.
Aberrations from baseline occurred nearly every year in the
beginning of January and end of December, mostly after
Christmas. Only in December 2005 and January 2006 no
signals have occurred.
Furthermore, signals occurred always in one or two spring
and summer months. Signals during autumn occurred in
2003, 2006, 2007, and 2008.
GermanyGermanyGermanyGermany: County of: County of: County of: County of GöppingenGöppingenGöppingenGöppingen
EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)EP: County of Göppingen EP(DE)
Unstratified baselineUnstratified baselineUnstratified baselineUnstratified baseline
Data are available for July 2005 to 2008. Table 7 shows the
signals which could be detected in this period. Several signals
occurred over the years 2005 to 2008.
In 2005 in total, fifteen signals occurred, four C1 signals,
three C2 signals and eight C3 signals. These signals occurred
in mid July, at the end of September and at the beginning of
October. Thus, most of the year’s signals are in summer and
autumn of 2005 (Table 7, Figure 24).
Developing Algorithms for Early Public Health Threat Detection in Europe 15
© SIDARTHa 2010 DRAFT January 2010
Summary 2005Summary 2005Summary 2005Summary 2005
Signals occurred in summer as well as in autumn 2005.
In 2006 in total, 47 signals occurred. Twelve C1 signals, nine
C2 signals and twenty-six C3 signals have been detected
(Table 7, Figure 25).
These signals broadly occurred in the end of January, in the
beginning of March, in mid April, in the beginning of August, in
mid August, in the end of September, in the end of November
and in the beginning of December. Thus, most of the year’s
signals occurred over the whole year, except for the early
summer. As can be seen in Table 7, signals are occurring on
mainly every day of the week, but a higher rate of signals
occurred on weekdays.
Summary 2006Summary 2006Summary 2006Summary 2006
Signals occurred over the whole year. There are a little bit
more signals on weekdays.
In 2007 in total, 45 signals occurred. Nine C1 signals, ten C2
signals and twenty-six C3 signals have been detected (Table
7, Figure 26).
These signals broadly occurred at the beginning of January, at
the end of March, at the end of April, in mid June, at the end
of June, at the end of September, in mid October and at the
end of November. Thus, most of the year’s signals occurred
over the whole year, no clear pattern can be seen. Signals are
occurring on every day of the week, in this, also no clear
pattern can be found.
Summary 2007Summary 2007Summary 2007Summary 2007
Signals occurred over the whole year. No clear signal pattern
can be seen on months or weekdays.
In 2008 in total 36 signals occurred. Seven C1 signals, eight
C2 signals and twenty-one C3 signals have been detected.
These signals broadly occurred at the beginning of January, at
the end of March, at the end of May, at the end of August, at
the end of September and in mid/at the end of November.
Thus, most of the year’s signals occurred over the whole year.
The signals seem to occur mainly in the end of the months,
but no clear pattern could be seen in the week- or weekend-
days (Table 7, Figure 27).
Summary 2008Summary 2008Summary 2008Summary 2008
Signals occurred over the whole year. Mainly the signals
occurred in the end of the months.
In every year it has been observed that C1, C2, C3 (or C2, C3)
signal mostly together which is followed by a C3 signal on the
two consecutive days (Table 7).
Stratified baselineStratified baselineStratified baselineStratified baseline
After stratifying weekdays and weekend days for baseline
determination the overall number of C1, C2, C3 signals stayed
nearly the same. 95 signals occurred before stratification and
105 signals occurred after. On average C1 signals 4.5 times
per year, C2 signals nine and C3 signals 25.5 times per year.
All signals also from the analysis with unstratified baseline can
be found in a calendar format in Figure 28.
In 2005 in total 22 signals occurred. Four C1 signals, four C2
signals and twelve C3 signals have been detected. These
signals occurred over the whole year, from July to November,
except for December. Thus, no clear pattern in months can be
seen in the signals in 2005.
In 2006 in total 49 signals occurred. Nine C1 signals, ten C2
signals and 30 C3 signals have been detected. These signals
occurred broadly in all the months of the whole period. Thus,
no clear pattern in months can be seen in the signals in 2006.
In 2007 in total 36 signals occurred. Eight C1 signals, seven
C2 signals and 21 C3 signals have been detected. These
signals occurred over the first months of the year, from the
beginning of October the amount of signals declines. Thus,
most of the signals occur before October 2007.
In 2008 in total 44 signals occurred. Eight C1 signals, ten C2
signals and 26 C3 signals have been detected. These signals
occurred over the whole year but the amount of signals seems
to be less in spring.
