USE OF ANIMAL DATA IN PUBLIC HEALTH SURVEILLANCE FOR EMERGING ZOONOTIC DISEASES by Linda Vrbova Hon.B.Sc., University of Toronto, 2000 M.Sc., University of Toronto, 2003 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in The Faculty of Graduate Studies (Population and Public Health) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) April 2013 © Linda Vrbova, 2013 Abstract Infectious agents transmitted between animals and humans (zoonoses) are important causes of emerging infectious diseases with major societal, economic, and public health implications. In order to prevent and control emerging zoonotic diseases (EZDs), they should ideally be identified in animals before they affect the human population. The utility of animal data for public health EZD surveillance was investigated in this thesis in four studies: a systematic literature review of current EZD surveillance systems and three critical examinations of pilot agricultural animal health surveillance systems. The first critical examination used expert-elicited criteria of EZD surveillance needs to evaluate a sentinel clinical pre-diagnostic system. The other two studies used statistical modeling to assess the ability of a laboratory-based system and an integrated system with both human and animal data to detect known patterns and outbreaks. The systematic review identified few evaluated surveillance systems, hence an evidence base for successful systems could not be obtained. Experts identified diagnostic data from laboratories and information on potential human exposures as important for public health action. While the sentinel animal surveillance system was not deemed useful on its own, identified gaps and biases in laboratory submissions suggest that sentinel veterinarians could inform animal laboratory surveillance. Seasonal trends and expected events of public health importance were identified in animal diagnostic laboratory data, however, statistical surveillance in either pre-diagnostic or diagnostic ii data streams did not provide adequate early warning signals for action. While the integrated surveillance for Salmonella bacteria allowed for the examination of the relationship between human and animal data, statistical alerts did not correlate with expert-identified investigations. Laboratory surveillance is likely the best candidate for EZD surveillance in animals, however, this information needs to be supplemented with potential human exposure information, as well as knowledge of data gaps and biases inherent in the data. Without this additional risk information to convert the animal data into risk for humans, the best use of animal laboratory data at this time is to help generate hypotheses in epidemiological investigations and in helping evaluate programs by examining longer-term trends. iii Preface The central chapters of this thesis are constructed as scientific manuscripts that have either been published (Chapter 2) or will be submitted for publication (Chapters 3-5) in peer-reviewed journals. These manuscripts have or will have co-authors as detailed below, but it should be understood that I, as primary author, take full responsibility for this thesis work. Ethical Approval was required for Chapter 5 of this thesis, as this was the only chapter that included human data; approval was obtained through the UBC Behavioural Research Ethics Board (UBC BREB Number: H09-00601). A version of Chapter 2 in this thesis has been published: Vrbova L, Stephen C, Kasman N, Boehnke R, Doyle-Waters M, Gibson B, Chablitt-Clark A, Patrick D. Systematic Review of Surveillance Systems for Emerging Zoonoses. Transboundary and Emerging Diseases; 2010 Jun; 57(3):154-61. I conceived of the study along with Craig Stephen, wrote the proposal to obtain the necessary funds, identified collaborators, conducted all analyses, and wrote the manuscript. Co-authors were involved in designing and conducting the library search (60% Doyle-Waters, 40% Vrbova), refining the inclusion and exclusion criteria (all authors), reviewing papers for inclusion in study and extracting data from papers (70% Vrbova and 30% Kasman), reviewing study results and manuscript (all authors). Overall contribution: 90%. Chapter 3 (Animal Health Surveillance Using Sentinel Veterinarians: a Public Health Perspective on a Pilot Project), first author: Vrbova L, proposed co-authors: Craig Stephen, David Patrick, Mieke Koehoorn, Aidan Nikiforuk. I designed the expert consensus meeting as well as the sentinel surveillance pilot, recruited experts and sentinels, constructed the web-based reporting system, and conducted all analyses. I designed a sentinel evaluation survey along with summer student, iv Aidan Nikiforuk, who conducted the survey and analyzed the results. I wrote the manuscript that was edited and revised with the aid of my supervisory committee. Overall contribution: 95%. Chapter 4 (Detection of Emerging Infectious Disease Trends and Clusters in Animal Laboratory Data for Public Health Surveillance), first author: Vrbova L, proposed co-authors: Craig Stephen, David Patrick, Mieke Koehoorn, John Berezowski, Rick White. I designed the study, extracted, cleaned, and analyzed all data. Data coding of laboratory diagnoses into etiologic agents and syndromes was done with the help of veterinary epidemiologists John Berezowski and Craig Stephen. Statistician, Rick White, developed the generalized additive model used in the analyses and edited the relevant methods section of the manuscript. I wrote the manuscript that was edited and revised with the aid of my supervisory committee. Overall contribution: 95%. Chapter 5 (Utility of Surveillance Algorithms in the Analyses of Multi-Species Salmonella Surveillance Data in British Columbia, Canada), first author: Vrbova L, proposed co-authors: Craig Stephen, David Patrick, Mieke Koehoorn, Eleni Galanis, Jane Parmley, Nancy DeWith, Colin Roberston. I designed the study with members of the British Columbia Integrated Salmonella Surveillance program: medical epidemiologist Eleni Galanis, and veterinary epidemiologists Jane Parmley and Nancy DeWith. I cleaned and analyzed all data, with the exception of statistical analyses for signals in time series, designed by me, but run in the statistical program R by Colin Roberston. I wrote the manuscript that was edited and revised with the aid of my supervisory committee and Eleni Galanis. Overall contribution: 90%. v Table of Contents Abstract ................................................................................................................................................ ii Preface .................................................................................................................................................iv Table of Contents .................................................................................................................................vi List of Tables ....................................................................................................................................... xii List of Figures ..................................................................................................................................... xiv List of Abbreviations .......................................................................................................................... xvi Acknowledgements .......................................................................................................................... xvii 1 Introduction ...................................................................................................................................... 1 1.1 Emerging Zoonotic Diseases (EZDs) ...................................................................................... 4 1.2 Examples of Recent EZDs Linked with Agricultural Animals ................................................. 8 1.2.1 Socio-economic Significance.................................................................................. 8 1.2.2 The Role of Agricultural Animals in EZDs ............................................................. 11 1.2.3 Identification of EZDs in Agricultural Animals Before People are Affected ........ 13 1.3 Public Health Surveillance ................................................................................................... 17 1.4 Animal Health Surveillance for Public Health ..................................................................... 19 1.5 EID Surveillance in British Columbia ................................................................................... 23 1.6 Gaps in Understanding and Thesis Structure...................................................................... 24 1.7 Figures ................................................................................................................................. 28 vi 2 Systematic Review of Surveillance Systems for Emerging Zoonoses ............................................. 32 2.1 Introduction ........................................................................................................................ 32 2.2 Materials and Methods ....................................................................................................... 34 2.3 Results ................................................................................................................................. 37 2.4 Discussion ............................................................................................................................ 40 2.5 Tables and Figures ............................................................................................................... 45 3 Animal Health Surveillance Using Sentinel Veterinarians: a Public Health Perspective on a Pilot Project ................................................................................................................................................ 49 3.1 Introduction ........................................................................................................................ 49 3.2 Methods .............................................................................................................................. 52 3.2.1 Sentinel Veterinarian Recruitment ...................................................................... 52 3.2.2 Case Definition, Data Collection, and Reporting ................................................. 53 3.2.3 Timeliness ............................................................................................................ 56 3.2.4 Animal Species Seen by Sentinels ........................................................................ 56 3.2.5 Reasons Animals Were Seen by Sentinels and Suspected Infections and Syndromes ............................................................................................................................ 58 3.2.6 3.3 Laboratory Submissions by Sentinels .................................................................. 58 Results ................................................................................................................................. 62 3.3.1 Sentinel Veterinarians: Recruitment and Reporting ........................................... 62 vii 3.3.2 Animal Species Seen, Reasons They Were Seen, Suspected Infections and Syndromes ............................................................................................................................ 65 3.4 3.3.3 Laboratory Submissions by Sentinels .................................................................. 67 3.3.4 Use of Animal Data in Public Health Practice ...................................................... 71 Discussion ............................................................................................................................ 73 3.4.1 Conducting Surveillance Using Sentinel Veterinarians........................................ 74 3.4.2 Animal Species Seen, Reasons They Were Seen, Suspected Infections and Syndromes ............................................................................................................................ 76 3.4.3 Use of Sentinel Animal Health Data for Public Health Practice .......................... 80 3.4.4 Limitations ........................................................................................................... 90 3.5 Conclusions ......................................................................................................................... 92 3.6 Tables and Figures ............................................................................................................... 93 4 Detection of Emerging Infectious Disease Trends and Clusters in Animal Laboratory Data for Public Health Surveillance ............................................................................................................... 103 4.1 Introduction ...................................................................................................................... 103 4.2 Methods ............................................................................................................................ 106 4.2.1 Creation of Surveillance Database: Data Extraction, Cleaning, and Coding ..... 106 4.2.2 Descriptive Analyses and Overall Trends........................................................... 108 4.2.3 Signals in Time ................................................................................................... 109 viii 4.2.4 4.3 4.4 Trends, Events, and Outbreaks .......................................................................... 111 Results ............................................................................................................................... 112 4.3.1 Creating the Surveillance Database and Descriptive Analysis ........................... 112 4.3.2 Descriptive Analysis and Correlation with Expected Trends ............................. 113 4.3.3 Signals in Time Series ......................................................................................... 117 4.3.4 Correlation of Signals to Events and Outbreaks ................................................ 119 Discussion .......................................................................................................................... 123 4.4.1 Feasibility: Data Extraction and Coding ............................................................. 123 4.4.2 Data Validity: Seasonality and Expected Trends ............................................... 125 4.4.3 Signals in Pre-Diagnostic and Diagnostic Data Streams .................................... 127 4.4.4 Ability of Data to Predict Known Outbreaks...................................................... 130 4.5 Conclusion ......................................................................................................................... 133 4.6 Tables and Figures ............................................................................................................. 135 5 Utility of Surveillance Algorithms in the Analyses of Multi-Species Salmonella Surveillance Data in British Columbia, Canada ................................................................................................................. 145 5.1 Introduction ...................................................................................................................... 145 5.2 Methods ............................................................................................................................ 150 5.2.1 Human Data: Provincial Public Health Reference Laboratory ........................... 151 5.2.2 Animal Data: Provincial Animal Health Diagnostic Laboratory ......................... 152 ix 5.3 5.2.3 Food Data: Canadian Integrated Program on Antibiotic Resistance Surveillance153 5.2.4 Descriptive Analyses .......................................................................................... 153 5.2.5 Statistical Signals in Individual Time Series ....................................................... 154 5.2.6 Cross-Sectoral Investigations and Signals.......................................................... 157 Results ............................................................................................................................... 159 5.3.1 Descriptive Analyses and Statistically Significant Signals in Separate Sectors .. 159 5.3.2 Cross-Sectoral Analyses: Comparison of Proportions, Investigations and Statistically Significant Signals across Sectors .................................................................... 161 5.3.3 Integrated Surveillance Working Group Investigations .................................... 162 5.3.4 Statistically Significant Signals across Sectors ................................................... 164 5.3.5 Comparison of Working Group Investigations with Statistically Significant Algorithm Signals Across Sectors ....................................................................................... 166 5.4 Discussion .......................................................................................................................... 167 5.4.1 Cross-Sectoral Serotypes ................................................................................... 168 5.4.2 Cross-Sectoral Signals: Working Group versus Algorithms ............................... 170 5.4.3 Discussion of the Data Sources and Methods Used .......................................... 172 5.4.4 Public Health Significance of Cross-Sectoral Signals ......................................... 176 5.5 Conclusion ......................................................................................................................... 180 5.6 Tables and Figures ............................................................................................................. 181 x 6 Conclusion ..................................................................................................................................... 185 6.1 Risk-Based Animal Surveillance for Public Health ............................................................ 186 6.2 The “Foundation” of Animal Health Surveillance: Laboratories ....................................... 189 6.3 Specificity of an Animal Signal: What Does It Mean? ....................................................... 192 6.4 Sentinel Surveillance: The “Weird” Network .................................................................... 194 6.5 Policy Implications............................................................................................................. 196 6.6 Conclusion ......................................................................................................................... 197 References ....................................................................................................................................... 198 Appendices....................................................................................................................................... 211 Appendix A: Appendices for Chapter 2......................................................................................... 212 Appendix B: Appendices for Chapter 3 ......................................................................................... 227 Appendix C: Appendices for Chapter 4 ......................................................................................... 247 xi List of Tables Table 2.1 MeSH search terms to describe surveillance. .................................................................... 45 Table 3.1 Variables collected from sentinel veterinarians. ............................................................... 93 Table 3.2 Reasons why veterinarian sentinels stated they submitted and did not submit samples to an external laboratory for mammals and birds by suspected infection. .................................... 94 Table 3.3 Suspected infections and laboratory sampling by species, as reported by sentinel veterinarians between March 1, 2009 and March 31, 2010. ...................................................... 95 Table 3.4 Submissions to an external laboratory for ill and infectious mammals and birds by syndrome. .................................................................................................................................... 96 Table 3.5 Logistic regression results for cattle submissions to any external laboratory and to the provincial diagnostic animal health laboratory. .......................................................................... 97 Table 3.6 Logistic regression results for equine submissions to any external laboratory. ............... 98 Table 3.7 Expert focus group sentinel animal health surveillance evaluation themes..................... 99 Table 4.1 ICD-10 Diagnosis Code Categories (based on WHO, 2006). ............................................ 135 Table 4.2 All chicken diagnostic codes made at the BC Animal Health Centre mapped to ICD-10 diagnostic categories (n=14,321)............................................................................................... 136 Table 4.3 All chicken infectious and parasitic diagnoses made at the BC Animal Health Centre by likely etiologic agent(s) and zoonotic status by Year 1998-2007 (n=3,661).............................. 137 Table 4.4 All cattle diagnostic codes mapped to ICD-10 diagnostic categories made at the BC Animal Health Centre (n=8,221). ............................................................................................... 138 Table 4.5 All cattle infectious and parasitic diagnoses by likely etiologic agent(s) made at the BC Animal Health Centre and zoonotic status by Year 1998-2007 (n=1,552). ............................... 139 xii Table 4.6 Statistically significant signals in more than one time series in chicken by commodity group and type of data stream. ................................................................................................. 140 Table 4.7 Statistically significant signals in more than one time series in cattle by commodity group and type of data stream. ........................................................................................................... 141 Table 5.1 Investigations into serotypes that were present in at least two of the three sectors (human, animal and food) conducted by the BC Integrated Salmonella Surveillance Working Group in 2010. ........................................................................................................................... 181 Table 5.2 Statistically significant signals across sectors in 2010. .................................................... 182 xiii List of Figures Figure 1.1 Global examples of emerging and re-emerging infectious diseases. ............................... 28 Figure 1.2 Illustration of the five stages through which pathogens of animals evolve to cause disease confined to humans. ....................................................................................................... 29 Figure 1.3 The continuum of emerging disease surveillance, highlighting the decreasing sensitivity and increasing specificity ............................................................................................................. 30 Figure 1.4 Animal health data flow for surveillance of infectious diseases. ..................................... 31 Figure 2.1 Peer-reviewed Articles Describing Emerging Zoonoses Surveillance Systems by Year of Publication between 1992 and 2006, by Type of Data in the System (N=201*). ....................... 48 Figure 3.1 Proportions of large animal species seen by sentinels and those counted in the agricultural census. .................................................................................................................... 100 Figure 3.2 Proportions of poultry species seen by sentinels and those counted in the agricultural census. ....................................................................................................................................... 101 Figure 3.3 Numbers of infectious syndromes (n=177) and non-infectious syndromes (n=287) suspected in ill animals by sentinels. ......................................................................................... 102 Figure 3.4 Numbers of cases with non-infectious syndromes (n=49) and infectious syndromes (n=5) suspected in ill poultry by sentinels. ......................................................................................... 102 Figure 4.1 Time series of all chicken submissions from April 1, 1998 to March 31, 2007 to the BC Animal Health Centre................................................................................................................. 142 Figure 4.2 Seasonality in beef cattle submissions from April 1, 1998 to March 31, 2007 to the BC Animal Health Centre................................................................................................................. 143 xiv Figure 4.3 Time series of all cattle whole animal submissions from April 1, 1998 to March 31, 2007 to the BC Animal Health Centre................................................................................................. 144 Figure 5.1 Comparison of serotypes across the sectors between 2008 and 2010. ........................ 183 Figure 5.2 Salmonella Enteritidis (SE) isolates in 2010 from animals (chicken), food (chicken meat) and humans (endemic human cases). ....................................................................................... 184 xv List of Abbreviations AI: Avian influenza BSE: Bovine spongiform encephalopathy C$: Canadian Dollar E. coli: Escherichia coli EID: Emerging infectious disease EZD: Emerging zoonotic disease HIV: Human immunodeficiency virus HPAI: Highly pathogenic avian influenza OIE: World Organisation for Animal Health US$: United States Dollar vCJD: Variant Creutzfeldt-Jakob disease WHO: World Health Organization WNV: West Nile virus xvi Acknowledgements I want to thank my thesis supervisory committee for their contributions of time, energy, and ideas. First, I would like to thank my superb co-supervisors, David Patrick and Craig Stephen for their unwavering support. I thank David for providing direction in murky waters, by helping me distil complex ideas into testable scientific hypotheses, and providing coherent answers to my endless questions. I owe thanks to Craig for his infectious enthusiasm, penetrating questions and insights that taught me to question more deeply, ongoing critical analysis of my work, and financial support. I thank my invaluable committee member, Mieke Koehoorn, for her clarity of thought, ensuring that my work was clear and coherent. I am appreciative of the educational environment afforded by the UBC School of Population and Public Health. I thank the staff for their administrative assistance, my fellow students for the enlightening experiences we shared, and my professors for their patient instruction. I acknowledge financial support from the National Science and Engineering Research Council, and UBC’s Four Year Fellowship, Graduate Fellowship, Cordula and Gunter Paetzold Fellowship, as well as the Pacific Century Graduate Scholarship. While there are too many people to thank individually, my research would not have been possible without the following: animal and public health experts that contributed to consensus meetings on animal disease surveillance evaluation and utility of animal data for public health surveillance; sentinel veterinarians in British Columbia who reported on animals seen as part of their work for the duration of the year–long sentinel surveillance pilot project; people at the British Columbia Ministry of Agriculture Animal Health Centre, British Columbia Centre for Disease Control, British xvii Columbia Ministry of Environment, Centre for Coastal Health, Canadian Cooperative Wildlife Health Centre, and the Alberta Veterinary Surveillance Network. I offer my enduring gratitude to my health research colleagues that were excellent sounding boards and sources of motivation: Anne Harris, Kate Sawford, Colin Robertson, Sue Pollock, David Roth, Michelle Anholt, and Elaine Fuertes. Special thanks are owed to my family and friends, whose have provided me with much needed moral support throughout my years of education. First and foremost I thank my mom Lenka, who was instrumental in my finishing this dissertation. Thank you as well to the rest of my family, my dads Ludek and Ivo, sisters Dominika, Tereza, and Anna, and my beloved friends Mirela Cara and Zara Comer. My final thanks go to my husband, Mark Pahulje, who unwaveringly supported me in taking all the time and energy I needed to devote to this work. xviii 1 Introduction In the 1960s, high-ranking public health officials concluded that it was time close the book on infectious diseases and move on to other public health problems (McDade, 1997). Advances in microbiology, antibiotics, immunization, vector control, sanitation, along with notable accomplishments such as smallpox eradication, led to well-earned optimism about ending a large source of suffering that had been present throughout human history. However, over the last decades a resurgence of so-called emerging and re-emerging infectious diseases (EIDs) has occurred (Word Health Organization, 2005, Jones et al., 2008). “Emerging infectious diseases are diseases of infectious origin whose incidence in humans has increased within the recent past or threatens to increase in the near future. These also include those infections that appear in new geographic areas or increase abruptly. The new infectious diseases and those which are re-emerging after a period of quiescence are also grouped under emerging infectious diseases.” (WHO, 2005) While very useful, this definition of EIDs is very anthropocentric: EIDs are a global phenomenon shared among species, and therefore should not be studied in humans alone. EIDs remain a significant cause of morbidity and mortality in both human and animal populations. In 2010, an estimated 34 million people were living with human immunodeficiency virus (HIV) worldwide, up 17% from 2001 (UNAIDS, 2011), and there were about 650,000 cases of emerging multi-drug-resistant (MDR) tuberculosis (TB) in the world with 150,000 deaths (WHO, 2012). The economic impacts of EIDs have been significant: highly pathogenic avian 1 influenza (HPAI) has cost the Asian economy US$10 billion (Elci, 2006), and Severe Acute Respiratory Syndrome (SARS) cost the East Asian economies US$200 billion (World Bank, 2005), and the Canadian economy C$1.5 billion (Darby, 2003). Figure 1.1 shows the location of some of the recent EIDs, illustrating that EIDs have emerged on every continent. Infectious diseases vary widely, since they are illnesses resulting from the presence and/or growth of a wide diversity of microbial organisms that can act as pathogens in a host, such as protozoa, helminthes, fungi, bacteria, rickettsiae, viruses and prions. Hosts for pathogens can be all living organisms, including humans, other animals, fungi and plants. Microorganisms, including pathogens, adapt to changing environments and therefore change the shared risk of living hosts acquiring (new) infectious diseases. Infectious disease emergence is an inevitable consequence of evolution, especially as environments change. The task for epidemiologists is more accurate prediction, prevention and response. A large number of factors linked to the emergence of infectious diseases have been identified, including: ecological changes (e.g. agricultural or economic development, land use, climate change), human demographic changes and behavior (e.g. urbanization, civil conflict, commercial sex trade), travel and commerce (e.g. worldwide movement of people and goods, air travel), technology and industry (e.g. food production and processing, globalization of food supplies, new medical devices and procedures, immunosuppressive drugs and antibiotics), microbial adaptation and change (e.g. microbial evolution, response to selection pressure), and the breakdown of public health measures (e.g. curtailment or reduction in disease prevention programmes, inadequate sanitation or vector control measures) (Morse, 2004). Many of these 2 factors are not new; they are well known from history when large epidemics were associated with concentration of people around production hubs, developments in agriculture (in particular animal husbandry), commerce, wars, migration, and conquests of new territories (Nelson and Williams, 2007, Diamond, 1999). Some of these factors related to emergence of infectious diseases can also be viewed as population risk factors for illness from EIDs and are closely related to general determinants of health, in general consisting of the social and economic environment, physical environment, and a person’s individual characteristics and behaviours (World Health Organization, 2012). Prevention of EIDs is therefore likely to be affected by upstream determinants that affect the resilience of a population, such as income inequality, education, social support networks, nutrition, sanitation, and safe housing and employment. There are a number of possible reasons why, to date, we have not been able to adequately forecast EIDs. One possibility is that prediction requires a more quantitative understanding of the drivers of emergence (Woolhouse, 2011). While there are a number of predictive methods available, including risk factor analysis, risk modeling and dynamic modeling, useful predictions using such methods ultimately depend on the quality and availability of input data (Woolhouse, 2011). While quantifying drivers such as investment in public health, population displacement, natural disasters and war is currently difficult (Woolhouse, 2011), other drivers of emergence, such as climate and population density have been quantified and used in EID predictive models (Jones et al., 2008). While such analyses show promise that EID prediction may be possible, especially for risk assessments of specific emerging diseases, such as predicting the risk of 3 Hantavirus pulmonary syndrome (Glass et al., 2000) and Sin Nombre virus (Glass et al., 2002) using remotely sensed geographic data, they are based on numerous assumptions, resulting in questionable accuracy and overall usefulness. Ultimately, accurate prediction of EIDs may not be possible due to the complex nature of emergence, since chaos plays an important role in both the biological sense (e.g. mutation and gene exchange are random) as well as the broad socio-ecological sense (e.g. civil conflict, technological and economic development). The interactions between people and ecosystems may in fact reflect a complexity that cannot be captured by even our most sophisticated complex systems models (Funtowicz and Ravetz, 1994). 1.1 Emerging Zoonotic Diseases (EZDs) Historically, a number of important human infectious diseases originated in animals as zoonoses, diseases involving the transmission of pathogens between animals and humans. Smallpox, a virus that has likely killed more humans than any other virus in history, is thought to have come from camels, and malaria may have emerged from apes (Wolfe, 2011). The measles virus is thought to have emerged in humans about 10,000 years ago, from an ancestor of rinderpest in cattle and peste des petites ruminants (Dobson and Carper, 1996). More recently, a hypothesis was put forward that HIV-1 was first introduced into humans from non-human primates (likely chimpanzees), through handling of infected animal carcasses (‘bush meat’) by hunters possibly around the year 1900 (Wolfe, 2011, Wolfe et al., 2004). The estimated proportion of recent EIDs that are zoonoses range from 60% to 75% (Jones et al., 2008, Woolhouse and Gowtage-Sequeria, 2005, Taylor et al., 2001, Smolinski et al., 2003) 4 An emerging zoonotic disease (EZD) is: a zoonosis that is newly recognized or newly evolved, or that has occurred previously but shows an increase in incidence or expansion in geographical, host or vector range (Word Health Organization et al., 2004). Of particular concern are a number of recent EZDs lethal to humans (e.g. Ebola virus), capable of spreading rapidly (e.g. SARS coronavirus), and causing pandemics (e.g. HIV-1 and Influenza viruses that have undergone reassortment in animal hosts) (Daszak et al., 2006). Further, most of the organisms listed as potential bio-terrorism and bio-warfare agents are zoonoses, including anthrax, plague, tularemia and the various hemorrhagic fever viruses (Morse, 2004). For many EZDs, there are few effective therapies, vaccines or other strategies available to us (Daszak et al., 2006). Therefore, surveillance and control programs and development of drug and vaccine candidates are a high priority for public health programs (Smolinski et al., 2003), requiring an understanding of the causes of emergence and of host and pathogen population dynamics (Daszak et al., 2006). In general, infectious disease emergence involves the introduction of an agent into a new host population, followed by the dissemination of the agent within the new host population (Morse, 2004). However, EIDs may also arise differently, for example from novel exposure pathways, or endemic opportunistic agents already present in a population, with factors such as immune deficiency caused by HIV or cancer treatments, or waning host immunity due to aging, contributing to their spread and impact (Morris and Potter, 1997, Morens et al., 2004). Despite the great number of possible zoonoses, few have successfully established themselves in the human population through efficient person-to- 5 person spread (Morse, 2004). For example, dramatic EZD outbreaks caused by Ebola and Nipah viruses have caused human disease with a rapid course and high mortality; luckily, they were transmitted poorly from person to person (Morse, 2004). The factors leading to the transition from a pathogen that is only transmitted from animals to humans to a pathogen that can also sustain human-to-human transmission are numerous and not fully understood (Wolfe et al., 2007). Some of the biological factors contributing to the species barrier are non-specific and induced immunity, receptor specificity, and host body temperature (Morse, 2004). Wolfe et al. (2007) outline five stages of infectious disease emergence from a stage where a pathogen exclusively infects animals to a pathogen that exclusively infects humans (Figure 1.2). Figure 1.2 illustrates the complexity of EZDs: diseases differ widely in terms of their host animals as well as their transmissibility to humans and within the human population. Since we want to prevent EIDs from becoming entrenched in human populations, the timeliest action would be for us to look at such EZDs that are not (yet) agents transmitted exclusively among humans. The close proximity of humans and domesticated animals provides multiple opportunities for pathogen sharing. Domesticated animals can act as mixing vessels (e.g. Influenza), amplifying hosts (e.g. Nipah and Hendra), as well as widely-distributed sources of the original infection (e.g. bovine spongiform encephalopathy (BSE), Escherichia coli (E. Coli) O157:H7). While it has been estimated that 72% of the EZDs that emerged over the past six decades originated in wildlife (Jones et al., 2008), in many of these cases domesticated animals played a critical role in transmission of disease to humans. This is because domestic animals provide 6 more exposure opportunities to humans, both direct (animal contact) and indirect (animalderived products such as food), resulting in a larger public health impact. The large number of endemic zoonoses (e.g. campylobacteriosis, salmonellosis, E. Coli O157:H7) still present in Europe (European Food Safety Authority and European Centre for Disease Prevention and Control, 2011), the United States (Scallan et al., 2011) and Canada (Public Health Agency of Canada, 2011b), suggest that domestic animals continue to be a source of exposure for humans and are important to watch for re-emergence of such diseases. Additionally, in the event that domestic animals harbor an infectious disease agent that originated in wildlife (e.g. Nipah and Hendra), domestic animals can help to identify these agents as potential EZDs, since they not only show the agent was able to cross the species barrier, but also that the agent may be transmissible in a setting different than that of its original wildlife host. EIDs arising in agricultural animals with large economic impacts have resulted in significant investments in infrastructure for diagnosis and surveillance in animals; for example, following the emergence of avian influenza (AI) worldwide (Fouchier et al., 2003), BSE and pathogenic strains of E. coli in the United Kingdom (DEFRA, 2011), and BSE and AI in Canada (LeBlanc, 2008, Pasick et al., 2009). With the exception of limited wild bird surveillance for AI (Parmley et al., 2008), the study of wildlife has not benefited in the same way, making wildlife data less available and reliable for surveillance (Stitt et al., 2007). Moreover, without clear exposure pathways to humans, or an ability to quantitatively assess the pathways if they are known, it is difficult to turn wildlife data into actionable public health risk information at this time. Therefore, in order to test a hypothesis that animal data can be used for public health EID 7 surveillance, it is better to work with the richest data possible at this time, as well as data that provide information on animals with known exposure pathways for people: domestic animals. 1.2 Examples of Recent EZDs Linked with Agricultural Animals 1.2.1 Socio-economic Significance Agricultural animal species have been linked to a number of EZDs of socio-economic importance that have caused significant morbidity and mortality in outbreaks in both animals and people, pandemics in people, incurred significant economic costs, and resulted in large responses from governments worldwide. The recent pandemic Influenza A H1N1 virus (pH1N1), a virus that is hypothesized to have originated in swine (Smith et al., 2009), spread highly effectively from person to person to cause a global human pandemic in 2009. Even prior to this latest pH1N1, swine were suggested to be ‘mixing vessels’ for influenza viruses coming from avian reservoirs such as waterfowl (e.g. ducks), and hence a potential source of influenza strains affecting the human population (Lipatov et al., 2004). While the emergence of pandemic influenza was historically thought to have occurred in China due to their integrated pig-duck agriculture system (Scholtissek and Naylor, 1988), current high intensity agriculture and the movement of livestock across borders in Europe and elsewhere may also present suitable conditions for influenza emergence (Lipatov et al., 2004). Highly pathogenic avian influenza (HPAI) has proved effective at spreading globally in the bird population, causing severe outbreaks in both poultry and humans in Asia and Africa 8 (Keusch et al., 2009). Outbreaks of HPAI in humans have suggested that poultry farming provides opportunities for exposure to novel influenza viruses (Li et al., 2004). Although EZDs are often rare, morbidity and mortality from these diseases has been significant in both humans and animals. Since 2003, there have been 608 human cases and 359 deaths (59% case-fatality rate) of H5N1 AI reported to the WHO (Word Health Organization, 2012a), and 7,279 separate outbreaks (each involving hundreds to many millions of birds that may have been culled) of H5N1 AI in poultry have been reported to the OIE (World Organisation for Animal Health (OIE), 2012c). Variant Creutzfeldt-Jakob disease (vCJD) is a fatal human neurodegenerative condition associated with consumption of meat contaminated by the BSE prion, with 214 cases worldwide reported to the WHO between October 1996 to March 2011 (Word Health Organization, 2012c). Currently, 190,302 cases of BSE in cattle have been reported to the OIE worldwide since the disease was identified in 1989 (World Organisation for Animal Health (OIE), 2012a). The recently emerged pH1N1 influenza virus did not remain rare, causing a pandemic infecting a large proportion of people; it is estimated that it caused infection in 32%-41% of the population in the Canadian province of Ontario (Achonu et al., 2011). The global economic impacts of EZDs from agricultural animals have been considerable: HPAI has cost the Asian economy US$10 billion (Elci, 2006), and bovine spongiform encephalopathy (BSE) cost the United Kingdom between $858 and $936 million per year between 1986 and 2000 (Organisation for Economic Co-operation and Development and World Health Organization, 2003). 9 The response to these EZDs by the media and the public has been substantial. Media coverage of EIDs in the USA was found to be relatively high in American elite newspapers (e.g. the New York Times and the Washington Post), peaking with newly reported human infections or infected animals, with public attention to news coverage following this trend (Ho et al., 2007). Around the times of increased media focus on AI between 1998 and 2006, half of Americans surveyed reported following the issue closely: when SARS emerged in 2003 more than 70% of Americans reported paying close attention to SARS-related news coverage, and when West Nile virus (WNV) increased its geographic spread in North America in 2002, 79% of Americans reported paying close attention to the news coverage (Ho et al., 2007). The Canadian public was so worried about eating beef in the midst of the BSE crisis, that Prime Minister Jean Chrétien made a show of lunching on beef steak at a restaurant in an effort to allay fears about Canadian beef (Canadian Broadcasting Corporation (CBC), 2004). In response to such intense public attention and devastating economic impacts, governments have invested in their agricultural animal surveillance programs. The first active BSE surveillance program in cattle was established at the height of the BSE outbreak in the United Kingdom in 1993, however, in Canada an active surveillance program was formally established in 2004, following the first Canadian BSE case identified in 2003 (LeBlanc, 2008). In 2004 the Canadian government invested C$92.1 million over five years to enhance measures for identification, tracking, tracing, and to enhance BSE surveillance (LeBlanc, 2008), which subsequently led to the current enhanced BSE surveillance program (Canadian Food Inspection Agency, 2012b). 10 Similarly with AI in poultry: prior to the first Canadian AI outbreak in 2004 there was no active surveillance for AI in poultry, only following this outbreak did the government invest C$3.25 million in the development of active surveillance programmes, enhanced on-farm biosecurity and emergency disease response planning (Pasick et al., 2009); with additional investments this ultimately led to the current enhanced AI surveillance program (Canadian Food Inspection Agency, 2012a, Canadian Food Inspection Agency, 2012b). While the focus of this thesis is on EZDs, it should be noted that many of the results will be relevant for endemic zoonoses as well. Although endemic zoonoses remain an issue in highincome countries in Europe and North America (European Food Safety Authority and European Centre for Disease Prevention and Control, 2011, Scallan et al., 2011, Public Health Agency of Canada, 2011b), the burden of endemic zoonoses is greatly underestimated in lower income countries (Maudlin et al., 2009). Endemic zoonoses in lower-income countries garner less attention than they deserve, and can be considered ‘neglected diseases’, despite the fact that they affect both the health and livelihoods of people in the poorest communities (Maudlin et al., 2009). 1.2.2 The Role of Agricultural Animals in EZDs Modern agricultural practices, food production and processing methods, as well as the proximity of agricultural animal populations to both wildlife and humans, have all played a role in the emergence of zoonoses. The feeding of bone meal to cattle is one example of an agricultural practice that contributed to EZD emergence: it has been hypothesized that BSE emerged through an interspecies transfer of the disease scrapie from sheep to cattle through 11 incomplete inactivation of the scrapie agent in sheep by-products that were fed to cattle (Morse, 1990, Wilesmith et al., 1991). Food animals, such as cattle and chicken, can now provide regular exposure routes to humans, by allowing pathogens present in a limited animal population to become more widespread. Modern food production and processing methods increase the chances that an accidental contamination will be amplified (Blancou et al., 2005). Pathogenic strains of E. coli, such as serotype O157:H7, for whom cattle (Karmali, 1989) and other ruminants (Caprioli et al., 2005) are considered major animal reservoirs, cause not only gastrointestinal illnesses, but also the more serious hemolytic uremic syndrome (HUS) in people. The effect that modern food processing plants can have on widespread contamination of meat products is illustrated by a recent outbreak of E. Coli O157:H7 linked to large beef processing plant occurring in the fall of 2012 in Canada; while to date only 18 human cases have been associated with this outbreak in four Canadian provinces (Public Health Agency of Canada, 2012c), a vast meat product recall was issued (1,800 products including steaks, ground beef and roasts spanning Canada and most USA states) and the plant was temporarily closed (Canadian Food Inspection Agency, 2012c, ProMED-mail, 2012). Another emerging foodborne bacteria present in the foodchain worldwide is Salmonella Enteritidis, thought to be acquired largely from hen eggs and chicken meat (Marcus et al., 2007). The evolution of antimicrobial resistance (AMR) is now a major cause of EIDs in developed countries (Jones et al., 2008), resulting in the emergence of antimicrobial-resistant Salmonella bacteria such as S. Typhimurium DT104 (Helms et al., 2005). The emergence of AMR in 12 Salmonella, has been hypothesized by some authors to be associated with the (mis-) use of antimicrobials in agriculture (Cohen and Tauxe, 1986, Angulo et al., 2000). While this hypothesis remains tenuous with respect to zoonoses in general, a recent Canadian study provides evidence that use of the antimicrobial ceftiofur in chicken resulted in extendedspectrum cephalosporin resistance in Salmonella Heidelberg from chicken and humans (Dutil et al., 2010). Three recently discovered paramyxovirus EZDs, the Nipah, Hendra, and Menangle viruses, illustrate how agricultural animals can act as intermediaries between people and wildlife (Chua et al., 2000, Mackenzie, 1999). Serological and viral isolation suggest that while all three viruses have bats as their natural reservoir hosts (Hyatt et al., 2004), spillover into human populations occurred following cases or outbreaks in livestock “amplifier” hosts (pigs for Nipah and Menangle viruses; horses for the Hendra virus) that were the sources of human exposure to these viruses (Philbey et al., 1998, Yob et al., 1999, Halpin et al., 2000, Chua et al., 2000, Chua et al., 2002). 1.2.3 Identification of EZDs in Agricultural Animals Before People are Affected Identification of an EZD in animals prior to (or concurrently with) the identification of disease in humans may help initiate early prevention and control measures, as well as help develop hypotheses on the origin and transmission patterns of the disease preventing further spread. A number of important EZDs were either first identified in animals, or subsequently found to have been present in animals prior to the first human case. 13 BSE emerged in the United Kingdom (UK) in 1986 as a serious neurological disorder of cattle, causing substantial livestock loss and trade restrictions, followed a full ten years later by the implication that BSE was the likely cause of a severe new variant of Creutzfeld Jacob Disease (vCJD) in people (DEFRA, 2011). The emergence of BSE illustrates a failure to capitalize on an animal ‘signal’ that existed prior to human cases, one that may have forewarned human risk. This likely occurred due to great uncertainty in the scientific community regarding the transmissibility and pathogenicity of the newly discovered prion pathogen responsible for BSE, especially with respect to humans (Ridley and Baker, 1999). When politicians were forced to make decisions in the face of this scientific uncertainty, beef in the UK was deemed safe prior to the identification of human vCJD cases (Ridley and Baker, 1999). Hendra virus was first identified in September 1994, when a sudden outbreak of an acute respiratory syndrome in thoroughbred horses in a training complex in Brisbane, Australia, caused a shutdown of the horse racing industry in the region (Field et al., 2001). As with BSE in cattle, the disease in horses was severe, characterised by severe respiratory signs and high (65%, 13/20) mortality (Field et al., 2001). Although the outbreak was contained and the causative agent was identified within days, two people involved with the nursing of the index equine case became ill with a severe influenza-like illness within one week of the death of the index case and prior to clinical signs in other horses (Field et al., 2001). Infection with the Hendra virus was demonstrated in both human cases, and one subsequently died after respiratory and renal failure (Field et al., 2001). 14 The first human case of HPAI H5N1 was identified in May 1997 in Hong Kong, followed by 18 more cases with six deaths (Tam, 2002). The H5N1 virus was isolated prior to these human cases from outbreaks of influenza in chicken farms in Hong Kong three months earlier, in March 1997 (Tam, 2002). Unfortunately, the outbreaks in chickens were not seen as a risk for humans: at the time, HPAI outbreaks devastating to the poultry industry had occurred at intervals in many parts of the world, but, as they did not affect humans, the concept of a species barrier between the avian and human hosts with need for reassortment between avian and human viruses in an intermediate host such as the pig was accepted at the time (Shortridge et al., 2003). This emergence of HPAI suggests that there is a lack of acceptance of early animal signals as presenting risk to humans without associated human illness, questioning their value from a public health perspective. Without a demonstrated jump by a pathogen from an animal to a person, a species barrier seems to be assumed. Nevertheless, this outbreak was “considered to be an incipient pandemic situation, the chicken being the source of virus for humans and, if so, was the first instance where a pandemic may have been averted” (Shortridge et al., 2003). Nipah virus was first identified in concurrent outbreaks in pigs and humans in Malaysia between September 1998 and April 1999, and was initially ascribed to Japanese encephalitis in humans (Chua et al., 2000) and to Classical Swine Fever in pigs (Chua, 2003). The outbreak resulted in 265 cases of encephalitis and 105 deaths in humans and the culling of approximately 1 million pigs (Chua et al., 2000). The disease in pigs was a non-specific febrile respiratory disease with occasional encephalitis, while humans had febrile encephalitis with high mortality (Chua et al., 2000). The majority of human cases had a history of direct contact with live pigs, most being pig farmers (Field et al., 2001). Retrospective investigations have suggested that 15 Nipah virus was likely causing disease in swine prior to the identification of the index human case as early as 1996, however, was not identified as a new syndrome because the clinical signs were not markedly different from those of several endemic diseases, and because morbidity and mortality were not remarkable (Mohd Nor et al., 2000, Aziz et al., 1999). These emergences suggest there are two general ways to identify EZDs in domestic animals: on the “frontlines” and in laboratories. Early clinical detection of diseases in domestic animals on the “frontline” can be done by people in direct contact with animals, such as owners/producers and veterinarians. Owners and producers have the opportunity to know the health status of their animals very well and can detect any changes or unusual signs and symptoms with a high degree of sensitivity, especially when the conditions affect productivity. Veterinarians, while not in constant contact with all domestic herds or flocks in their catchment areas as owners (and therefore not having as high a sensitivity to detect issues as owners), may be able to diagnose potential health issues with higher specificity due to their medical training, and may be able detect unusual syndromes or disease presentations across a larger area (i.e. in more than one farm) if they see animals from various owners. Alternatively, or in complement to frontline surveillance, samples can be sent to animal health (veterinary) laboratories to greatly enhance both the sensitivity and specificity of a particular diagnosis. In the event that a sample is sent to a laboratory, the laboratory is further able to perform specific testing to identify which pathogen is potentially responsible for the disease (if it is infectious in nature). Importantly, it is in the laboratory that both novel pathogens and known pathogens in a novel area or species that are responsible for EIDs can be 16 identified. While initial identification of syndromes on the frontlines may be more suitable for diseases with severe or distinct clinical syndromes, such as Hendra virus, BSE, and HPAI in poultry, laboratory testing may be more suited to initially detect novel pathogens responsible for milder diseases or those without distinct clinical syndromes, such as Nipah and influenza viruses in swine. Therefore, for EZD surveillance, both frontline reporting and laboratories are potentially useful sources of information. 1.3 Public Health Surveillance Public health surveillance is the ongoing and systematic collection and analysis of data for public health action (Thacker, 2000, Buehler, 2008). The objectives of surveillance can be to detect cases of disease, outbreaks of disease, or other factors that predispose to disease (Wagner et al., 2006c). Public health surveillance is seen as a useful defense against EIDs (Keusch et al., 2009, Food and Agriculture Organization of the United Nations, 2011, Word Health Organization et al., 2004), providing decision makers with timely data upon which to initiate prevention and control measures. Public health surveillance comes in many forms, related to the objectives of the surveillance and classified by the type of data sources used. These types in general include: reportable disease surveillance (Velikina et al., 2006), healthcare system surveillance (Wagner et al., 2006b), laboratory surveillance (Brokopp et al., 2006), sentinel surveillance (Wagner et al., 2006a), and syndromic surveillance (Henning, 2003). Reportable disease surveillance largely consists of population-based data on diseases reported to public health, mandated by local health regulations. Laboratory surveillance is often a key source of data for reportable disease 17 reporting, however, it can also be used to monitor pathogens that are not legally reportable to public health, and for surveillance of pathogen subtypes in the catchment area of the laboratory (which may be population-based if the laboratory is a regional reference centre). In sentinel surveillance, data are not collected from the entire population (or a catchment area), but from a select number of previously-identified individuals (e.g. family physicians), who report on a regular basis to public health authorities on cases with the specific condition under surveillance. Sentinel surveillance is used when population-based surveillance is not possible or advantageous, and allows public health to get a snapshot of the condition under surveillance. An infectious syndrome is a characteristic constellation of signs and symptoms in a host that will often be caused by a largely predictable array of pathogens. Syndromic surveillance is now a term used to describe surveillance using data that are not based on specific microbiologic diagnosis, but rather on the observation of a characteristic syndrome (e.g. influenza-like-illness). Syndromic surveillance systems, also called early warning systems, are specifically designed to increase timeliness, largely at the expense of specificity. Another way to classify surveillance systems is to divide them into ‘statistical’ surveillance systems and ‘atypical’ surveillance systems, with either of these being able to use various types of data, including syndromic (e.g. clinical) and laboratory data. Statistical surveillance in this context means a system that compares current numbers/proportions of submissions/syndromes/diseases with ‘expected’ numbers/proportions and produces statistically significant alerts when these are not equal. Atypical surveillance means only focusing on the odd and the strange, and only when such an event is identified would an alert be generated (e.g. phone call to a colleague). 18 Surveillance for endemic infectious diseases differs from surveillance for EIDs largely due to the specificity of the case definitions required for each (see Figure 1.3 for a schematic representation on EID surveillance illustrating the trade-off between sensitivity and specificity in case definitions). Without a well-delineated case definition, and EID is difficult to study and characterize from an epidemiological standpoint. Endemic infectious diseases usually have specific (well-defined) case definitions and associated laboratory tests. While such specificity is possible for some EIDs, such as previously known pathogens increasing their host range (e.g. WNV), or evolving pathogens (e.g. antimicrobial-resistant Salmonella), this is not the case for previously unknown pathogens (e.g. SARS, BSE, Nipah virus). Public health surveillance for EIDs has mostly focused on trying to find index cases and/or outbreaks of an EID in humans, with interest expanding in response to bioterrorist concerns in the preceding decade (Buckeridge et al., 2004, Hutwagner et al., 2003, Lober et al., 2002, Bravata et al., 2004). Such approaches, however, have not been effective at preventing initial cases of disease in humans by definition (i.e. human cases are needed for an initial signal). A true early warning system to detect EZDs as early as possible should, therefore, look to identify EZDs in animals prior to their detection in humans. 1.4 Animal Health Surveillance for Public Health Early warning systems using animal data for public health strive to identify timely ‘signals’ in animals prior to ‘signals’ in humans. The issue with such (early) animal signals is low positive predictive value, a problem shared with human syndromic surveillance. This is because surveillance in animals can be seen as a type of syndromic surveillance from the public health 19 perspective i.e. surveillance looking for signals prior to the detection of diagnoses in humans. Animal health surveillance can also be viewed as “risk factor” surveillance, where the animal signals can be seen as possible risk factors for human disease. Whether seen as “syndromic” or “risk factor” surveillance, animal health surveillance could examine clinical syndromes in animals, submissions to animal health laboratories, as well as specific (microbiologicallyconfirmed) diagnoses in animals, all in an effort to associate them with specific human diagnoses. The two sources of animal surveillance data most analogous to traditional public health surveillance data sources are veterinarians (e.g. animals seen, clinical syndromes, diagnoses) and laboratories (e.g. laboratory submissions, tests preformed, microbiological isolations). There is a long history of disease surveillance conducted in domestic animals, though not primarily for public health purposes. Such surveillance has been instrumental in the control and reduction of a large number of infectious diseases in animals, including rinderpest, the only infectious disease other than smallpox to have been successfully eradicated through human effort (Food and Agriculture Organization of the United Nations, 2011). While animal health surveillance is generally similar to public health surveillance in terms of the data types used (e.g. reportable disease surveillance, laboratory surveillance), it differs in that 1) the number of diseases under surveillance for individual animal species is lower than for people, and 2) the mandates of organizations conducting animal health surveillance are not the same as those of human surveillance (e.g. the impact of animal disease on economics and trade). 20 Veterinarians are important players in animal surveillance and can take on a number of different roles. Practicing veterinarians on the frontlines can not only report syndromes in the animals they encounter, they can report unusual diseases or disease clusters, and submit samples to laboratories for further diagnostics. Veterinarians also work in laboratories (e.g. veterinary pathologists) in order to diagnose diseases in submitted samples/carcasses though necropsy and various microbiological disciplines and their tools (e.g. virology, mycology, parasitology, bacteriology, immunology, molecular microbiology). Additionally, veterinary epidemiologists work in numerous governmental agencies, including agriculture, environment (e.g. wildlife), and public health. In these roles, veterinarians are critically situated to develop, support, and use the outputs of animal surveillance. Veterinary public health is defined by the WHO as “the sum of all contributions to the physical, mental and social well-being of humans through an understanding and application of veterinary science" (Word Health Organization, 2012d), and is therefore not limited to diseases important in agricultural production and other such economically important diseases. Veterinary public health surveillance can therefore be viewed as surveillance that involves microbiological hazards to human health of animal origin: new, emerging and re-emerging zoonotic diseases, and foodborne diseases, including those due to antimicrobial resistant bacteria. A recent report from the United Kingdom estimated that if BSE were to occur with the current cattle surveillance system in place, the time taken from the detection of the first case through to the recognition of the emergence of a new disease and communication to stakeholders, would shorten from the two years it took then, to a far more acceptable 3-12 months now (Department of Environment and Rural Affairs, 2011). 21 It is also possible that the Nipah virus could have been characterized earlier if appropriate surveillance existed in swine. Nipah presented as a non-specific syndrome in pigs without remarkably high morbidity and mortality, the disease was initially thought to be Classical Swine Fever (Chua, 2003, Aziz et al., 1999). Since the Nipah virus was likely causing disease in swine prior to the identification of the index human cases (Mohd Nor et al., 2000), had syndromic (or outbreak) surveillance been present for swine, with a laboratory component to confirm etiology, Nipah virus could have been isolated as a disease-causing agent prior to the first human cases. Although it would not have been known that the Nipah virus could be a zoonosis in the absence of human cases, the natural reservoir (bat) could have been identified, and once the first human cases appeared in pig farmers, Nipah could have been suspected. This could have led to the implementation of more appropriate control measures, such as those reducing contact with infected swine, rather than the mosquito control measures that were implemented that are more appropriate for Japanese encephalitis, which was the initial suspected diagnosis for human cases (Chua et al., 2000). There have been many calls toward a more integrated way to conduct EZD surveillance, continuously recommending strengthening animal health surveillance for the purposes for public health, even declaring animal disease surveillance a public good (Word Health Organization et al., 2004, Keusch et al., 2009, Food and Agriculture Organization of the United Nations, 2011). The hope of these recommendations is to strengthen the use of surveillance in animals in order to identify an issue as early as possible to initiate mitigating measures, prevent illnesses, and limit the economic impacts of disease outbreaks or epidemics. However, in order for animal surveillance to be of value to public health, the warnings from the system likely need 22 to be at least one of: 1) early – likely involving frontline animal health workers such as veterinarians, 2) specific – a disease agent has been identified in the laboratory, and 3) involving a shared pathogen occurring in both humans and animals. 1.5 EID Surveillance in British Columbia British Columbia (BC) is the westernmost province of Canada, with a population of 4,622,600 people (Statistics Canada, 2012). The British Columbia Centre for Disease Control tracks diseases reportable to public health, including potential cases of zoonotic diseases (reportable and non-reportable identified in the laboratory) in the province by identifying cases of human illness and carrying out detailed follow-up to obtain information on travel and animal contact. In Canada there are federally reportable diseases (Public Health Agency of Canada, 2005), as well as provincially reportable diseases that vary from province to province; in BC the list of reportable diseases includes 19 potential zoonoses (British Columbia Centre for Disease Control, 2012). Animal disease surveillance in BC consists largely of passive surveillance (or monitoring) of diagnostic samples submitted to laboratories. There are federally and provincially reportable diseases, with the addition of some targeted industry-specific reporting on production-limiting diseases. Surveillance is mostly conducted through the British Columbia (BC) Ministry of Agriculture Animal Health Centre (AHC), the provincial veterinary diagnostic laboratory, whose mandate is to diagnose, monitor, and assist in controlling and preventing animal disease in BC (British Columbia Ministry of Agriculture, 2012a). Some monitoring of animal diseases analyzed 23 at the federal level (e.g. BSE) or not diagnosed at the AHC (e.g. rabies), is done by the Canadian Food Inspection Agency. Animal diseases that are federally reportable in Canada are listed under the Health of Animals Act (Canadian Food Inspection Agency (CFIA), 2012), and are largely informed by internationally reportable diseases that are related to security of animal food production systems and international trade, including a number of zoonoses (e.g. Q Fever, Tularemia, Avian Influenza) (World Organisation for Animal Health (OIE), 2012b). Provincially reportable animal diseases vary from province to province in Canada. In BC, there are four provincially reportable animal diseases, including influenza in poultry and swine (British Columbia Ministry of Agriculture, 2012b). At the time of writing, H5 and H7 influenzas are the only animal diseases directly reportable to public health in BC. The AHC also works in cooperation with agriculture industries (e.g. poultry, dairy, and fish farms) on targeted surveillance and monitoring projects. There are also integrated surveillance projects that span both human and animal diseases, such as the BC Integrated Salmonella Surveillance program (see Chapter 5 for more details). 1.6 Gaps in Understanding and Thesis Structure While there are many appeals for using animal data for EID and/or EZD surveillance (Keusch et al., 2009, Food and Agriculture Organization of the United Nations, 2011, Word Health Organization et al., 2004), there is less information describing how such surveillance should be conducted. Stephen et al. state that “while it is accepted that animals play a role in the 24 emergence and spread of human diseases, the nature, magnitude and importance of animal determinants is based on opinion rather than evidence” (Stephen et al., 2004). This highlights an important gap: despite the promise of animal health surveillance providing an early warning signal for EZDs, most current literature in this area is at the hypothesis stage, with little critical evaluation of the actual data, such as data availability, quality, representativeness, reliability, acceptability (especially of an animal signal without associated human cases), and the positive predictive value of an animal signal. This thesis deals with this gap by addressing a number of limits in the current literature. The first question addressed in this thesis is what EZD surveillance is currently being done, and what evaluations of such systems suggest are the most useful features of such systems. In order to answer this question, Chapter 2 of this thesis is a systematic review of EZD surveillance systems, focusing on what type of data is being collected (animal, human, both, other), and whether evaluations of such systems can give us clues as to the best ways forward in terms of what worked and what did not. The focus of the remainder of the thesis is on ‘statistical’ surveillance systems (as opposed to ‘atypical’ surveillance systems), using both clinical and laboratory animal data. The questions center on what type of data we should be collecting about animal health and from where. The examples above show that agricultural animals can be sources of infection, that they have many direct routes of infection to humans, and that they can be amplifying hosts and bridging species who transmit diseases either directly or indirectly (e.g. food of animal origin) to people. Moreover, animal surveillance in agricultural species is more developed than that for wildlife 25 species. For these reasons, the scope of this thesis is limited to examining surveillance of agricultural animals. Figure 1.4 illustrates the data flow for animal health surveillance, highlighting the main sources of data evaluated in this thesis: sentinel veterinarians and the provincial laboratory; agricultural census data is used for animal population estimates in the province. Chapter 3 looks at using the timeliest form of animal health information, that coming from sentinel veterinarians. In order to assess this data source, a pilot sentinel veterinary surveillance system is designed, implemented, and evaluated using a public health lens. Various limitations and biases to the data are examined, with specific focus on the types of samples sentinels submit to laboratories. Further, the chapter explores what type of animal health information public health practitioners want in order to make decisions about EZDs. Chapter 4 examines more specific animal health data for surveillance, namely diagnostic laboratory data, and how similar the data are to (human) laboratory data regularly used for public health surveillance. The chapter focuses on the utility of applying detection algorithms to animal diagnostic laboratory data, specifically in their ability to detect known animal health events that are of interest to public health. To date there have been very few studies quantitatively linking animal and human disease data (Scotch et al., 2009). Chapter 5 attempts to integrate animal and human diagnostic laboratory data for one emerging foodborne zoonotic pathogen, Salmonella, to examine whether statistical surveillance of data from each sector could be meaningfully integrated, i.e. whether a signal in animal data is likely to be correlated with a signal in human data for the 26 same pathogen. We used Salmonella for this investigation because it is an emerging antimicrobial resistant organism, and antimicrobial resistance is a leading cause of EID emergence in developed countries (Jones et al., 2008); additionally, comparable laboratory diagnostic data from humans and animals were available in BC for this pathogen. 27 1.7 Figures Figure 1.1 Global examples of emerging and re-emerging infectious diseases. Red represents newly emerging diseases; blue, re-emerging/resurging diseases; black, a ‘deliberately emerging’ disease. Reprinted with permission from MacMillan Publishers Ltd: NATURE. Morens DM, Folkers GK, and Fauci AS. The challenge of emerging and re-emerging infectious diseases, 430: 242-249 doi:10.1038/nature02759, copyright 2004. 28 Figure 1.2 Illustration of the five stages through which pathogens of animals evolve to cause disease confined to humans. The four agents depicted have reached different stages in the process, ranging from rabies (still acquired only from animals) to HIV-1 (now acquired only from humans). Reprinted by permission from Macmillan Publishers Ltd: NATURE. Wolfe ND, Panosian Dunavan C, and Diamond J. Origins of major human infectious diseases, 447(7142):279-283, copyright 2007. *Note on dengue transmission - although humans are the main reservoir for dengue, this is primarily a mosquito-borne disease that is not transmitted directly from person-to-person, except for rare transmission from mother to fetus, or via blood transfusion, or organ donation. 29 Figure 1.3 The continuum of emerging disease surveillance, highlighting the decreasing sensitivity and increasing specificity Information surveillance: information about disease outbreaks, often through the Internet, e.g. Program for Monitoring Emerging Diseases e-mail list for sharing news about emerging diseases (ProMED-mail), and the Global Public Health Intelligence Network (GPHIN) that uses automatically searches the Internet for selected disease-specific words. Syndromic surveillance: automatic capture, transmission and analysis of non-specific patterns of information in prediagnostic health data. Animal and environmental surveillance: Once epidemiology of an emerging pathogen is understood, surveillance in host animals or vectors can identify where human infections are most likely to occur. Laboratory surveillance: Specific diagnostic methods allow for highly specific laboratory surveillance, using automated surveillance of positive laboratory results. Adapted from Buckeridge et al., 2006. 30 Figure 1.4 Animal health data flow for surveillance of infectious diseases. The surveillance flow shows how an agent from an infected animal could be identified by a veterinarian or laboratory and used for surveillance purposes. Red dashed circles indicate the three sources of animal data used in this thesis: the agricultural census (for population estimates), sentinel veterinarians, and the provincial animal diagnostic laboratory. 31 2 Systematic Review of Surveillance Systems for Emerging Zoonoses 2.1 Introduction Good surveillance is the first major tool in preventing emerging infectious diseases (EIDs) that arise naturally or through terrorist activities (Sosin, 2003, Koplan, 2001, Centers for Disease Control (CDC), 1998). Estimates of the proportion of EIDs that involve pathogens transmitted from animals to humans, or zoonoses, range from 60% to 75% (Jones et al., 2008, Woolhouse and Gowtage-Sequeria, 2005, Taylor et al., 2001). Society might be better prepared to detect and prevent EIDs if we can get “ahead of the curve”, identifying risky situations before the first cluster of hospital cases are identified (Brilliant, 2008). Specifically, for emerging zoonoses, it has been suggested that animal health information should be used in surveillance systems for early warning purposes (Kruse et al., 2004). Current trends are to integrate human and animal data in one surveillance initiative (Leslie and McQuiston, 2007, Rabinowitz, 2008, Kahn, 2006), often under the flag of “One Health” (One Health Initiative, 2012). Unfortunately, surveillance of animal disease and zoonotic disease in animals is often not legally mandated to the same extent as in humans, particularly in wildlife. This lack of a legal mandate, structured reporting mechanism and funding often means that EID surveillance attempting to include animal and human data are challenging to design, interpret, and operate. The significant investments being made in EID surveillance makes the evaluation and design of EID surveillance systems of prime importance. The recent emergence of swine-origin influenza A H1N1 illustrates this point: it is 32 possible that had there been a clear signal from animal surveillance prior to wide-spread human-human transmission, a pandemic may have been averted. Surveillance systems for various diseases have proliferated over the past 50 years, with many focused on EIDs in the past decade (Bravata et al., 2004). In North America, EID systems have increased since West Nile virus (WNV) was first identified on the continent in 1999 and after the terrorist attacks of September 11, 2001 (Bravata et al., 2004). Surveillance systems collect and analyze morbidity, mortality, and other relevant data on a routine basis, and facilitate the timely dissemination of results to decision makers (German et al., 2001) (see Appendix A.6 for illustration of surveillance system data flow). Preventing or restricting the impact of an EID is dependent on rapid detection of the first cases (WHO et al., 2004), making timeliness critical. Timely decision making and response based on data interpretation makes surveillance different from monitoring and more than just a tool for event detection. Surveillance for zoonoses is necessarily a multi-disciplinary endeavor, crossing not only human and animal health, but also environmental health and public health practice and policy. The interconnected roles of agricultural animals, pets, wildlife, human populations and their environment for zoonosis transmission and pathogenesis creates a number of distinct challenges for surveillance (Leslie and McQuiston, 2007). Information is needed on how best to structure these interdisciplinary surveillance efforts. There are published recommendations for evaluating various types of surveillance systems available (including syndromic surveillance systems) (German et al., 2001, Wagner et al., 2006c, Buehler et al., 2004), but little attention has been focused on whether or not EID surveillance requires a different set of criteria for 33 design and evaluation when compared to systems intended to keep endemic and noninfectious diseases under surveillance. This review documents the extent of EID surveillance system evaluation and determines what criteria have been used to evaluate these systems. 2.2 Materials and Methods Three questions guided this review: (1) What public health surveillance initiatives for emerging zoonotic diseases exist worldwide? (2) Have these surveillance initiatives been evaluated? (3) What criteria were used to evaluate the surveillance initiatives? Subject terms and keyword terms were identified for key concepts of surveillance systems and zoonotic diseases which were then combined for the search. The definition of surveillance employed was: systematic ongoing collection, collation, and analysis of data with the timely dissemination of information to responsible decision makers (Last, 2001). MEDLINE MeSH terms were hand-searched for relevance under each component (see Table 2.1 for search terms used for ‘surveillance’). We defined an emerging zoonosis as: a zoonosis that is newly recognized or newly evolved, or that has occurred previously but shows an increase in incidence or expansion in geographical, host or vector range (Word Health Organization et al., 2004). The 111 emerging zoonoses used for the literature search included 51 viruses and prions, 29 bacteria and rickettsia, 9 helminths, 11 protozoa and 11 fungi (see Appendix A.1). Diseases were searched by the names of the causative agents as well as their various common names (see Appendix A.2 for disease names). 34 To ensure a very high degree of sensitivity, both subject and keyword searches were used. MEDLINE, EMBASE, AGRICOLA, several subsets of databases under Environmental Sciences and Pollution Management, and Zoological Record were searched to include publications from medicine, public health, zoology, biology, environmental studies and agriculture. The exact search strategy was unique for each database due to differences in subject thesauri or subject terminology (see Appendix A.5 for MEDLINE search strategy). All search strategies were recorded at each step and citations from database searches were downloaded or manually entered into RefWorks (RefWorks, LLC) and duplicates removed. The review was limited to English language papers published between 1992 and 2006. The Canadian Field Epidemiology Program (CFEP) in the Public Health Agency of Canada (PHAC) provided surveillance system evaluations completed by their trainees between 1999 and 2007. These CFEP reports provided non peer-reviewed literature available on the evaluation of public health surveillance of infectious diseases in Canada. The local, provincial/territorial and federal agencies that hosted the CFEP epidemiologist’s placement and commissioned the reports were contacted to obtain permission to use the reports. A condition of the data sharing agreement struck with these placements prohibits identification of individual systems under evaluation in this report. Results were grouped to preserve this anonymity. The first stage of identifying articles to be included was based on titles, subject headings and abstracts (if available) of the articles (Table 2.2). Two reviewers assessed the reliability of the initial inclusion/exclusion decision process using a sub-section of the total MEDLINE search. 35 In this pilot, the two researchers applied the initial inclusion/exclusion criteria separately, and then compared their selections. The degree of agreement was tested using the Cohen’s Kappa statistic. Where there was disagreement regarding a specific paper, consensus was reached after discussion. The new consensus criteria were used thereafter to include or exclude articles independently by the two reviewers based on the full article texts (Table 2.2). Articles were included if they described and/or evaluated emerging zoonoses surveillance systems. We included systems such as diagnostic, management and reporting and/or communications systems if they could potentially be classified or used as surveillance systems. The data extracted from the articles about the system included: purpose, location, population, year started, organizations involved, disease(s) under surveillance, whether the agent(s) is known/defined, data collected, collection and analysis methods and what evaluation was performed (see Appendix A.3 for a complete list of data extracted from articles). An evaluation of a system was considered to have been conducted if the paper stated that an evaluation was conducted and/or if the paper contained at least two of the following four criteria: sensitivity, positive predictive value, specificity, or timeliness. These three criteria were chosen from accepted evaluation criteria of simplicity, flexibility, data quality, acceptability, sensitivity, positive predictive value, representativeness, timeliness and stability (German et al., 2001) because the authors deemed them most pertinent for an EID surveillance system. The authors did not have to specify that they were indeed addressing the specific criterion. For example, if the authors included the time taken from data capture to analysis, then the “timeliness” criterion was considered to be “assessed”. 36 2.3 Results The literature search identified 2,263 articles from the various databases (see Appendix A.7 for figure showing citation counts from all databases and subsequent inclusions and exclusions of articles). After applying the initial inclusion/exclusion criteria, 603 articles were selected. All but 20 articles were obtained in full (96.7%). The Cohen’s Kappa measuring inter-rater agreement of the pilot MEDLINE search was 0.47 or moderate, arising due to initial difficulty in agreement on papers that described the systems sufficiently for inclusion (see Appendix A.8 for Figure illustrating inter-rater agreement in the pilot search). The pilot phase Kappa does not provide an estimate of agreement that should be extended to the rest of the study because subsequent selection criteria were modified based on the results of the pilot. Of the initial 583 full text articles, 212 contained systems meeting the study inclusion criteria. More than half (55%) of the articles were published in the four-year period from 2003 to 2006 (Figure 2.1). The Figure also shows that the number of human-only surveillance systems in the literature increased in 1998, leveled off, then increased again in 2003-5; animal-only surveillance systems were relatively steady with a marked jump in 2006; and systems containing both animal and human data show an a steady rate, a jump in 2000, followed by a leveling off. Data extraction from the 212 articles resulted in 221 different systems, with some articles describing more than one system. Data extraction was made difficult due to unclear terminology in the papers. The structure and components of the systems, in general, were poorly described. For example, we could not distinguish monitoring systems from surveillance systems, as both were often called surveillance systems. All articles contained information on 37 whether the system collected data on one versus multiple diseases (100%), what country or continent the system was in (both 99%), if the system was evaluated (93%), whether the system collected known and/or unknown pathogen data (88%) type of data collected (human, animal, other) (88%), specific syndromes or diseases under surveillance (80%), organizations involved (79%), purpose of the system (79%), population information (72%), type of data collected (66%), data collection and analysis methods (66%), and the year started (60%). Very few articles contained enough information to determine system type (28 %), or specific evaluation criteria (from 10-26%, depending on the criterion). Table 2.3 shows the results of the systems included by continent. Most systems were from North America (40%) and Europe (29%). Most (70%) were designed to detect ‘known’ pathogens, followed by systems targeting both known and unknown pathogens (22%), and the least (8%) for only unknown pathogens. The proportion of systems focusing on unknown pathogens was much higher in North America (45%) than in Europe (20%). The systems primarily examined human data (56%), followed by animal data (25%) with the least evaluating both human and animal data (19%). Finally, most systems looked at multiple diseases (65 %) (Table 2.3). Only 17 of the 221 (8%) systems were considered evaluated according to our criteria with most (65%) in North America. Eleven papers reported timeliness, one sensitivity, one specificity, one positive predictive value, and three stated they conducted an evaluation but did not present any results. Although these papers also looked at the other evaluation criteria outlined by the CDC evaluation framework (German et al., 2001), they were not used consistently. 38 There was not enough information in the reports to determine whether these evaluations were conducted on an ad hoc basis or as an ongoing part of the systems. The general usefulness of systems for detection of disease was part of the evaluation of four of the systems in various publications (Carrat et al., 1998, Myers et al., 2000, McKenna et al., 2003, Cooper et al., 2006, Miller et al., 2004, Aguilera et al., 2003, Letrilliart et al., 2005, Parsons et al., 1996, Toubiana and Flahault, 1998, Valleron and Vidal, 2002). All looked at human disease data only. Three of the four systems targeted both known and unknown pathogens while one looked exclusively at known pathogens. Since none of the four systems had detected an emerging disease, retrospective data or modeled data were used to assess the detection capability of the system. Three of the four evaluated systems concluded in a number of publications that the system was useful for outbreak detection (Carrat et al., 1998, Myers et al., 2000, McKenna et al., 2003, Aguilera et al., 2003, Letrilliart et al., 2005, Parsons et al., 1996, Toubiana and Flahault, 1998, Valleron and Vidal, 2002, Miller et al., 2004), while one concluded that it was not (Cooper et al., 2006). A total of 45 “surveillance of health events” evaluations conducted by epidemiologists in the Canadian Field Epidemiology Program (CFEP) were identified between 1999 and 2007. Although ten reports fit the initial inclusion criteria, only seven were included in this review, as two were incomplete, and one was not available. Since two of the reports looked at the same system, descriptive statistics are calculated for only six of the seven reports. Half the reports featured true surveillance systems (vs. monitoring systems); three of the systems were provincial, two were national and one was local. All of the CFEP reports described the systems in detail. Three were started in 2001, the others in 1997, 1998 and 2005. Most (4/6) of the 39 systems were for known pathogens, the remaining two were for both known and unknown agents. Most (4/6) gathered information on multiple diseases, while two focused on one only. Half looked at only human data, one looked at human and animal data, one looked at human and other data, while one looked at all three. All of the CFEP reports contained evaluations of the selected systems per our criteria. The reports often used multiple evaluation criteria depending on the system attributes, data availability and the specific objectives of each evaluation. The most common evaluation criterion was timeliness (5/7), followed by acceptability (4/7), utility or relevance (4/7), flexibility (3/7), sensitivity, specificity or positive predictive value (3/7), data quality (2/7), simplicity (2/7), and sustainability (1/7) (see Appendix A.4 for a table of the results of the evaluations from CFEP reports). 