As it has been described for the unstratified analysis in most
cases C1, C2, C3 (or C2, C3) signal together and C3 signals
on the two consecutive days.
Developing Algorithms for Early Public Health Threat Detection in Europe 16
© SIDARTHa 2010 DRAFT January 2010
4444 Summary Summary Summary Summary &&&& Next StepsNext StepsNext StepsNext Steps
In the last section data from two test sites have been
described. These data sets have been generated from three
different emergency medical care services, namely EMD
events, EP responses and ED admissions.
It turned out that most events per day occurred in the
EMD(AT) data set (weekdays: 245, weekend: 186), followed
by the ED(AT) data set (weekdays: 97, weekend: 52) and the
smallest amount have been seen in the EP(DE) data set (11
to 12 events).
On these data sets an easy-to-use detection algorithm (C1,
C2, C3) has been applied. We analysed the overall occurrence
of cases or events and tested for aberration from the
expected numbers based on the frequencies of the previous
days.
When applying C1, C2, C3 on the total amount of events per
day the lowest amount of signals occurred in the ED(AT) data
set. There were no signals when using an unstratified baseline
and two signals in the stratified analysis.
The highest amount of signals occurred in the EP(DE) data
set, the data set with the lowest amount of cases per day. The
amount of signals in this data set stayed stable irrespective if
the baseline was calculated after stratification in weekdays
and weekend days or without stratification. Furthermore,
signals identified by stratified analysis in the EP(DE) data set
happened quite often on different days compared to the
unstratified analysis. Whereas in the EMD(AT) data, signals of
the unstratified analysis were mostly confirmed by the
stratified one. However, in the Tyrol dispatch data set the
number of signals increased five-fold when calculating a
stratified baseline.
In the EP(DE) data set signals occurred all over the year
without a clear pattern. In the EMD(AT) data set a clear
pattern of signals could be observed. Signals occurred every
year – despite 2005 – at the end of December. Furthermore
in almost every year a clear cluster of signals occurred in
summer, in 2003 and 2008 several signals in autumn have
occurred. Signals in spring were more scattered in the
investigated years.
After description of data sets and the first application of an
detection algorithm investigation of background information is
necessary in order to be able to interpret the signals identified
in the data sets.
Furthermore, it is intended to apply other detection algorithms
and to analyse aberrations by syndrome.
Developing Algorithms for Early Public Health Threat Detection in Europe 17
© SIDARTHa 2010 DRAFT January 2010
ReferencesReferencesReferencesReferences
1.1.1.1. Siegrist D, Pavlin J. BioSiegrist D, Pavlin J. BioSiegrist D, Pavlin J. BioSiegrist D, Pavlin J. Bio----ALIRT biosurveALIRT biosurveALIRT biosurveALIRT biosurveillance detection algorithm evaluation. MMWR Morb Mortal Wkly Rep. 2004 Sep 24;53 illance detection algorithm evaluation. MMWR Morb Mortal Wkly Rep. 2004 Sep 24;53 illance detection algorithm evaluation. MMWR Morb Mortal Wkly Rep. 2004 Sep 24;53 illance detection algorithm evaluation. MMWR Morb Mortal Wkly Rep. 2004 Sep 24;53
Suppl:152Suppl:152Suppl:152Suppl:152----8.8.8.8.
2.2.2.2. Mandl KD, Reis B, Cassa C. Measuring outbreakMandl KD, Reis B, Cassa C. Measuring outbreakMandl KD, Reis B, Cassa C. Measuring outbreakMandl KD, Reis B, Cassa C. Measuring outbreak----detection performance by using controlled feature set simulations. MMWR Morb detection performance by using controlled feature set simulations. MMWR Morb detection performance by using controlled feature set simulations. MMWR Morb detection performance by using controlled feature set simulations. MMWR Morb
Mortal Wkly Rep. 2004 Sep 24;53 Suppl:Mortal Wkly Rep. 2004 Sep 24;53 Suppl:Mortal Wkly Rep. 2004 Sep 24;53 Suppl:Mortal Wkly Rep. 2004 Sep 24;53 Suppl:130130130130----6.6.6.6.