2.4 Discussion Decision makers need to be cautious when making decisions based on systems that have not been adequately evaluated. The developers of EID surveillance systems are hampered by the lack of a systematic accounting of the necessary elements for integrated EID surveillance and are thus left to use anecdotal information and/or trial and error when developing and evaluating their programs. Our systematic review identified 221 surveillance and monitoring systems that tracked emerging zoonoses worldwide in the peer-reviewed literature, and 6 systems in the selected non peer-reviewed literature. Of the 221 surveillance and monitoring systems, only 17 were evaluated, and most of the evaluations were limited. Few papers used more than two of the standard criteria for evaluations of surveillance systems (German et al., 2001) and no paper addressed all of the criteria. Only four systems explicitly used an evaluation 40 to assess the utility of their systems (Carrat et al., 1998, Myers et al., 2000, McKenna et al., 2003, Cooper et al., 2006, Miller et al., 2004, Aguilera et al., 2003, Letrilliart et al., 2005, Parsons et al., 1996, Toubiana and Flahault, 1998, Valleron and Vidal, 2002). A systematic review of surveillance systems that focused on early detection systems for bioterrorism-related diseases found 29 evaluated systems in the 115 identified, however none of the evaluated systems collected zoonotic disease information (Bravata et al., 2004). The quality of the CFEP evaluations was much higher than those in most of the peer-reviewed literature articles, containing many more of the elements that comprise an evaluation. The general lack of evaluation data may be due to unwillingness to publicly report negative evaluation results, or because the agencies that operate the surveillance systems prefer to publish internal reports rather than scientific articles. It may also be due to the relative novelty of many of these systems: more than half of the articles used for this review being published in the last four years of the study period (2003-6) (see Figure 2.1). Independent data that would enable comparisons and establishment of ‘gold’ standards for evaluations are lacking for EID surveillance systems. The lack of available gold standards makes comparisons very difficult and complicates calculations of measures such as sensitivity and specificity. One method researchers have used to measure imprecision in their surveillance efforts without an independent collection mechanism providing them a gold standard, was by comparing cause of death data from hospitals with nationally collected surveillance data (Paddock et al., 2002). One limitation of this review was the inconsistent application of the term ‘surveillance’. Although a number of articles stated that they described surveillance systems, the information 41 provided suggested they were monitoring systems with no timely analysis or ongoing dissemination of data. Most of the articles did not contain enough information to correctly distinguish whether the systems were surveillance systems or monitoring systems. This lack of specificity in the term ‘surveillance’ likely reflects the fact that, to date, there has been very little ‘surveillance theory’ in public health, resulting in the term not being consistently defined or applied. Surveillance in general has not been subjected to much academic scrutiny, perhaps because it is considered philosophically dichotomized away from research. This distinction should not, however, preclude surveillance methodology from the scrutiny of research and evaluation. While the majority of systems in this review (70%) were designed to detect ‘known’ pathogens only, a minority were syndromic surveillance systems, such as systems designed to detect both known and unknown pathogens (22%), or unknown pathogens alone (8%). The importance of these syndromic surveillance systems which use a broader case definition for surveillance is that these systems may be able to detect completely unexpected diseases. One such instance of serendipitous detection using syndromic surveillance systems comes from a public health surveillance system designed to capture anthrax cases in New York that resulted in the detection of another zoonosis, Rickettsialpox (Koss et al., 2003, Paddock et al., 2006). This review shows that there have been constant attempts to integrate human data with animal data in surveillance initiatives for zoonoses (Figure 2.1). While over half of the systems looked at human data alone, and about a quarter looked at animal data only, almost a fifth tracked both human and animal data. Unfortunately, since none of the evaluated systems in 42 this review were those that captured both human and animal data in one system, it is difficult to assess how well these integrated systems perform. Currently, few human or animal health agencies have an explicit mandate to compare animal and human disease data in a ‘One Health’ manner (Rabinowitz, 2008). Without formal legislation, these ‘integrated’ surveillance systems will remain in the hands of key motivated individuals, susceptible to disuse or complete collapse if these individuals take on new responsibilities or leave their positions. Our systematic review has three main limitations: 1) the scope of and search terms used in the search strategy, 2) the lack of necessary data in the included articles, and 3) the focus on peer-reviewed literature. Our search strategy did not pick up papers discussing new areas of research, such as the use of spatial data to determine risk of zoonotic diseases. For example, remotely sensed data have been used to predict risk of Hantavirus pulmonary syndrome (Glass et al., 2000) and Sin Nombre virus (Glass et al., 2002). These methods hold promise either in conjunction with other data in a surveillance system or to help evaluate surveillance systems. Further, since this review was limited to articles in English, the results were biased towards including systems from North America and Europe. The broad definition of emerging zoonoses used in this study may have resulted in the inclusion of some surveillance systems for endemic zoonoses because the definition excluded geographic location. For example, although WNV is an emerging disease in North America, it is not an emerging disease in the Middle East or Northern Africa. Nevertheless, an article on surveillance for WNV in the Middle East or Northern Africa would have been included in this review, since the articles were chosen based on the causative agent and not location. 43 Missing data were an issue since many articles did not contain basic descriptive information of the systems. Duplicate counts of systems occurred when 1) a system changed over the study time, either in name or in scope, or 2) a system was described at a local level in one paper, and at a regional or national level in another paper. With the information provided in the articles, it was often not possible to conclusively state whether a particular system was the same as another. The review focused mainly on published and peer-reviewed literature. Our assessment of the grey literature suggests that it may contain many quality evaluations. Although future reviews should include internet searches for reports, such as those by various government agencies, these agencies often do not post their reports. Detailed descriptions and evaluations of surveillance systems are scarce in peer-reviewed literature and the definition of surveillance was unclear in both peer-reviewed and non peerreviewed literature. There is a need for further research into the science of surveillance: surveillance needs to be studied, defined and standardized. Evaluation must be built into surveillance systems as an ongoing component. Since many EID surveillance systems are still in their infancy, we anticipate proper evaluations in the future when the necessary data are collected. Government agencies and epidemiology training programs should be encouraged to publish their surveillance evaluation reports, including detailed descriptions of their surveillance programs, in peer-reviewed literature. 44 2.5 Tables and Figures Table 2.1 MeSH search terms to describe surveillance. Surveillance Component Information Technology Public Health Organizational Structure MeSH Terms decision making, computer-assisted; decision techniques; clinical laboratory informatics systems; decision support systems, clinical; hospital information systems; integrated advanced information management systems diagnosis, computer-assisted; epidemiologic methods; disease outbreaks; disease reservoirs; disease transmission; environmental medicine; environmental microbiology; environmental monitoring; food contamination; communicable disease control; mandatory reporting; "disease management communication; decision making; information dissemination” inter-professional relations; public health administration; organization and administration; and health care organization 45 Table 2.2 Inclusion and exclusion criteria for selection of articles in the review. Initial Exclusion Criteria* Secondary Exclusion Criteria† Language Non-English Non-English Time Period Prior to 1987 Prior to 1992 Study Type Basic research articles Organ transplant articles Basic research articles Organ transplant and blood transfusion articles Diseases Does not relate to an emerging/reemerging zoonotic disease Does not relate to an emerging/re-emerging zoonotic disease Reports of the results of a surveillance system only, not discussing the system No statement of purpose or no description of system Reports of the results of a surveillance system, not discussing the system General listserves, e-mail distribution lists, chat rooms, electronic versions of textbooks or Web sites that provide information on emerging zoonoses without a moderator or peer-review process System Description/ Type *Initial exclusion criteria were used to select articles using only the titles, subject headings and abstracts (if available) †Secondary exclusion criteria were used to select articles when full texts were available 46 Table 2.3 Emerging zoonoses surveillance systems by Continent (N=221), pathogen(s) under surveillance, type of data collected, and number of diseases under surveillance. Continent of System† No. Systems for Known and Unknown Pathogens (N=190) No. Systems Total No. Systems Collecting Collecting One No. Human and Animal Data Disease versus Systems (N=194) Multi-disease Data Included (N=216) (N=221) Only Known Pathogens Only Unknown Pathogens Known & Unknown Pathogens Human Data Animal Data Africa 8 0 0 4 3 Human & Animal Data 4 Asia 11 1 3 11 0 13 0 2 9 3 0 0 48 1 41 Australia and Oceania Central and South America Europe North America International† Unknown Total* One Disease MultiDisease Total 7 4 11 1 8 8 16 4 4 6 12 18 2 0 1 1 2 3 11 30 24 7 26 35 63 11 23 46 16 15 21 65 88 12 2 4 7 2 4 6 14 21 0 0 1 0 0 0 0 1 1 136 15 44 109 49 36 75 141 221 † The Continent of a system was determined by the country in which the system was located, systems spanning more than one country (even on the same continent) were classified as International *Totals for the systems by pathogen(s) under surveillance, type of data collected, and number of diseases under surveillance do not always add to 221 due to missing values 47 35 Human Animal Human and Animal 30 Number of Articles 25 20 15 10 5 0 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Year of Publication Figure 2.1 Peer-reviewed Articles Describing Emerging Zoonoses Surveillance Systems by Year of Publication between 1992 and 2006, by Type of Data in the System (N=201*). *The total number of articles in this time period was 212, 11 were excluded from the figure because they either contained ‘other’ data (3), and/or the type of data could not be conclusively established (10). For those articles that contained descriptions of more than one system, if one of the systems contained both animal and human data, then the whole article was categorized as having both; for articles that contained only systems describing either human or animal data, the data type of data in the majority of systems was used to categorize the article. 48 3 Animal Health Surveillance Using Sentinel Veterinarians: a Public Health Perspective on a Pilot Project 3.1 Introduction Infectious agents transmitted between animals and humans (zoonoses) are important for public health both as causes of endemic disease (e.g. gastrointestinal agents such as Salmonella and E. coli), and emerging infectious disease (EID) (e.g. Avian Influenza, Bovine Spongiform Encephalopathy). Animals have been implicated as the source of 60% of new EIDs (Jones et al., 2008). To date, animal disease surveillance has largely been used for tracking and managing diseases of importance to animal health and trade, while public health surveillance of zoonoses has largely focused on detecting disease in humans. Less commonly have animal health surveillance data been directly incorporated into public health surveillance systems for the purpose of forecasting and preventing human zoonotic diseases. In an era of emerging zoonotic infections in people, it is therefore worthwhile to investigate whether surveillance of animal health events could be of value to public health for disease prevention and control. The World Health Organization (WHO) has recognized that animal health surveillance is a critical element of public health surveillance (Word Health Organization, 2002). The World Organisation for Animal Health (OIE) has developed the World Animal Health Information System and Database (WAHIS and WAHID) (Food and Agriculture Organization of the United Nations et al., 2010). The WHO, Food and Agriculture Organization of the United Nations (FAO), and the OIE are working on creating common surveillance and reporting systems (Word Health Organization et al., 2004), and have already created the Global Early Warning System for Major Animal Diseases, 49 including zoonoses (GLEWS) (Food and Agriculture Organization of the United Nations et al., 2010). While there are many forms of animal information, for the purposes of infectious disease surveillance, we can divide the data into three general categories: 1) clinical or sentinel information from practitioners in the field (e.g. veterinarians), 2) etiologic and pathologic information from laboratories, and 3) information about animal populations or animal-derived products (e.g. meat, eggs). The choice of data stream largely depends on the objectives of the surveillance. From a public health perspective, the first important distinction is whether the objective of surveillance of animal information is for tracking known or endemic diseases, or unknown and emerging diseases. A surveillance system focused on endemic infectious disease should strive to be as representative as possible, as a representative surveillance population would allow for identification of the best targets for interventions (i.e. specific populations), and evaluation of their effectiveness (Buehler et al., 2004). A surveillance system for EIDs, on the other hand, may focus on populations or situations deemed at high risk for disease emergence or transmission of pathogens to people in order to achieve prevention and/or early detection and containment of an EID. It is currently unclear what type of animal data would lend itself best to either of these objectives, and what type of uncertainties and biases exist in the animal data that would be relevant to a public health practitioner. Recent history has also shown the perils of focusing in the wrong place: prior to the most recent influenza pH1N1 pandemic that emerged from pigs (Neumann et al., 2009), 50 extensive surveillance of influenza was underway in poultry while much less effort was focused on pigs. In this study we focus on agricultural animals for a number of reasons: 1) food industry changes are an important driver of disease emergence due to the globalization of the food industry and potential for a significant extent of human exposure to zoonotic pathogens in food (WHO et al., 2004), 2) population estimates are available for a number of agricultural animal species, enabling calculations of proportions and rates, and 3) there are concerns over emerging antimicrobial resistance in zoonotic pathogens that are present in agricultural animals (Jensen et al., 2009). There have been a number of novel sentinel surveillance efforts in agricultural animals developed recently, including two regional Canadian provincial systems (Gouvernement du Québec, 2010, Government of Alberta, 2010), a state regional system in the United States (Wyoming Department of Health, 2010), and three national systems (Barnouin, 2010, Vourc'h et al., 2006, DeGroot, 2005, Robertson et al., 2010). These systems were either not developed for use as public health surveillance systems, or evaluated from a public health perspective. In this study we developed a pilot sentinel animal health surveillance system to describe the data that sentinels could provide and used expert focus groups to assess the potential use of such data for public health practice or in public health surveillance. Specifically we examined: 1) the feasibility of conducting surveillance using sentinel veterinarians by examining recruitment and engagement of participants, as well as sentinel characteristics; 51 2) the animal species seen by the sentinels and how they compare to the species in the province; 3) the reasons why sentinel veterinarians saw animals, as well as suspected infections and syndromes in those animals; 4) the reasons why sentinels submit samples to laboratories, giving us information about biases in laboratory submission practices and other factors associated with submissions, such as animal demographics; and 5) the type of animal data that public health needs for action regarding emerging diseases, such as disease control, prevention and surveillance. We hypothesized that we would be able to recruit sentinels and retain more than half of them for the duration of the study, that the sentinels would see a non-representative proportion of species present in the province, and that while most animals would be seen by sentinels for non-infectious diseases issues, such as vaccination, there would be animals seen for suspected infectious diseases. 3.2 Methods 3.2.1 Sentinel Veterinarian Recruitment Participation in the Emerging Zoonoses Sentinel Animal Health Surveillance (EZ Surveillance) project was solicited from veterinarians across the province of British Columbia (BC), Canada. Eligible veterinarians were identified from the 2008 BC Veterinary Medical Association 52 Commercial Directory that lists all practicing veterinarians in the province (N=890). Veterinarians in this directory were classified by the type of practice based on the species of animals seen and/or agricultural production systems they were involved in. Each veterinarian could be classified into a number of non-mutually exclusive practice types (e.g. one veterinarian could identify as being in small animal practice (i.e. pets) as well as mixed animal practice (i.e. pets and agricultural animals), as well as equine practice. Potential sentinels were identified as those belonging to practices that were any of the following: avian (n=31), dairy (n=12), equine (n=91), food animals (n=17), fur bearing animals (n=2), goats (n=10), large animals (e.g. cattle, horses, goats) (n=84), llamas or alpacas (n=10), mixed animals (n=73), poultry (n=5), rabbits (n=38), sheep (n=9), or swine (n=4). Veterinarians that focused only on smaller pets (e.g. cats and dogs), fish, zoo animals or wildlife were excluded. Since some practices included veterinarians who were eligible (i.e. saw eligible animals) as well as those who were not eligible (i.e. specialized in small animals only), and these differences were not listed in the directory, all veterinarians in eligible practices had to be contacted. A total of 428 veterinarians from 386 eligible practices were sent a letter introducing the project and called a couple of weeks later inviting them to participate. 3.2.2 Case Definition, Data Collection, and Reporting A case was defined as an encounter by a sentinel veterinarian with agricultural birds and mammals, seen for the same reason, regardless of how many animals were seen in that encounter. For example, a sentinel could see one cow on one visit and twenty cows on another visit, but if the twenty cows are all seen for the same reason (e.g. vaccination), then the one 53 cow and the twenty cows were each counted as one case. Same for chickens – a case could either be one chicken or a flock seen for the same reason. Case data were gathered for a oneyear period for each sentinel; data collection occurred between March 1, 2009 and April 31, 2010, though exact start and end dates differed for each sentinel (maximum difference in starting dates was 1.5 months). Sentinels were classified into two different reporting types: those that reported once a week on either all cases seen that week, or, if the sentinel’s average case load exceeded 15 cases in a week, that sentinel reported only on cases seen on one day of the week (same day each week, e.g. Wednesday). Cases were reported by sentinels either using a secure web-based application (Appendix B.1) or by completing and faxing a one-page form containing the same data fields (Appendix B.2). The data collected were adapted from the Alberta Veterinary Surveillance Network (Government of Alberta, 2010), and included: demographics of the animal, the reason the animal was seen, suspected diagnoses, and whether laboratory samples were sent for testing (Table 3.1; the case form in Appendix B.2 shows all variables and categories). Table 3.1 shows the mandatory fields that had to be completed for sentinels to submit a case report using the web application. Missing mandatory fields on faxed case report forms were obtained from sentinels prior to case entry. Diagnostic information was based on clinical presentation; syndromes were based on the organ systems affected (e.g. respiratory, gastrointestinal, multisystemic). Sentinels were given reporting instructions in person or over the phone, as well as in a written format for reference, and were paid a small compensation for their participation to offset the time spent reporting. The University of British Columbia Behavioral Research Ethics Board approved this study (H08-02093). Location was only collected for each sentinel, since 54 the location of the animal (or owner) could not be collected due to privacy concerns. The distance from the sentinel’s closest city/town to the provincial laboratory was calculated using the ‘direction’ function within Google Maps (Google, 2012). Sentinel practices were described in terms of numbers of reports per month, cases suspected of having an infection, and specimens sent to an external laboratory; this information was fed back to sentinels in the form of monthly reports for the first six months of the project. The number of reports per month for each sentinel and type of veterinary practice was calculated for each month the sentinel reported. Linear regression was used to quantify and test the significance of linear trends in the numbers of cases seen by sentinels per month; normality of residuals was assessed using residual histograms and Kolmogorov-Smirnov and Shapiro-Wilk tests, and homoscedasticity was assessed by inspection of the studentized residuals versus predicted values; regression models and diagnostics were conducted using SPSS for Windows, Rel. 14.0.0 (Chicago: SPSS Inc.). A survey of veterinarians was conducted part-way through the project to gather information on the reasons for participating and for not participating. All participating veterinarians were asked to participate via email, and asked why they chose to participate in the project, what they thought was going well and not going well with the project, whether there was an incentive that would persuade them to participate in a similar project in the future, and finally whether they were still interested in the project. A random sample of 100 veterinarians was chosen from the original list of veterinarians contacted who did not choose to participate; they were faxed a questionnaire asking why they did not participate and whether 55 there was an incentive that would persuade them to participate in a similar project in the future. 3.2.3 Timeliness Timeliness was assessed using the time between disease occurrence (date case seen) and the availability of the data for analysis by the surveillance program (the date the case was entered into the online database) as done in a previous study (del Rocio Amezcua et al., 2010). Timeliness was further assessed for each data report type (web data entry or fax) and separately for cases that were submitted and not submitted to an external laboratory. The mean time taken to complete the reports online was assessed using the difference between the date and time the report was created to the last date and time it was modified. 3.2.4 Animal Species Seen by Sentinels The total numbers of cases seen by sentinels over the one year period for each species were compared to estimates of the population of each species in the province obtained from the 2006 Canadian Census of Agriculture (Statistics Canada, 2006). This is a national census done every four years that collects information from every farm and agricultural operation that produces at least one of the following products for sale: crops, livestock (e.g. cattle, pigs, sheep, horses, game animals), poultry (e.g. hens, chickens, turkeys, chicks, game birds), animal products (e.g. milk or cream, eggs, wool, furs, meat), or other agricultural products (e.g. mushrooms, sod, honey, maple syrup products) (Statistics Canada, 2007). The number of animals of each species on the farm covered by the census was enumerated on May 16, 2006; 56 the response rate was 95.7%, with an estimated 3.4% under-coverage of farms, with most of these being farms with sales under $10,000 in 2005 (Statistics Canada, 2007). Proportions of species seen by sentinels were calculated using the number of cases of a species as the numerator, and the total number of mammal cases seen as the denominator for mammals, and the total number of bird cases seen as the denominator for birds. Proportions of species reported in the agricultural census were calculated as the number of a particular species in the province as the numerator, and the total number of mammal species in the province as the denominator for mammals, and the total number of bird species as the denominator for poultry. Proportions were also re-calculated with one species (mink) excluded from the population denominator to test the effect on the results. Mink was chosen for this purpose, because despite their large numbers in the overall population in the province, they are reared only on 13 farms in the province (Statistics Canada, 2006), making it unlikely that the sentinels in our study would come into contact with these animals. In order to test whether the animals seen by sentinels mirrored the distribution of animals in the province, proportions of numbers for each species that were seen by sentinels were compared to those reported in the census using the 2-sided Pearson’s χ2 or the 2-sided Fisher’s exact test when there were small numbers of observations in any cell of the 2x2 table constructed for each test. Pearson’s χ 2 and Fisher’s exact tests were calculated using SPSS for Windows, Rel. 14.0.0 (Chicago: SPSS Inc.). 57 3.2.5 Reasons Animals Were Seen by Sentinels and Suspected Infections and Syndromes Animal species were described by total number seen, the reason they were seen (e.g. routine health examination, disease investigation), the syndromes suspected in the animals, whether an infection was suspected, and whether laboratory samples were sent to an external laboratory. Cases where sentinels suspected an infection were classified as infectious; cases classified as non-infectious were those with a non-infectious disease and those where there was no suspected diagnosis (e.g. animals were seen for reproduction services or immunization). Proportions of syndromes and infectious syndromes were calculated using the number of cases with assigned syndromes for the denominators in order to limit the analysis to diagnostic cases. 3.2.6 Laboratory Submissions by Sentinels Two binary outcome variables were created for all sentinel cases: 1) whether an external laboratory sample was sent by the sentinel, and 2) whether an external laboratory sample was sent to the provincial animal health laboratory specifically. Case-level and veterinarian-level factors (Table 3.1), as well as the factors cited by sentinels as the reasons for submitting or not submitting samples, were described in relation to both outcomes for all species together, and stratified by mammals and birds. Further analyses were performed for the two species with the highest number of submissions and with more than two sentinels reporting them: cattle and horses. Other species were excluded from these models because their small case numbers resulted in models that would not converge on most parameter estimates. 58 The date the case was seen was grouped into months and seasons (spring: March-May, summer: June-August, fall: September-November, winter: December-February). Age was kept continuous, as well as coded into categories for each species separately. Age categories for cattle were: < 1 year, 1-2.5 years, and > 2.5 years (after which point they are adult cattle and may be tested for BSE at slaughter) (Berezowski, 2010). Age categories for horses were: <2.5 years (age before horses are typically ridden, when they are typically only moved, sold and trained), 2.5-15 years (their main ‘career’ age at which time they are most valuable, move around the most and have lots of contact with other horses in sporting circuits), >15-25 years (age when chronic issues begin, they are less active in the sport they participated in, and travel less), and >25 years (‘retired’ horses, little contact with other ‘outside world’ except with younger horses on the property) (Stitt, 2010). Possible predictors (Table 3.1 and categorical variables listed above) were tested against each outcome using a χ2 test of significance for categorical variables and a t-test (with equal variances not assumed) for continuous variables. The reason stated by the sentinels for either submitting or not submitting laboratory samples, as well as diagnostic test type were excluded from statistical tests because they were by definition different for cases with submitted samples and those without submitted samples. Logistic regression models were created for both outcome variables: external lab sample sent and external lab sample sent to provincial laboratory for variables that were significant at the p < 0.10 level in bivariate analyses, with the following exceptions: 59 For variables measuring the same parameter but coded differently (i.e. continuous vs. categorical), the variable that yielded the lowest p-value (or the highest t-value or χ2 value) was chosen For variables that correlated significantly with each other using the Pearson correlation coefficient (such as the number of animals in the pen/group and number of animals in the herd), the variable with the lowest p-value (or the highest t-value or χ2 value) was chosen If variables included in models resulted in unstable estimates and/or very high standard errors (due to low cell counts), they were excluded from the final models. Interactions between all pairs explanatory variables were tested for statistical significance, and effect modification and confounding were assessed in multiple step-wise addition/deletion of variables and interaction terms in all possible combinations of variables by examining >10% changes in effect estimates in other variables in the model as well as stratified analyses. All descriptive statistics, χ2 test, t-tests, as well as logistic regression models were calculated using SPSS for Windows, Rel. 14.0.0 (Chicago: SPSS Inc.). 22.214.171.124 Use of Animal Data in Public Health Practice An emerging infectious disease (EID) for the purposes of this study was defined as either 1) a completely new infectious disease agent, including antimicrobial resistant organisms, 2) an introduction of a previously known disease agent from a different location, or 3) an unexpected 60 significant increase (e.g. outbreak) of previously known disease agent (i.e. an outbreak of an agent not seen in a longer time interval). An expert panel focus group was provided narrative scenarios to develop general rules for zoonotic EID decision-making used by public health for animal signals with no human cases. This panel developed a framework to assess the utility of data being derived from sentinel EID surveillance in animals as there are currently no formal means of assessing emerging zoonoses surveillance data for public health (see Chapter 2 in this thesis). The experts were veterinarians in research, agriculture, wildlife and public health, public health medical practitioners, biologists, and epidemiologists. The experts were split into three groups of three experts each, discussing scenarios of specific zoonotic agents in animals: Avian Influenza, Bovine Spongiform Encephalopathy, and Hantavirus. First, the groups developed a core “biology and epidemiology” model, including various data types that could be gathered and their relative importance, for their specific agent. The groups then developed three narrative scenarios each (Fischoff et al., 2006) based on given initial conditions, ensuring the groups examined the behavior of the agent in various circumstances. Lastly, the groups came together and discussed the results, exploring similarities and differences between the different agents and scenarios, and developing general rules for the data types needed in order to make public health decisions. Heuristic algorithm trees using themes from the discussions were created iteratively after the meeting and validated by experts using 35 specific scenarios for each tree endpoint. The experts arrived at a specific endpoint in the trees for each scenario, assigning one of three response levels: (1) No action; (2) Public health to be notified - investigation suggested; or (3) Public health to be notified - immediate action needed. Three experts were assigned each scenario; agreement 61 between experts on the public health action suggested for each scenario was assessed using Fleiss’ kappa (κ) (Fleiss, 1971), and the resulting κ-values interpreted using tables by Landis and Koch (Landis and Koch, 1977). A follow-up focus group was held at a symposium to discuss the use of animal data for surveillance of emerging zoonoses on November 23, 2009 in Vancouver, Canada. The participants were 30 experts in the fields of animal and human health from Canada, United States and France. Discussion centered around two main questions from a public health perspective: 1) how are and how should sentinel animal health surveillance systems be evaluated and 2) what data should be collected for animal health surveillance. 3.3 Results 3.3.1 Sentinel Veterinarians: Recruitment and Reporting Thirteen (3.0% [13/428] of the veterinarians contacted or 3.4% [13/386] of eligible practices) participated in the project: seven mixed animal practitioners (9.6%, 7/73), four large animal practitioners (4.8%, 4/84), one equine veterinarian (1.1%, 1/91), and one poultry veterinarian (20.0%, 1/5). Five other veterinarians originally signed up for the project but did not start reporting. During the study period, one large animal practitioner changed to a different practice and ceased reporting, while another interrupted reporting for two months while changing practices, therefore twelve sentinels were reporting at the end of the study period. The mean number of reports per month over the entire time interval for individual sentinels was highly variable, ranging from 1.4 to 37.3 cases/month (mean 8.9 cases/month) (Appendix 62 B.3). The mean number of reports did not differ significantly between sentinels that reported on all cases seen in a week (mean 10.9 cases/month) and those that reported on cases seen only on one day per week (mean 8.2 cases/month) (p=0.67). The mean number of reports for each type of veterinary practice was 4.3 cases/month for equine, 4.8 cases/month for poultry, 8.7 cases/month for mixed, and 11.1 cases/month for large animal practice. The total number of cases that sentinels reported decreased at a rate of 0.5 cases per month per practice (p<0.001) over the time period. When stratified by type of veterinary practice, the decrease was 1.1 cases per month for each large animal practice (p<0.001), 0.5 cases per month for each equine practice (p=0.01), 0.2 cases per month for each mixed practice (p=0.11), and 0.1 cases per month for each poultry practice (p=0.39) (Appendix B.3). Twelve sentinels reported using the web reporting system, and one sentinel reported by faxing the one-page reporting form; 96.6% of the total reports were submitted through the web-based system. The median time to enter a report into the web-based system was 3.0 minutes. For cases entered into the webbased system, the time from the date the case was seen to the date the case was entered was a median of 5.0 days (mean: 11.7 days, standard deviation: 17.7 days, minimum: 0 days, maximum: 114 days), with 67.7% of cases entered within one week. For cases faxed in, the time from the date the case was seen to the date the case was entered was a median of 26.5 days (mean: 40.8 days, standard deviation: 39.0 days, minimum: 3 days, maximum: 152 days), with 11.5% of cases entered within one week. The median time from the date the case was seen to the date it was entered was 9.0 days when sentinels reported that they sent samples to an external laboratory for testing (mean: 22.5 days, standard deviation: 26.2 days), and 4.0 days 63 when sentinels did not report sending any samples (mean: 11.2 days, standard deviation: 17.9 days). The response rate for the survey on reasons for participation was 61% (11/18) for sentinels who initially signed up for the project. Nine of the respondents were sentinels that completed the project, and two were sentinels that signed up but never started reporting. The nine participating sentinels reported ten reasons for participating: most (6/10) of the reasons were because they wanted to help, with the others being that the project was either interesting (2/10), or important (2/10). Participating sentinels thought it was too early to say what was going well with the project (3/9), though some found the summaries fed back to them interesting (2/9), one found the web interface easy to use and one was just satisfied that the project existed. A minority of sentinels struggled with the fact that the relevance of the project was not clear (3/9), one found it difficult to find the time to participate, one had issues using the website for data entry, and one thought that the emphasis was not on animals that represented the highest risk of zoonoses to humans. Clearer relevance (3/9), reminders (1/9), and easier reporting (1/9) were given as the incentives they would need to participate in a similar project in the future, with only two responding they would be fine with the same incentives they received in the current project. Despite this, seven (78%) reported that they remained interested in the project. The two sentinels who signed up but never started reporting initially signed up in order to help, one reporting that they didn’t end up participating because of lack of time, while the other found the data being collected too general. They did think the project was interesting and potentially useful, however they thought the data 64 collected should be more detailed (e.g. should include whether the suspected disease was zoonotic or not), and should be collected retrospectively and automatically. The response rate for the survey for veterinarians who did not participate in the project was 8% (8/100). The reasons veterinarians did not participate were: they were too busy and could not find the time (4/8), they did not see eligible animals (i.e. they only saw small pets such as cats and dogs) (3/8), and that they did not hear about the project (1/8). They stated that personal contact (3/8), financial compensation (2/8), more appropriate and relevant results (2/8) and being given more time to respond (1/8) would make them more likely to participate in a similar project in the future. 3.3.2 Animal Species Seen, Reasons They Were Seen, Suspected Infections and Syndromes The sentinels saw 1,281 animals from 9 different agricultural species groupings. The large animal (mammal) cases seen were: cattle (N = 798), horses and donkeys (N = 375), goats (N = 12), sheep (N = 12), alpacas and llamas (N = 11), deer (N = 1), and pigs (N = 1). The poultry (bird) cases seen were: chicken (N = 60) and turkey (N = 9). The census of agriculture reported a total of 21,796,379 agricultural birds and mammals in the province in 11 different species groups, and the overall population estimates for the province per species are listed in Appendix B.4. Figures 3.1 and 3.2 compare proportions of species seen by sentinels and counted in the agricultural census over the study period for mammals and birds respectively. There were significantly higher proportions of cattle and horses seen by sentinels versus in the population, and a significantly lower proportion of sheep (p<0.001) (Figure 3.1). When the total population 65 of mammals was calculated without mink, the proportions and significance levels remained similar for all species except for cattle, where the increased proportion of cattle in the population (73.6%) resulted in a significantly lower proportion seen by sentinels (p<0.001) (Appendix B.4). There were significantly higher proportions of turkeys seen by sentinels versus the population (p<0.001) (Figure 3.2). Sentinels saw animals mostly for health promotion and prevention (43.3%), followed by investigations of clinical illness and/or decreased productivity (29.5%), traumatic conditions (16.2%), and post-mortem (3.1%), and other (7.8%). The reasons were different when stratified by mammals and birds: birds were seen mostly for investigations of clinical illness and/or decreased productivity (40.6%), followed by post-mortems (34.8%), health promotion and prevention (21.7%), and other (2.9%). Mammals were seen for health promotion and prevention (44.6%), followed by investigations of clinical illness and/or decreased productivity (28.9%), traumatic conditions (17.2%), post-mortems (1.3%), and other (8.1%). For individual species, see Appendix B.5. Overall 17.6% (226/1,281) of cases seen by sentinels were suspected of having an infectious disease, 14.6% (177/1,212) of mammal cases and 71.0% (49/69) of poultry (bird) cases. For mammals the most common syndromes were gastrointestinal and reproductive, and the most common infectious syndromes were respiratory and gastrointestinal (Figure 3.3). For poultry sudden death was the most common specified syndrome both overall and for infectious syndromes, however a very large proportion of syndromes were classified as other (Figure 3.4). The ‘other’ syndrome for poultry was predominantly specified to be cases seen due to high or 66 increased mortality (explained or unexplained): 87.5% (21/24) of overall syndromes and 95.5% (21/22) of infectious syndromes. For individual species, see Appendix B.6. 3.3.3 Laboratory Submissions by Sentinels 126.96.36.199 Reasons for Laboratory Submissions by Suspected Infection Table 3.2 lists the reasons sentinels gave for sending or not sending samples to the laboratory. The top reason sentinels reported for sending mammal samples to an external laboratory was to obtain a diagnosis (27.5%), and this increased for mammals suspected of having an infection (45.5%). The top reason sentinels reported for sending bird samples to an external laboratory was to confirm a diagnosis (67.2%), and this increased for those birds thought to be infectious (90.9%). The top reasons for not sending a mammal sample to the laboratory was that submission was not necessary since the animal was not ill (53.2%) and confidence in diagnosis for those animals suspected of being infectious (68.4%). The top reason for not submitting bird samples was confidence in diagnosis both for all birds (50%) and for birds suspected of infection (60%). 188.8.131.52 Laboratory Submissions by Infection and Syndrome Laboratory samples were sent to external laboratories for 13.3% of all cases seen by sentinels; 5.2% of cases had laboratory samples sent for suspected infections. This differed widely between mammals (1.8%) and birds (63.8%), as well as between individual species (Table 3.3). There were a total of 206 cases in which the sentinels specified the laboratory used: 93 (45.1%) were sent to the public provincial animal health laboratory, 45 tests (21.8%) were done in-house, 33 (16.0%) were sent to a large private laboratory, 22 (10.7%) to a federal 67 laboratory, 3 (1.5%) to other provincial laboratories in Canada, and seven (3.4%) were sent to other laboratories. Fifty-six of the total 109 (51.4%) mammal submissions were not associated with disease, and were therefore not associated with any syndromes. For all diseased mammals, 11.4% of samples were sent to external laboratories; the syndromes with the highest proportion of submissions were sudden death (71.4%) and neurological (66.7%) (Table 3.4). For mammals with suspected infectious diseases, 12.4% had samples sent to an external laboratory; the highest proportions of infectious syndromes sent for testing were respiratory (19.4%) and gastrointestinal (18.2%) (Table 3.4). Fourteen of the total 61 (23.0%) avian submissions were not associated with disease, and were therefore not associated with any syndromes. For all diseased avian cases, 87.0% had samples sent to an external laboratory; for suspected infectious avian cases the number sent to the laboratory increased to 89.8% (Table 3.4). 184.108.40.206 Laboratory Submissions by Sentinel Practice Type and Location Looking at all species together, there were many differences in sentinel practices, ranging from the numbers of species seen (range: 2-7 different species seen), proportions of infections suspected (range: 4% - 86%), in-house laboratory testing (range: 0%-19%), and external laboratory testing (range: 0% - 85%). Poultry practice submitted the highest proportion of samples to the provincial public laboratory (57.9%), followed by large animal practices (3.3%), mixed animal practices (3.2%), and equine practice (1.9%). While there were differences in the overall submitting practices by region, and linear regression showed a decrease of 2% in submissions to the provincial public laboratory with 68 every 100 km increase in distance from the laboratory, this decrease was not statistically significant (p=0.25) (Appendix B.8). 220.127.116.11 Cattle Submissions to the Laboratory Significant variables associated with the submission of cattle samples to any external laboratory in bivariate analysis were age group of the animal (p=0.001), number of animals affected (p=0.009), number of animals in the pen (p=0.001), reason for examination (p<0.001), syndrome (p<0.001), outcome (p<0.001), month (p=0.002) and season (p<0.001) the case was seen, whether an in-house post-mortem was performed (p<0.001), and the closest city or town to the sentinel (p<0.001) (Appendix B.9). Significant variables associated with the submission to the provincial animal health laboratory specifically were the total number of animals in the herd (p=0.001), the number of animals in the pen (p=0.006), reason for examination (p=0.080), month the case was seen (p=0.029), and the closest city or town to the sentinel (p<0.001) (Appendix B.9). Two predictors were included in the final multivariate logistic regression model looking at factors related to the submission of bovine samples to any external laboratory: age and number of animals in the pen. Animals less than 1 year old had higher odds of having samples sent to the laboratory versus animals older than 2.5 years of age (OR: 2.9; 95% CI: 1.2, 6.7); animals in smaller pens (1-10 animals) had lower odds of having samples sent to the laboratory than animals in larger pens (101 – 1,000 animals) (OR: 0.2; 95% CI: 0.1, 0.5) (Table 3.5). The model looking at submission of cattle samples to the provincial laboratory included only the number of animals in the pen; animals in smaller pens (1-10 animals) had lower odds of having samples 69 sent to the laboratory versus animals in larger pens (101 – 1,000 animals) (OR: 0.2; 95% CI: 0.1, 0.6) (Table 3.5). 18.104.22.168 Equine Submissions to the Laboratory Significant variables on bivariate analysis associated with the submission to any external laboratory were number of animals in the herd (p=0.067), reason for examination (p<0.001), syndrome (p<0.001), if an infection was suspected (p=0.002), outcome (p<0.001), whether an in-house test was performed (p=0.009), the closest city to the sentinel (p=0.048), and the type of veterinary practice (p=0.09) (Appendix B.10). Significant variables associated with the submission to the provincial animal health laboratory specifically were age (p<0.001), number of animals in the herd (p=0.063), the month the case was seen (p=0.029), and the distance to the provincial laboratory (p=0.005) (Appendix B.10). Two predictors were included in the logistic regression model looking at factors related to the submission of equine samples to any external laboratories: reason for examination and the type of veterinary practice. Animals seen for health promotion or investigation had higher odds of having samples sent to the laboratory (health promotion OR: 4.2, 95% CI: 1.4, 12.7; investigation OR: 14.3, 95% CI: 4.6, 44.9) versus animals seen for trauma (Table 3.6). Equine samples had higher odds of being submitted by large animal practitioners (OR: 3.0, 95% CI: 1.4, 6.2) than mixed animal practitioners (Table 3.6). A model with significant predictors looking at submission of horse samples to the provincial laboratory could not be created. 70 3.3.4 Use of Animal Data in Public Health Practice 22.214.171.124 Heuristic Algorithms Two heuristic algorithm trees were created based on the thematic analysis of expert group discussions to describe public health decision-making based on animal signals: one for the scenario where an agent has been identified in a laboratory, the other for where no agent has been identified (Appendix B.11). Decision nodes included severity of disease in humans (if known), analogies to known agents for unknown or new agents, and the level of exposure of humans. Immediate public health action was suggested for animal signals where an agent was isolated and where severity of disease in humans and the potential of human exposure were seen as high. For animal signals with more uncertainty, specifically those without a laboratory isolate (i.e. die-offs, outbreaks, changes in distribution or abundance of host species) experts usually did not think public health needed to be contacted. There was agreement between experts on the level of public health action required in the scenarios: overall the Fleiss’ kappa was κ =0.49, “moderate agreement”, for scenarios where an agent had been identified by the laboratory, and κ =0.30, “fair agreement’, for scenarios where no agent had been identified. Two or more experts agreed on the level of response for 91% for the 35 scenarios; there were no differences in level of responses based on the profession of the expert, or the length of time they have been in practice. 126.96.36.199 Evaluating Sentinel Animal Health The focus group reported that while sentinel animal health surveillance systems are not always being evaluated from a public health perspective, valuable information is being 71 collected. Although evaluation criteria designed for human surveillance systems (CDC, 2001; CDC, 2004) were not seen as appropriate criteria for the evaluation of sentinel animal surveillance systems, the focus group felt that running through some of the CDC evaluation criteria was a good exercise. There was no consensus as to how these systems should be evaluated for public health purposes, but four general themes for evaluation emerged: descriptive statistics, information and knowledge translation, public health action, and outcomes (Table 3.7). Two important issues plaguing sentinel animal health surveillance system evaluations were (1) the availability of data to conduct evaluations (e.g. missing denominators for animal populations for a particular time and area, or stratified by age and sex), and (2) aligning the goals of animal surveillance and public health surveillance. 188.8.131.52 Sentinel Data to be Collected Similar to the focus group that created heuristic algorithm trees, the second group found that public health needs to know whether people have been exposed (i.e. that it is likely that transmission has occurred), and something about the agent that would suggest it is a potential zoonosis. Laboratory diagnostic data were seen by the group as the cornerstone of animal surveillance, since, despite limitations, they offer the necessary specificity. However, syndromic data were still deemed important, as sometimes clusters of syndromes get identified long before a causative agent is identified (e.g. HIV). The group thought that sentinel data (such as syndromes, unusual morbidity/mortality) should be used in addition to “traditional” lab surveillance data and can even be reported informally, such as through a network of people who know who to call when they see something strange. One way the panel suggested that the difficulty in obtaining necessary denominators to calculate rates could be overcome, would be 72 for sentinel animal health surveillance to focus on detecting change (e.g. animal movement, development of a new food industry, food distribution systems). Veterinarians were seen as people in the best position to collect front-line animal health information, as they are trained to look at animals at both the individual and population level. Sustainable surveillance systems using veterinarians would need to be designed to collect information that is useful to the practitioner, automatically sending a standard subset of this data to governments for surveillance purposes. The focus group recognized that other frontline sentinels for animal health could be people on farms (farmers, brand inspectors), landfill operators, dead-stock pick-up people, auctioneers, slaughterhouse operators, or animal control officers. 3.4 Discussion This pilot study provided important insights into a number of key relationships that would affect the design of a sentinel system. While sample sizes were too small to conclusively identify how variables were related to types of practice, reasons for submissions, species involved and other factors affect sentinel data utility and quality, this pilot has generated useful information for future work and hypotheses for future study. Our expert group identified that it is important to include in an evaluation how well a system performed and the population under surveillance, therefore, we examine the issues associated with conducting surveillance using sentinel veterinarians, the species seen by the veterinarians, and why veterinarians saw these animals. 73 Etiological information was seen as the foundation of animal health surveillance by both of our focus groups, and therefore we examine in detail the factors we found to be related to laboratory submissions and how these would affect using such data for surveillance. We then discuss the other themes emerging from our focus groups: timeliness, human exposure information, and two main types of possible animal health surveillance (i.e. statistical surveillance vs. ‘atypical’ surveillance) in the context of expert opinion on the use of such animal data for public health practice and surveillance. 3.4.1 Conducting Surveillance Using Sentinel Veterinarians The proportion of eligible veterinarians in the province who participated in the project was very low (3.0%); however, the number who participated (n=13) is similar to the numbers of participants in other clinical or sentinel veterinary surveillance projects: seven veterinarians in VetPAD in New Zealand, twenty-seven in RSVP-A in two states in the USA, and twelve in the first year of “émergences” in France (Vourc'h et al., 2006). Veterinarians who did not choose to participate in the project and who answered our follow-up questionnaires stated that the main reason they did not participate was that they did not have the time. This suggests that if the necessary information could be extracted automatically from the practitioner’s records, the participation rate in such surveillance projects could be much higher. We found a large degree of variation in sentinel reporting behavior making it very difficult to extrapolate based on these findings as to who would make the best type of sentinel. We also found evidence of a mild reporting fatigue among most sentinels over the study period, both in the first half of the year when feedback reports were distributed as well as in the 74 second half when feedback reports were no longer provided. Authors evaluating a veterinary syndromic surveillance system in swine did not find such reporting fatigue in their study, finding instead that compliance among their sentinel veterinarians increased after the first initial three months of their pilot (del Rocio Amezcua et al., 2010). Similarly, Robertson et al. found an increasing trend in number of cases reported over time in their sentinel veterinarian pilot study (Robertson et al., 2010). While reporting is at the bottom of the priority list for practicing veterinarians, they are more likely to discuss interesting findings with a fellow veterinarian (Musgrave, 2009). This again suggests that either automatic extractions from data currently being collected in practices or sentinel network reporting may be better alternatives. The use of such automatic data extracts was also suggested by one of the sentinels who originally signed up for our project but did not end up participating. As electronic management software becomes increasingly used in veterinary practices, this may be an excellent option, as it would not only eliminate the time needed to do double entry, but it could allow for the collection of more detailed information. This information could then be used for the purposes of both animal health practice and public health practice, based on how the data is analyzed. Unfortunately, even veterinary practices that have started using electronic medical records in their offices use a variety of software and data fields, making automatic data extraction difficult, with confidentiality issues further complicating the process as no identifiable or financial information can be extracted from the systems (Anholt, 2011). Surveillance efforts in human public health relying on automatic data extracts from hospitals have encountered similar issues (Tsui et al., 2003). 75 The veterinarians in this study reported that they wanted to see a clearer relevance of the project. This could be achieved with feedback, either in the form of timely reports that are relevant to their practice, as is done by the Alberta Veterinary Surveillance Network (Government of Alberta, 2010), or regular communication with peers discussing changes or odd and unusual clinical presentations, as is done by Québec’s Le Réseau d'alerte et d'information zoosanitaire (Gouvernement du Québec, 2010) and Wyoming (Wyoming Department of Health, 2010). This theme of timely feedback of information and knowledge translation was also identified by our expert focus group, who saw new contacts, creation of data sharing agreements, knowledge and perception of animal and human disease among participants, and/or in the community as an integral component to sentinel system evaluation. 3.4.2 Animal Species Seen, Reasons They Were Seen, Suspected Infections and Syndromes The differences in proportions of animal species seen by the sentinel veterinarians in this project compared to their populations in the province highlight the importance of choosing the appropriate species for surveillance, as well as problems capturing rarer species in such a system. The appropriate target species for surveillance could be based on the susceptibility of the animal to the chosen infectious agent(s) and whether it generates a measurable clinical or immunological response to the agent(s) (McCluskey, 2003), and/or on the risk posed by the animals and the selected agent(s) to humans (Stark et al., 2006). However, another consideration should be who is to provide the source of the data. If a species is seen more by veterinarians than would be expected from the number in the population, then veterinarians may be a good source of surveillance data. We saw almost ten times the proportion of horses 76 seen by sentinels than was estimated to be in the population (Figure 3.1). This bias may be related to horses being more companion animals than food animals, inherently having a different value for their owners than other agriculture species, who then use veterinary services differently for such species. For more rare species, such as goats and sheep, owners and producers could make better sentinels. Alternatively, some of the more rare species in province, such as swine (raised in much larger numbers in other provinces in Canada), may not have been seen by our volunteer sentinel group because swine veterinary services may be more specialized, in the same manner as poultry veterinary services are, and would need to be captured using a more targeted approach (i.e. target veterinarians based on the target species). The top reason veterinarians saw animals was for health promotion and prevention reasons, with less animals seen for investigation, including suspected infectious diseases. A surveillance system focused on animals with clinical disease would then exclude all healthy animals, as well as those seen for other reasons such as traumatic conditions. Unfortunately, we were not able to find a reference for the proportion of infectious diseases suspected in human patients at general medical practice and compare it to the proportions suspected by veterinarians in this study. However, a well-established sentinel physician surveillance system for influenza-likeillness (ILI) in the province suggests that weekly proportions of suspected ILI ranges from 0% to 2% (peak influenza season) of patient visits (British Columbia Centre for Disease Control (BCCDC), 2010). We saw that 0% to 31% of cases in different species seen were suspected of having an infectious respiratory disease. This difference in suspected infectious respiratory diseases seen human and veterinary medicine suggests that this difference might scale to 77 infectious diseases in general: veterinarians may be more likely to either see or suspect infectious diseases than physicians. Overall we found that 18% of cases seen were suspected of having an infectious disease, and that this proportion varied widely between large animals (15%) and poultry (71%). This difference is likely related to the higher number of large animal cases seen for health promotion and prevention and traumatic conditions as compared to poultry, and may reflect the different roles a veterinarian has in large animal production systems versus poultry production systems. The longer life span and higher monetary value of individual large animals as compared to individual poultry means that these animals may receive more vaccinations and/or treatments for a variety of conditions from veterinarians. In contrast, a poultry producer often vaccinates flocks without a veterinarian, and may find it more cost-effective to destroy ill or injured poultry, consulting a veterinarian only in more severe situations, where many birds could be affected. We found that syndromes differed greatly between mammals and birds, and between individual species. In future planning of syndromic surveillance in animals, such syndrome proportions could help determine the expected numbers of syndromes per sentinel based on the types of species they see, and help in calculating necessary sample sizes. Unfortunately, while such calculations may be possible for more common species (e.g. cattle), they may not be for rarer species (e.g. goats) not seen by many sentinels, for whom such calculation of expected baselines would be very difficult. Further, there were signs that the syndrome categories could not be used across all species: the high use of the ‘other’ syndrome category for poultry cases 78 by sentinels when a higher than expected mortality was seen suggests that such a syndrome category would be a useful addition for this commodity group, possibly because chickens often do not present with distinctive clinical signs. Since syndromes cannot be consistently tracked across all species, and cannot be interpreted the same way, future syndromic surveillance in animals must be species-specific. Just as in human clinical presentation, there is a spectrum of disease for an individual animal species when it is infected with a particular etiologic agent; the situation gets more complex, however, for different species, as each species may have a different clinical spectrum upon infection with the same etiologic agent. Further, some zoonotic gastrointestinal disease agents that pose a risk to human health, such as Salmonella, Campylobacter, verotoxigenic Escherichia coli O157, are found in measurable quantities in animals that are asymptomatic (Hutchison et al., 2004), calling into question syndromic surveillance in those animals for such diseases for public health. Problems in the consistency of coding syndromes across species were found in analysis of animal health laboratory data, where various people (e.g. pathologists, veterinary epidemiologists) coded conditions differently from each other, and for different species (see Chapter 4 in this thesis). The expert focus group results suggest that public health currently prefers etiologic information to syndromic data, however, that such data could be useful for providing context, especially communications via sentinel networks. Therefore, syndromic surveillance in animals, even when species-specific, may be limited for use by animal health practitioners at this time, unless a significant exposure route to humans is identified. 79 3.4.3 Use of Sentinel Animal Health Data for Public Health Practice 184.108.40.206 Laboratory Diagnoses and Etiology Both focus groups agreed that laboratory information is the cornerstone of animal health data for public health at this time, since public health needs specific diagnoses in order to take action. Despite the various biases in laboratory submissions highlighted by this study, laboratories do generate etiological information that sentinels do not and such information was highly valued by our focus groups. Laboratories perform many tests on unusual samples, and so are not only looking for, but can find, the ‘unknown’. When something strange is detected, it is thought that samples usually get sent to the lab at some point (especially if the event becomes large in scope or serious). Unfortunately, we were not able to support this with the data in this pilot study, since very few cases (13%) had samples sent to the laboratory, and we did not collect information on the confidence of sentinels in all of their diagnoses (nor was suspected diagnosis a mandatory field), except in relation to submission of laboratory samples. The top reason sentinels did not send samples for animals suspected of having an infection was that they were confident in the diagnosis and did not require laboratory confirmation (68% for mammals, 60% for poultry). While only 5% of all cases seen were suspected of having an infection and had samples sent to the laboratory, almost 30% of suspected infectious diseases were sent to an external laboratory for testing. This suggests that while infectious diseases are relatively rare in terms of what sentinels see on a regular basis, sentinels see some infectious diseases that the laboratory does not, and that would therefore not be captured by an ‘official’ laboratory-based surveillance stream. The largest proportion of these may be infectious 80 diseases that are endemic and common, since a large proportion of samples were not sent to the laboratory because sentinels were confident in their diagnoses. Future studies could try to examine which specific diseases sentinels were most likely to be confident in diagnosing and that could therefore be under-represented in laboratory data. Adding a “confidence” field to the suspected diagnosis, with categories such as “unsure”, “somewhat sure”, “very sure” would allow for such analyses. It would be critical for any syndromic or diagnostic surveillance system to understand whether this level of confidence is warranted. In this study we found that only 1.4% of cattle with suspected infections get laboratory tests, and this has implications for the rational use of antibiotics and for antimicrobial resistance. For example, a veterinarian confident that an animal has pneumonia caused by a specific bacterium may choose an antibiotic course based on that belief alone, with the chosen treatment possibly contributing to misuse of antibiotics and an increase in antimicrobial resistance (Stephen, 2011). Studies comparing veterinary clinical diagnoses to pathological and microbiological outcomes could inform interpretation of clinical and etiologic surveillance signals. The proportion of suspected infections in human patients seen by general practitioners that have samples sent to the laboratory for diagnostic testing should also be compared to those sent for suspected animals by veterinarians, in order to determine whether some of the work done in human diagnostic laboratory surveillance could be applied to animal diagnostic laboratory data. While there were differences in the laboratory submission rates when examined by syndrome, the numbers of submissions in most categories were so small that the statistical difference between these proportions could not be examined, particularly for individual species. This issue is unlikely to be resolved with the addition of more sentinels, particularly for the rarer 81 species (e.g. sheep, llamas); without targeted surveillance aimed at specifically such species, statistical significance is unlikely to be reached. This is an important difference between human health surveillance and animal health surveillance – a general animal health surveillance system for all species is unlikely to capture rare species in meaningful numbers, whether the system is one based on sentinels or on laboratory submissions. Even in human public health, the syndromic surveillance systems demonstrated to be successful are those tracking specific syndromes of notable threat, for example those tracking influenza-like-illnesses (ILI) (Miller et al., 2004, Sakai et al., 2004, Marsden-Haug et al., 2007). Although there were no infectious neurological cases, 67% of all neurological cases were submitted to the laboratory, suggesting that infectious neurological cases would have also had a high likelihood of being submitted for testing. This high submission rate for neurological cases may be due to veterinarians not being confident in the etiology of rare neurological cases and needing diagnostic help from the laboratory, or that these diseases are seen as potentially more serious (e.g. BSE), and veterinarians would not want to err on such a diagnosis. Further, the Canadian Food Inspection Agency (CFIA) has a national BSE surveillance program that provides financial support to offset the costs of veterinary examinations and carcass disposal for BSE (Canadian Food Inspection Agency, 2010b). While the submission of samples for suspected BSE may be influenced by financial incentives, it is also possible that veterinarians may be motivated to submit samples for such diseases if they believe the diseases are important. 82 Financial consideration (i.e. that testing is too expensive) was not observed to be the main impediment to sending samples: only 3% of all mammals and 9% of mammals with suspected infections were not sent to the laboratory due to the expense of testing. This has important implications for policy, as it suggests that a surveillance strategy subsidizing testing costs may not necessarily result in great increases in submissions, unless practitioners were also encouraged to submit samples for animals where they were relatively confident in their diagnosis, or for those that are seen for other reasons (e.g. vaccination) and are otherwise apparently healthy. Of course, if owners/producers involved veterinarians on a more regular basis, meaning the veterinarians would see more animals, this could also increase numbers of submissions. One important area where subsidization of testing costs could result in more submissions is for those cases where veterinarians are not confident in their diagnosis: in these instances sentinels reported they did not send samples due to the testing expense for 30% of cases. These particular cases may include emerging diseases, where the veterinarians may be unsure of their diagnosis due to unusual or unfamiliar disease presentation, or cases initially treated who did not get better prompting the veterinarian to second-guess their initial diagnosis; government subsidies in these instances could result in more samples being submitted to the laboratory to diagnose new or rare diseases. In fact, this could be a system focused explicitly on the ‘odd’ diagnostic challenges that veterinarians have not seen before; however, such a system would need careful consideration of the types of samples that would be accepted, as it could be vulnerable to inappropriate submissions. There was a lot of heterogeneity in lab submission rates by type of veterinary practice, with poultry veterinarians submitting the highest proportion of samples and mixed animal practice 83 the least. This reflects the historical way diseases are diagnosed by veterinarians. It is common practice in poultry medicine to minimize clinical examination and send birds off to pathology laboratories to be euthanized and examine post-mortem to obtain diagnoses. Horses, with a much higher monetary and emotional value, are subject to more extensive clinical investigations and clinical pathology studies. In our study area, the provincial laboratory had less than half (45%) of the market share for clinical pathology services, and house laboratory tests and private laboratories were used to different extents in various types of practices. People designing future animal health surveillance systems must understand species-specific clinician behaviours and client expectations to understand how and when laboratories are used. While private laboratories received 14% of all external laboratory submissions in our study, they are not required to report etiological findings, unless they are reportable animal diseases. Reportable animal diseases are not the same as reportable human diseases; reportable animal diseases are often those that affect trade in animals and animal products, with only a few that are potential zoonoses and therefore of interest to public health. In Canada, there are 32 federally reportable animal diseases that are “usually of significant importance to human or animal health or to the Canadian economy” (Canadian Food Inspection Agency, 2010a); the list is based on diseases that are reportable internationally to the World Organisation for Animal Health (OIE). The OIE is encouraging countries to report “emergence of new diseases and relevant epidemiological events” in animals (World Organisation for Animal Health (OIE), 2011), as Canada did with the first detection of H1N1 swine influenza, a non-reportable disease. However, legislation has not yet changed to compel the reporting of such events. Therefore, private laboratories are not a practical source of data unless sentinel veterinarians were to 84 enter diagnostic laboratory results from those laboratories as part of their surveillance reporting protocol. Interestingly, no owners of animals requested the sentinel veterinarians in this study to conduct tests for suspected infections over the study period, while this was one of the most common reasons for sending samples to an external laboratory for cases that were not infectious. This underscores the important role that owners and producers play in the decision to send samples to the laboratory to isolate an infectious etiologic agent. Unfortunately, concerns over data privacy (and data ownership) and possible economic harm coming from a positive test result were not investigated in this study. Factors related to submission of samples differed for each of the two species examined (cattle and horses), highlighting that such factors must always be investigated for each species separately. For cattle, the odds of submission increased when animals were younger, and when there were more animals in the herd. These findings mirror our current cattle industry, where younger animals have a longer production life and therefore a higher value, and larger herds are more likely to be in farms with higher overall value than in smaller hobby farms. Two important implications for surveillance using laboratory data flow from these results. First, although smaller hobby farms might be of interest as they may present a risk of disease emergence due to interactions between various species including humans, cattle from such farms are likely not well represented in the laboratory data. Second, older cattle are likely underrepresented, since they are near the end of their productive life cycle and do not warrant farmer investment in diagnostics unless linked to disease events that threaten the herd 85 (Stephen, 2010). Since some diseases manifest more often in older cattle (e.g. BSE), this submission behavior represents a potential bias to not detecting emerging infectious diseases in this animal group when using laboratory data. For horses, the odds of submission increased when the reason for examinations was a disease investigation (versus trauma), and if the type of veterinary practice was a large animal practice (versus mixed practice). It should be noted that we did not have sufficient sample sizes to check appropriately for individual sentinel effects such as clustering by the individual sentinel in these models; these individual effects are potential confounders in these models, and warrant further investigation. Since the province of British Columbia has many sparsely populated areas that are separated by great distances, we examined whether distance to the public laboratory affected laboratory submission rates. While there were slight decreases in submission rates with increasing distance, the decreasing trend was not statistically significant. This suggests that veterinary practices in remote areas could still be good candidates for surveillance initiatives that rely on submissions of laboratory samples. 220.127.116.11 Timeliness Timeliness is a persuasive argument for the existence of early warning systems (Wagner et al., 2001) and was also included in the four ‘evaluation themes’ discussed in our expert panel (Table 3.7), both for evaluation of data collected and for possible public health action based on the data. We found that timeliness of case reports was good for sentinels who reported via the web (median 5.0 days), versus those who reported via the fax (26.5 days). A pilot syndromic surveillance study in swine found that timeliness of sentinel veterinarians was an average of 86 22.3 days, with most of sentinels reporting via fax or email, concluding this was too slow to identify disease trends on a weekly basis (del Rocio Amezcua et al., 2010). This highlights the importance of conducting surveillance using web-based data entry in order to ensure timeliness for action. Since 96.6% of our case reports were done via the web, we believe that our efficient and streamlined web data entry design allowed sentinels to enter cases quickly (3 minutes per case) and enhanced compliance with this reporting method. Mobile devices that streamline data entry have also been successfully used to conduct front-line surveillance in animals (Robertson et al., 2010). We saw a decrease in timeliness of 5 days for cases where sentinels reported sending samples to external laboratories compared to those who did not. The lack of laboratory submissions numbers and other identifiers in the sentinel data (e.g. owner names) meant we were not able to link the sentinel data with data from the laboratory in order to obtain diagnostic information consistently for all cases. In order to perform such analyses, sentinel surveillance initiatives should include a mandatory ‘laboratory submission ID’ field to allow for such a linkage. Including such a field necessarily means the sentinels would have to either report cases only after receiving the laboratory results, or revisit their reported cases to add the appropriate laboratory submission ID once they receive the laboratory reports, adding to their reporting load. 18.104.22.168 Exposure to Humans The heuristic algorithms developed in this study show that besides wanting to know the specific etiologic agent responsible for a disease, experts wanted to know about potential human 87 exposures in order to initiate public health measures. Our sentinel system did not collect such data. Veterinarians will vary in terms of how much data they collect on human-animal interactions while undertaking clinical investigations. This will vary with commodity group, species and circumstances. Such data may be collected if they have implications for animal disease control or involve known or suspected zoonotic diseases that impact the animal’s owner or for animals that will enter the food chain. Therefore, animal disease surveillance, whether based on sentinel reporting of individual cases, sentinel networks, or laboratory data, should strive to collect such exposure information if they want public health actions to be taken based on their findings. Without such information, public health will likely remain more reactive than proactive at this time, using animal health or veterinary surveillance information only for retrospective investigations or hypothesis generation. 22.214.171.124 Statistical Signal Surveillance Versus ‘Atypical’ Surveillance Integrating animal health data into a public health surveillance system is very complex. If we are to consider using animal data for disease detection algorithms, similar to those being used in human disease surveillance, we would need data to come from surveillance systems with a large and representative sample, such as Alberta (Government of Alberta, 2010), whose main stakeholders are animal producers and veterinarians. This type of data could be used to assess the risk to humans from endemic diseases in animals, by quantifying the relationship between the disease burden in animals and humans, using such information to detect changes in the burden of illness or emergence of new disease in both animals and humans. Spatiotemporal models could be created to understand the entire food supply system and develop risk management systems that would help the industry make the systems more 88 efficient, hence allowing for surveillance along the food chain. As mentioned above, however, a more representative sample of sentinels may still not capture rare species or rare conditions. The necessary investment in time, personnel and resources to support and sustain such a system has not yet been calculated, precluding decisions on whether such systems would be justified investments of public funds. Further, the Alberta system remains unevaluated from a public health perspective, notably in terms of its utility, effectiveness or efficiency. Rare syndromes, especially in uncommon animals, will always be difficult to find. A system tracking ‘rates’ with no denominators and not representing all animal populations is unlikely to work for all possible species, whether the system is based on syndromes or laboratory diagnoses. Our results support the design of the Québec and émergences2 models (Gouvernement du Québec, 2010, Barnouin, 2010) that specifically focus on timely sharing of ‘the unusual’. The importance of data sharing, collaboration and contact building, identified as a second ‘evaluation theme’ by our expert panel, directly addresses the shortcoming of our study and other pilot studies in this area. Fragmented data collection with small numbers of data from veterinary practice surveillance all struggle with similar obstacles of low participation rates, nonexistent databases or lack of linkages among various databases, and – last but not least – economic considerations. Labour-intensive surveillance systems may have a future in the Québec-type network where interesting and unusual events are shared among veterinarians; and where personal contact, sense of community, and obvious relevance of the work create strong rewards for their efforts. 89 126.96.36.199 Public Health Outcomes The idea of evaluating these systems based on outcomes (e.g. detection of an emerging disease or changing burden of illness in humans or animals) and the resulting interventions by public health are the last two ‘evaluation themes’ suggested by our expert panel. While we did not identify an emerging disease or a situation needing further investigation, collaborations between those working within animal health surveillance and public health surveillance that result in public health actions should be documented for use in evaluations. 3.4.4 Limitations The small number of cases and laboratory submissions reported in this pilot study resulted in the inability to attain significance in various statistical tests, and the exclusion of a number of variables of interest (i.e. those statistically significant on bivariate analyses) from the multivariate regression models, limiting our regression analyses to the most numerous species, cattle and horses. Additionally, we were not able to study clustering by individual sentinel, and even more importantly by farm in our analyses due to both small numbers and not being able to collect farm identification information: since disease can cluster by farm, this is an important limitation. In our study sentinels chose one syndrome for each case at the time of reporting, they were not required to specify the diagnosis either at the time of reporting or after receiving laboratory confirmation of disease, nor were they asked to report their confidence level in their suspected diagnoses. It was therefore not possible to examine the syndrome categories in more detail, and should be attempted in future studies. 90 We did not conduct formal validity checks of the sentinel reports by auditing a sample of the reports, as we did not have another data source to use as a comparison. Consequently, we cannot comment on data validity. Another limitation stems from use of the federal agricultural census to obtain the estimated numbers and proportions of species that are in the province. In the agricultural census, population estimates are derived from owners/producers entering the number of individual animals they have on their premises on ‘census day’ (Statistics Canada, 2009). These numbers are affected by the nature of the ‘all-in, all-out’ animal production system: large numbers of animals are moved at various stages of their life cycle from one facility to another, for example from a reproduction facility where they are born to another facility for growing, causing large changes in population numbers on farms and in regions, even potentially excluding them from the census when they are ‘in transit’ on census day. The census only collects data from farms that sell agricultural goods (Statistics Canada, 2007), smaller hobby farms are excluded, potentially resulting in lower population estimates for animals such as horses that are often kept as companion animals on such farms. Finally, these population data are available only for census administrative regions and not by sentinel practice area (i.e. their catchment area). For all of these reasons, there would likely be large inaccuracies in denominator estimates at the time of case reporting. This necessarily precludes the creation of a rate-based surveillance system and re-enforces our suggestion that animal surveillance for public health should be focused on numerators (the weird and the unusual). 91 3.5 Conclusions We conclude that syndromic surveillance in animals using sentinel veterinarians as conducted in this pilot project is not likely to be useful for public health practice at this time. Focus should be placed on animal health surveillance using diagnostic laboratories, collaborative networks of veterinarians with links to public health, or targeted automatic data extracts from veterinarian practices focusing on species and diagnoses of interest to public health. An ‘atypical’ surveillance system focused on collecting information on odd and unusual cases or clusters, can both identify and help interpret other surveillance data in order to detect emerging disease threats. All of these surveillance efforts should try to include information on potential human exposures, in order to give public health the information it needs to take action. 92 3.6 Tables and Figures Table 3.1 Variables collected from sentinel veterinarians. Level Type Demographic Date Quantity Case Diagnostic Laboratory Demographic Veterinarian Location * Mandatory fields Variable Species* Age Sex Date Case Seen* Number of Animals Affected Number of Animals in the Herd/Flock Number of Animals in Pen/Group Reason for Examination* Syndrome* Infection Suspected* Suspected Diagnosis Outcome In-House Laboratory Test or Post-Mortem External Laboratory Test* Laboratory Test Submissions and Results Reason for Submitting/Not Submitting Laboratory Sample* ID* Type of Practice* Address of Practice* 93 Table 3.2 Reasons why veterinarian sentinels stated they submitted and did not submit samples to an external laboratory for mammals and birds by suspected infection. Reason samples were: Submitted Obtain a diagnosis Request by owner Confirm a diagnosis Response to an important outbreak Severity of disease Other Not submitted Animal not ill Confident in diagnosis Samples not easily obtained Testing too expensive Necropsy done in-house Other *treated and no longer considered ill Mammals Suspected All (N=109) Infections (N=22) 27.5% 45.5% 24.8% 12.8% 27.3% 1.8% 0.9% 32.1% All (N=1,107) 53.2% 33.8% 3.8% 3.2% 2.5% 3.5% 27.3% Suspected Infections (N=155) 1.3%* 68.4% 5.8% 9.0% 3.2% 12.3% All (N=61) 6.6% 67.2% - Birds Suspected Infections (N=44) 4.5% 90.9% - 3.3% 23.0% 2.3% 2.3% All (N=8) Suspected Infections (N=5) 50.0% 50.0% 60.0% 40.0% 94 Table 3.3 Suspected infections and laboratory sampling by species, as reported by sentinel veterinarians between March 1, 2009 and March 31, 2010. Species Total N Mammals Alpaca Cattle Deer Donkey Goat Horse Llama Pig Sheep Unknown Large Animal All Mammals Birds Chicken Turkey All Birds Overall Total Suspected Infections N N% Total Sent to Lab N N% Suspected Infections Sent to Lab N N% 4 798 1 3 12 372 7 1 12 2 1,212 1 137 0 0 4 29 0 1 5 0 177 25.0% 17.2% 0.0% 0.0% 33.3% 7.8% 0.0% 100.0% 41.7% 0.0% 14.6% 1 57 1 0 1 47 0 0 2 0 109 25.0% 7.1% 100.0% 0.0% 8.3% 12.6% 0.0% 0.0% 16.7% 0.0% 9.0% 1 11 0 0 0 9 0 0 1 0 22 25.0% 1.4% 0.0% 0.0% 0.0% 2.4% 0.0% 0.0% 8.3% 0.0% 1.8% 60 9 69 1,281 45 4 49 226 75.0% 44.4% 71.0% 17.6% 52 9 61 170 86.7% 100.0% 88.4% 13.3% 40 4 44 66 66.7% 44.4% 63.8% 5.2% 95 Table 3.4 Submissions to an external laboratory for ill and infectious mammals and birds by syndrome. Syndrome Ill Mammals* N N (%) sent to external laboratory 5 (71.4%) 2 (66.7%) 8 (20.0%) 7 (16.7%) 3 (12.5%) N Ill Infectious Mammals** N (%) sent to external laboratory 0 (0.0%) 0 (0.0%) 7 (19.4%) 2 (14.3%) 2 (15.4%) Ill Birds* N N (%) sent to external laboratory 15 (75.0%) 1 (100.0%) 2 (100.0%) 4 (100.0%) Ill Infectious Birds** N (%) sent N to external laboratory 19 15 (78.9%) 1 1 (0.0%) 1 1 (100.0%) 4 4 (100.0%) Sudden Death 7 2 20 Neurological 3 0 1 Respiratory 40 36 2 Musculoskeletal 42 14 4 Decreased 24 13 Production Multi-systemic 55 6 (10.9%) 30 3 (10.0%) 2 2 (100.0%) 2 Gastrointestinal 118 10 (8.5%) 33 6 (18.2%) Dermatologic 16 1 (6.3%) 5 0 (0.0%) Reproductive 82 2 (2.4%) 28 1 (3.6%) 1 0 (0.0%) 0 Other 77 9 (11.7%) 16 1 (6.3%) 24 23 (95.8%) 22 Total 464 53 (11.4%) 177 22 (12.4%) 54 47 (87.0%) 49 *Ill mammals/birds: cases with an associated syndrome **Ill infectious mammals/birds: cases with an associated syndrome and suspected of infection 2 (100.0%) 0 (0.0%) 21 (95.5%) 44 (89.8%) 96 Table 3.5 Logistic regression results for cattle submissions to any external laboratory and to the provincial diagnostic animal health laboratory. Predictors Submission to all external laboratories Age < 1 year 1 – 2.5 years > 2.5 years Number of animals in the pen 1-10 11-100 101-1,000 Submission to provincial laboratory Number of animals in the pen 1-10 11-100 101-1,000 Bivariate P-value* OR Logistic Regression 95 % CI 0.004 0.142 0.008 2.89 0.41 Ref (1.24, 6.73) (0.12, 1.38) - <0.001 0.125 0.001 0.21 0.59 Ref (0.09, 0.48) (0.29, 1.22) - 0.008 0.645 0.006 0.24 0.53 Ref (0.10, 0.60) (0.23, 1.21) - *P-value for individual category vs. all other categories OR: Odds Ratio, 95% CI: 95% Confidence Interval, Ref: reference category. 97 Table 3.6 Logistic regression results for equine submissions to any external laboratory. Predictors Submission to all external laboratories Reason for Examination Trauma Health promotion Investigation Other Type of practice Mixed Equine Large animal Bivariate P-value* OR Logistic Regression 95 % CI <0.001 0.120 <0.001 0.783 Ref 4.16 14.29 6.68 (1.36, 12.72) (4.55, 44.91) (1.32, 33.90) 0.002 0.176 <0.001 Ref 0.70 2.96 (0.18, 2.64) (1.42, 6.15) *P-value for individual category vs. all other categories OR: Odds Ratio, 95% CI: 95% Confidence Interval, Ref: reference category. 98 Table 3.7 Expert focus group sentinel animal health surveillance evaluation themes. Evaluation Themes 1. Descriptive Statistics 2. Information/ Knowledge Translation 3. Public Health Action 4. Outcomes Description Examples Documentation on how good the system is and a description of the population under surveillance Number of reports submitted, number of participants reporting, how often reporting is done, coverage, types of diagnoses or animal health issues, number of cases, and number or proportion of samples sent to laboratories, timeliness of signals (also under “Public Health Action”) Interagency and interdisciplinary communication and collaboration and public awareness Investigations and interventions by public health arising from animal health signals Detection of a new emerging disease or changes in the burden of human illness Numbers of new contacts, creation of data sharing agreements, knowledge and perception of animal and human disease among participants, and/or in the community Numbers and depth of investigations and/or interventions by public health arising from animal health signals, time to follow-up (also under “Descriptive Statistics”) Detection of emerging disease, changes in an emerging disease, changes in the burden of illness in animals, changes in the burden of illness in humans 99 100.0% 90.0% 80.0% Percent of Mammals 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% % Sentinel % Population Alpacas & Llamas 0.9% 0.4% Bison Cattle Deer Goats Horses‡ 0.0% 66.0% 0.1% 1.0% 31.0% 0.9% 59.3% 0.4% 1.0% 3.9% Mink Pigs Sheep 0.0% 0.1% 1.0% 19.4% 10.1% 4.5% Figure 3.1 Proportions of large animal species seen by sentinels and those counted in the agricultural census. A (*) indicates sentinel proportions significantly different from proportions in the population at the p<0.001 level for each species; significance versus the population could not be calculated for bison, deer, mink and pigs for sentinels due to small counts. ‡ Horses include ponies and donkeys. 100 100.0% 90.0% 80.0% Percent of Birds 70.0% 60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% Chicken Turkey Other Poultry % Sentinel 87.0% 13.0% 0.0% % Population 93.3% 4.0% 2.6% Figure 3.2 Proportions of poultry species seen by sentinels and those counted in the agricultural census. A (*) indicates sentinel proportions significantly different from proportions seen in the population at the p<0.001 level for each species; significance versus the population could not be calculated for other poultry due to small counts. 101 140 120 Number of Mammal Cases 100 80 60 40 20 0 sudden death neurological respiratory musculoskeletal decreased production reproductive other Infectious 2 0 36 14 13 30 33 5 28 16 Not infectious 5 3 4 28 11 25 85 11 54 61 multi-sytemic gastrointestinal dermatologic Figure 3.3 Numbers of infectious syndromes (n=177) and non-infectious syndromes (n=287) suspected in ill animals by sentinels. 30 25 Number of Bird Cases 20 15 10 5 0 sudden death neurological respiratory musculoskeletal multi-sytemic reproductive other Infectious 19 1 1 4 2 0 22 Not infectious 1 0 1 0 0 1 2 Figure 3.4 Numbers of cases with non-infectious syndromes (n=49) and infectious syndromes (n=5) suspected in ill poultry by sentinels. 