3.3.3.3. Hutwagner L, Thompson W, Seeman GM, Treadwell T. The bioterrorism preparedness and response Early Aberration Reporting Hutwagner L, Thompson W, Seeman GM, Treadwell T. The bioterrorism preparedness and response Early Aberration Reporting Hutwagner L, Thompson W, Seeman GM, Treadwell T. The bioterrorism preparedness and response Early Aberration Reporting Hutwagner L, Thompson W, Seeman GM, Treadwell T. The bioterrorism preparedness and response Early Aberration Reporting
System (EARS). J Urban Health. 2003 Jun;80(2 Suppl 1):i89System (EARS). J Urban Health. 2003 Jun;80(2 Suppl 1):i89System (EARS). J Urban Health. 2003 Jun;80(2 Suppl 1):i89System (EARS). J Urban Health. 2003 Jun;80(2 Suppl 1):i89----96.96.96.96.
4.4.4.4. Jackson ML, Baer A, Painter I, Duchin J. A simulation study comparinJackson ML, Baer A, Painter I, Duchin J. A simulation study comparinJackson ML, Baer A, Painter I, Duchin J. A simulation study comparinJackson ML, Baer A, Painter I, Duchin J. A simulation study comparing aberration detection algorithms for syndromic surveillance. g aberration detection algorithms for syndromic surveillance. g aberration detection algorithms for syndromic surveillance. g aberration detection algorithms for syndromic surveillance.
BMC Med Inform Decis Mak. 2007;7:6.BMC Med Inform Decis Mak. 2007;7:6.BMC Med Inform Decis Mak. 2007;7:6.BMC Med Inform Decis Mak. 2007;7:6.
5.5.5.5. Zhu Y, Wang W, Atrubin D, Wu Y. Initial evaluation of the early aberration reporting systemZhu Y, Wang W, Atrubin D, Wu Y. Initial evaluation of the early aberration reporting systemZhu Y, Wang W, Atrubin D, Wu Y. Initial evaluation of the early aberration reporting systemZhu Y, Wang W, Atrubin D, Wu Y. Initial evaluation of the early aberration reporting system--------Florida. MMWR Morb Mortal Wkly Rep. Florida. MMWR Morb Mortal Wkly Rep. Florida. MMWR Morb Mortal Wkly Rep. Florida. MMWR Morb Mortal Wkly Rep.
2005 Aug 26;54 Suppl:1232005 Aug 26;54 Suppl:1232005 Aug 26;54 Suppl:1232005 Aug 26;54 Suppl:123----30.30.30.30.
6.6.6.6. Najmi AH, Magruder SF. An adaptive prediction and detection algorithm for multistream syndromic surveillance. BMC Med Inform Najmi AH, Magruder SF. An adaptive prediction and detection algorithm for multistream syndromic surveillance. BMC Med Inform Najmi AH, Magruder SF. An adaptive prediction and detection algorithm for multistream syndromic surveillance. BMC Med Inform Najmi AH, Magruder SF. An adaptive prediction and detection algorithm for multistream syndromic surveillance. BMC Med Inform
Decis Mak. 2005;5:33.Decis Mak. 2005;5:33.Decis Mak. 2005;5:33.Decis Mak. 2005;5:33.
7.7.7.7. Kleinmann KKleinmann KKleinmann KKleinmann KPPPP, Lazarus R, Platt R. , Lazarus R, Platt R. , Lazarus R, Platt R. , Lazarus R, Platt R. A Generalized Linear Mixed Models Approach for Detecting Incident Clusters A Generalized Linear Mixed Models Approach for Detecting Incident Clusters A Generalized Linear Mixed Models Approach for Detecting Incident Clusters A Generalized Linear Mixed Models Approach for Detecting Incident Clusters of Disease in Small of Disease in Small of Disease in Small of Disease in Small
Areas, with an Application tAreas, with an Application tAreas, with an Application tAreas, with an Application to Biological Terrorism. Am Jo Biological Terrorism. Am Jo Biological Terrorism. Am Jo Biological Terrorism. Am J Epidemiol. 2004;Epidemiol. 2004;Epidemiol. 2004;Epidemiol. 2004;159159159159((((3333))))::::217217217217----224224224224....
8.8.8.8. Kleinmann KP, Abrams A, Mandl K, Platt R. Kleinmann KP, Abrams A, Mandl K, Platt R. Kleinmann KP, Abrams A, Mandl K, Platt R. Kleinmann KP, Abrams A, Mandl K, Platt R. Simulation and Other Evaluation Approaches Simulation for Assessing Statistical Simulation and Other Evaluation Approaches Simulation for Assessing Statistical Simulation and Other Evaluation Approaches Simulation for Assessing Statistical Simulation and Other Evaluation Approaches Simulation for Assessing Statistical
Methods of Biologic TerroMethods of Biologic TerroMethods of Biologic TerroMethods of Biologic Terrorism Surveillance. MMWRrism Surveillance. MMWRrism Surveillance. MMWRrism Surveillance. MMWR Morb Mortal Wkly Rep.Morb Mortal Wkly Rep.Morb Mortal Wkly Rep.Morb Mortal Wkly Rep. 2005 Aug 26;54(Suppl):2005 Aug 26;54(Suppl):2005 Aug 26;54(Suppl):2005 Aug 26;54(Suppl):101101101101----108108108108....