102 4 Detection of Emerging Infectious Disease Trends and Clusters in Animal Laboratory Data for Public Health Surveillance 4.1 Introduction The emergence of infectious diseases from animals, or zoonoses, such as avian influenza (AI), bovine spongiform encephalopathy (BSE), and Nipah virus have had major societal implications such as the economic cost of control measures, lost production, disruptions in trade, and increased public health concerns. Most (60%) of emerging infectious diseases (EIDs) are zoonoses, and the incidence of EID events has increased significantly from 1940 to 2004 (Jones et al., 2008). Animal disease surveillance is a core veterinary public health activity (Doherr and Audige, 2001). Improved surveillance in animals is thought to reduce the impacts of emerging diseases by allowing for early detection and containment of diseases and their associated effects on society (Dufour, 1999, Doherr and Audige, 2001). There are numerous data sources that are being used in animal surveillance, such as pet insurance data (Egenvall et al., 1998, Penell et al., 2007), auction markets (Van Metre et al., 2009), abattoirs (Benschop et al., 2008), and clinical data (Checkley et al., 2009, del Rocio Amezcua et al., 2010, DeGroot, 2005). Among this variety of potential data sources, animal laboratory data continues to be the cornerstone of animal disease surveillance (Dorea et al., 2011b). Although animal laboratories differ in their diagnostic testing capabilities and coverage, laboratory data form the foundation for diagnostics needed for international trade and agricultural surveillance. Should they prove reliable, it could be that new systems to collect data (e.g. sentinel veterinary surveillance 103 described in Chapter 3) are not needed. Biases inherent in laboratory data, such as species underrepresentation, gaps in coverage, and different diagnostic practices by individual pathologists do not necessarily preclude use of these data for surveillance as long as the biases are known and remain constant over time. Early research into using animal laboratory data to detect outbreaks (Kosmider et al., 2006, Shaffer et al., 2008) and emerging diseases (Kosmider et al., 2011, Gibbens et al., 2008) suggest that they may be useful. In this study we examine whether agricultural animal diagnostic laboratory data can detect animal health events of importance, especially those of potential relevance to public health. The objectives of this study were to 1) extract data in an appropriate format and timely manner for surveillance, and 2) investigate the validity of the data for identifying seasonal trends and events in animal health outcomes of relevance to public health. While data quality is still a major concern, there have been advancements in statistical approaches to the surveillance of infectious diseases in veterinary public health (Hohle et al., 2007). Since we did not have information about the distribution of our data and expected it to vary across agricultural groups and time, we chose generalized additive models (GAM) (Hastie and Tibshirani, 1990) as they are flexible models that do not require strong assumptions about the distribution of the data that are implicit in standard parametric regression. We used a GAM currently in use at the British Columbia Centre for Disease Control Epidemiology Services to monitor abnormal increases in human weekly disease counts as part of their surveillance efforts (British Columbia Centre for Disease Control (BCCDC), 2010). We hypothesized that 104 statistically significant signals in animal health events of possible public health relevance would be identified in our data time series using the GAM analyses. Within laboratory data as a whole, it is possible to assess both pre-diagnostic (e.g. total submissions, microbiology orders) and diagnostic data from an emerging disease surveillance perspective. Diagnostic data are more specific and should have less variation in the background, i.e. noise, than pre-diagnostic data. However, since specific diagnoses can only signal an emerging disease once it is clearly identified, diagnoses classified in syndromes may provide an opportunity to identify an unexpected increase in a new emerging pathogen if it has been classified into other diagnoses that have a similar disease presentation. Pre-diagnostic data streams should provide another opportunity to identify signals of unknown or emerging diseases, since they are not dependent on known diagnoses. Additionally, pre-diagnostic data may provide timelier detection of disease outbreaks, as they are available prior to diagnoses (Shaffer et al., 2008, Dorea et al., 2011b). We hypothesized that statistically significant signals in both pre-diagnostic and diagnostic data streams will correlate with known animal health events of interest to public health. Further, we hypothesized that the pre-diagnostic data streams will correlate with the diagnostic data streams, and that signals in pre-diagnostic data streams will be timelier than in the diagnostic data streams. 105 4.2 Methods 4.2.1 Creation of Surveillance Database: Data Extraction, Cleaning, and Coding The British Columbia (BC) Ministry of Agriculture Animal Health Centre (AHC) is a full-service veterinary diagnostic laboratory, whose mandate is to diagnose, monitor, and assist in controlling and preventing animal disease in BC (British Columbia Ministry of Agriculture, 2012a). The AHC provides a full range of diagnostic testing, including pathology, bacteriology and virology, and is frequently involved in investigative projects addressing emerging disease problems in production animals, poultry, and fish (British Columbia Ministry of Agriculture, 2012a). Nine years of laboratory diagnostic data, from April 1, 1998 to March 31, 2007 from the Oracle VetLab system used by the AHC diagnostic laboratory were extracted as tab delimited files using AQT (www.querytool.com), a tool that provides a query environment with multidatabase support and the ability to extract millions of rows of data. Associated look-up tables were extracted using Microsoft Access 2007. The two animal species with the greatest number of submissions, chicken and cattle, were chosen for further analyses as the large numbers make them more amenable to statistical analyses. These species were further stratified by agricultural commodity groups. The chicken commodity groups were: all chicken, layers (egg-laying hens), broilers (chicken for meat production), and broiler-breeders (breeders of chicken for meat production); for cattle they were: all cattle, beef, and dairy. The pre-diagnostic (sometimes also called “syndromic” in the 106 literature, e.g. Dorea et al. (Dorea et al., 2011b)) data streams were total submissions and submissions by sample type. For chicken the sample types were: whole bird, tissue, blood and serum, swabs/water samples/feed, and other or unknown. For cattle the sample types were: whole animal, milk, fecal, tissue, blood and serum, swabs/water samples/feed, and other or unknown. Diagnostic codes assigned to submissions were used to create time series by etiologic agent and body systems based on the International Statistical Classification of Diseases and Related Health Problems (ICD) tenth edition (ICD-10) categories. ICD codes were chosen because they are widely used and internationally endorsed by the World Health Organization (Word Health Organization, 2007). Diagnosis codes were assigned by veterinary pathologists for cases using the results of diagnostic tests and/or pathology results. A total of 922 diagnosis codes in the data were classified into 22 broad ICD-10 categories (Table 4.1; (Word Health Organization, 2007)). For infectious diseases with less than three separate etiologic agents associated with the diagnosis, the diagnoses were further classified by the etiologies (Pan American Health Organization (PAHO), 2003). All coding of diagnosis codes into both clinical syndromes and etiologic agents responsible was done initially by an epidemiologist, with some sections coded by veterinary epidemiologists, and then validated by a third veterinary epidemiologist. The first (primary) diagnosis code was used to categorize each submission for descriptive analyses, and all diagnoses linked to a submission were used for statistical signal analyses, where each diagnosis was treated as a separate submission. The ICD-10 categories used for both chicken and cattle for most time series analyses were: infectious, respiratory, 107 gastrointestinal, and neurological diseases. Specific etiologic agents chosen for analysis were those with at least one diagnosis per year in 8 of the 9 years, and/or those of significant public health importance. The potential zoonoses chosen for chicken were: Clostridium perfringens, Escherichia coli, Pasteurella spp., Pasteurella multocida, Staphylococcus spp., Salmonella, and Avian influenza (the last two added for their agricultural and public health importance). The non-zoonotic agents chosen for chicken were: Avian adenovirus, Eimeria spp., Gallid herpesvirus, and Marek’s disease virus. The potential zoonoses chosen for cattle were: Escherichia coli, Pasteurella spp., and Salmonella (added for its veterinary and public health importance). The non-zoonotic agents chosen for cattle were: Clostridium chaovei (feseri) and Eimeria or Isospora. Data cleaning was done using SAS software, version 9.0 of the SAS System for Windows (SAS Institute Inc, Cary, NC, USA); coding was conducted using SPSS for Windows, Release 17.0.0 (SPSS Inc, Chicago, 2008). 4.2.2 Descriptive Analyses and Overall Trends The descriptive statistics for all submissions by agricultural group were: number of submissions through time (including monthly time series graphs), number of submissions by diagnosis and year, and number of submissions by ICD-10 category and year. Mean submission numbers (and standard deviations) by month and by season were calculated to assess seasonal patterns. A two-sample, two-tailed T-test, with unequal variance (heteroscedastic) was used to test for differences in numbers of submissions. Linear regression was used to test for overall trend in the submission time series data for each group. Population estimates for select species 108 were obtained from the 1996, 2001, and 2006 Statistics Canada Agricultural Census (Statistics Canada, 1996, Statistics Canada, 2001, Statistics Canada, 2006). Time series, monthly and seasonal graphs, T-tests, as well as control chart analyses were done using Microsoft Excel 2007. 4.2.3 Signals in Time A generalized additive model (GAM) developed for the British Columbia Centre for Disease Control Epidemiology Services as part of their surveillance efforts (British Columbia Centre for Disease Control (BCCDC), 2010) was used to identify statistically significant signals over the study period by agricultural group, sample type, ICD-10 code, and etiologic agent. Data were aggregated into weekly time series of counts and run for the full time period (9 years). We used the model to identify statistically significant alerts (p-value ≤ 0.001) over the time period, where each alert signifies a count above the calculated expected value of the model. The alert assessment starts from the most recent record, assessing the aggregated observed count backward up to four previous weeks. If only the current week is above expected, a point alert is generated. If instead some combination of the last two, three or four weeks are above expected, then a group alert is issued. Descriptive statistics of alerts were calculated, including mean, minimum, and maximum number over the time period. Assuming that actionable signals were more likely to be those where more than one time series showed an alert, weeks with statistically significant alerts across time series for a particular agricultural commodity were also identified. In a GAM (Hastie and Tibshirani, 1990) the response (outcome) is additively related to the independent (predictor) variables as in the equation 109 where g is the link function (a log link in our model) and ε is a random error term. Each is assumed to be in some predefined space of functions defined by smoothers such as loess (locally estimated polynomial regression). We assume that weekly counts, y(t), come from a family of distributions with an expected value of λ(t) and variance equal to σ2λ(t), where σ2 is the dispersion parameter. The dispersion parameter determines the distribution of the data: when σ2 > 1, we assume λ(t) follows a negative binomial distribution, when σ2 < 1, we assume λ(t) follows a binomial distribution, and if σ2 = 1 λ(t) we assume a Poisson distribution. The mean structure of our model is assumed to be where α is a constant, f(t) is a temporal trend, and c(t) is an seasonal cyclical trend (with a period of one year). Both f(t) and c(t) are non-parametric functions fit by loess. The cyclical trend c(t) is excluded from the model if there are less than two years of data or if there are less than 10 weeks per year on average with an observed count greater than zero. The temporal trend f(t) is calculated if the weekly time series contains at least two years of data and at least 10 non-zero observations. If f(t) and c(t) are both excluded from the model, then the mean is assumed to be constant and is estimated by the average of the data, . An alert in the data is identified through two components: the log odds ratio and delta. The odds of obtaining the difference between observed and expected counts in the last four weeks of data are expressed 110 as the log odds ratio. Delta is an assessment of the variability in the long term trends in the data, by calculating the difference in the fitted function between subsequent time points (f1(t) – f1(t1)) over the given time interval. Delta compensates for edge effects by being added as a penalty term when calculating the z-score for the log odds at the edge (last time point or most recent week) of the fitted model. When the mean of the time series was constant, only log odds were used. The GAM models backfit each component of the model by holding the others constant while that component is updated, cycling through each component until all have converged. Details on the model are available in the GAM outbreak alert algorithm’s user manual (British Columbia Centre for Disease Control (BCCDC), 2010). We used GAMs as our main signal detection method, as they do not require strong assumptions about the distribution of the data that are implicit in standard parametric regression, assumptions that may force the fitted relationship away from its natural path at critical points. While in our GAM we do not assume independence of observations, we do assume that the size of the population of the cohort of interest increases/decreases stably over time, taking this population trend into account during the model fitting. 4.2.4 Trends, Events, and Outbreaks In order to assess how well the data correlate to known animal health patterns or clusters over the time period, critical informants (laboratory personnel, including veterinary pathologists and laboratory technicians) were polled for cattle and chicken trends they would expect to see in the data as well as disease ‘events’ that occurred between 1998 and 2007. The personnel were asked to list possible reasons for the patterns/clusters/signals as well as events 111 and outbreaks in chicken and cattle using a survey on conducted prior to seeing initial descriptive data analyses. These survey results were then compared qualitatively to the descriptive analyses of the weekly laboratory data streams. Case studies were then used to identify time periods and data streams where statistically significant signals would be expected. Case studies were included in this study if they were identified on the initial critical informant survey, and if corroborating information on the event was found by searching past issues of the laboratory’s newsletter (British Columbia Ministry of Agriculture, 2009-2012), searching the internet and ProMed using keywords describing the outbreaks. The case studies then qualitatively compared to all possible data streams that could exhibit the event (e.g. a signal corresponding to a known poultry outbreak was searched for in all poultry streams that were consistent with the disease in question) in the four weeks preceding the event, the week(s) of the event, and four weeks following the event. 4.3 Results 4.3.1 Creating the Surveillance Database and Descriptive Analysis The animal diagnostic laboratory data were extracted as a combination of tab-delimited files (“main table”), for a total of 17 million lines, and Microsoft Access spreadsheets (all other tables including look-up tables). The data had to be extracted separately for each year due to the size of the files. The main data table was cleaned for issues with date, identifiers of text (data enclosed in single [‘] or double quotes [“]) and was recombined into species-group files that became the raw data files for all subsequent analyses. 112 The extraction process took approximately 70 person-hours. Data were transformed into analyzable (e.g. one line per case) databases for each specific purpose (e.g. coding of diagnosis, time series analyses), with cases subsequently linked to variables in other databases by their case ID. Data cleaning and coding took approximately 840 person-hours, with the coding of diagnoses into meaningful ICD-10 categories and etiologic agents taking another 420 personhours. There were sufficient numbers of submissions to conduct meaningful analyses, with 9,762 cattle submissions and 12,305 chicken submissions. 4.3.2 Descriptive Analysis and Correlation with Expected Trends 188.8.131.52 Chicken Descriptive Analyses and Trends There were 12,305 individual chicken submissions between 1998 and 2007, with a mean 114 submissions per month, and a standard deviation of 34. This total included 2,805 layers, 6,690 broiler, 1,920 broiler-breeders, and 890 unclassified or other chicken. There was a statistically significant decrease in submission numbers over the time period for all chicken (β= -0.79, p<0.001), broiler-breeders (β= -0.14, p<0.001), broilers (β= -0.48, p<0.001), as well as layers (β= -0.11, p=0.005). Conversely, the population of total hens and chickens in BC increased during this time: in 1996 the number was 13,759,261 (number of farms reporting: 4,840), in 2001 the number was 18,820,347 (number of farms reporting: 5,198), and in 2006 the number was 18,341,907 (number of farms reporting: 4,460). The number of broilers increased over the time period, from 9,656,204 in 1996 to 14,120,577 in 2006; the number of laying hens remained relatively constant, with 3,523,249 in 1996 and 3,855,093 in 2006. 113 No seasonality was apparent in chicken submissions: there were no significant differences seen between the month with highest mean number of submissions (September mean=123) and the lowest (December mean=102) (p=0.26). Similarly, no significant seasonality was seen in the specific commodity groups (broiler-breeders, broilers or layers). At least one diagnosis code was assigned to 4,657 (37.8%) submissions, with a total of 7,418 diagnostic codes assigned. The 7,418 diagnostic codes mapped to 14,321 ICD-10 code categories (since one diagnosis could map to more than one ICD-10 code, e.g. AI maps to four ICD-10 codes: 1, 6, 10, and 11 – see Table 4.1 for explanation of codes); the most common were infectious diseases (n=3,661), followed by non-specific symptoms and laboratory findings (n=2,411), followed by codes for special purposes (see Table 4.2 for diagnosis codes by ICD-10 category by year for all chicken). There were 3,661 infectious disease diagnoses of which 2,122 had specific etiologic agents associated with them (see Table 4.3 for etiologic agents associated with infectious disease diagnoses by year for all chicken). Detailed descriptive analyses for each chicken commodity group are shown in Appendices C.1-C.4. Experts at the laboratory expected three trends in chicken data. The first was a seasonal trend in respiratory submissions, presenting as a bi-modal increase in spring and winter, and a nadir in the fall. There was, however, little evidence of seasonality in respiratory diseases in chicken: there was no significant difference between the season with the highest mean number of submissions (spring mean: 21, SD: 13), and the lowest (winter mean: 15, SD: 7) (p=0.56). The second was an expected decrease in submissions in broiler-breeders due to a reduction in the broiler-breeder monitoring program. Indeed, a statistically significant decrease in monthly 114 submissions in broiler-breeders was seen over the time period (β= -0.135; p<0.001). Thirdly, a general decrease in overall chicken submissions following the 2004 AI outbreak was expected: the mean monthly submissions prior to the outbreak were significantly higher than after the outbreak (1999-2003 monthly mean: 162, SD: 28; 2005-6 monthly mean: 91, SD: 17; p<0.0001) (Figure 4.1). This decrease was more pronounced in broilers and broiler-breeders, than in layer chickens (see Appendices C.2-C.4). 184.108.40.206 Cattle Descriptive Analyses and Trends There were 9,762 individual cattle submissions between 1998 and 2007, with a mean 90 submissions per month, and standard deviation of 24. This total includes 2,013 beef, 6,580 dairy, and 1,169 unclassified or other cattle. There was a statistically significant decrease in submission numbers over the time period for all cattle (β= -0.19, p=0.007), and moderately in beef cattle (β= -0.08, p=0.07), but not in dairy cattle (β= 0.01, p=0.79). The population of cattle in BC decreased slightly during this time: in 1996 the number of cattle was 814,103 (number of farms reporting: 9,185), in 2001 the number was 814,949 (number of farms reporting: 7,726), and in 2006 the number was 800,855 (number of farms reporting: 6,996). The drop in numbers may have been largely attributable to the dairy cattle population (that decreased from 82,008 in 1996 to 72,756 in 2006), whereas the beef cattle population in the province remained relatively constant (273,217 animals in 1996 and 276,897 in 2006). All cattle together showed evidence of seasonality, with the highest submissions in the spring (April mean=117), and the lowest in the summer (July mean=68) (p<0.0001). Beef cattle 115 showed a similar seasonal pattern as all cattle, with the highest number of submissions in the spring (March mean = 38), and the lowest in the summer (August mean = 6) (p<0.0001) (Figure 4.2). There was no seasonality found in dairy cattle submissions. At least one diagnosis code was assigned to 2,865 (29.3%) submissions, with a total of 4,751 diagnostic codes assigned. The 4,751 diagnostic codes mapped to 8,221 ICD-10 code categories (since one diagnosis could map to more than one ICD-10 code, e.g. BSE maps to two ICD-10 codes: 1 and 6 – see Table 4.1 for explanation of codes); the most common were non-specific symptoms and laboratory findings (n=1,629), followed by infectious diseases (n=1,552), and diseases of the digestive system (n=1,250) (see Table 4.1 for diagnosis codes by ICD-10 category by year for all cattle). There were 1,552 infectious disease diagnoses of which 427 had specific etiologic agents associated with them (see Table 4.5 for etiologic agents associated with infectious disease diagnoses by year for all cattle). Detailed descriptive analyses for each cattle commodity group are shown in Appendices C.5-C.7. Laboratory experts expected four trends in the cattle data. The first was an expected seasonality in overall submissions in beef cattle, with more submissions in spring and fall, and a nadir in the summer. Clear seasonality was seen in the beef cattle, the highest in the spring (highest - spring mean: 139, SD: 37) with a significant drop in mean submissions in the summer (lowest: summer mean: 35, SD: 10; p<0.0001) (Figure 4.2). Seasonality in respiratory diseases, with more submissions in the winter, was expected: this pattern was seen in dairy cattle (highest - winter mean: 15, SD: 5; lowest: spring mean: 6, SD: 3; p=0.002), however, not in beef cattle (highest - spring mean: 13, SD: 4; lowest: summer mean: 4, SD: 4; p=0.0006). The third 116 expected event was a drop in whole animal submissions after 2003 due to BSE concerns: the monthly submissions of whole animals prior to the outbreak were significantly different from the monthly submissions after the outbreak (1999-2002 monthly mean: 28, SD: 12; 2004-6 monthly mean: 19, SD: 9; p=0.0002) (Figure 4.3). Finally, an increase in milk submissions was expected in 2005 due to special milk monitoring projects in dairy cattle: the two months with the greatest number of submissions in dairy cattle over the whole time period were February 2005 and May 2005 (Appendix B.7). 4.3.3 Signals in Time Series 220.127.116.11 Chicken Overall there were more group alerts (mean: 19 range: 11-25) than point alerts (mean: 10, range: 5-15) based on the GAM model analyses. All data streams produced at least one signal over the time interval in at least one of the chicken commodity groups except for the etiological data streams of E. coli and Pasteurella sp. Over the time interval, there were a total of 38 point signals (average 4.2 per year) and a total of 75 group signals (average 8.3 per year). The total number of alerts (counting all signals, including overlapping signals in one week) for all chicken data streams was 113 (average of 12.6 per year). All chicken time series resulted in a total of 33 alerts, layer chicken time series resulted in a total of 35 alerts, broiler chicken time series resulted in a total of 29 alerts, and broiler-breeder chicken time series resulted in a total of 16 alerts. Appendices C.8.1-C.8.4 show all of the point and group alerts generated over the time interval for chicken by commodity groups. In all chicken there were two weeks that had signals 117 in more than one time series, in layers there were four such weeks, in broilers there were five such weeks, and finally in broiler-breeders there were three such weeks (Table 4.6). In the individual time series by production type, 73% of alerts could not be found in the overall chicken time series. The time series for broiler-breeders in particular had 87% signals not detected in the overall chicken time series. Still, the overall time series, not broken up into production types, yielded 40% of alerts not found in the individual production type time series. One third of these signals in the overall chicken time series are likely ‘late’ signals that had earlier signals in the individual production type time series. Within the overall chicken time series there were 7% of submissions that were not coded by production system. There were 41 signals in the pre-diagnostic data streams, of these 36 (on average 4 per year) were not concurrent with a signal in an etiologic data stream. Within the ICD 10 data streams, there were 28 signals, of these 17 (average 1.9 per year) were not concurrent with a signal in an etiologic data stream. 18.104.22.168 Cattle Overall there were more group alerts (mean: 16 range: 9-21) than point alerts (mean: 14, range: 6-18) based on the GAM model analyses. All data streams produced at least one signal over the time interval in at least one of commodity group, except for the etiologic data stream for Listeria monocytogenes. Over the time interval, there was a total of 41 point signals (average 4.6 per year), and a total of 49 group signals (average 5.4 per year). All cattle time series resulted in a total of 39 alerts, beef cattle time series resulted in a total of 15 alerts, and dairy cattle time series resulted in a total of 36 alerts. The maximum number of alerts (ignoring 118 overlapping signals in one week) for all cattle data streams was 90 (average of 10 per year). Appendices C.8.5-C.8.7 show all of the point and group alerts generated over the time interval for cattle by commodity group. In all cattle there were three weeks with a signal in more than one time series, there were no such weeks in beef cattle, and there were six such weeks in dairy cattle (Table 4.7). In the individual time series by production type, 44% of alerts could not be found in the overall cattle time series. Conversely, the overall time series, not broken up into production types, had 25% of alerts not found in the individual beef and dairy time series. Within the overall cattle time series there were 12% of submissions that were not coded by production system. There were 52 signals in the pre-diagnostic data streams, and none (on average 5.8 per year) were concurrent with a signal in an etiologic data stream. Within the ICD 10 data streams, there were 18 signals, of these 16 (average 1.8 per year) were not concurrent with a signal in an etiologic data stream. 4.3.4 Correlation of Signals to Events and Outbreaks Three events were chosen from those identified by laboratory experts to be case studies in this investigation, based on ability to find corroborating and more detailed information from other sources: avian influenza (AI) in chicken in 2004, Salmonella pullorum in chicken in 1997-8 and 2001, and bovine spongiform encephalopathy (BSE) in cattle in 2006. 119 22.214.171.124 Case Study 1: Highly Pathogenic Avian Influenza in Chicken in 2004 Highly pathogenic avian influenza (HPAI) H7N3 was detected in a broiler-breeder chicken operation in British Columbia on February 16, 2004. The entire population of approximately 16,000 birds on this farm was destroyed on February 19-20, at which point a heightened surveillance program, based on oropharyngeal and cloacal swabs, was initiated on commercial farms within 5 km of the infected flock (Canadian Food Inspection Agency, 2004). On the index farm the first signs of illness, prior to confirmation of HPAI, were a slight increase in mortality, and a mild drop in egg production and feed consumption (Canadian Food Inspection Agency, 2004). On March 11, another flock approximately 3 km from the initial premises was confirmed as infected with HPAI H7N3; clinical signs included increased mortality in one barn, and all birds were destroyed after laboratory confirmation of HPAI (Canadian Food Inspection Agency, 2004). More infections were confirmed over the next two months, resulting in the decision to depopulate the entire Control Area (approximately 19 million birds), with the last positive commercial premises detected on May 13, and the last backyard poultry on May 18 (Canadian Food Inspection Agency, 2004). At the end of the outbreak, H7 viruses had been isolated from 28 commercial farm premises and two backyard flocks (Canadian Food Inspection Agency, 2004). The first human case confirmed for H7 was on March 18 in a worker involved in culling chickens, the second case of H7 was April 5 in a poultry worker; by mid-May of the 643 workers involved in culling/cleaning activities, 21 reported mild respiratory symptoms and two were confirmed for H7 (Buck, 2004). In our study HPAI was coded as a diagnosis of Avian Influenza, due to etiologic agent Orthomyxovirus A, and in the following ICD-10 categories: an infectious and parasitic disease, a 120 disease of the nervous system, a disease of the eye and adnexa, and a disease of the respiratory system. There were 12 signals generated between February and April of 2004 for all chicken submissions, 10 in layer chicken, 9 in broiler chicken, and 9 in broiler-breeders (Appendices C.8.1-C.8.4). The first signal was on February 14, 2004, specifically for Orthomyxovirus A, in both all chicken submissions and the broiler-breeders. There was no signal in the prediagnostic or syndromic data streams (e.g. all submissions, submission by sample type), or the ICD-10 groups, prior to this alert. The date with the largest number of signals (n=19) was April 3, 2004, consisting of signals in all submissions, submissions by sample type (whole bird), etiologic agent (Orthomyxovirus A) and ICD-10 categories (respiratory, nervous, and digestive). See Figure 4.1 for the time series of all chicken submissions in the study period, showing the weeks with statistically significant signals. 126.96.36.199 Case Study 2: Salmonella pullorum in Chicken in 1997-8 and in 2001 In October 1997, the laboratory diagnosed pullorum disease caused by Salmonella pullorum in a backyard poultry flock on southern Vancouver Island, with three new cases identified by November (Bowes, 2009). With ongoing testing, a new positive flock was identified in June, unrelated to previous cases (Bowes, 2009). The testing program was expanded and by September 30, 1998, 38,964 premises were visited, 53,316 birds were tested and of the 72 flocks with reactors, 18 were confirmed positive by culture and depopulated (Bowes, 2009), with a total of 27 cases identified by the end of the investigation (British Columbia Ministry of 121 Agriculture, 1999). Subsequently, two more backyard flocks were found positive on Vancouver Island sometime in 2001 (Canadian Poultry Magazine, 2010). In our study neither Pullorum disease nor S. pullorum was a diagnostic code in the original laboratory data, and were therefore not included in our study database. The only diagnosis in the original laboratory data related to a Salmonella spp. was Salmonella septicemia. The diagnosis of oophoritis, which laboratory experts reported that Pullorum disease cases could have been coded under, was not coded as infectious, had no etiologic agent associated with it, and was classified as a disease of the genitourinary system. As our data only started on April 1 1998, we cannot examine signals at the start of this outbreak. The following signals were in the pre-diagnostic data streams in 1998: in all chicken: 10/31/1998 in all submissions and 11/14/1998 in whole bird submissions; in layers: 9/26/1998; 10/3/1998, 10/10/1998 and 10/31/1998 in all submissions, and 9/26/1998; 10/3/1998, and 10/31/1998 in whole bird submissions; and none in either broilers or broiler-breeders. In 2001, there were the following signals in the pre-diagnostic data streams: none in all chicken; in layers: 12/15/2001 in swabs, water samples and feed submissions; none in broilers; and in broiler-breeders: 6/30/2001 and 7/7/2001 in blood and serum submissions. See Appendices C.8.1-C.8.4 for all signals in the chicken commodity groups, and Figure 4.1 for the time series of all chicken submissions in the study period, showing the weeks with statistically significant signals. 188.8.131.52 Case Study 3: Bovine Spongiform Encephalopathy (BSE) in Cattle in 2006 On April 8, 2006 a 6-year old Holstein dairy cow was euthanized and sampled under Canada’s BSE Surveillance Program (Canadian Food Inspection Agency, 2006). BSE was 122 confirmed on April 16, 2006 and no part of the carcass entered the food chain (ibid). BSE was not included as a specific diagnostic code in our data; the only prion disease in the database was scrapie. BSE was not diagnosed at the provincial laboratory during the study period, it was only diagnosed at the federal laboratories. There was only one signal in cattle in all of 2006, and this was after the date of diagnosis, on April 22, for the time series of blood and serum samples for all cattle (Appendix C.8.5). Figure 4.3 shows the weekly time series for whole cattle submissions during the whole study interval. 4.4 Discussion Animal health laboratory data can be extracted, coded, and analyzed for surveillance purposes. With standardized coding, substantial resources, and commitment to investigate signals, statistical analyses of the data could yield a feasible number of alerts for investigation. We found that the data was able to identify expected trends such as seasonality, as well as a sustained decline in the number of laboratory submissions following the occurrence of major health events. Unexpectedly, our case studies did not show early warning signals in prediagnostic and syndromic data streams. 4.4.1 Feasibility: Data Extraction and Coding Our study shows that the entire process of extracting, cleaning and analyzing data was exceedingly complex, requiring a wide range of expertise and personnel-hours, resulting in a process not timely enough for surveillance purposes in this initial phase. While it may seem that the ongoing time burden would be lower, since some of the necessary coding (e.g. converting diagnoses to ICD-10 categories and associating them with pathogens) has been done, 123 our case studies suggest that ongoing modification of codes is needed. Further, significant investment of time and resources is required to fully automate data cleaning and coding, to allow for sufficient time for analyses. Recent descriptions of laboratory surveillance using animal data in the literature (Gibbens et al., 2008, Shaffer et al., 2008, Kosmider et al., 2006, Danan et al., 2011) did not contain a discussion of the necessary person-time needed to create or maintain the surveillance databases. In order for animal data to be used for surveillance, and for practitioners and researchers to be able to learn from others who have built animal health surveillance systems, standard coding of pre-diagnostic and clinical syndromes and diagnoses needs to be implemented (Dorea et al., 2011b).There are two levels of standardized coding schemes needed for implementation of laboratory surveillance, since the objectives of diagnostic animal laboratories and surveillance are different. First, standardized laboratory codes (for tests, results and diagnoses) are needed, followed by the mapping of these codes to standardized surveillance categories. The importance of standardized laboratory codes is highlighted by our S. pullorum case study: since new diagnosis codes were not present (or easily added) in the laboratory information system, S. pullorum was not classified appropriately for surveillance. In this study ICD-10 codes were used to classify laboratory data for surveillance primarily because of their prevalent use in human medicine, reporting, and public health surveillance. There are currently no ICD coding schemes designed for surveillance in veterinary medicine, and existing standardized veterinary coding schemes are not used universally (Wurtz and Popovich, 2002). None of the systems covered in a recent review of veterinary syndromic 124 surveillance used a standard classification system (Dorea et al., 2011b), suggesting that there is no consensus on appropriate (and user-friendly) categories in veterinary surveillance to date. 4.4.2 Data Validity: Seasonality and Expected Trends A number of expected seasonal patterns and trends were observed in the data. Seasonality was pronounced in beef cattle, due to the spring calving season and the occurrence of neonatal and juvenile animal diseases (Figure 4.2). The expected seasonality in respiratory diseases in cattle (highest submissions in the winter) was confirmed for dairy cattle only; beef cattle exhibited a different seasonality (highest submissions in the spring). Given the different production systems for these animals, such differences are not unexpected. Other expected trends were also found in the data, such as a decrease in submissions in broiler-breeders due to a reduction in a broiler-breeder monitoring program. This highlights the significant effect that such enhanced testing may have on overall trends in laboratory data, and the potential benefit of classifying and analyzing submissions by the type of submission (e.g. diagnostic, monitoring program, enhanced surveillance due to a known outbreak) in future studies. The marked and lasting drop in overall chicken submissions after the 2004 outbreak of Avian Influenza (Figure 4.1) demonstrates that animal health events cause lasting changes in submission patterns. A similar decline in whole animal submissions was seen in cattle in following the first case of BSE in Alberta in 2003 (Figure 4.3), with mean numbers of submissions differing significantly before and after the outbreak. The drop in submissions in chickens may be due to a number of factors. There may be decreased disease and hence need for testing due to enhanced biosecurity following the AI outbreak (Bowes, 2007). Another 125 reason may be a decrease in the overall number of farms involved in production. The restructuring of the poultry sector favoring larger commercial producers following an AI outbreak has been observed in other countries (McLeod et al., 2005), and was seen in the decreased number of poultry farms reporting in BC on the agricultural census between 2001 (5,198 farms) and 2006 (4,460) (Statistics Canada, 2001, Statistics Canada, 2006). Similarly, there was a decrease in the number of cattle operations between 2001 (7,726) and 2006 (6,996) (Statistics Canada, 2001, Statistics Canada, 2006). Finally, the drop may be due to changes in farmer behavior. After the AI outbreak in BC, the level of compensation post-outbreak was perceived to be low by the farmers, decreasing their willingness to submit samples for AI specifically (Kitching, 2011). Notably, we could not find reports of drops in submissions following outbreaks in human surveillance, therefore such drops appear to be a feature of animal surveillance only. The cost of laboratory testing is a known barrier for animal owners in the agricultural sector (Kosmider et al., 2006), an issue heightened by a large outbreak: the BSE outbreak in 2003 in the province of Alberta changed profit margins for farmers quite drastically, with farmers experiencing average losses of 33% compared to the previous year (Mitura and di Pietro, 2004). An anthropological study looking at the effects of the 2003 outbreak on farmers in Alberta found changes in on-farm animal management after the outbreak, such as more “do-it-yourself” diagnosis and treatment of animal health problems, as well as a higher frequency of terminating sick animals on the farm instead of consulting veterinarians (Smart, 2007). These findings support the hypothesis that the overall decreases seen in laboratory submissions in our study after the AI and BSE outbreaks were driven at least in part by changes in farmer behavior. 126 4.4.3 Signals in Pre-Diagnostic and Diagnostic Data Streams We were able to identify alerts in weeks with counts above those expected by the GAM model in a large variety of data streams. A possible investigation into the signals on a prospective basis seems feasible, as the number of signals was, on average, 13 per year for chicken and 10 per year for cattle. If only signals with an alert in more than one time series in a particular production type in one week were investigated, then the workload would be two investigations per year in chicken, and one per year in cattle. However, the AI case study highlights that the earliest signal for this outbreak was a point signal in one time series alone, suggesting that focusing only on correlation between time series may not be the best way to achieve early detection. Another approach may be to focus only on those signals where potential emerging issues are more likely. One of these would be when an alert exists in either a pre-diagnostic or a diagnostic ICD-10 data stream with no corresponding alert in an etiologic stream (i.e. alerts with no potential known cause associated with them). In our study there was a feasible amount of these for potential investigation: on average four of these per year in chicken and six per year in cattle in the pre-diagnostic streams, and two per year for both chicken and cattle for the ICD-10 streams. It is also possible to focus on the more ambiguous ICD-10 categories, i.e. ICD-10 number 18 “symptoms and abnormal findings not elsewhere classified” and ICD-10 number 22 “codes for special purposes”. In our study the ICD-10 number 18 category was the most common ICD-10 category in cattle and second most common for chicken, suggesting that further investigation 127 of these data may be warranted. Moreover, since only 38% of chicken and 29% of cattle submissions were associated with a diagnosis code, both specific and non-specific (i.e. grouped in ICD-10 numbers 18 and 22) submissions with no diagnoses and non-specific diagnoses together represent the majority of all chicken and cattle submissions. Focusing on these types of submissions would require an in-depth re-categorization, based not on diagnoses but on clinical information submitted with the laboratory samples, a task beyond the scope of this study, and perhaps most surveillance systems. Gibbens et al (Gibbens et al., 2008) designed a livestock surveillance system focused specifically on these instances where diagnoses were not reached: they separated specific categories (infectious and non-infectious) where diagnoses could not be reached. Since they coded these ‘unknown’ cases by body system (similar to our ICD-10 codes) based on the presenting clinical sign, they were able to focus on how the proportions of ‘unknown’ cases vary through time in relation to total submissions, and compare these proportions to previous years to produce signals indicating issues needing investigation. While neither Gibbens et al. (Gibbens et al., 2008), nor Kosmider et al. (Kosmider et al., 2006), who conducted analyses of the same data source, were able to detect a new or emerging infectious disease, they found the data useful for supporting disease-free status for international trade, epidemiologic investigations, and general situational awareness. Future animal surveillance systems would benefit from differentiating between submissions where a diagnosis could not be reached (i.e. a real unknown) from those where a diagnosis code was not present in the database (i.e. a coding issue), something we were unable to do with our data. 128 In our study, alerts occurred at different times in different species and type of production system (e.g. broiler vs. layer in chicken, dairy vs. beef in cattle). Further, there were alerts that were identified in the overall time series (i.e. all chicken, all cattle) that were not found in the individual production groups, and vice versa. We were also unable to determine the relative utility of the types of data stream based on correlations: alerts that appeared in two or more time series for a particular production type occurred between all combinations of prediagnostic, ICD-10, and etiologic streams. Since this is a proof-of-concept study, similar to other studies of statistical signals generated in animal health surveillance data (Kosmider et al., 2006, Shaffer et al., 2008), we were not able to investigate the signals prospectively. Without investigating these alerts in detail it is difficult to determine which streams yielded the most useful and meaningful alerts. The GAMs seem to be a suitable method to detect anomalies in animal health submission patterns. The conspicuous change in baseline submission rates after the 2004 AI outbreak underscores the usefulness of statistical models that can quickly adjust to shifts in overall submission numbers, such as the GAM. Since we used all nine years to create our GAM models, we did not precisely re-create prospective surveillance in our analyses (i.e. model would need to be run separately for each week in the time period using only the preceding cumulative weekly data to create the model), only time series without enough data to fit temporal or seasonal models are likely have been affected. It is unclear, however, whether our model was impacted by our assumption that the size of our underlying populations increase/decrease stably over time; such an assumption is not likely to be valid for an all-in all-out agricultural system such as poultry in BC. Although we ran a large number of models (n=137), leading to 129 possible issues related to multiple testing, we chose a very stringent level (p<=0.001) for our alerts. Another issue with the GAM alerting model is that only counts within a one to four week interval are assessed, therefore diseases with longer incubation/transmission periods may be missed by such models. Our analyses were based on all diagnosis codes associated with each submission, resulting in multiple counts for a submission with more than one diagnosis. Excluding these duplicate submissions could have resulted in different signals, particularly for overall number of submissions and submissions by sample type. While inclusion of these additional diagnoses as separate cases is a non-differential bias and would not result in changes in overall trends and seasonality, they may increase the probability of the GAM model detecting a more pronounced signal whenever multiple diagnoses are assigned to submissions in the syndromic data streams. However, the pattern in the time series graph for all submissions in chicken used in the GAM (Figure 4.1) looks very similar to the time series graph for all chicken submissions where multiple submissions for additional diagnoses were not included (Appendix C.1). 4.4.4 Ability of Data to Predict Known Outbreaks In our case studies, we looked at correlations of three known outbreaks with the alerts generated using GAM analyses of our data. Our results suggest that such analyses are unlikely to be useful in timely detection of outbreaks. We hypothesized that both the pre-diagnostic and diagnostic data streams would correlate with known outbreaks, however, we found that they only correlated with the Avian Influenza outbreak. Further, we hypothesized that the prediagnostic data streams and ICD-10 streams would be able to detect the outbreaks in our case 130 studies in a more timely fashion than the etiologic data streams. We found, however, that the etiologic data resulted in timelier alerts. The pre-diagnostic and ICD-10 data streams alerted one to seven weeks after the etiologic data stream, presumably picking up increased submission activity in many poultry sectors related to the outbreak. This goes against the prevailing expectation that ‘syndromic’ data should lead to timelier alerts than diagnostic data (Shaffer et al., 2008, Dorea et al., 2011b). Detecting events in surveillance data can be seen as a signal-to-noise problem. Good detection algorithms are designed to reduce the background noise (i.e. variation of the data stream background) while preserving the signal as much as possible. The definition of noise is context-dependent and therefore flexible, as it includes factors that are not of interest to the surveillance system in question (Burkhom, 2007). There are two main categories of noise: the expected and the unexpected. Expected noise can be included in a model, and in our study it included random variability from historical data, linear and seasonal trends. Unexpected noise cannot be included in models without extensive research of retrospective data and resource intensive auxiliary feeds (Burkhom, 2007). In our study, these factors could include changes in application of disease case definitions and coding, in testing availability and prices, underlying populations and submission practices. Within the more non-specific data streams, such as overall submission numbers and ICD-10 categories, the specific signal we were looking for was only a small proportion of the data, the rest being noise. In our Avian Influenza example we found that the specific etiologic data stream (Orthomyxovirus A) showed a ‘true’ signal, while the other data streams potentially 131 signaled increased noise from our perspective (e.g. changing submission patterns due to enhanced surveillance). Unfortunately, if we are truly looking for unknown diseases, there will not be an etiologic data stream – we would be forced to look only at the more non-specific data streams. While these streams do include relevant signals, our study shows they are likely to identify noise. The problem with EID surveillance based on such data is that a ‘true’ signal is difficult to define in advance. It follows then, that this type of analysis is not likely to result in EID signals in non-specific data streams. The Salmonella pullorum case study highlights a situation where an outbreak was known to be occurring, however there were no corresponding alerts in the diagnostic data streams (as no specific diagnostic code existed in original lab data). The pre-diagnostic data streams did not generate an alert correlated to the identification of new positive flock, and instead generated alerts later, likely related to an expanded testing program (‘noise’). The BSE case study shows that certain diseases, especially those that are new, or not diagnosed at the laboratory due to unavailability of tests or proper level of containment, can be under surveillance only in select laboratories. At the time of this study, BSE could not be diagnosed in the provincial laboratory; it could only be diagnosed at the federal CFIA laboratory. The one signal we found in the 2006 cattle data (in the blood and serum data stream) was 6 days after the diagnosis of BSE was publicized. Since the required sample for BSE is the whole animal, head, or brain, this signal was likely either coincidental or, again, related to increased surveillance efforts after an outbreak was declared. The confounding effects of the response in the field by producers and veterinarians to a diagnosis or outbreak seen in all three cases studies underscore the need to 132 know what is happening in the industry when interpreting signals associated with more submissions. Since limited retrospective records are kept of investigations, the best way to evaluate the clinical significance of the types of signals such as the ones generated in this study would be to conduct a prospective surveillance study investigating each signal. Our results, however, would not support large investments in an emerging disease surveillance system based on this type of laboratory data. It is more likely that the first indicator of an emerging disease will be one (statistically insignificant) positive test, which is nonetheless significant both clinically and in the context of animal health, industry, trade, and public health. Therefore, in disease emergences with a small number of initial cases, astute clinicians (Shaffer et al., 2008) or laboratory personnel may be more likely to detect the event. 4.5 Conclusion Our study looked at using cattle and chicken laboratory data in British Columbia, Canada, for emerging disease surveillance. While we found that the data were largely valid in detecting seasonal trends and expected events, we found significant challenges in preparing the data for analysis in terms of time invested and availability of resources for coding and classification. In order for such data to be used in ongoing surveillance, consensus on animal EID surveillance case definitions, both nationally and internationally, is needed. This consensus may be especially difficult for animal health data, as it is very context-specific: it varies by region, as well as by species and type of agricultural production system. 133 In contrast to our expectations based on current literature on EID surveillance, we found that statistical analyses of pre-diagnostic and syndromic data did not result in timely and valid early warning signals. These data streams seemed to contain more noise than signal. Statistically significant alerts did not correlate with events of epidemiological significance. It may be that EIDs in animals are not amenable to classical statistical surveillance approaches: rather than initially presenting as large outbreaks that could be detected by algorithms, they may present as isolated cases. It is therefore unlikely that statistical surveillance of animal health laboratory data, as conducted in this study, will be useful for EID surveillance at this time. 134 4.6 Tables and Figures Table 4.1 ICD-10 Diagnosis Code Categories (based on WHO, 2006). ICD-10 Category 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. ICD-10 Description Certain infectious and parasitic diseases Neoplasms Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism Endocrine, nutritional and metabolic diseases Mental and behavioural disorders Diseases of the nervous system Diseases of the eye and adnexa Diseases of the ear and mastoid process Diseases of the circulatory system Diseases of the respiratory system Diseases of the digestive system Diseases of the skin and subcutaneous tissue Diseases of the musculoskeletal system and connective tissue Diseases of the genitourinary system Pregnancy, childbirth and the puerperium Certain conditions originating in the perinatal period Congenital malformations, deformations and chromosomal abnormalities Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified Injury, poisoning and certain other consequences of external causes External causes of morbidity and mortality Factors influencing health status and contact with health services Codes for special purposes 135 Table 4.2 All chicken diagnostic codes made at the BC Animal Health Centre mapped to ICD-10 diagnostic categories (n=14,321). Year* Total 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Infectious & parasitic diseases 279 548 520 562 438 412 306 277 252 67 3661 Neoplasms 117 176 81 50 42 60 28 33 34 12 633 Blood & blood-forming organs diseases & certain disorders 92 152 77 92 92 108 60 43 35 12 763 involving the immune mechanism Endocrine, nutritional & 45 73 45 48 34 23 11 8 10 1 298 metabolic diseases Nervous system diseases 12 4 8 14 10 15 17 7 4 91 Eye & adnexa diseases 4 9 3 1 10 6 7 5 1 46 Ear & mastoid process 0 diseases Circulatory system diseases 33 58 55 71 39 23 35 23 22 4 363 Respiratory system diseases 49 70 89 94 67 73 76 47 44 7 616 Digestive system diseases 100 186 215 172 143 127 94 107 99 28 1271 Skin & subcutaneous tissue 25 51 26 33 17 21 18 10 12 1 214 diseases Musculoskeletal system & 127 193 176 156 111 103 61 46 40 8 1021 connective tissue diseases Genitourinary system diseases 19 48 39 40 38 42 24 13 23 2 288 Pregnancy, childbirth & the 4 5 4 4 4 3 5 4 1 3 37 puerperium Conditions originating in the 12 38 48 58 59 22 20 18 42 6 323 perinatal period Congenital malformations, deformations & chromosomal 4 4 4 4 4 2 5 1 1 1 30 abnormalities Symptoms, signs & abnormal clinical & laboratory findings, 231 372 336 363 276 289 214 157 137 36 2411 not elsewhere classified** Injury, poisoning & certain other consequences of 27 73 75 60 48 43 27 29 33 3 418 external causes External causes of morbidity 27 82 83 50 35 44 19 24 36 5 405 & mortality Codes for special purposes 146 219 220 202 172 167 98 99 82 27 1432 Total 1353 2361 2104 2074 1639 1583 1125 951 908 223 14321 *incomplete years: 1998 is from April 1-December 31 1998, 2007 is from January 1-March 31, 2007 **Examples of diagnosis codes grouped in the more ambiguous category 18 “Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified” included non-specific diagnoses such as ‘hemorrhage’, ‘anemia’, edema’, ‘dehydration’, and ‘inflammation’; category 22 “Codes for special purposes” included diagnoses such as ‘autolysis’, ‘foreign body’, ‘normal tissue’, and ‘specimen unsuitable’. ICD-10 Category 136 Table 4.3 All chicken infectious and parasitic diagnoses made at the BC Animal Health Centre by likely etiologic agent(s) and zoonotic status by Year 1998-2007 (n=3,661). Etiologic Agents Associated Year* § with Infectious Diagnoses 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Possible Zoonoses Aspergillus 3 1 1 4 Capillaria sp. 2 3 1 1 Clostridium perfringens 4 26 24 18 19 11 6 14 8 1 Cryptosporidium 1 Erysipelothrix rhusiopathiae 3 1 (insidiosa) Escherichia coli 60 117 91 102 73 53 37 31 10 3 Escherichia coli ‘and viruses’ 2 Listeria monocytogenes 1 1 Mycobacterium avium var 1 1 1 2 3 avium Newcastle disease virus 1 1 2 (NDV)/ Avian paramyxovirus-1 Orthomyxovirus A 1 7 Pasteurella multocida 4 9 2 3 1 4 2 3 3 Pasteurella spp. 1 2 Salmonella spp. 2 5 1 2 1 3 Staphylococcus aureus, S. 31 60 60 55 46 33 20 21 14 5 hyicus, S. epidemidis, S. gallinarium Not Zoonoses Avian Adenovirus Group I 8 16 22 26 5 12 37 38 43 11 Avian Bornavirus 1 Avian encephalomyelitis virus 1 Clostridium colinum 1 1 1 1 Eimeria or Isospora 1 Eimeria sp. 31 42 62 50 28 48 23 17 18 4 Gallid herpesvirus 1 (GaHV-1)/ 3 8 31 8 1 14 3 2 15 2 Avian herpesvirus 1 Histomonas meleagridis 1 2 1 1 2 Infectious bronchitis virus 1 1 1 2 (IBV) Infectious bursal disease virus 1 1 1 4 3 (IBDV) Marek's disease virus (MDV)/ 16 57 45 32 24 40 18 20 22 9 gallid herpesvirus 2 (GaHV-2) Mycoplasma spp 1 2 1 2 1 1 1 Ornithonyssus sylviarum 5 3 5 8 1 3 1 1 No specific agent 111 205 160 236 230 188 143 119 118 29 Total 279 548 520 562 438 412 306 277 252 67 § Agents were associated with a diagnosis code if the diagnosed condition was an infectious or parasitic disease and caused by three or fewer etiologic agents. These data do not represent actual isolation or serologic test positives for the specific agents listed, nor are they corrected for coding errors. *incomplete years: 1998 is from April 1-December 31 1998, 2007 is from January 1-March 31, 2007 Total 9 7 131 1 4 577 2 2 8 4 8 31 3 14 345 218 1 1 4 1 323 87 7 5 10 283 9 27 1539 3661 137 Table 4.4 All cattle diagnostic codes mapped to ICD-10 diagnostic categories made at the BC Animal Health Centre (n=8,221). ICD-10 Category 1998 149 1999 245 2000 200 2001 213 Year* 2002 2003 210 161 2004 162 2005 107 2006 69 2007 36 Infectious & parasitic diseases Neoplasms 7 7 8 6 10 4 6 6 4 1 Blood & blood-forming 36 37 32 45 52 44 42 29 25 12 organs diseases & certain disorders involving the immune mechanism Endocrine, nutritional & 28 45 42 45 54 55 36 26 13 15 metabolic diseases Nervous system diseases 15 19 17 20 17 22 3 8 10 4 Eye & adnexa diseases 1 2 2 2 1 1 Ear & mastoid process 1 1 diseases Circulatory system diseases 18 24 14 12 33 30 19 20 24 9 Respiratory system diseases 58 93 95 94 100 85 60 68 47 15 Digestive system diseases 101 186 162 164 195 139 128 92 62 21 Skin & subcutaneous tissue 10 10 7 10 12 17 20 16 5 4 diseases Musculoskeletal system & 5 13 8 12 12 14 10 11 8 1 connective tissue diseases Genitourinary system 21 35 29 24 29 42 27 22 15 1 diseases Pregnancy, childbirth & the 86 144 133 86 98 81 61 52 47 29 puerperium Conditions originating in the 3 8 5 16 10 9 5 3 8 4 perinatal period Congenital malformations, 4 5 5 7 9 5 8 5 1 deformations & chromosomal abnormalities Symptoms, signs & abnormal 134 200 164 182 264 221 163 137 120 44 clinical & laboratory findings, not elsewhere classified** Injury, poisoning & certain 20 41 33 29 37 42 17 19 24 10 other consequences of external causes External causes of morbidity 15 24 18 22 20 25 13 12 17 10 & mortality Codes for special purposes 13 11 5 14 36 9 8 12 8 3 Total 724 1147 979 1003 1200 1006 788 646 509 219 *incomplete years: 1998 is from April 1-December 31 1998, 2007 is from January 1-March 31, 2007 **Examples of diagnosis codes grouped in the more ambiguous category 18 “Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified” included non-specific diagnoses such as ‘hemorrhage’, ‘anemia’, edema’, ‘dehydration’, and ‘inflammation’; category 22 “Codes for special purposes” included diagnoses such as ‘autolysis’, ‘foreign body’, ‘normal tissue’, and ‘specimen unsuitable’. 138 Total 1552 59 354 359 135 9 2 203 715 1250 111 94 245 817 71 49 1629 272 176 119 8221 Table 4.5 All cattle infectious and parasitic diagnoses by likely etiologic agent(s) made at the BC Animal Health Centre and zoonotic status by Year 1998-2007 (n=1,552). Year* Etiologic Agents Associated Total § with Infectious Diagnoses 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Possible Zoonoses Clostridium perfringens 1 1 1 1 4 Clostrodium heamolyticum 1 1 Coxiella burnetti 1 1 2 Cryptosporidium 5 3 4 8 5 6 1 32 Erysipelothrix rhusiopathiae 1 1 1 0 2 1 4 1 11 Escherichia coli 2 8 9 10 12 9 4 2 1 1 58 Listeria monocytogenes 1 1 1 2 5 Mycobacterium avium 0 2 5 1 1 2 11 subspecies paratuberculosis Pasteurella spp. 8 12 18 9 11 7 4 2 4 1 76 Salmonella spp. 4 10 12 9 8 2 5 5 1 56 Sarcocystis spp. 2 1 1 1 1 6 Not Zoonoses Actinobacillus spp. 1 1 Bovine herpesvirus 1 (BHV-1) 1 2 1 2 1 7 Bovine papillomavirus 1 & 2 1 1 Bovine Respiratory Syncytial 2 6 3 5 16 Virus Bovine viral diarrhea virus 7 7 6 4 4 4 6 2 40 Clostridium chauvoei (feseri) 2 6 3 5 1 1 2 4 3 27 Eimeria or Isospora 3 7 5 5 5 2 4 5 2 2 40 Haemonchus placei and H. 0 1 1 4 6 contortus Histophilus somnus 3 2 1 6 Histoplasma farciminosum 1 1 Pestivirus 2 2 3 5 1 1 0 14 Rhadinovirus 1 1 2 Ureaplasma diversum 1 1 2 4 No specific agent 111 178 131 150 148 122 125 84 50 26 1125 Total 149 245 200 213 210 161 162 107 69 36 1552 § Agents were associated with a diagnosis code if the diagnosed condition was an infectious or parasitic disease and caused by three or fewer etiologic agents. These data do not represent actual isolation or serologic test positives for the specific agents listed, nor are they corrected for coding errors. *incomplete years: 1998 is from April 1-December 31 1998, 2007 is from January 1-March 31, 2007 139 Table 4.6 Statistically significant signals in more than one time series in chicken by commodity group and type of data stream. Commodity group All chicken All chicken Layers Layers Layers Layers Broilers Broilers Broilers Broilers Broilers Broiler-breeders Broiler-breeders Broiler-breeders Data streams Date of signal Diseases of the digestive system and Eimeria sp. All submissions, whole bird submissions, Orthomyxovirus A, diseases of the nervous system, diseases of the respiratory system, and diseases of the digestive system All submissions and whole bird submissions All submissions and whole bird submissions All submissions and whole bird submissions All submissions and whole bird submissions, Orthomyxovirus A, diseases of the nervous system, and diseases of the respiratory system All submissions and diseases of respiratory system All submissions and diseases of respiratory system Diseases of the nervous system and diseases of the digestive system Avian Adenovirus and Eimeria sp Salmonella sp. and infectious diseases Orthomyxovirus A and diseases of the respiratory system Orthomyxovirus A, Staphylococcus sp., and infectious diseases All submissions, whole bird submissions, and diseases of the respiratory system December 19, 1998 April 3, 2004 September 26, 1998 October 3, 1998 October 31, 1998 April 3, 2004 March 13, 2004 March 27, 2004 April 3, 2004 October 9, 2004 November 5, 2005 February 14, 2004 February 21, 2004 March 6, 2004 140 Table 4.7 Statistically significant signals in more than one time series in cattle by commodity group and type of data stream. Commodity group All cattle All cattle All cattle Dairy Dairy Dairy Dairy Dairy Dairy Data streams Date of signal All submissions, other sample type submissions, and diseases of the respiratory system Cryptosporidium sp., and diseases of the digestive system All submissions and fecal sample submissions Infectious diseases and diseases of the respiratory system Cryptosporidium sp., and diseases of the digestive system All submissions and infectious diseases All submissions and fecal sample submissions All submissions and fecal sample submissions All submissions and fecal sample submissions December 22, 2001 January 24, 2004 May 21, 2005 February 27, 1999 January 24, 2004 July 24, 2004 February 26, 2005 May 14, 2005 May 21, 2005 141 Figure 4.1 Time series of all chicken submissions from April 1, 1998 to March 31, 2007 to the BC Animal Health Centre. The blue line shows the generalized additive model; weeks with statistically significant group alerts signify weeks with counts above the modeled expected value at three levels: p≤0.001 (high alerts – red), p≤0.01 (medium alert – orange), and p≤0.05 (low alert – green). The first set of high alerts correlates with the enhanced Salmonella pullorum surveillance in the fall of 1998, the second set of high alerts correlate with the Avian Influenza outbreak in spring 2004. 142 A) A ) B) 50 B ) Mean Number of Submissions per Month 45 40 35 30 25 20 15 10 5 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Figure 4.2 Seasonality in beef cattle submissions from April 1, 1998 to March 31, 2007 to the BC Animal Health Centre. A) Time series of beef cattle submissions. The blue line shows the generalized additive model, illustrating regular yearly seasonality. Red dots indicate recent weeks being evaluated by the model. B) Mean number of beef cattle submissions per month showing seasonality with high submissions in the spring, and a nadir in the summer; error bars show ± 1 standard deviation. 143 Figure 4.3 Time series of all cattle whole animal submissions from April 1, 1998 to March 31, 2007 to the BC Animal Health Centre. The blue line shows the generalized additive model, illustrating the gradual decrease in submissions after Bovine Spongiform Encephalopathy was first identified in Canada in the spring of 2003. The most current week is represented by a red dot, indicating it is being evaluated by the model. 144 5 Utility of Surveillance Algorithms in the Analyses of MultiSpecies Salmonella Surveillance Data in British Columbia, Canada 5.1 Introduction Salmonella species are major bacterial pathogens that cause enteric infections in people and result in economic losses for the livestock and food industry. In Canada, the rate of human salmonellosis is estimated to be between 2.5 to 7.1 illnesses per 1000 people (Thomas et al., 2006), with 1,145 laboratory-confirmed cases of Salmonella spp. reported in the province of British Columbia in 2010 (Public Health Agency of Canada, 2011b). The reservoirs of Salmonella spp. include a wide range of domestic and wild animals, and humans; transmission occurs through ingestion of food derived from infected animals, food contaminated with animal or human feces, or contact with infected animals, people, and their environment (Heymann, 2008). Numerous serotypes are pathogenic to both animals and humans, with 2,500 serotypes of Salmonella identified worldwide to date (Heymann, 2008). Information on laboratory isolates of Salmonella is systematically collected by human health laboratories; analysis of such surveillance data is conducted by public health authorities at local, regional, national and international levels, in the hopes of identifying outbreaks or emerging sources of infection as early as possible and implementing proper control measures to limit the number of infections. There are many surveillance efforts focused on human Salmonella serotypes, including the Public Health Agency of Canada’s National Enteric Surveillance Program (Public Health Agency of Canada, 2012b), the Center for Disease Control’s Laboratory145 based Enteric Disease Surveillance in the United States (Centers for Disease Control and Prevention, 2012), and the Enter-net International Surveillance Network in Europe (European Centre for Disease Prevention and Control, 2012). Further subtyping of Salmonella serotypes, such as phage typing (PT) or pulse-field gel electrophoresis (PFGE) can be done on select isolates and common serotypes to detect matching subtypes across jurisdictions and help identify cross-jurisdictional outbreaks (see for e.g. (Centers for Disease Control and Prevention, 2010)). Information on Salmonella PFGE subtypes in human samples is shared internationally through networks such as PulseNet (PulseNet International, 2012). There are now also initiatives that examine Salmonella surveillance data from a number of sectors, namely human, animal and food. The World Health Organization’s Global Foodborne Infections Network (GFN) (formerly called Global Salm-Surv) (Word Health Organization, 2012b) collects information on global Salmonella serotypes from both human and non-human sources, and while not timely enough for outbreak detection, this database has been useful for hypothesis generation and international collaborations to control foodborne diseases (Galanis et al., 2006). An international investigation of a human outbreak of S. Typhimurium in Norway, Denmark and Sweden used national animal and meat Salmonella surveillance databases to supplement human surveillance data and identify Danish pork as the source of infection (Bruun et al., 2009). A Canadian surveillance initiative, C-EnterNet, collects and analyzes Salmonella data from the human, animal, environment and food sectors in two pilot sites, with the purpose of providing a more reliable assessment of risks posed by enteric pathogens to Canadian communities (Public Health Agency of Canada, 2012a). The Canadian Integrated Program for Antimicrobial Resistance Surveillance, CIPARS, (Public Health Agency of Canada, 146 2011a) and the Danish Programme for surveillance of antimicrobial consumption and resistance in bacteria from animals, food and humans, DANMAP, (DANMAP, 2012) are two other surveillance initiatives that examine Salmonella surveillance data across sectors. All of these surveillance initiatives have largely focused on identifying trends in the data and targeting and evaluating measures to reduce Salmonella incidence and/or antimicrobial resistance in Salmonella; their primary focus has not been on outbreak detection. There are now numerous methods available for analyzing laboratory surveillance data for detection of statistically significant clusters for further investigation (Sonesson and Bock, 2003). The use of such methods to identify clusters of Salmonella in people has become part of routine public health laboratory surveillance, and the use of these methods has been suggested in animal surveillance (Hohle et al., 2007). Current methods used in surveillance to detect signals in time series include time-series analysis, regression analysis, scan statistics, cumulative sums (Hohle et al., 2007), and generalized additive models or GAMs (see Chapter 4 in this thesis). Recently, there have been successful studies that have used univariate surveillance algorithms for detection of signals of Salmonella in animals (Danan et al., 2011, Kosmider et al., 2006), however, these approaches did not explicitly compare signals in animal and human data. For example, while Danan et al. observed that a succession of statistical signals in an emerging Salmonella serotype, S. I 4,12:i:-, in the agro-food chain coincided with an increased number of human cases in France (Danan et al., 2011), this association was not detailed further. We therefore wanted to assess whether use of such algorithms would identify signals in animals and/or food that would correlate with signals in human Salmonella data, or, in other words, whether such algorithms could identify “cross-sectoral” signals across the animal, food and 147 human sectors. We looked to identify statistically significant signals in each sector (not only the agro-food chain), the time between the signals in the different sectors, and whether additional subtype information in correlated signals would further support an epidemiological link between the signals in the sectors. In British Columbia, Canada, the Integrated Salmonella (IS) surveillance program collects laboratory Salmonella surveillance data from three sectors - humans, animals and food; data are analyzed on a regular basis by the IS epidemiology working group (IS WG). The IS WG members consist of public health and animal health practitioners from provincial and federal levels. IS WG analyses consist of a bi-monthly qualitative review of the data from each sector (i.e. monthly counts by animal species, commodity, serotype, PFGE/PT patterns), and across the three sectors looking for matching strains, clusters (increased number of isolates above expected based on historical data), and trends over time (Galanis et al., 2012). Because these analyses require inspection of a large number of descriptive data, they are very laborious and time-consuming in nature. We therefore wanted to investigate whether surveillance algorithms could be used as an automatic alerting system to aid the IS WG in their analyses of the data. The purpose of our study was to assess whether less resource-intensive univariate surveillance algorithms were appropriate for identifying clusters in the IS data, validate if the algorithms identified the same cross-sectoral clusters as those that were identified by the IS WG (assumed to be the current gold standard), and if they identified additional cross-sectoral clusters that could be relevant for public health. 148 In order to answer these questions, we examined time series of Salmonella serovars in the IS database through: 1) examination of the assumptions and limitations of each sector’s data; 2) comparison of serovars identified across the human, food, and animal sectors; 3) identification statistically significant signals (i.e. weeks with number of isolates above expected) within the laboratory isolates across sectors in 2010 using univariate surveillance algorithms; 4) identification of events relevant to public health through outbreak and cluster investigations conducted by the IS WG in 2010 for each serovar; 5) relation of statistically significant signals across sectors to IS WG outbreak/cluster investigations; and 6) examination of the statistically significant cross-sectoral clusters in terms of additional subtype data and human exposure data to comment on potential public health relevance. We hypothesized that the data, despite limitations, would be amenable to analysis using the less resource-intensive univariate surveillance algorithms for individual serotype time series, that there would be matching serotypes across the three different sectors (animal, food, human), and that there would be agreement between the statistically significant signals identified using the univariate surveillance algorithm and the IS WG investigations identified using descriptive analyses. 149 5.2 Methods Salmonella serovars from three sectors, namely human, animal, and food (specifically meat), isolated between January 2008 and December 2010 in the province of British Columbia (BC), Canada, were used in this study. The data in the BC IS database came from three different sources: 1) human laboratory diagnostic data from the BC Centre for Disease Control Public Health and Microbiology Reference Laboratory, the provincial public health reference laboratory, 2) animal laboratory diagnostic data from the BC Ministry of Agriculture Animal Health Centre, the provincial animal health diagnostic laboratory, and 3) food testing data from the Canadian Integrated Program on Antibiotic Resistance Surveillance (CIPARS), operated by the Public Health Agency of Canada (Public Health Agency of Canada, 2011a), a federal food sampling and testing program. In order to look for commonalities across the data sources, the isolates needed to be matched. Salmonella serotypes reported in the databases from all sources were re-named where needed based on the most recent CDC Salmonella naming conventions (Brenner et al., 2000, Centers for Disease Control (CDC), 2007). Serotypes with letter and number designations (e.g. “4,5:12:i:-“) were matched as follows: serotypes with the same number/letter combinations in the same order were considered matches, underscores and dashes were considered the same, terminal blanks and terminal sections identified with underscores or dashes were considered the same (e.g. “4:5:12:i:-“ and “4:5:12:i“), numbers in brackets were considered the same as those not in brackets (e.g. “” vs. “5”), “Rough O” and “RO” and “OR” were considered the same, leading “O” was not required for a match (e.g. “O4:5:12:i:-“ and “4:5:12:i:-“), and leading “SS I enterica” was not required for a match (e.g. “SS I enterica 150 O4:5:12:i:-” and “O4:5:12:i:-”). Since S. Typhimurium was not distinguished into S. Typhimurium var. Copenhagen in humans, the S. Typhimurium var. Copenhagen serotype was combined with S. Typhimurium for our analyses. 5.2.1 Human Data: Provincial Public Health Reference Laboratory The human Salmonella isolates used in this study were population-based diagnostic samples; in the province of BC Salmonella isolates are sent to the provincial laboratory for further differentiation and sub-typing, therefore all human Salmonella isolates from the province should be captured in this dataset. Subtyping by PFGE was done for all human isolates of Salmonella, and PT was done on all serotypes received in the first 15 days of each month. Since human exposure information was not contained within the BC IS database, each human isolate from 2008-2010 was linked to information in the integrated Public Health Information System (iPHIS) that records travel exposure information from case interviews in order to separate cases by travel status (i.e. travel history, no travel history, and unknown travel history). Cases that were successfully linked with iPHIS exposure information and, based on the exposure information did not have any relevant travel history, were classified as “endemic” cases. The cases were then separated into separate weekly time series for “all human cases” and “endemic human cases” by serotype. 151 5.2.2 Animal Data: Provincial Animal Health Diagnostic Laboratory Submissions with a laboratory diagnosis of Salmonella for all animal species were obtained. The provincial laboratory does not receive samples from all animal species or from all areas in the province, and therefore cannot be considered to be a source of population-based data. While the Canadian Food Inspection Agency (CFIA) also conducts laboratory surveillance of animals in abattoirs, they do not have clinical isolates involving food animals, nor were their data available to the IS WG for analysis, hence their data were not included in this study. Submissions to the provincial laboratory largely come from the agricultural sector, with only limited submissions from other animals such as companion animals and wildlife. Largely due to these biases, the animal species included in our analyses were limited to chicken, cattle, swine, and turkeys. The remaining species in the database, excluded also due to small counts or insufficient information on the actual species of the animal, were: cats, dogs, horses, domestic ducks/geese, reptiles, pigeons, wildlife species, exotic/zoo species, and unspecified species; these excluded species accounted for 10.1% of the animal isolates in the database. The animal isolates were stratified by type of submission where available: diagnostic samples (disease cases and investigations), monitoring samples (ongoing, routine animal health programs, available for chicken only), and targeted project samples; targeted project samples were excluded from further analysis due to their intermittent sampling protocols. Samples submitted for unknown reasons (i.e. not identified as diagnostic, monitoring, or project) were included (16% of all chicken samples). Monitoring samples included environmental (e.g. chicken ‘fluff’) samples from healthy flocks. The cases were then stratified into weekly time series by species and serotype. For chicken, two time series were created for each serotype: 152 “all chicken” and “diagnostic chicken”, where “all chicken” included diagnostic, monitoring, and samples submitted for unknown reasons, and are also referred to as samples from “live chicken and their environment”. 5.2.3 Food Data: Canadian Integrated Program on Antibiotic Resistance Surveillance Data on Salmonella identified in retail meat samples obtained at the point of sale (i.e. not at a wholesaler or abattoir) in the province of BC were used. Retail sampling occurred approximately every two weeks throughout the year, with 8 samples of pork and 8 samples of chicken collected in two regions of the province selected through a population-weighted random sampling strategy in a sampling week, for a total of 32 samples. The meat samples were stratified by meat type (chicken and pork), and were then further stratified into separate weekly time series by serotype. 5.2.4 Descriptive Analyses For all three sectors the total number of isolates, the number of different serotypes, and the proportions of the top three serotypes were calculated. For animal data these were stratified by species, and for chicken additionally stratified by diagnostic and all (live chicken diagnostic and environmental monitoring) cases. The human data were further stratified by travel exposure information, into “all human” cases and “endemic human” cases. Meat data were stratified by type of meat: chicken or pork. Serotype proportions across sectors were presented for the serotypes that spanned at least two sectors. Proportions for S. Enteritidis, S. Typhimurium, S. Heidelberg, S. Kentucky, and all other serotypes together were compared 153 across sectors using the Chi-square test. Descriptive analyses were conducted using SPSS for Windows, Rel. 17.0.0 (Chicago: SPSS Inc.). 5.2.5 Statistical Signals in Individual Time Series We used three different univariate surveillance algorithms based on Danan et al (Danan et al., 2011), who analyzed Salmonella isolates from various agricultural animal species to identify clusters. We chose these methods because Danan et al. were able to identify unusual agrofood chain contamination as well as the emergence of a serotype in the agro-food chain that correlated with an emergence of the same serotype in humans (Danan et al., 2011). While Danan et al. (Danan et al., 2011) combined all the weekly isolates for a particular Salmonella serotype for all animal species together, we created weekly time series for each serotype for each animal species and meat types separately; all time series with at least two submissions in the three-year time interval were included in our analyses. The three statistical algorithms used to detect clusters in the individual time series (within, not across, individual sectors) were: 1) Farrington method (also used by Kosmider et al. (Kosmider et al., 2011) for detection of Salmonella clusters in animal data), 2) a Bayesian algorithm, and 3) the Robert Koch Institute (RKI) algorithm (all in Hohle et al. (Hohle et al., 2007)). The three algorithms were used in order to increase the specificity of the signals in any particular week (i.e. all three algorithms have to agree a signal is present), while allowing for increased sensitivity in detecting smaller but more sustained increases (i.e. a signal identified in only one algorithm but in two consecutive weeks). Since Salmonella infections are often found to be seasonal in both humans and animals, these algorithms account for seasonality using reference values. Reference values are a moving 154 window of a pre-specified subset of past counts, which are then used for constructing the expected values, thereby directly accounting for seasonality. All three algorithms assume that the past counts (reference values) follow a Poisson distribution. The Farrington calculates an expected value by fitting a regression model to the past counts, followed by a creation of confidence intervals (using a transformed Normal distribution) and a threshold to allow for a statistical comparison of the observed count to the expected value. The Bayes algorithm creates a posterior (probability) distribution by combining a Gamma distribution with the Poisson distribution from past counts, and uses a negative binomial distribution to determine the threshold. The RKI algorithm calculates an expected value using the mean of the reference distribution, calculates a Poisson confidence interval around that estimate, and compares whether the current value is within the interval. The years 2008-2009 were used to create the baseline (reference values), with statistical signals evaluated for weeks in 2010. All analyses were done using R, a freely available statistical program (WU Wien Institute for Statistics and Mathematics, 2012), using the ‘surveillance’ package in R (Hohle et al., 2007)). The following assumptions were made for analyses of the individual time series using these three algorithms: Salmonella isolates are independent, samples are submitted at a constant rate (hence the denominator can be considered constant over the time period and does not need to be included), and isolation counts follow Poisson distributions. We acknowledge that violations of these assumptions by our data would affect our results. The data from each of the three sectors used in this study (human, animal, food) may violate our assumptions for different reasons. The data coming from the human sector were least likely to violate these 155 assumptions: the data are population-based, likely submitted at a constant rate, and all isolates were diagnostic cases that can be considered independent as each isolate likely represents only one case. The only exception to the independence assumption would be repeat testing for an individual, however, this is not likely to be an issue for the laboratory data as they are routinely checked for repeat isolates that are not subsequently reported to the IS database, and even less of an issue for human cases linked with iPHIS data (i.e. domestic cases), since iPHIS data are all case-based (rather than sample-based). The data coming from the meat sector are based on a statistical and convenience sampling strategy, therefore the isolates were independent by design, and while some changes in sampling occurred throughout 2008-9, we assumed a constant submission rate. The situation is more complex for animal data: agricultural animal species are part of species-specific agricultural production systems (e.g. chicken for meat or egg-production, cattle for milk or meat production), that differ widely in the reasons samples are sent to the laboratory. The data coming from the agricultural sector are not population-based (a biased fraction of samples are submitted the provincial laboratory), and are susceptible to differential testing due to seasonality and other changes within their particular production system affecting submission rates. Further, isolates may not always be independent, since one isolate can represent a group of animals such as a flock or herd (or conversely, a number of samples from one group can be sent at one time), isolates coming from animals and their environment can be sent in at the same time (i.e. chicken monitoring samples), and isolates from the same barn or animal can be sent sequentially until infection is no longer found in the samples. 156 5.2.6 Cross-Sectoral Investigations and Signals Information on investigations conducted and outbreaks identified by the IS WG during 2010 was obtained from IS WG meeting minutes and through supplemental interviews with the IS WG members. The IS WG meeting minutes recorded information from the group’s bi-monthly meetings to review data from each sector looking for matching strains, clusters, and trends. These reviews consisted of examining a number of tables and graphs in Microsoft Excel. The tables included monthly counts of serotypes for each sector (and species for animals and meat) for the previous three months and previous years (grouped together), looking for matching serotypes spanning the sectors. Monthly graphs for the more common serotypes (e.g. S. Enteritis, S. Heidelberg, S. Typhimirium) were examined for patterns (e.g. seasonality), clusters (increases above expected) and trends across sectors. When available, subtype information (PFGE/PT patterns) were also examined to identify matching strains. No statistical tests were used to identify patterns, trends or clusters. The information extracted from the minutes was: date investigation started, reason investigation started, sector(s) involved, the type of investigation, and details of the results of the investigation. Investigation type 1 was an identified situation (e.g. an increasing trend or cluster), with no further action taken. Investigation type 2 was an identified situation where an investigation was launched. Investigation type 3 was where an outbreak investigation was launched through interviews of the human cases involved with a specific outbreak investigation questionnaire. Cross-sectoral investigations were defined as those where the group noted an increase or another pattern of interest in least two of the three sectors. 157 Statistically significant signals across the three sectors were identified from the results of the univariate algorithm analyses conducted in each sector for each serotype. We considered as statistically significant weeks in 2010 with the following criteria: 1) Weeks with statistically significant signals identified by all three algorithms (for greater specificity), or 2) Weeks with statistically significant signals (using one or more algorithm) persisting for two or more consecutive weeks in humans and animals and either consecutive or alternating weeks for meat samples (due to their bi-weekly sampling strategy) Serotypes with a significant week in more than one sector (animal, food, human) were selected for further examination. For each serotype with a cross-sectoral signal, all available information on the subtypes by PT or PFGE, as well as travel information for human cases was extracted from the IS database. For examination of cross-correlations, “all chicken” time series were chosen over “diagnostic chicken” time series, since subclinical infection with Salmonella is common in many animals (Humphrey et al., 1998) and we assumed that inclusion of these asymptomatic isolates should correlate better with human risk. Graphs showing the time series of isolates in 2010 from each sector, dates of investigations by the IS WG, statistically significant signals, as well as associated subtype information were used to examine the relation of the signals to each other. 158 5.3 Results 5.3.1 Descriptive Analyses and Statistically Significant Signals in Separate Sectors 184.108.40.206 Human Over the three years there were a total of 3,335 isolates, of which 65% (2,178/3,335) were linked to exposure information: 1,403 were classified as endemic (domestic) cases, 721 as travel-related, and 54 could not be classified as either. There were 150 different serotypes for human cases overall, while for endemic cases there were 90. For all human samples the top serotype was S. Enteritidis (1,477/3,335, 44%), followed by S. Typhimurium (301/3,335, 9%), and S. Typhi (158/3,335, 5%). For endemic samples the top serotype was S. Enteritidis (693/1,403, 49%), followed by S. Typhimurium (167/1,403, 12%), and S. Heidelberg (89/1,403, 6%). In 2010, the ‘test year’, 68% (67/99) of “all human” serotype time series and 62% (33/53) of “endemic human” serotype time series had at least one isolate and were therefore tested for signals. For the 67 “all human” time series, 66 had a signal using the Farrington algorithm, 63 using the Bayes, and seven using the RKI algorithm. For the 33 “endemic human” time series, 32 had a signal using the Farrington algorithm, 30 using the Bayes, and four using the RKI algorithm. 220.127.116.11 Animal Over the three years there were a total of 750 positive animal Salmonella samples, in 68 different serotypes. There were 595 positive samples in chicken, 40 in cattle, 23 in swine, and 159 16 in turkeys. The top serotype in “all chicken” (live chicken and their environment) was S. Enteritidis (269/595, 45%), followed by S. Kentucky (168/595, 28%), and S. Heidelberg (43/595, 7%). Eighty-four percent (497/595) of all chicken cases that had information on type of sample, 94 were identified as diagnostic cases, 403 as monitoring cases. The top diagnostic serotype was S. Enteritidis (48/94, 51%), followed by S. Kentucky (22/94, 23%), S. Heidelberg (5/94, 5%) and S. I 4,,12:i:- (5/94, 5%). The top monitoring serotype was S. Enteritidis (165/403, 41%), followed by S. Kentucky (119/403, 30%), and S. Heidelberg (38/403, 9%). The top serotype in cattle was S. Typhimurium (32/40, 80%), followed by S. Dublin (6/40, 15%) and S. Worthington (1/40, 3%). There were 78% (31/40) of cattle cases identified as diagnostic cases. The top diagnostic serotype was S. Typhimurium (27/31, 87%), followed by S. Dublin (3/31, 10%), and S. Worthington (1/31, 3%). In swine the top serotype was S. Typhimurium (9/23, 39%), followed by S. Derby (4/23, 17%), and S. Enteritidis (3/23, 13%). There were 87% (497/595) of swine cases identified as diagnostic cases. The top diagnostic serotype was S. Typhimurium (9/20, 45%), followed by S. Enteritidis (3/20, 15%), S. Derby (2/20, 10%), and S. Worthington (2/20, 10%). The top serotype in turkeys was S. Schwarzengrund (5/16, 31%), followed by S. Worthington (4/16, 25%), and S. Hadar (4/16, 25%). There were 94% (15/16) of turkey cases identified as diagnostic cases. The top diagnostic serotype was S. Schwarzengrund (5/15, 33%), followed by S. Worthington (4/15, 27%), and S. Hadar (3/15, 20%). In the ‘test year’ 2010, there were 31 time series created for serotypes with at least one positive isolate tested for signals: 15 “all chicken”, 9 “diagnostic chicken”, 3 cattle, 2 swine, and 2 turkey. For each of the 31 time series, there was at least one corresponding signal in the 160 Bayes (total number of signals: 62) and Farrington (total number of signals: 76) algorithms. There were 12 signals in four time series using the RKI algorithm. 18.104.22.168 Meat Over the three years there were a total of 169 positive food (meat) samples, in 20 different serotypes. There were 162 positive samples in chicken and 7 in pork. The top five serotypes in chicken meat were S. Enteritidis (68/162, 42%), followed by S. Kentucky (41/162 25%), S. Hadar (14/162, 9%), S. Heidelberg (13/162, 8%), and S. Typhimurium (5/162, 3%). In pork, all 7 samples were of different serotypes. In the ‘test year’ 2010, 6 of 11 serovars, all only in chicken, had positive isolates, and all 6 had at least one signal using the Farrington algorithm (total number of signals: 21), five had signals using the Bayes algorithm (total number of signals: 13), and there were no signals using the RKI algorithm. 5.3.2 Cross-Sectoral Analyses: Comparison of Proportions, Investigations and Statistically Significant Signals across Sectors Differences in proportions of serotypes seen in the various sectors are compared for S. Enteritidis, S. Heidelberg, S. Kentucky, and S. Typhimurium in Figure 5.1. Figure 5.1 shows that the chicken meat samples and samples from live chicken and their environment have similar proportions of S. Enteritidis (42% vs. 45%), S. Heidelberg (8% vs. 7%), S. Typhimurium (3% vs. 1%), and S. Kentucky (25% vs. 28%). The figure also shows that human samples have similar proportions of S. Enteritidis (49%), and S. Heidelberg (6%), with higher proportions of S. Typhimurium (12%), and lower proportions of S. Kentucky (0.1%). There was a high proportion 161 of S. Typhimurium in cattle (63%), and to a lesser extent swine (13%) (Figure 5.1). The proportions were significantly different between all pair-wise combinations of sectors at the p<0.001 level, except for chicken meat, and live chicken and their environment (Chi-square: 3.15, p=0.533). When comparing humans to all other sectors, the lowest Chi-square value was between humans and chicken meat (Chi-square: 362.54, p<0.001). 5.3.3 Integrated Surveillance Working Group Investigations The IS WG met five times in 2010: in March (week 10), June (week 23), July (week 30), September (week 37) and November (week 45). The group investigated four serotypes across at least two sectors: S. Enteritidis, S. Heidelberg, S. Typhimurium, and S. I 4,,12:i:- (Table 5.1). Two of these serotypes, S. Enteritidis and S. Heidelberg, were identified by the IS WG as present in all three sectors. 