9.9.9.9. Fricker RD, Jr., Hegler BL, Dunfee DA. Fricker RD, Jr., Hegler BL, Dunfee DA. Fricker RD, Jr., Hegler BL, Dunfee DA. Fricker RD, Jr., Hegler BL, Dunfee DA. Comparing syndromic surveillance detection methods: EARS' versus a CUSUMComparing syndromic surveillance detection methods: EARS' versus a CUSUMComparing syndromic surveillance detection methods: EARS' versus a CUSUMComparing syndromic surveillance detection methods: EARS' versus a CUSUM----based based based based
methodology. Stat Med. 2008 Jul 30;27(17):3407methodology. Stat Med. 2008 Jul 30;27(17):3407methodology. Stat Med. 2008 Jul 30;27(17):3407methodology. Stat Med. 2008 Jul 30;27(17):3407----29.29.29.29.
10.10.10.10. Hutwagner Hutwagner Hutwagner Hutwagner LC, Thompson WW, Seeman GM, Treadwell T. A simulation model for assessing aberration detection methods used in LC, Thompson WW, Seeman GM, Treadwell T. A simulation model for assessing aberration detection methods used in LC, Thompson WW, Seeman GM, Treadwell T. A simulation model for assessing aberration detection methods used in LC, Thompson WW, Seeman GM, Treadwell T. A simulation model for assessing aberration detection methods used in
public health surveillance for systems with limited baselines. Stat Med. 2005 Feb 28;24(4):543public health surveillance for systems with limited baselines. Stat Med. 2005 Feb 28;24(4):543public health surveillance for systems with limited baselines. Stat Med. 2005 Feb 28;24(4):543public health surveillance for systems with limited baselines. Stat Med. 2005 Feb 28;24(4):543----50.50.50.50.
11.11.11.11. Tokars JI, Burkom H, Xing J, English R, Bloom Tokars JI, Burkom H, Xing J, English R, Bloom Tokars JI, Burkom H, Xing J, English R, Bloom Tokars JI, Burkom H, Xing J, English R, Bloom S, Cox K, et al. Enhancing timeS, Cox K, et al. Enhancing timeS, Cox K, et al. Enhancing timeS, Cox K, et al. Enhancing time----series detection algorithms for automated series detection algorithms for automated series detection algorithms for automated series detection algorithms for automated
biosurveillance. Emerg Infect Dis. 2009 Apr;15(4):533biosurveillance. Emerg Infect Dis. 2009 Apr;15(4):533biosurveillance. Emerg Infect Dis. 2009 Apr;15(4):533biosurveillance. Emerg Infect Dis. 2009 Apr;15(4):533----9.9.9.9.
12.12.12.12. GarciaGarciaGarciaGarcia----Castrillo Riesgo L, Krafft T, Ziemann A, Fischer M, Lippert F, Vergeiner G, Krämer A, Brand H for Castrillo Riesgo L, Krafft T, Ziemann A, Fischer M, Lippert F, Vergeiner G, Krämer A, Brand H for Castrillo Riesgo L, Krafft T, Ziemann A, Fischer M, Lippert F, Vergeiner G, Krämer A, Brand H for Castrillo Riesgo L, Krafft T, Ziemann A, Fischer M, Lippert F, Vergeiner G, Krämer A, Brand H for tttthe SIDARTHa project group he SIDARTHa project group he SIDARTHa project group he SIDARTHa project group
(ed(ed(ed(edssss.). .). .). .). The SIDARTHa Coding Manual The SIDARTHa Coding Manual The SIDARTHa Coding Manual The SIDARTHa Coding Manual –––– How to generate syndromes based on routinely collected emergency care data for the How to generate syndromes based on routinely collected emergency care data for the How to generate syndromes based on routinely collected emergency care data for the How to generate syndromes based on routinely collected emergency care data for the
European syndromic surveillance system SIDARTHa. European syndromic surveillance system SIDARTHa. European syndromic surveillance system SIDARTHa. European syndromic surveillance system SIDARTHa. 2009, 2009, 2009, 2009, Bad Honnef.Bad Honnef.Bad Honnef.Bad Honnef.
top related