22.214.171.124 S. Enteritidis The dates of the five IS WG investigations in 2010 are shown in Figure 5.2 as stars. The IS WG identified that there were three S. Enteritidis subtypes present in all three sectors in 2010. The group identified an emergence of PT 51 in animals (live chicken) and chicken meat in March (week 10), becoming the dominant PT in animals. The group continued to see this serotype in animals, food, and humans throughout the year, however near the end of the year (November, week 45) they concluded that there was no apparent increase in PT 51 in humans. In June (week 23) the IS WG identified PT 13a in animals and humans, and in all three sectors in July (week 30) and November. In July and November the IS WG identified PT 8 as the third PT that crossed all three sectors. 162 An ongoing investigation into an increase in S. Enteritidis in people since 2008 in the province by public health authorities (Galanis et al., 2012) meant the IS WG did not initiate separate investigations into S. Enteritidis based on their review of data. The results of the larger investigation are not yet published; however, the main hypothesis for that investigation was that the source of the illness in humans was due to eggs, with the contribution of chicken meat unknown (Taylor, 2011). 126.96.36.199 Other Serotypes The IS WG started a S. Heidelberg outbreak investigation (investigation Type III) at the beginning of 2010 based on similar PFGE patterns in late 2009 and early 2010. The IS WG noticed another increase in S. Heidelberg in humans in the fall 2010, however they did not find any common subtypes to warrant further investigation. Two other small (i.e. of Type I or II) investigations were conducted by the group: S. I 4,,12:i:- and S. Typhimurium (Table 5.1). The investigation into S. I 4,,12:i:- was initiated because there was a seasonal trend noted in human isolates. The S. Typhimurium investigation was initiated because there was a cluster identified in one animal sector (swine) and a seasonal trend in another animal sector (cattle). Neither of these investigations progressed to a Type III investigation. 163 5.3.4 Statistically Significant Signals across Sectors The statistical algorithm analyses identified three serotypes that had statistically significant signals in at least two sectors in 2010: S. Enteritidis, S. Hadar, and S. Kentucky (Table 5.2). Two of these serotypes, S. Enteritidis and S. Kentucky, had statistically significant signals in all three sectors. 188.8.131.52 S. Enteritidis S. Enteritidis had statistically significant signals in: “all chicken” (weeks 1, 12, 24, and 44), “diagnostic chicken” (weeks 29, 31, 39, 40, 42, 44, and 47), chicken meat (week 3), “all humans” (weeks 6, 47, 48, 49, and 50), and ‘”endemic humans” (weeks 6, 10, 14, 47, and 48). We investigated the signals across the sectors in two time periods (Figure 5.2). We focused our cross-sectoral analyses on “all chicken” (see Methods), chicken meat and “endemic humans”. We chose to use the endemic human time series since that meant we did not have to include human signals with high proportions of travel-related exposures. When we investigated the signals in “all humans” in the weeks that were not also identified in the “endemic human” time series, we found that the percent travel in week 49 was 43% and in week 40 it was 40% (out of cases with linked exposures). The first time period shows overlap of subtypes (by PT) across sectors for PT 8, PT 13a, and PT 13: PT 8: The first signal in the time period was an animal (chicken) signal (shown by “A” arrow in Figure 5.2) in week 1, and included 43% PT 8 (3/7 typed isolates), while the animal signal in week 12 (n=7) (shown by the second “A” arrow in Figure 5.2) consisted 164 of 14% PT 8 (1/7 typed isolates). All three of the subsequent human signals (shown by the “H” arrows in Figure 2) included at least one PT 8 isolate: week 6 (n=7), and consisted of 40% of PT 8 (2/5 typed isolates), week 9-10 (n=12) and consisted of 55% PT 8 (6/11 typed isolates), and week 14 (n=9) consisted of 13% PT 8 (1/8 typed isolates). PT 13a: The animal (chicken) signal in week 1 and included 14% PT 13a of (1/7 typed isolates). The food (chicken) signal (weeks 1 and 3) (shown by “F” arrow in Figure 5.2) consisted of 67% PT13a (4/6 typed isolates). All three of the subsequent human signals included at least one PT 13a isolate: week 6 (n=7) consisted of 40% PT 13a (2/5 typed isolates), weeks 9-10 (n=12) consisted of 9% PT 13a (1/ 11 typed isolates), and week 14 (n=9) consisted of 13% PT 13a (1/8 typed isolates). PT 13: The animal (chicken) signal in week 12 (n=7) consisted of 29% PT 13 (2/7 typed isolates). The first two human signals in this time period with a PT 13 were prior to the animal signal, in week 6 (n=7), consisting of 20% PT 13 (1/5 typed isolates), with the second signal in weeks 9-10 (n=12) consisting of 18% PT 13 (2/11 typed isolates), and the third in week 14 (n=9) consisting 75% PT 13 (6/8 typed isolates). The second time period did not show any overlap of subtypes (by PT) across sectors. The animal (chicken) signal in the third time period in week 24 (n=5) consisted of 4 PT 51, and one PT 23. The animal signal in week 44 (n=8) consisted of 5 PT 51, 3 atypical PTs, and one PT13. The human signal in week 47 (n=10) had no PT subtype information. The human signal in week 48 (n=11) consisted of 7 PT8 and one PT13a. 165 184.108.40.206 Other Serotypes The signals in S. Hadar were in: chicken meat (week 11) and 36 weeks (over 8 months) later in “all humans” (weeks 47, 48, and 49). There were no PT results for these isolates. Of the five human isolates in the cluster, three had no exposure information and two were related to travel outside of the province. The signals in S. Kentucky were in: “all chicken” (weeks 35 and 43), “diagnostic chicken” (weeks 35, 43, and 44), chicken meat (weeks 12, 41, and 50), and “all humans” (week 38). There were no PT results for these isolates. Of the five human isolates in the cluster in week 38, three had no exposure information and two were related to travel outside of the province. 5.3.5 Comparison of Working Group Investigations with Statistically Significant Algorithm Signals Across Sectors In 2010, the IS WG investigated four serotypes, S. Enteritidis, S. Heidelberg, S. Typhimurium, and S. I 4,,12:i:- (Table 5.1), while the algorithms identified three serotypes that had statistically significant signals across sectors: S. Enteritidis, S. Hadar, and S. Kentucky (Table 5.2). The only serotype that the IS WG and the statistical algorithms both identified was S. Enteritidis. However, when we examined this serotype further by PT, the IS WG identified three PTs of interest (PT 8, PT 13a, PT 51), while the statistical analyses identified two of the same three (PT 8, PT 13a) and one additional one (PT 13). The IS WG records suggest that PT 51 was identified by the group because of a gradual increase of the subtype, particularly among animals. This sort of trend is not likely to be picked up using our methods since this emergence was gradual in the animal sector and not as pronounced in the other sectors. 166 The S. I 4,,12:i:- serotype was identified by the IS WG due to differences in seasonality patterns among humans and animals. While a week with a significant signal in S. I 4,,12:i:was identified in “endemic humans”, seasonal trends in animals were not flagged and no crosssectoral signal for this serotype was generated. S. Typhimurium was identified by the IS WG based on seasonality in animals and an increase in humans. While the increase in humans was also identified using the automatic algorithms, there were no signals in the animal or meat sectors, and no cross-sectoral signal was generated. The two serotypes identified by the algorithms that were not investigated by the IS WG were unlikely to have warranted investigation. The algorithms found a signal in S. Hadar in food early in 2010 and in humans late in 2010; however, the signals were 36 weeks apart, the number of human cases was low (n=5), and two of the cases had relevant travel histories outside of the province. The S. Kentucky signals were identified across sectors in chicken, chicken meat and humans, and while the time between one of the animal signals and the human signal was only 3 weeks, and the time between one of the meat signals and the human signal was 26 weeks, the number of cases in humans was again quite low (n=5), and two of these cases had relevant travel histories outside the province. 5.4 Discussion We were able to generate cross-sectoral signals that we could compare to investigations conducted by the IS WG and comment on their public health relevance. The limitations we identified in using univariate algorithms for analyzing Salmonella surveillance data were most prominent for the animal sector. Although we found 18 serotypes in at least two different 167 sectors, there was more agreement between the proportions of serotypes found in live chickens and chicken meat than between humans and any other sector. There was little agreement between the investigations conducted by the IS WG and the cross-sectoral signals generated by the univariate surveillance algorithms. It was difficult to assess the public health significance of the cross-sectoral signals, since there were no conclusive outbreak investigations during the time period that linked human cases to either meat or live animals. 5.4.1 Cross-Sectoral Serotypes We found 18 Salmonella serotypes present in at least two sectors (animal, food and human), and focused on these since we assumed these had the highest likelihood of representing contamination in the food chain. The proportions of serotypes differed between humans and the animal and food sectors and among animal species (Figure 5.1). This was also found in a study that compared the Salmonella serotype distribution between agricultural animals at slaughter and humans, concluding that there was not a match between the two (Sarwari et al., 2001). One limitation of our study, as well as that conducted by Sarwari et al. (Sarwari et al., 2001), is that we focused only on agricultural animals and meat. While Salmonellae are considered to be zoonotic bacteria, they are also known to be transmitted to humans from other sources, such as other humans (i.e. person to person), other animals, such as household pets, and various contaminated fruits and vegetables (King, 2008). We did find similar proportions of Salmonella serotypes in chicken meat and in samples from live chickens and their environment for S. Enteritidis, S. Heidelberg, S. Kentucky, and S. Typhimurium (Figure 5.1), suggesting that they may be measuring similar populations. Galanis 168 et al. stated that fresh chicken meat consumed in BC is largely produced within the province (Galanis et al., 2012). While human samples had similar proportions of S. Enteritidis and S. Heidelberg to chicken, they had lower levels of S. Typhimurium, and much lower proportions of S. Kentucky (Figure 5.1). S. Typhimurium has been found in high proportions in cattle and swine in our study and by others in Canada (Guerin et al., 2005b). Several explanations have been proposed to account for such differences. For example, S. Kentucky may have a lower pathogenicity than other Salmonella serotypes (Sarwari et al., 2001), and the same may be true of other Salmonella serotypes present in high proportions in animals but not in humans (e.g. S. Dublin comprised of 15% of cattle isolates and was not found at all in humans in our study). Hald et al. (Hald et al., 2004) quantified the contribution of various food sources to human salmonellosis in Denmark comparing the number of Salmonella serotypes in humans to those predicted from the frequency the serotypes isolated from food sources. Their findings suggest that S. Enteritidis exhibits the best ability to survive food processing/cause disease of all serotypes, and S. Dublin one of the lowest. It seems that Salmonella serotypes should not all be treated equally: an increase in certain serotypes (e.g. S. Enteritidis) in animals and food may be more important for human disease than increases in other serotypes (e.g. S. Kentucky and S. Dublin). An alternative explanation might be that these other serotypes are not coming from food animals, and that there is another source of these serotypes in humans. A study using a rigorous ecological approach to assessing the epidemiology of Salmonella Typhimurium DT 104 isolates in human and animal populations concluded that while the two populations were ecologically connected, the S. Typhimurium DT 104 communities were distinguishable in their prevalence, linkage and diversity (Mather et al., 2012). Their findings 169 call into question any sort of link between agricultural animal and human isolates, especially a causal link, even when they are of the same subtype. As whole genome sequencing becomes more widely available and affordable, novel genotyping methods based on sequence-based identification will result in better resolution of subtypes (Fournier et al., 2007); better resolution of human and animal isolate subtypes will allow for further investigations into the overlap and origins of isolates in these populations. 5.4.2 Cross-Sectoral Signals: Working Group versus Algorithms In 2010, the IS WG investigated four different serotypes for possible links across the sectors. Using the univariate surveillance algorithms we were able to identify three serotypes with significant signals across sectors. The only serotype with an investigation by the IS WG and a statistically significant cross-sectoral signal was S. Enteritidis. However, when examining the signals within S. Enteritidis in detail, the algorithms only identified three of the four S. Enteritidis subtypes that were noted by the IS WG. It is possible that S. Heidelberg (not identified by the detection algorithms) would have been identified if data from late 2009 was used, since the S. Heidelberg investigation by the IS WG was initiated at the beginning of 2010 based on similar PFGE patterns in late 2009 and early 2010, and on the knowledge that the serotype was present in the other sectors in late 2009. Since the algorithms found statistically significant signals in humans in 2010, they could have identified signals in animals and/or food in late 2009. The two other serotypes (S. Hadar and S. Kentucky) identified only by the algorithms had a low number of ‘associated’ human cases (n=5), a number that likely would have been too low 170 to warrant an investigation. Additionally, when we examined the human exposures for these two serotypes, two of the five cases were associated with travel outside the province for each serotype. This suggests that, on their own, univariate surveillance algorithm analyses cannot replace current IS WG analyses. However, our lack of information on exactly what data the IS WG had at each of their meetings (i.e. what was the latest date they had complete data for), limits our comparisons: we could not assess whether the algorithms could have found the cross-correlations sooner than the IS WG. The statistical detection algorithms generated a large number of signals in individual sector time series; however, we found that focusing only on those weeks where all three algorithms generated a signal in a serotype and/or if there was a signal in at least one algorithm in consecutive weeks (as done in Danan et al (Danan et al., 2011)) appeared to be an effective way to identify potential signals to investigate. One advantage to using this methodology is that it is easy to increase the sensitivity of finding signals in data streams without re-running the analyses, for example, by decreasing the need for all three algorithms to signal in the same week to only two. Since the meat time series had the smallest number of isolates, it is possible that this data stream would benefit from such an increased sensitivity. Examining longer-term trends in serotype increases in animals and humans has led to initiation of mitigating measures that have reduced the level of the particular serotype in both animals and humans (Poirier et al., 2008). Guerin et al. (Guerin et al., 2005a) found similar multi-year trends and temporal clusters of S. Heidelberg from humans and chickens over a ten year period, as well as a 10-month temporal cluster in S. Typhimurium var. Copenhagen that 171 ended in chickens 9 months prior to the start of a human temporal cluster, suggesting their results point to a possible association between human illness and exposure to chicken products. The IS WG used trends and seasonality within and across sectors to identify potential areas for investigation. Two of the serotypes identified only by the IS WG (S. I 4,,12:i:- and S. Typhimurium) were identified using such trends. Examining the longer-term changing levels of PT distribution in Salmonella Enteritidis in BC has been instrumental in generating hypotheses for investigations of increases in human cases, as well as in identifying an emerging PT (PT 51) in animals, food, and humans (Galanis et al., 2012). Because the IS WG did not require a statistically significant signal in humans (or meat) associated with PT 51 in order to identify this emergence, they were able to look out for the emergence of PT 51 in humans before any crosscorrelated signals using automatic algorithms would have alerted them to its presence. Future integrated surveillance studies could try to examine the correlation between smoothed weekly or monthly counts, statistically significant linear and non-linear trends for identification of potential clusters for investigation., such as those done previously in animals and humans separately (Guerin et al., 2005a, Guerin et al., 2005b). 5.4.3 Discussion of the Data Sources and Methods Used Analysis of surveillance data using univariate surveillance algorithms, whether from human, or animal (or food sources if they are not obtained using a statistical sample as in our study), would benefit from denominator data such as the number of samples submitted for Salmonella testing (Guerin et al., 2005b). Having such denominator data would allow us to assess whether 172 samples are submitted at a constant rate, and therefore whether we are justified in assuming such a constant rate for all sectors. In order to identify statistically significant signals we assumed all isolates were independent. However, in animal data, multiple isolates from the same flock/herd are often submitted sequentially to determine whether the infection is still in the flock/herd, hence such samples cluster together. Other authors have aggregated samples from the same “epidemiological unit” within 30 days (Guerin et al., 2005b, Kosmider et al., 2011); unfortunately, we did not have the data (e.g. submitter information) to determine whether isolates were from the same epidemiological unit. The reasons for testing for Salmonella vary widely within animal data, and can include samples submitted due to clinical illness (similar to human data), as well as routine surveillance and monitoring to comply with various legislative requirements (e.g. table egg monitoring in layer chicken). These different reasons for submission also vary by species and industry and potentially bias the estimates of Salmonella burden in the population they come from. We tried to minimize these potential biases by focusing mostly on clinical animal samples to closely approximate the human data for all species except chicken. For chicken we stratified the analyses into diagnostic and all types of samples, the latter including monitoring samples from chickens and their environment. Indeed, we found that the two types of chicken time series produced different signals, and more work is needed to determine which time series is more amenable to such analyses, including examining the effect of different denominators and aggregation of isolates into epidemiological units. Animal data are also influenced by changing circumstance within the livestock industry, with large outbreaks affecting surveillance data 173 (Kosmider et al., 2011), since submission of animal samples for testing is driven largely by costs, while submission of human samples is driven primarily by awareness (see Chapter 4). Our results show that the three algorithms performed very differently, with Farrington and Bayes algorithms detecting more statistically significant weeks than the RKI algorithm, with the RKI algorithm identifying a statistically significant week in 17% of time series, while the Farrington identified one in 99% of time series, and the Bayes one in 94% of time series. This underscores the importance of choice of detection algorithm with the appropriate sensitivity and specificity, tailored to the investigation capacity within a surveillance system. Had we relaxed our requirements and included weeks with signals only in either the Farrington or the Bayes algorithms, we would have ended up evaluating a large number of cross-sectoral clusters in weeks with only one isolate. Due to the low sensitivity of the RKI algorithm, and our requirement for a statistical signal to be present in all three algorithms to be considered significant overall, the RKI algorithm drove the majority of our signals that we tested across sectors. However, using only the RKI would have eliminated signals that were significant in sequential weeks with the Farrington and/or Bayes. Notably, the Farrington algorithm allows the user to ignore low weekly counts; however, we did not use this option since it was not available for the other two algorithms. Use of this option would be preferable in future analyses, and may in fact render the Farrington algorithm as a happy medium of all three algorithms, allowing the Farrington to be used for identifying signals in individual weeks as well as subsequent weeks. A further limitation of the univariate surveillance algorithm analyses is that the time period available was three years, with only one year of data available for examination of signals. If 174 these methods were applied prospectively, as cross-sectoral signals are identified, information (e.g. subtyping, exposure information in humans, production system information from animals such as time to slaughter, spatial information on meat and human samples) could be gathered to help assess whether the cross-sectoral signals warrant further investigation, and therefore if such methods could be useful complementary tools for integrated Salmonella surveillance systems. Moreover, with more years of data, different time intervals could be used, such as monthly counts of isolates. Our algorithms may have been less appropriate for the more sparse serotype time series with few isolates over the time period, and larger time aggregations could be beneficial in such instances. The IS WG used monthly counts in their analyses, and identified more serotypes for cross-sectoral investigation than the (weekly count) algorithms. Larger time aggregation units (and associated higher counts in the time units) may lead to a better performance of surveillance algorithms due to more robust calculations of expected counts and confidence intervals. Kosmider et al. (Kosmider et al., 2006) successfully applied the Farrington algorithm to monthly Salmonella time series in animals over ten years, and statistically significant temporal clusters of monthly counts have been found in Salmonella in both humans (Guerin et al., 2005a) and animals (Guerin et al., 2005b) using time scan statistics. Danan et al. (Danan et al., 2011) used a different strategy to increase weekly counts in the serotype time series and improve the performance of the detection algorithms: they combined all of the data from different species together, and were still able to find statistically significant signals that identified unusual events linked to contamination in the agro-food chain that were confirmed upon investigation. 175 5.4.4 Public Health Significance of Cross-Sectoral Signals Causal links between the signals among the sectors, e.g. whether an animal signal was the ‘cause’ of a subsequent food or human signal, could not be evaluated using the univariate surveillance algorithm analyses. We were, however, able to examine subtyping and/or epidemiological information to see whether further work looking into possible relationships between the signals would have been warranted. If we work under the assumption that all the signals we found in animals identify a real increase in Salmonella in the animal population, we saw that this increase did not seem to translate into an increase in food and/or humans after the signal. Overall our results show that there is questionable value in looking at short term correlation of significant weekly signals in animal, food, and human data. Had we limited our analyses to only endemic human time series, the two serotypes identified only by the algorithms would not have yielded cross-sectoral signals at all. We propose that, when possible, analyses focused on identifying contamination in the food chain should use endemic human time series to reduce the probability of identifying clusters related to travel. Further work comparing the signals generated by all human time series to endemic human time series is necessary, for example by comparing the proportion of travel cases in the signals generated in the all human time series to the signals generated by the endemic human time series. Unfortunately, 35% of human cases could not be linked to exposure information using our linkage procedure, and were therefore excluded from the endemic time series. Since the inclusion of exposure information for humans (not currently in the IS database) would require additional time and resources, quantifying the relative costs and benefits of including such information before making such changes to a surveillance system would be important. 176 We chose to use the signals from the ‘all chicken’ time series which included monitoring samples from live chicken and their environment over ‘diagnostic chicken’ time series in our cross-correlation analyses based on our a priori assumption that ‘all chicken’ would provide a better predictor of risk to humans based on subclinical infections in chicken (see Methods). We did not have additional epidemiological information for our chicken isolates that could help determine which time series may be less affected by biases (e.g. which has more clusters of related samples from one epidemiological unit) and therefore more appropriate for future analyses using surveillance algorithms. We did find that the ‘diagnostic chicken’ time series, with a much smaller number of isolates, generated more statistically significant signals overall than the ‘all chicken’ time series. At first sight, Figure 5.2 offers a promising appearance of signals in animals (and food in one instance) preceding human infections of S. Enteriditis. However, upon closer inspection, the PTs within the signals did not convincingly match across sectors, except possibly in weeks 1-14 (Figure 5.2). We were unable to investigate the possible correlations between these signals further than subtypes (PT) and travel information for human cases, when available. We did not have a defined time period that we could use to determine when there could have been a risk to humans (or meat) following a signal in animals, and thus conducted our statistical analyses in each sector separately and then compared the results qualitatively. In general it is unclear what the best analytic methods using surveillance algorithms are for integrated surveillance programs. A recent review of research articles published between 1996 and 2007 linking animal and human health data found only 29 studies (6%) that attempted a quantitative linkage between the two (Scotch et al., 2009). The review showed that the most 177 successful linkages were for diseases that had a clear spatio-temporal component, such as West Nile virus. A pathogen such as Salmonella that is transmitted through a complex food production and distribution chain may only show a spatiotemporal association between human cases and the final products in the marketplace – e.g. meat on grocery store shelves. In order to quantitatively test whether there was correlation among the animal and human time series (e.g. specificity or positive predictive value of animal/food signals to predict human signals), we would need to determine what a reasonable time interval between the two signals (animal and human or animal and food or food and human) should be. To do this, we could either use the results of previous outbreak investigations where human cases were linked to known food/animal contamination events, or, we would need to know information that is currently lacking, such as the animal production type (e.g. egg production vs. meat production for chicken, milk vs. beef for cattle), age of the animal (to estimate time to slaughter or to determine whether sample is taken at beginning or end of a layer’s life), and other epidemiological information on the flock/herd. For example, the egg production and meat production industries in poultry are very different, from the age of the chickens involved to the shelf life and distribution of the products. Further, a positive S. Enteritidis test in egg production (layer chicken) diverts all subsequent eggs the flock produces to be pasteurized, effectively eliminating the risk from the market. Complicating this is that some table egg monitoring programs sample flocks near the end of their production cycle when the probability of finding a positive is expected to be greatest (Cox, 2010), therefore, a positive test could mean that there were months of contaminated eggs on the market prior to the test. Our dataset did not include tests of the final products (eggs), but only of layer chicken and their environment. 178 Another complication is that Salmonella contamination can occur during transport and slaughter, whereby previously Salmonella-free animals may result in Salmonella-contaminated meat cuts (Heyndrickx et al., 2002). Additionally, since salmonellosis has been associated with consumption of frozen products (Currie et al., 2005), it is even more difficult to estimate the time interval of human risk from frozen products. Finally, foods eaten in a jurisdiction are not necessarily from that jurisdiction, for chicken this is especially true for processed and frozen chicken products, which are imported from various world regions into the province. All of these complexities raise the issue of the actual specificity of an animal signal for an enteric agent such as Salmonella in relation to public health. A signal in animals could mean any of the following: a real signal of an event that is relevant to public health, a real signal that is relevant to animal health, an artefact of the surveillance data, or a statistical anomaly. Without detailed outbreak investigations that link human disease to specific food products and then to the specific animals that such products originated from, we cannot assess causal links between signals in animals, food, and humans. Notably, while we focused on the relevance of signals to public health, it is possible that animal signals such as those identified in our study may be relevant to animal health, as has been found in Danan et al. (Danan et al., 2011). In light all of the issues associated with animal data, correlations of public health significance may be more likely to be found between meat at point of sale (more so than at a wholesaler or abattoir) and human time series. In the case of meat, a positive Salmonella isolate is a clearer indication of human risk. In general, a signal in meat should have a higher specificity than a signal in animals in terms of predicting risk of disease in humans, since meat is much closer to humans in the food chain than animals. Our results seem to support this: the 179 only cross-sector clusters with matching subtypes in S. Enteritidis were in weeks 1-14, and this was the only time interval with a signal in meat (Figure 5.2). Despite these promising results, a recent Canadian study found that contaminated chicken meat did not determine seasonality in human salmonellosis cases, and suggested that human activities such as barbequing and gardening are more likely to be the driving factors (Ravel et al., 2010). 5.5 Conclusion We found little evidence that statistical signals generated using univarite surveillance algorithms on Salmonella serotype data from agricultural animals, meat, and humans correlated with investigations conducted by an integrated Salmonella surveillance expert working group in the same time interval. Analyses of animal data using univariate surveillance algorithms based on weekly data are not likely to provide public health with Salmonella early warning signals. It is likely that our inability to identify actionable alerts is due to the complexity of Salmonella dynamics within the food chain, specifically quantifying the risk to human health from a live animal with Salmonella, as well as the lack of specificity and quantity of data in certain sectors. Animal and food surveillance data of this type may be more amenable to generating hypotheses in epidemiological investigations and in helping evaluate programs by examining longer-term trends. 180 5.6 Tables and Figures Table 5.1 Investigations into serotypes that were present in at least two of the three sectors (human, animal and food) conducted by the BC Integrated Salmonella Surveillance Working Group in 2010. Serotype Enteritidis Subtype PT 51 All PT 11b PT 13a Wk* 10 23 Inv† II I II II PT 13a PT 8 30 PT 51 PT 13 All PT 51 All PT 13a PT 51 PT 8 PT 15a 37 45 Sector‡ H, A, F H A A Details PT 51 is increasing in animals (chicken) and food (chicken meat) Changing trend: increase in winter months, not seen in previous years Two isolates in cats An isolate in an exotic reptile II II Reason Subtype in 3 sectors Increase Rare subtype Public health investigation Subtype in 3 sectors Subtype in 3 sectors H, A, F H, A, F II Subtype in 3 sectors H, A, F Review of human exposure info in July revealed little. Review of human exposure info in July revealed little. PT used to be dominant in animals in 2008 and 2009. PT 51 has become dominant in animals instead of PT 8. Review of human exposure info in July revealed little. I I I II I I I I Subtype in 1 sector Decrease in July Subtype in 3 sectors Increase in August Subtype in 3 sectors Subtype in 3 sectors Subtype in 3 sectors Subtype in 2 sectors H H H, A, F H H, A, F H, A, F H, A, F H, A Associated with PFGE 0003 and 0007. PT 51 still dominant PT in animals Restaurant cluster (6 cases) identified by public health in August PT 51 has not increased in humans Associated with a cluster at a party, where common foods were cake, pasta salad, and a fruit platter; also in one chicken isolate. PT 19 10 III Increase in humans, H, A, F Standardized case follow-up of human cases in Jan-Feb, no common source subtype in 3 sectors found. Same PT found in animals and food in previous 6 months Heidelberg All 30 II Increases/decreases H No common PT or PFGE in humans All 37 II Increase in July H No common PT or PFGE in humans All 23 II Cluster in swine A Three isolates in swine Typhimurium All 23 I Seasonality in cattle A Seasonal increase in winter, likely due to beef cattle calving season (Jan-Mar) All 37 I Increase in July H No common PT or PFGE in humans I 4,,12:i:All 23 I Serotype in 2 sectors H, A Seasonal trend in humans (increase in winter), no such trend in animals *Wk: Week investigation started in 2010; †Inv: Type of Investigation: Type I: a potential situation is identified, with no further action taken; Type II: a situation was identified and an investigation was launched; Type III: a situation was identified and an outbreak investigation was launched; ‡Sectors: H: human, A: animal, F: food 181 Table 5.2 Statistically significant signals across sectors in 2010. Serotypes with at least one statistically significant week in 2010 by all three algorithms or in one algorithm in consecutive weeks are designated as “Signal”, with the rest designated as “No Signal”. Serotypes with significant signals across at least two of the three sectors (human, animal, and meat) as designated as “Signal” in the Cross-Sectoral Column. Serotype Human Animal Meat Cross-Sectoral S. I 4,,12:i:No Signal No Signal -* No Signal S. Agona No Signal No Signal No Signal S. Albany No Signal No Signal No Signal S. Braenderup Signal No Signal No Signal S. Derby No Signal No Signal No Signal S. Enteritidis Signal Signal Signal Signal S. Hadar Signal No Signal Signal Signal S. Heidelberg Signal No Signal No Signal No Signal S. Infantis No Signal No Signal No Signal S. Kentucky Signal Signal Signal Signal S. Mbandaka Signal No Signal No Signal S. Newport Signal No Signal No Signal S. Rissen No Signal No Signal No Signal S. Schwarzengrund No Signal No Signal No Signal No Signal S. Senftenberg No Signal No Signal No Signal S. Tennessee No Signal No Signal No Signal S. Typhimurium Signal No Signal No Signal No Signal S. Worthington No Signal No Signal No Signal *”-“: No time series for serotype included in analyses, since serotype have less than 2 submissions in 2010. 182 Figure 5.1 Comparison of serotypes across the sectors between 2008 and 2010. Humans with no travel history (endemic human cases) (N=1,403), chicken meat (N=162), live chicken and their environment (N=595), swine (diagnostic cases) (N=23) and cattle (diagnostic cases) (N=40). Differences in proportions between all sectors were significant at the p<0.001 level, except between meat and chicken (p=0.53). 183 Chicken all PT8: 43% Chicken meat Human endemic 0% 40% 55% 14% 13% 0% 0% -- 88% PT13a: 14% 67% 40% 9% 0% 13% 0% 0% -- 13% PT13: 0% 0% 20% 18% 29% 75% 0% 11% -- 0% A F H H A H A A H H 12 Number of isolates 10 8 6 4 2 0 1 I 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 II Week in 2010 1 2 3 4 5 Figure 5.2 Salmonella Enteritidis (SE) isolates in 2010 from animals (chicken), food (chicken meat) and humans (endemic human cases). Black bars are the weekly number of live chicken (and their environment) SE isolates, grey bars are the weekly number of chicken meat SE isolates, and the red line indicates the weekly number of endemic human isolates. Black “A” arrows indicate statistically significant signals in animals (chicken), grey “F” arrow indicates a statistically significant signal in food (chicken meat), and red “H” arrows indicate statistically significant signals in humans. The year is split into two time periods (I and II) to facilitate subtype analyses. Stars at the bottom of the graph indicate times when the IS WG reviewed data and initiated investigations. Time period I: The animal (chicken) signal in week 1 (n=7) (shown by “A” arrow) consisted of 3 PT8, 3 PT51, and 1 PT13a. Food (chicken) signal in week 3 was a combined signal with week 1 (n=3 in each week) (shown by “F” arrow) consisted of 4 PT13a, and 2 atypical PTs in the two weeks. The human signal in week 6 (n=7) (shown by arrow “H”) consisted of 2 PT13a, 2 PT 8, 1 PT13; the samples by PFGE were 5 SENXAI.0003, 1 SENXAI.0038, 1 SENXAI.0068. The second human signal was in week 9 (n=6) (shown by second arrow “H”), and was a combined signal with week 10 (n=6), and consisted of 6 PT8, 1 PT 13a, 1 PT 1, 2 PT 13, and 1 atypical PT; the samples by PFGE were 2 SENXAI.0003, 1 SENXAI.0068, 1 SENXAI.004, 2 SENXAI.0038. The animal signal in week 12 (n=7) (shown by the second “A” arrow) consisted of 4 PT 51, 2 PT13, and 1 PT8. The human signal in week 14 (n=9) (shown by the third “H” arrow) consisted of 6 PT 13, 1 PT13a, 1PT8; the samples by PFGE were 5 SENXAI.0038, 1 SENXAI.0003, 1 SENXAI.0007. Time period II: The animal signal in week 24 (n=5) (shown by the third “A” arrow) consisted of 4 PT 51, and 1 PT 23. The animal signal in week 44 (n=8) (shown by the fourth “A” arrow) consisted of 5 PT 51, 3 atypical PTs, and 1 PT13. The human signal in week 47 (n=10) (shown by the fourth “H” arrow) consisted of the following samples by PFGE: 7 SENXAI.0003, 1 SENXAI.0068, and 2 SENXAI.0007. The human signal in week 48 (n=11) (shown by the fifth and last “H” arrow) consisted of 7 PT8, 1 PT13a; the samples by PFGE were 5 SENXAI.0003, 5 SENXAI.0007, and 1 SENXAI.0006. 6 Conclusion The main goal of this thesis has been to improve our understanding of how animal data could be used for EZD surveillance. I have addressed this goal through a systematic literature review of current EZD surveillance systems, expert elicitations of EZD surveillance needs, and a critical examination of three pilot animal surveillance systems: a sentinel clinical pre-diagnostic system, a laboratory-based system, and an integrated system that included human data in addition to animal data. Important features of using animal data for surveillance were uncovered, including low positive predictive value of an animal signal and important biases inherent in both clinical and laboratory data. Using these results, it is possible to identify a number of possible ways forward for EZD surveillance using animal data. The systematic review of current EZD surveillance systems (Chapter 2) showed that while such systems are increasing in number, very few have been evaluated, leaving those developing and using such systems without the necessary information on what works best. Further, the review showed that most systems focused only on human disease, with very few systems integrating animal and human data. It was therefore necessary to turn to public health and animal health experts in order to learn what information decision-makers would need in order to use animal health surveillance data (Chapter 3). The experts agreed that etiological information gathered from laboratories was the foundation of animal health surveillance, since it offers the necessary specificity needed for a public health to respond. While they saw value in sentinel or syndromic data, they saw it more as ancillary to ‘traditional’ laboratory surveillance. They thought that an important component of animal surveillance from a public health perspective was information on human exposures (e.g. how likely 185 it is that people were in contact with the animal(s)/animal product(s) in question, how many people may have been exposed). 6.1 Risk-Based Animal Surveillance for Public Health We propose that taking laboratory data and exposure information together might, in effect, allow for a rapid risk assessment of a signal/event in animal data, giving public health the information it needs to take action. This type of system would be an etiology-based (i.e. laboratorybased) “public health animal risk assessment surveillance system”, with information on the zoonotic potential of the disease and the presence and magnitude of human exposure to the disease. In the field of risk analysis, risk is seen as the synthesis of the probability of occurrence of an undesired event and the consequences or costs of this event (Society for Risk Analysis, 2008). An animal surveillance system using this concept could be seen as ‘risk-based veterinary surveillance’, defined as: “a surveillance programme in the design of which exposure and risk assessment methods have been applied together with traditional design approaches in order to assure appropriate and cost-effective data collection” (Stark et al., 2006). In this context, the necessary exposure and outcome data to be collected for each pathogen would be defined upfront, as would the ongoing risk analyses and responses. Such an EZD surveillance system with integrated risk assessments can also be seen as an EZD “Intelligence” system, as it would combine data from various sources to provide the context to determine the significance of an animal signal related to risk for human health (Sawford et al., 2011). Other roles for risk assessment in risk-based veterinary surveillance would be to determine what type of pathogens should be under surveillance (i.e. hazard identification), risk-based selection of sampling strata as well as sample size calculation 186 based on risk considerations, all with the goal of more efficient systems in terms of their costeffectiveness ratio (Stark et al., 2006). While this type of system would likely be useful from the public health perspective, it would require a significant amount of work upfront as well as on an ongoing basis, requiring constant case investigation. Upfront needs would include: 1) specifying the diseases to be tracked (e.g. those ranking highest based on numerous factors including disease severity, communicability, socioeconomic impact, preventability, potential to cause outbreaks and risk perception, as in (Doherty, 2006), 2) the type of human exposure data to be collected (e.g. the number of humans that could potentially be exposed if the animal in question entered the food chain, potential contact to humans by other routes of transmission), 3) quantification of the socioeconomic impacts of the disease (e.g. costs to the producer and to the industry from consumer fears), and 3) sets of pre-defined public health actions (e.g. food recalls, risk messaging). Unfortunately, such detailed exposure and impact data are not easily obtainable without comprehensive investigation of each case, a matter largely reserved for “reportable” diseases. If animal disease(s) were made reportable to public health, rather than to agricultural agencies, public health could follow up cases in a manner similar to human reportable diseases, e.g. interviewing cases using standardized questionnaires. Possible candidates for this sort of investigation could be specially-trained environmental health officers (public health inspectors), who often investigate cases of foodborne or environmental disease in humans in their local health units. Unfortunately, in the current environment of fiscal restraint that is not likely to change anytime soon, adding new work for public health personnel, when the specificity of animal signals is questionable, is likely not advisable. 187 Despite current fiscal restraint, closer integration of human and animal surveillance would not only facilitate risk assessments, it would allow for cross-pollination between experts in human and animal health disciplines, as well as heightening awareness of the bidirectional nature of transmission of zoonoses from animals to humans and humans to animals. The importance of such bidirectional transmission is highlighted by an outbreak of pandemic influenza virus (pH1N1) in swine on a farm in Alberta, Canada: the swine were originally infected by an ill person working on the farm ventilation system (Howden et al., 2009). This outbreak not only caused clinical illness in swine, it resulted in the depopulation of the entire swine herd, as well as suspected subsequent transmission “back” to two people that responded to the outbreak (Howden et al., 2009). If Salmonella in animals was made a reportable disease to public health, the core laboratory data could be quite similar to that currently gathered by BC Integrated Surveillance (Chapter 5). The addition of the magnitude of potential human exposures to the mandatory reporting data fields would add much more context or “intelligence” to the data than was available for this thesis. In the context of Salmonella, this could include whether the animal was a layer hen producing eggs or a broiler chicken destined for meat production, the age of the chicken or hen, whether its “epidemiological unit” entered the food chain; if it did enter the food chain, then additional information on how much entered the food chain, in what form of product(s), and location of distribution would need to be added as well. Such data would allow for the calculation of the relevant space-time interval when an increased risk for humans could be expected; with a defined geographic region and time interval, an increased incidence in human cases could be evaluated for possible correlation with the animal signal. If such correlations were identified, then this would allow for the estimation of risk associated with an animal signal prospectively. Without this type of contextual data, the results of the Salmonella study in Chapter 5 suggest that statistically significant 188 increases in Salmonella in animals are not necessarily actionable, lending little evidence for meaningful investigation of clusters as well as individual cases of Salmonella in animals. Public health action in relation to laboratory isolations of Salmonella in animals without a solid assessment of risk to human health is likely to have negative consequences. The most important repercussions would likely be for those submitting the samples to the laboratories: the owners and producers. Besides causing significant harm to the industry, there might also be an overall reduction of submission of relevant samples to the laboratory for diagnostics, as was found following HPAI and BSE in Chapter 4 of this thesis. 6.2 The “Foundation” of Animal Health Surveillance: Laboratories In Chapter 3, the consultations with animal and public health experts identified laboratory surveillance as the foundation of animal surveillance. Animal health laboratories generally function as ‘atypical’ surveillance systems, in the sense that if a rare or new etiologic agent is isolated or if an odd sample/carcass is submitted, laboratory personnel can choose to share such information with various stakeholders. Hence, in this thesis, I focused on whether animal laboratories could function as a ‘statistical’ surveillance system (Chapter 4). In general, we found that events of epidemiological significance were not associated with statistically significant alerts in either diagnostic or pre-diagnostic laboratory data. This suggests that animal lab data are not amenable to classical statistical surveillance approaches: rather than initially presenting as large outbreaks that could be detected by statistical algorithms, significant disease events may present as isolated cases. This puts the human element back in the forefront: laboratory personnel who recognize that they’ve seen something new or odd, and who know who 189 they should inform. This is especially important since animal laboratory data face many challenges to statistical surveillance that are not likely to change in the near future, such as lack of species and geographic representativeness, and numerous biases on submissions associated with economic costs and confidence in the diagnosis. It is nevertheless possible that statistical surveillance of animal laboratory data prove useful for animal health issues that are not relevant to public health, since these types of outcomes were not assessed in this thesis. Further, the bidirectional nature of zoonotic disease transmission was not assessed in this thesis, i.e. the potential role of a human signal on increased risk of disease in animals. Our finding that statistical analyses of pre-diagnostic data did not result in timely and valid early warning signals was unexpected, since current literature on EID surveillance suggests this type of data should be useful (Shaffer et al., 2008, Dorea et al., 2011b). However, results from the pilot sentinel surveillance project (Chapter 3) suggest there are a number of factors that could potentially bias laboratory surveillance data and even act as confounders in analyses, including animal species, age of the animal, number of animals affected, reason for examination, type of veterinary practice, confidence in a diagnosis, and the distance to a diagnostic laboratory. Other confounders of veterinary laboratory submissions have been found, an interesting one is clinical rotations of senior veterinary students (Shaffer, 2007), suggesting that increases in laboratory submissions may be related to new personnel (including animal owners and producers) that are more likely to submit samples for diseases they are not yet familiar with, rather than with true increases in disease. Another important finding in the analysis of laboratory data was the sustained decline in the number of laboratory submissions following the occurrence of major health events (HPAI in chicken 190 and BSE in cattle). While the reasons for these declines are likely multi-factorial, to my knowledge, there are no reports of analogous drops in submissions in human laboratory-based public health surveillance systems after outbreaks. Therefore, this appears to be a unique feature of animal surveillance, one that again underscores the impact a large outbreak has on the animal health sector, one with direct consequences for both interpretation of historical surveillance data and the potential to lower sensitivity to detect new outbreaks due to lower submission rates. Although the issues with statistical analysis of animal laboratory data outlined above cannot be easily fixed, two useful pieces of information could be added relatively easily: epidemiological unit and submission type. The analysis of laboratory data in Chapter 4 was hampered by the inability to create “epidemiological units” to identify consecutive or concurrent samples from the same population of animals (e.g. on the same farm); unique identifiers for locations of farms could be used for this purpose and in order to protect privacy they could be scrambled and not reported out in publications. The differences between diagnostic and monitoring sample time series for Salmonella serotypes in Chapter 5 underscore the importance of such information on the type of submission when analyzing laboratory surveillance data. While animal health laboratory data could be extracted and analyzed for surveillance purposes, the process was hampered by the absence of standardized coding of diagnoses and syndromes. Once consensus on animal surveillance case definitions, both nationally and internationally, is reached, there is now promise in reducing the time needed for manual classification by using automatic classification of laboratory data into syndrome groups (Dorea et al., 2011a). 191 6.3 Specificity of an Animal Signal: What Does It Mean? The Salmonella study in Chapter 5 provided a unique opportunity to examine the relationship between signals in animal data and signals in human data for a specific pathogen. The study showed that while it was possible to generate cross-sectoral statistically significant alerts (i.e. alerts in at least two of: agricultural animals, meat, and humans), their public health relevance was questionable. It is likely that this is due to the complexity of Salmonella dynamics within the food chain, specifically quantifying the risk to human health from a live animal with Salmonella, rather than a limitation in the statistical methodology used. This was supported by the lack of agreement we found in the Salmonella serotypes in animals, meat, and people over the study period. The Salmonella case study highlights a much larger issue: what does a certain animal ‘signal’ really mean? It could be a real signal relevant to public health, a real signal relevant for animal health, a real signal of no significant relevance, an artefact of the surveillance data, or a statistical anomaly. It is exceedingly difficult to assess causality, i.e. that a particular event in animals likely caused another event (an EZD) in humans, just as it is in epidemiology studies in general, where Hill’s guidelines are often used. The overarching implicit questions that Hill’s guidelines address are whether confounding and bias are reasonable alternative explanations for a particular (statistical) association, and if not, whether a cause-effect relationship can be inferred, using experimental evidence, temporality, strength of association, dose-response, biologic plausibility, and consistency (Szklo and Nieto, 2007). Possible causal associations in the case of a foodborne pathogen, such as Salmonella, could be investigated using 1) detailed outbreak investigations retrospectively linking human disease to pathogens in a particular food, and ultimately to the animal sources of that pathogen, 2) analytical 192 studies such as case-control studies identifying foods associated with increased risk in humans, and 3) laboratory studies identifying the pathogen (e.g. in greater numbers) in the implicated foods versus other foods (e.g. matching genotypes or full genomes). While not practical for everyday surveillance, recent advances in whole genome sequencing may prove uniquely useful in defining transmission links and directionality of transmission, as demonstrated by a recent outbreak of tuberculosis in BC (Gardy et al., 2011). Such detailed genetic information may provide us with convincing evidence for links between human and animal disease, in addition to, or in the absence of, supporting epidemiological and field data. Analysis of laboratory surveillance data on its own, without detailed genome sequence information, is not likely to provide the necessary level of evidence to inform causal links. An interesting outcome in the Salmonella study was, however, that there was significant agreement between the proportions of serotypes found in live chickens and chicken meat, and that the correlation between signals in meat prior to human was supported by subtyping data. Since meat at retail is much closer on the food chain to human consumption, contamination in meat likely represents a better-definable risk for humans than contamination found in a live animal. It follows, then, that tracking meat may be a good way to link the risk from an animal signal to human health outcomes. The difficulty of assessing relevance of an animal signal in terms for human disease is amplified for EZDs, since for such diseases, there is even less evidence than that for Salmonella. Similarly with syndromic surveillance in animals – how can public health react to a signal when its significance is unclear? Indeed, we found that statistical analyses of pre-diagnostic laboratory data did not result in timely and valid early warning signals; these data streams seemed to contain more noise than 193 signal. It remains to be seen whether ancillary epidemiologic and other risk information can help clear up syndromic signals into usable risk signals with sufficiently narrow confidence intervals. The situation appears to be better for diseases that are directly communicable from a live animal to a person (e.g. HPAI and Nipah virus) instead of through a complex food production chain, or those that are spatiotemporally linked to risk (e.g. vector-borne diseases such as Lyme Disease and West Nile virus). This is supported by a recent review that found the majority of analytic studies quantifying the relationship between animal and human disease were those for vectorborne diseases, notably West Nile virus (Scotch et al., 2009). 6.4 Sentinel Surveillance: The “Weird” Network The focus of this thesis was on ‘statistical’ surveillance systems, hence, I designed, implemented and evaluated a pilot sentinel surveillance system collecting ongoing pre-diagnostic clinical animal health data from practicing veterinarians. The project demonstrated that it is possible to find motivated veterinarians that could be sentinels, that the web-based interface was appropriate for such a project as it was used by all but one sentinel, and that timely reporting could be achieved. While the pilot generated data that were not seen as useful for public health practice at this time, the baseline information on the species of animals seen, the proportion seen for suspected infections and how many samples were sent to laboratories, provided not only data that could be used to design more specific sentinel or syndromic surveillance systems in certain animal species, but it also elucidated and quantified several biases that are associated with the use of laboratory data for surveillance. 194 Since this pilot surveillance system was relatively resource-intensive and produced no data that were ‘directly’ useful for EZD surveillance, it follows that future sentinel animal health surveillance should not be structured in this manner, i.e. as a ‘statistical’ surveillance system. An ‘atypical’ surveillance system focused on collecting information on odd and unusual cases or clusters, might be more useful to providing the necessary ancillary information to help interpret other (e.g. laboratory) surveillance data. This may include collection of human exposure and other risk data on select cases in lieu of investigating each and every case, as is done by public health for reportable diseases. This may be a more cost-effective strategy. There are two examples of ‘atypical’ animal health systems, each structured somewhat differently: a collaborative network of sentinel veterinarians, each focused on one agricultural animal species, with links to public health (Gouvernement du Québec, 2010), and a surveillance system where only odd and unusual cases are reported (Barnouin, 2010). “Weird networks” could also serve as platforms to conduct ad-hoc or ongoing projects searching for EZDs. For example, if a network of veterinarians note in their bi-monthly conference calls that they are increasingly seeing (or hearing about) a new syndrome in sheep, subsidized samples could be sent to the laboratory for a full diagnostic work-up. Another could be a project focused on uncovering various biases in laboratory data, either collecting similar information as what was collected in the pilot sentinel system in this thesis, or an ethnographic study focused on factors that influence veterinarians to submit cases to a diagnostic laboratory and the factors that affect the willingness of veterinarians to participate in surveillance programs, as has recently been done in Alberta, Canada (Sawford, 2011). Projects could also focus on specific syndromes in individual species, collecting information on suspected diagnoses and the confidence in such diagnoses. Identifying laboratory submission biases could uncover regions and species 195 underrepresented by laboratory surveillance, as well as conditions for which samples are not being sent to the lab (and why), all of which could be used to improve lab surveillance either by increasing certain types of submissions or by taking the biases into account during data analysis. Prior to the creation of a desired targeted surveillance system, for example one for the detection of West Nile virus in horses (which presents as a neurological disease in severe cases), initial data on the general proportion of horses with neurological syndromes could be gathered using the sentinel veterinary network, and be then used to calculate appropriate sample sizes needed to detect the disease. We found that the most important factors in getting sentinel veterinarians to participate in the pilot surveillance project were an interest in the work and a desire to help. It is through interested animal health and public health workers that such projects and enhanced surveillance systems will happen. Future pilot sentinel surveillance systems may want to place much more emphasis on enhancing and evaluating personal and professional networks. 6.5 Policy Implications There are a number of possible policy implications that arise from this work. Laboratories remain key sources of EZD information and the continued submission of samples to animal health laboratories from practitioners should be encouraged. Consideration could be given to subsidization of select laboratory testing, especially in instances where a new or unusual presentation of a suspected infectious disease is encountered. The establishment of sentinel animal health practitioner networks providing epidemic intelligence and situational awareness should be considered. Practitioners in the field and in laboratories should be encouraged to share unusual infectious disease findings with colleagues both within and outside their place of work and 196 discipline through collaborative networks of professionals interested in EZDs, maintained through regular in-person meetings and conferences. Finally, integrated surveillance efforts involving public and animal health practitioners should be encouraged and funded appropriately. 6.6 Conclusion Animal data, while a relevant data stream for EZD surveillance, is difficult to use by public health at this time, as it does not contain the necessary information to convert the data into risk for humans. Laboratory surveillance is likely the best candidate for EZD surveillance in animals, however, this information needs to be supplemented with additional epidemiological data including potential human exposure information, as well as knowledge of data gaps and biases inherent in the data. Gaps and biases in laboratory data could be gathered using sentinel networks of animal health practitioners. Without this additional risk information, the best use of etiologically-specific animal laboratory data at this time is to help generate hypotheses in epidemiological investigations and in helping evaluate programs by examining longer-term trends. 197 References ACHONU, C., ROSELLA, L., GUBBAY, J. B., DEEKS, S., REBBAPRAGADA, A., MAZZULLI, T., WILLISON, D., FOISY, J., MCGEER, A., JOHNSON, I., LAFRENIERE, M., JOHNSON, C., WILLMORE, J., YUE, C. & CROWCROFT, N. S. 2011. Seroprevalence of Pandemic Influenza H1N1 in Ontario from January 2009–May 2010. PLoS ONE, 6, e26427. AGUILERA, J. F., PAGET, W. J., MOSNIER, A., HEIJNEN, M. L., UPHOFF, H., VAN DER VELDEN, J., VEGA, T. & WATSON, J. M. 2003. Heterogeneous case definitions used for the surveillance of influenza in Europe.see comment. European journal of epidemiology, 18, 751. ANGULO, F. J., JOHNSON, K. R., TAUXE, R. V. & COHEN, M. L. 2000. Origins and consequences of antimicrobial-resistant nontyphoidal Salmonella: implications for the use of fluoroquinolones in food animals. Microbial Drug Resistance, 6, 77-83. ANHOLT, M. 2011. RE: Personal communication. Type to VRBOVA, L. AZIZ, J., OLSON, J., LEE, O. B., DANIELS, P., ADZHAR, A. B., BUNNING, M., SHARIHUDDIN FIELD, H. & JOHARA HOOPER, P. 1999. Nipah virus infections of animals in Malaysia. XIth Int. Cong. Virol. Sydney, Australia. BARNOUIN, J. 2010. emergences2.be [Online]. INRA and CODA-CERVA. Available: http://epiaweb.clermont.inra.fr/emergences/main.php [Accessed December 5, 2011. BENSCHOP, J., STEVENSON, M. A., DAHL, J., MORRIS, R. S. & FRENCH, N. P. 2008. Temporal and Longitudinal Analysis of Danish Swine Salmonellosis Control Programme Data: Implications for Surveillance. Epidemiology & Infection, 136, 1511-20. BEREZOWSKI, J. 2010. RE: Personal communication. Type to VRBOVA, L. BLANCOU, J., CHOMEL, B. B., BELOTTO, A. & MESLIN, F. X. 2005. Emerging or re-emerging bacterial zoonoses: factors of emergence, surveillance and control. Veterinary research, 36, 507. BOWES, V. 2009. Pullorum Disease on Vancouver Island [Online]. Western Meting of Poultry Clinicians and Pathologists Available: http://www.westvet.com/pullorum.htm [Accessed May 2, 2012. BOWES, V. A. 2007. After the Outbreak: How the British Columbia Commercial Poultry Industry Recovered after H7N3 HPAI. Avian Diseases, 51, 313-6. BRAVATA, D. M., MCDONALD, K. M., SMITH, W. M., RYDZAK, C., SZETO, H., BUCKERIDGE, D. L., HABERLAND, C. & OWENS, D. K. 2004. Systematic Review: Surveillance Systems for Early Detection of Bioterrorism-related Diseases. Annals of Internal Medicine, 140, 910-22. BRENNER, F. W., VILLAR, R. G., ANGULO, F. J., TAUXE, R. & SWAMINATHAN, B. 2000. Salmonella Nomenclature. Journal of clinical microbiology, 38, 2465-7. BRILLIANT, L. 2008. Emerging Threats: Can We Predict and Prevent Them? International Conference on Emerging Infectious Diseases (ICEID) Atlanta, GA. BRITISH COLUMBIA CENTRE FOR DISEASE CONTROL. 2012. British Columbia Annual Summary of Reportable Diseases 2011 [Online]. Available: http://www.bccdc.ca/NR/rdonlyres/B24C1DFD-3996-493F-BEC70C9316E57721/0/2011_CD_Annual_Report_Final.pdf [Accessed December 13, 2012. BRITISH COLUMBIA CENTRE FOR DISEASE CONTROL (BCCDC) 2010. PHIDO (Public Health Intelligence for Disease Outbreaks) for Windows: User manual, Victoria, BC. BRITISH COLUMBIA MINISTRY OF AGRICULTURE. 1999. Diagnostic Diary. Animal Health Centre Newsletter [Online], 9. Available: http://www.agf.gov.bc.ca/ahc/pdf/january1999.pdf BRITISH COLUMBIA MINISTRY OF AGRICULTURE 2009-2012. Animal Health Monitor. Victoria. BRITISH COLUMBIA MINISTRY OF AGRICULTURE. 2012a. The Animal Health Centre [Online]. Available: http://www.agf.gov.bc.ca/ahc/. BRITISH COLUMBIA MINISTRY OF AGRICULTURE. 2012b. Reportable Diseases [Online]. Available: http://www.agf.gov.bc.ca/ahc/rept_dis.htm [Accessed October 19, 2012. 198 BROKOPP, C., RESULTAN, E., HOLMES, H. & WAGNER, M. M. 2006. Chapter 8: Laboratories. In: WAGNER, M. M., MOORE, A. W. & ARYEL, R. M. (eds.) Handbook of Biosurveillance. Burlington, MA, USA: Elsevier Academic Press. BRUUN, T., SØRENSEN, G., FORSHELL, L. P., JENSEN, T., NYGARD, K., KAPPERUD G., LINDSTEDT, B. A., BERGLUND, T., WINGSTRAND, A., PETERSEN, R. F., MÜLLER, L., KJELSØ, C., IVARSSON, S., HJERTQVIST, M., LÖFDAHL, S. & ETHELBERG, S. 2009. An Outbreak of Salmonella Typhimurium Infections in Denmark, Norway, and Sweden, 2008. Eurosurveillance, 14, 1-6. BUCK, P. 2004. Public Health Agency of Canada and Health Canada Lessons Learned: Avian Influenza. Canadian Poultry Industry Forum. Abbotsford, BC. BUCKERIDGE, D. L., BURKOM, H., MOORE, A., PAVLIN, J., CUTCHIS, P. & HOGAN, W. 2004. Evaluation of Syndromic Surveillance Systems - Design of an Epidemic Simulation Model. Morbidity & Mortality Weekly Report. Recommendations & Reports, 53, 137-143. BUEHLER, J. W. 2008. Surveillance. In: ROTHMAN, K. J. & GREENLAND (EDS), S. (eds.) Modern Epidemiology. 3rd ed. ed. Philadelphia, PA: Lippincott Williams & Wilkins. BUEHLER, J. W., HOPKINS, R. S., OVERHAGE, J. M., SOSIN, D. M. & TONG, V. 2004. Framework for evaluating public health surveillance systems for early detection of outbreaks: Recommendations from the CDC working group. Morbidity & Mortality Weekly Report. Recommendations & Reports, 53, 1-11. BURKHOM, H. 2007. Chapter 4: Alerting Algorithms for Biosurveillance. In: LOMBARDO J.S. & BUCKRIDGE (EDS), D. (eds.) Disease Surveillance: A Public Health Informatics Approach. Hoboken, NJ: WileyInterscience. CANADIAN BROADCASTING CORPORATION (CBC). 2004. Mad Cow [Online]. Available: http://www.cbc.ca/canadavotes2004/politicalcanada/madcow.html [Accessed October 15, 2012. CANADIAN FOOD INSPECTION AGENCY. 2004. A Short Summary of the 2004 Outbreak of High Pathogenicity Avian Influenza (H7N3) in British Columbia, Canada [Online]. Canadian Food Inspection Agency. Available: http://www.collectionscanada.gc.ca/webarchives/20071122091743/http://www.inspection.gc.ca/e nglish/anima/heasan/disemala/avflu/2004sum/summaryhpaie.shtml [Accessed June 25, 2011. CANADIAN FOOD INSPECTION AGENCY. 2006. Report on the Investigation of the Fifth Case of Bovine Spongiform Encephalopathy (BSE) in Canada [Online]. Canadian Food Inspection Agency. Available: http://www.collectionscanada.gc.ca/webarchives/20071123005620/http://www.inspection.gc.ca/e nglish/anima/heasan/disemala/bseesb/bccb2006/5investe.shtml [Accessed June 25, 2011. CANADIAN FOOD INSPECTION AGENCY. 2010a. Federally Reportable Diseases in Canada – 2010 [Online]. Canadian Food Inspection Agency. Available: http://www.inspection.gc.ca/english/anima/disemala/rep/repe.shtml [Accessed December 5, 2011. CANADIAN FOOD INSPECTION AGENCY. 2010b. National BSE Surveillance Reimbursement Program [Online]. Canadian Food Inspection Agency. Available: http://inspection.gc.ca/english/anima/disemala/bseesb/surv/produce.shtml [Accessed December 5, 2011. CANADIAN FOOD INSPECTION AGENCY. 2012a. Avian Influenza Surveillance [Online]. Available: http://www.inspection.gc.ca/animals/terrestrial-animals/diseases/surveillance/avian-influenzasurveillance/eng/1329693810008/1329694298513 [Accessed October 15, 2012. CANADIAN FOOD INSPECTION AGENCY. 2012b. BSE Enhanced Surveillance Program [Online]. Available: http://www.inspection.gc.ca/animals/terrestrial-animals/diseases/reportable/bse/enhancedsurveillance/eng/1323992647051/1323992718670 [Accessed October 15, 2012. CANADIAN FOOD INSPECTION AGENCY. 2012c. CFIA Investigation into XL Foods Inc. (E. coli O157:H7) [Online]. Available: http://www.inspection.gc.ca/food/consumer-centre/food-safetyinvestigations/xl-foods/eng/1347937722467/1347937818275 [Accessed December 11, 2012. CANADIAN FOOD INSPECTION AGENCY (CFIA). 2012. Terrestrial Animal Diseases [Online]. Ottawa, ON. Available: http://inspection.gc.ca/animals/terrestrialanimals/diseases/eng/1300388388234/1300388449143. 199 CANADIAN POULTRY MAGAZINE. 2010. Available: http://www.canadianpoultrymag.com/content/view/1000/142/ [Accessed January 10, 2010. CAPRIOLI, A., MORABITO, S., BRUGERE, H. & OSWALD, E. 2005. Enterohaemorrhagic Escherichia Coli: Emerging Issues on Virulence and Modes of Transmission. Veterinary research, 36, 289-311. CARRAT, F., FLAHAULT, A., BOUSSARD, E., FARRAN, N., DANGOUMAU, L. & VALLERON, A. J. 1998. Surveillance of Influenza-like Illness in France: The Example of the 1995/1996 Epidemic. Journal of Epidemiology & Community Health, 52 Suppl 1, 32S-38S. CENTERS FOR DISEASE CONTROL (CDC) 1998. Preventing emerging infectious diseases: A strategy for the 21st century overview of the updated CDC plan. Morbidity & Mortality Weekly Report. Recommendations & Reports, 47, 1-14. CENTERS FOR DISEASE CONTROL (CDC) 2007. Salmonella Surveillance: Annual Summary, 2005. In: CENTERS FOR DISEASE CONTROL (ed.). Atlanta, GA. CENTERS FOR DISEASE CONTROL AND PREVENTION 2010. Salmonella Montevideo Infections Associated with Salami Products Made with Contaminated Imported Black and Red Pepper - United States, July 2009 - April 2010. Morbidity and Mortality Weekly Report, 59, 1647-1650. CENTERS FOR DISEASE CONTROL AND PREVENTION. 2012. National Salmonella Surveillance Overview [Online]. Available: http://www.cdc.gov/nationalsurveillance/salmonella_surveillance.html [Accessed October 9, 2012. CHECKLEY, S., BEREZOWSK, J., PATEL, J., CLARKE, R., PETERS, D., KEENLISIDE, J., MILLER, B., BYSTROM, J. & ANNETT, C. 2009. Emerging Disease Surveillance of Livestock Through the Alberta Veterinary Surveillance Network. International Symposia on Veterinary Epidemiology and Economics. Durban, South Africa. CHUA, K. B. 2003. Nipah virus outbreak in Malaysia. Journal of Clinical Virology, 26, 265-275. CHUA, K. B., BELLINI, W. J., ROTA, P. A., HARCOURT, B. H., TAMIN, A., LAM, S. K., KSIAZEK, T. G., ROLLIN, P. E., ZAKI, S. R., SHIEH, W., GOLDSMITH, C. S., GUBLER, D. J., ROEHRIG, J. T., EATON, B., GOULD, A. R., OLSON, J., FIELD, H., DANIELS, P., LING, A. E., PETERS, C. J., ANDERSON, L. J. & MAHY, B. W. 2000. Nipah Virus: a Recently Emergent Deadly Paramyxovirus. Science, 288, 1432-5. CHUA, K. B., KOH, C. L., HOOI, P. S., WEE, K. F., KHONG, J. H., CHUA, B. H., CHAN, Y. P., LIM, M. E. & LAM, S. K. 2002. Isolation of Nipah Virus from Malaysian Island Flying-foxes. Microbes and Infection, 4, 145-51. COHEN, M. L. & TAUXE, R. V. 1986. Drug-resistant Salmonella in the United States:an epidemiologic perspective. Science, 234, 964-969. COOPER, D. L., VERLANDER, N. Q., SMITH, G. E., CHARLETT, A., GERARD, E., WILLOCKS, L. & O'BRIEN, S. 2006. Can Syndromic Surveillance Data Detect Local Outbreaks of Communicable Disease? A Model Using a Historical Cryptosporidiosis Outbreak. Epidemiology & Infection, 134, 13. COX, W. 2010. RE: Personal communication. Type to VRBOVA, L. CURRIE, A., MACDOUGALL, L., ARAMINI, J., GAULIN, C., AHMED, R. & ISAACS, S. 2005. Frozen Chicken Nuggets and Strips and Eggs are Leading Risk Factors for Salmonella Heidelberg Infections in Canada. Epidemiology & Infection, 133, 809-16. DANAN, C., BAROUKH, T., MOURY, F., JOURDAN, D. A. S. N., BRISABOIS, A. & Y, L. E. S. 2011. Automated Early Warning System for the Surveillance of Salmonella Isolated in the Agro-food Chain in France. Epidemiology & Infection, 139, 736-41. DANMAP. 2012. DANMAP: the Danish Programme for surveillance of antimicrobial consumption and resistance in bacteria from animals, food and humans [Online]. Available: http://www.danmap.org/ [Accessed October 9, 2012. DARBY, P. M. 2003. The Economic Impact of SARS. Ottawa, ON: Conference Board of Canada. DASZAK, P., PLOWRIGHT, R. K., EPSTEIN, J. H., PULLIAM, J., ABDUL RAHMAN, S., FIELD, H. E., JAMALUDDIN, A., SHARIFAH, S. H., SMITH, C. S., OLIVAL, K. J., LUBY, S., HALPIN, K., HYATT, A. D., CUNNINGHAM, A. A. & (HERG), T. H. E. R. G. 2006. The Emergence of Nipah and Hendra Virus: Pathogen Dynamics Across a Wildlife-livestock-human Continuum. In: COLLINGE, S. K. & RAY, C. (eds.) Disease Ecology: Community Structure and Pathogen Dynamics. New York: Oxford University Press. 200 DEFRA, D. O. E. A. R. A. 2011. A Review of the implementation of the Veterinary Surveillance Strategy (VSS). London. DEGROOT, B. 2005. The Rapid Syndrome Validation Project for animals: Augmenting contact with the Network of Accredited Veterinarians. NAHSS Outlook. DEL ROCIO AMEZCUA, M., PEARL, D. L., FRIENDSHIP, R. M. & MCNAB, W. B. 2010. Evaluation of a Veterinarybased Syndromic Surveillance System Implemented for Swine. Canadian Journal of Veterinary Research, 74, 241-51. DEPARTMENT OF ENVIRONMENT AND RURAL AFFAIRS 2011. A Review of the Implementation of the Veterinary Surveillance Strategy (VSS). London: Department of Environment and Rural Affairs. DIAMOND, J. M. 1999. Guns, germs, and steel : the fates of human societies, New York, Norton. DOBSON, A. P. & CARPER, E. R. 1996. Infectious Diseases and Human Population History. Bioscience, 46. DOHERR, M. G. & AUDIGE, L. 2001. Monitoring and Surveillance for Rare Health-related Events: A Review from the Veterinary Perspective. Philosophical Transactions of the Royal Society B: Biological Sciences, 356, 1097-106. DOHERTY, J.-A. 2006. Final report and recommendations from the National Notifiable Diseases Working Group. Canada Communicable Disease Report, 32. DOREA, F. C., MUCKLE, C. A., KELTON, D., MCEWEN, B. J., MCNAB, W. B., SANCHEZ, J. & REVIEL, C. 2011a. Syndromic surveillance using veterinary laboratory diagnostic test requests. Épidémiologie et Santé Animale, 59-60, 128-30. DOREA, F. C., SANCHEZ, J. & REVIE, C. W. 2011b. Veterinary Syndromic Surveillance: Current Initiatives and Potential for Development. Preventive veterinary medicine, 101, 1-17. DUFOUR, B. 1999. Technical and Economic Evaluation Method for Use in Improving Infectious Animal Disease Surveillance Networks. Veterinary research, 30, 27. DUTIL, L., IRWIN, R., FINLEY, R., NG, L. K., AVERY, B., BOERLIN, P., BOURGAULT, A., COLE, L., DAIGNAULT, D., DESRUISSEAU, A., DEMCZUK, W., HOANG, L., HORSMAN, G. B., ISMAIL, J., JAMIESON, F., MAKI, A., PACAGNELLA, A. & PILLAI, D. R. 2010. Ceftiofur Resistance in Salmonella enterica Serovar Heidelberg from Chicken Meat and Humans, Canada. Emerging Infectious Diseases, 16, 48-54. EGENVALL, A., BONNETT, B. N., OLSON, P. & HEDHAMMAR, A. 1998. Validation of Computerized Swedish Dog and Cat Insurance Data Against Veterinary Practice Records. Preventive veterinary medicine, 36, 51-65. ELCI, C. The impact of HPAI of the H5N1 strain on economies of affected countries. First International Conference on Human and Economic Resources, May 24–25, 2006 Izmir, Turkey. Izmir University of Economics, 104–117. EUROPEAN CENTRE FOR DISEASE PREVENTION AND CONTROL. 2012. Enter-net [Online]. Available: http://ecdc.europa.eu/en/activities/surveillance/pages/enter-net.aspx [Accessed October 9, 2012. EUROPEAN FOOD SAFETY AUTHORITY & EUROPEAN CENTRE FOR DISEASE PREVENTION AND CONTROL 2011. Scientific Report of EFSA and ECDC: The European Union Summary Report on Trends and Sources of Zoonoses, Zoonotic Agents and Food-borne Outbreaks in 2009. The European Food Safety Authority Journal, 9, 2090. FIELD, H., YOUNG, P., YOB, M. Y., MILLS, J., HALL, L. & MACKENZIE, J. 2001. The natural history of Hendra and Nipah viruses. Microbes and Infection, 3, 307-314. FISCHOFF, B., DE BRUIN, W. B., GÜVENÇ, Ü., CARUSO, D. & BRILLIANT, L. 2006. Analyzing Disaster Risks and Plans: An Avian Flu Example. Journal of risk and uncertainty, 33, 131-149. FLEISS, J. L. 1971. Measuring Noiminal Scale Agreement Among Many Raters. Psychological Bulletin, 76, 378382. FOOD AND AGRICULTURE ORGANIZATION OF THE UNITED NATIONS. Challenges of Animal Health Information Systems and Surveillance for Animal Diseases and Zoonoses. Procedings of the internal workshop organized by FAO, 23-26 November 2010 2011 Rome. FAO Animal Production and Health Proceedings. 201 FOOD AND AGRICULTURE ORGANIZATION OF THE UNITED NATIONS, WORLD ORGANISATION FOR ANIMAL HEALTH (OIE) & WORD HEALTH ORGANIZATION 2010. Sharing Responsibilities and Coordinating Global Activities to Address Health Risks at the Animal-Human-Ecosystems Interfaces: A Tripartite Concept Note. Rome: Food and Agriculture Organization of the United Nations, World Organization for Animal Health, World Health Organization. FOUCHIER, R. A. M., OSTERHAUS, A. D. M. E. & BROWN, I. H. 2003. Animal influenza virus surveillance. Vaccine, 21, 1754-1757. FOURNIER, P. E., DRANCOURT, M. & RAOULT, D. 2007. Bacterial Genome Sequencing and Its Use in Infectious Diseases. The Lancet Infectious Diseases, 7, 711-23. FUNTOWICZ, S. & RAVETZ, J. 1994. Emergent complex systems. Futures, 26, 568-582. GALANIS, E., LO FO WONG, D. M. A., PATRICK, M. E., BINSZTEIN, N., CIESLIK, A., CHALERMCHAIKIT, T., AIDARA-KANE, A., ELLIS, A., ANGULO, F. J. & WEGENER, H. C. 2006. Web-Based Surveillance and Global Salmonella Distribution, 2000-2002. Emerging Infectious Diseases, 12, 381. GALANIS, E., PARMLEY, J., DE WITH, N. & GROUP, B. C. I. S. O. F. P. W. 2012. Integrated Surveillance of Salmonella Along the Food Chain Using Existing Data and Resources in British Columbia, Canada. Food Research International, 45. GARDY, J. L., JOHNSTON, J. C., HO SUI, S. J., COOK, V. J., SHAH, L., BRODKIN, E., REMPEL, S., MOORE, R., ZHAO, Y., HOLT, R., VARHOL, R., BIROL, I., LEM, M., SHARMA, M. K., ELWOOD, K., JONES, S. J. M., BRINKMAN, F. S. L., BRUNHAM, R. C. & TANG, P. 2011. Whole-Genome Sequencing and SocialNetwork Analysis of a Tuberculosis Outbreak. The New England Journal of Medicine, 364, 730-739. GERMAN, R. R., LEE, L. M., HORAN, J. M., MILSTEIN, R. L., PERTOWSKI, C. A. & WALLER, M. N. 2001. Updated Guidelines for Evaluating Public Health Surveillance Systems: Recommendations from the Guidelines Working Group. Morbidity & Mortality Weekly Report: Recommendations & Reports, 50, 1-35; quiz CE1-7. GIBBENS, J. C., ROBERTSON, S., WILLMINGTON, J., MILNES, A., RYAN, J. B., WILESMITH, J. W., COOK, A. J. & DAVID, G. P. 2008. Use of Laboratory Data to Reduce the Time Taken to Detect New Diseases: VIDA to Farmfile. Veterinary Record, 162, 771-6. GLASS, G. E., CHEEK, J. E., PATZ, J. A., SHIELDS, T. M., DOYLE, T. J., THOROUGHMAN, D. A., HUNT, D. K., ENSCORE, R. E., GAGE, K. L., IRLAND, C., PETERS, C. J. & BRYAN, R. 2000. Using Remotely Sensed Data to Identify Areas At Risk For Hantavirus Pulmonary Syndrome. Emerging Infectious Diseases, 6, 23847. GLASS, G. E., YATES, T. L., FINE, J. B., SHIELDS, T. M., KENDALL, J. B., HOPE, A. G., PARMENTER, C. A., PETERS, C. J., KSIAZEK, T. G., LI, C. S., PATZ, J. A. & MILLS, J. N. 2002. Satellite Imagery Characterizes Local Animal Reservoir Populations of Sin Nombre Virus in the Southwestern United States. Proceedings of the National Academy of Sciences of the United States of America, 99, 16817-22. GOOGLE. 2012. Google Maps [Online]. Available: https://maps.google.ca [Accessed October 24, 2012. GOUVERNEMENT DU QUÉBEC. 2010. Réseau d'alerte et d'information zoosanitaire (RAIZO) [Online]. Le ministère de l'Agriculture, des Pêcheries et de l'Alimentation de Québec (MAPAQ). Available: http://www.mapaq.gouv.qc.ca/Fr/Productions/santeanimale/raizo/Pages/raizo.aspx [Accessed December 5, 2011. GOVERNMENT OF ALBERTA. 2010. Alberta Veterinary Surveillance Network (AVSN) [Online]. Agriculture and Rural Development, Government of Alberta. Available: http://www1.agric.gov.ab.ca/$department/deptdocs.nsf/all/afs10440 [Accessed December 5, 2011. GUERIN, M. T., MARTIN, S. W. & DARLINGTON, G. A. 2005a. Temporal Clusters of Salmonella Serovars in Humans in Alberta, 1990-2001. Canadian Journal of Public Health, 96, 390-5. GUERIN, M. T., MARTIN, S. W., DARLINGTON, G. A. & RAJIC, A. 2005b. A Temporal Study of Salmonella Serovars in Animals in Alberta Between 1990 and 2001. Canadian Journal of Veterinary Research, 69, 88-99. HALD, T., VOSE, D., WEGENER, H. C. & KOUPEEV, T. 2004. A Bayesian Approach to Quantify the Contribution of Animal-Food Sources to Human Salmonellosis. Risk Analysis, 24, 255. 202 HALPIN, K., YOUNG, P. L., FIELD, H. E. & MACKENZIE, J. S. 2000. Isolation of Hendra Virus from Pteropid Bats: A Natural Reservoir of Hendra Virus. Journal of General Virology, 81, 1927-32. HASTIE, T. & TIBSHIRANI, R. 1990. Exploring the Nature of Covariate Effects in the Proportional Hazards Model. Biometrics, 46, 1005-16. HELMS, M., ETHELBERG, S. & MOLBAK, K. 2005. International Salmonella Typhimurium DT104 Infections, 1992-2001. Emerging Infectious Diseases, 11, 859-67. HENNING, K. J. 2003. What is syndromic surviellance? Morbidity & Mortality Weekly Report., 53(Suppl), 7-11. HEYMANN, D. (ed.) 2008. Control of Communicable Diseases Manual, Washington, DC: American Public Health Association. HEYNDRICKX, M., VANDEKERCHOVE, D., HERMAN, L., ROLLIER, I., GRIJSPEERDT, K. & DE ZUTTER, L. 2002. Routes for Salmonella Contamination of Poultry Meat: Epidemiological Study from Hatchery to Slaughterhouse. Epidemiology & Infection, 129, 253-65. HO, S. S., BROSSARD, D. & D.A., S. 2007. The Polls - Trends: Public Reactions to Global Health Threats and Infectious Diseases. Public Opinion Quarterly, 71, 671–692. HOHLE, M., PAUL, M. & HELD, L. 2007. Statistical Approaches to the Surveillance of infectious Diseases for Veterinary Public Health. Technical Reports 14. Department of Statistics, University of Munich. HOWDEN, K. J., BROCKHOFF, E. J., CAYA, F. D., MCLEOD, L. J., LAVOIE, M., ING, J. D., BYSTROM, J. M., ALEXANDERSEN, S., PASICK, J. M., BERHANE, Y., MORRISON, M. E., KEENLISIDE, J. M., LAURENDEAU, S. & ROHONCZY, E. B. 2009. An investigation into human pandemic influenza virus (H1N1) 2009 on an Alberta swine farm. Canadian Veterinary Journal, 50, 1153-1161. HUMPHREY, T., THRELFALL, E. & CRUIKSHANK, J. 1998. Salmonellosis. In: SR, P., EJL, S. & DIH, S. (eds.) Zoonoses: Biology, Clinical Practice, and Public Health Control. New York, NY: Oxford University Press. HUTCHISON, M., WALTERS, L., AVERY, S., SYNGE, B. & MOORE, A. 2004. Levels of Zoonotic Agents in British Livestock Manures. Letters in Applied Microbiology, 39, 207–214. HUTWAGNER, L., THOMPSON, W., SEEMAN, G. M. & TREADWELL, T. 2003. The Bioterrorism Preparedness and Response Early Aberration Reporting System (EARS). Journal of Urban Health, 80, i89-96. HYATT, A. D., DASZAK, P., CUNNIGHAM, A. A., GIELD, H. & GOULD, A. R. 2004. Henipaviruses: Gaps in the Knowledge of Emergence. Ecohealth, 1, 25-38. JENSEN, V. F., LARSEN, L. S., SEYFARTH, A. M., AGERSØ, Y., JENSEN, L. B., STRUVE, T. & CHRISTIANSEN, P. 2009. DANMAP 2009: Use of Antimicrobial Agents and Occurrence of Antimicrobial Resistance in Bacteria from Food Animals, Foods and Humans in Denmark. Søborg, Denmark: Danish Integrated Antimicrobial Resistance Monitoring and Research Programme. JONES, K. E., PATEL, N. G., LEVY, M. A., STOREYGARD, A., BALK, D., GITTLEMAN, J. L. & DASZAK, P. 2008. Global Trends in Emerging Infectious Diseases. Nature, 451, 990-3. KAHN, L. H. 2006. Confronting Zoonoses, Linking Human and Veterinary Medicine. Emerging Infectious Diseases, 12, 556-561. KARMALI, M. A. 1989. Infection by verocytotoxin-producing Escherichia coli. Clinical Microbiology Reviews, 2, 15-38. KEUSCH, G. T., PAPPAIOANOU, M., GONZALEZ, M. C., SCOTT, K. A., TSAI, P. & (EDS.) 2009. Sustaining Global Surveillance and Response to Emerging Zoonotic Diseases, Washington, DC, National Academies Press. KING, L. J. 2008. Committee on Agriculture Subcommittee on Horticulture and Organic Agriculture CDC Congressional Testimony. Washington, DC: Centers for Disease Control. KITCHING, P. December 30, 2011 2011. RE: Dr. Paul Kitching: Chief Provincial Veterinarian, British Columbia Animal Health Centre. KOPLAN, J. 2001. CDC's strategic plan for bioterrorism preparedness and response. Public health reports, 116, 9. KOSMIDER, R., KELLY, L., EVANS, S. & GETTINBY, G. 2006. A Stastistical System for Detecting Salmonella Outbreaks in British Livestock. Epidemiology & Infection, 134, 952-60. 203 KOSMIDER, R. D., KELLY, L., SIMONS, R. L., BROUWER, A. & DAVID, G. 2011. Detecting New and Emerging Diseases On Livestock Farms Using An Early Detection System. Epidemiology & Infection, 139, 147685. KOSS, T., CARTER, E. L., GROSSMAN, M. E., SILVERS, D. N., RABINOWITZ, A. D., SINGLETON, J., JR., ZAKI, S. R. & PADDOCK, C. D. 2003. Increased Detection of Rickettsialpox in A New York City Hospital Following the Anthrax Outbreak of 2001: Use of Immunohistochemistry for the Rapid Confirmation of Cases in An Era of Bioterrorism. Archives of Dermatology, 139, 1545-52. KRUSE, H., KIRKEMO, A. M. & HANDELAND, K. 2004. Wildlife as source of zoonotic infections. Emerging Infectious Diseases, 10, 2067-72. LANDIS, J. R. & KOCH, G. G. 1977. The Measurement of Observer Agreement for Categorical Data. Biometrics, 33, 159-74. LAST, J. M. 2001. A Dictionary of Epidemiology, Oxford ; New York, Oxford University Press. LEBLANC, M. 2008. Chronology of BSE-Related Events and Government Initiatives. In: DIVISION, P. O. C. E. (ed.). LESLIE, M. J. & MCQUISTON, J. H. 2007. Surveillance for Zoonotic Diseases. In: M'IKANATHA, N. M., LYNFIELD, R., VAN BENEDEN, C. A. & DE VALK, H. (eds.) Infectious Disease Surveillance. Malden, Massachusetts: Blackwell Publishing. LETRILLIART, L., RAGON, B., HANSLIK, T. & FLAHAULT, A. 2005. Lyme Disease in France: A Primary Care-Based Prospective Study. Epidemiology & Infection, 133, 935. LI, K. S., GUAN, Y., WANG, J., SMITH, G. J., XU, K. M., DUAN, L., RAHARDJO, A. P., PUTHAVATHANA, P., BURANATHAI, C., NGUYEN, T. D., ESTOEPANGESTIE, A. T., CHAISINGH, A., AUEWARAKUL, P., LONG, H. T., HANH, N. T., WEBBY, R. J., POON, L. L., CHEN, H., SHORTRIDGE, K. F., YUEN, K. Y., WEBSTER, R. G. & PEIRIS, J. S. 2004. Genesis of A Highly Pathogenic and Potentially Pandemic H5N1 influenza Virus in Eastern Asia. Nature, 430, 209-13. LIPATOV, A. S., GOVORKOVA, E. A., WEBBY, R. J., OZAKI, H., PEIRIS, M., GUAN, Y., POON, L. & WEBSTER, R. G. 2004. Influenza: Emergence and Control. Journal of Virology, 78, 8951-9. LOBER, W. B., KARRAS, B. T., WAGNER, M. M., OVERHAGE, J. M., DAVIDSON, A. J., FRASER, H., TRIGG, L. J., MANDL, K. D., ESPINO, J. U. & TSUI, F. C. 2002. Roundtable On Bioterrorism Detection: information System-Based Surveillance. Journal of the American Medical Informatics Association, 9, 105-15. MACKENZIE, J. S. 1999. Emerging viral diseases: An Australian perspective. Emerging Infectious Diseases, 5, 1-8. MARCUS, R., VARMA, J. K., MEDUS, C., BOOTHE, E. J., ANDERSON, B. J., CRUME, T., FULLERTON, K. E., MOORE, M. R., WHITE, P. L., LYSZKOWICZ, E., VOETSCH, A. C. & ANGULO, F. J. 2007. Re-Assessment of Risk Factors for Sporadic Salmonella Serotype Enteritidis infections: A Case-Control Study in Five Foodnet Sites, 2002-2003. Epidemiology & Infection, 135, 84-92. MARSDEN-HAUG, N., FOSTER, V. B., GOULD, P. L., ELBERT, E., WANG, H. & PAVLIN, J. A. 2007. Code-Based Syndromic Surveillance for influenzalike Illness By international Classification of Diseases, Ninth Revision. Emerging Infectious Diseases, 13, 207-16. MATHER, A. E., MATTHEWS, L., MELLOR, D. J., REEVE, R., DENWOOD, M. J., BOERLIN, P., REID-SMITH, R. J., BROWN, D. J., COIA, J. E., BROWNING, L. M., HAYDON, D. T. & REID, S. W. 2012. An Ecological Approach to Assessing the Epidemiology of Antimicrobial Resistance in Animal and Human Populations. The Royal Society Proceedings: Biological Sciences, 279, 1630-9. MAUDLIN, I., EISLER, M. C. & WELBURN, S. C. 2009. Neglected and endemic zoonoses. Philosophical Transactions of the Royal Society B: Biological Sciences, 364, 2777-2787. MCCLUSKEY, B. J. 2003. Use of Sentinel Herds in Monitoring and Surveillance Systems. In: SALMAN, M. D. E. (ed.) Animal Disease Surveillance and Survey Systems. Ames, IA: Blackwell. MCDADE, J. E. 1997. Global infectious Diseases: Surveillance and Response. Australian Journal of Medical Science, 18, 2. 204 MCKENNA, V. B., GUNN, J. E., AUERBACH, J., BRINSFIELD, K. H., DYER, K. S. & BARRY, M. A. 2003. Local Collaborations: Development and Implementation of Boston's Bioterrorism Surveillance System. Journal of Public Health Management & Practice, 9, 384. MCLEOD, A., MORGAN, N., PRAKASH, A. & HINRICHS, J. 2005. Economic and Social Impacts of Avian Influenza. Rome: Food and Agriculture Organization of the United Nations. MILLER, B., KASSENBORG, H., DUNSMUIR, W., GRIFFITH, J., HADIDI, M., NORDIN, J. D. & DANILA, R. 2004. Syndromic Surveillance for Influenzalike Illness in Ambulatory Care Network. Emerging Infectious Diseases, 10, 1806-11. MITURA, A. & DI PIETRO, L. 2004. Canada's Beef Cattle Sector and the Impact of Bovine Spongiform Encephalopathy (BSE) On Farm Family Income. Agriculture and Rural Working Paper Statistics Canada Agriculture Division. MOHD NOR, M. N., GAN, C. H. & ONG, B. L. 2000. Nipah virus infection of pigs in peninsular Malaysia. Revue scientifique et technique - Office international des epizooties, 19, 160-165. MORENS, D. M., FOLKERS, G. K. & FAUCI, A. S. 2004. The challenge of emerging and re-emerging infectious diseases. Nature, 430, 242-249. MORRIS, J. G., JR. & POTTER, M. 1997. Emergence of New Pathogens as a Function of Changes in Host Susceptibility. Emerging Infectious Diseases, 3, 435-441. MORSE, S. S. 1990. Looking for a Link. Nature, 344, 297. MORSE, S. S. 2004. Factors and Determinants of Disease Emergence. Revue scientifique et technique - Office international des epizooties, 23, 443-51. MUSGRAVE, K. 2009. Monitoring Animal Diseases and Their Impact on Public Health: The Wyoming Regional Veterinary Coordinator Program. Emerging Zoonoses Sentinel Animal Health Surveillance Symposium. Vancouver. MYERS, M. F., ROGERS, D. J., COX, J., FLAHAULT, A. & HAY, S. I. 2000. Forecasting Disease Risk for Increased Epidemic Preparedness in Public Health. Advances in Parasitology, 47, 309-30. NELSON, K. E. & WILLIAMS, C. F. M. 2007. Early History of Infectious Disease: Epidemiology and Control of Infectious Diseases. In: NELSON, K. E. & WILLIAMS, C. F. M. (eds.) Infectious Disease Epidemiology: Theory and Practice. 2nd ed. Burlington, MA: Jones and Bartlett Publishers. NEUMANN, G., NODA, T. & KAWAOKA, Y. 2009. Emergence and Pandemic Potential of Swine-origin H1N1 Influenza Virus. Nature, 459, 931-9. ONE HEALTH INITIATIVE. 2012. One Health Initiative will unite human and veterinary medicine [Online]. Available: http://www.onehealthinitiative.com [Accessed October 24, 2012. ORGANISATION FOR ECONOMIC CO-OPERATION AND DEVELOPMENT & WORLD HEALTH ORGANIZATION 2003. Food Borne Disease in the OECD Countries: Present State and Economic Costs, Paris, Organisation for Economic Co-operation and Development and World Health Organization. PADDOCK, C. D., HOLMAN, R. C., KREBS, J. W. & CHILDS, J. E. 2002. Assessing the Magnitude of Fatal Rocky Mountain Spotted Fever in the United States: Comparison of Two National Data Sources. American Journal of Tropical Medicine & Hygiene, 67, 349-54. PADDOCK, C. D., KOSS, T., EREMEEVA, M. E., DASCH, G. A., ZAKI, S. R. & SUMNER, J. W. 2006. Isolation of Rickettsia Akari from Eschars of Patients With Rickettsialpox. American Journal of Tropical Medicine & Hygiene, 75, 732-8. PAN AMERICAN HEALTH ORGANIZATION (PAHO) 2003. Zoonoses and Communicable Diseases Common to Man and Animals, Washington, DC. PARMLEY, J. E., BASTIEN, N., BOOTH, T. F., BOWES, V., BUCK, P. A., BREAULT, A., CASWELL, D., DAOUST, P., DAVIES, J. C., ELAHI, S. M., FORTIN, M., KIBENGE, F., KING, R., LI, Y., NORTH, N., OJKIC, D., PASICK, J., PRYOR, S. P., ROBINSON, J., RODRIGUE, J., WHITNEY, H., ZIMMER, P. & LEIGHTON, F. A. 2008. Wild Bird Influenza Survey, Canada, 2005. Emerging Infectious Diseases, 14, 84-87. PARSONS, D. F., GARNERIN, P., FLAHAULT, A. & GOTHAM, I. J. 1996. Status of Electronic Reporting of Notifiable Conditions in the United States and Europe. Telemedicine Journal, 2, 273. 205 PASICK, J., BERHANE, Y. & HOOPER-MCGREVY, K. 2009. Avian influenza: the Canadian experience. Revue scientifique et technique - Office international des epizooties, 28, 349-358. PENELL, J. C., EGENVALL, A., BONNETT, B. N. & PRINGLE, J. 2007. Validation of Computerized Swedish Horse insurance Data Against Veterinary Clinical Records. Preventive veterinary medicine, 82, 236-51. PHILBEY, A. W., KIRKLAND, P. D., ROSS, A. D., DAVIS, R. J., GLEESON, A. B., LOVE, R. J., DANIELS, P. W., GOULD, A. R. & HYATT, A. D. 1998. An Apparently New Virus (Family Paramyxoviridae) Infectious for Pigs, Humans, and Fruit Bats. Emerging Infectious Diseases, 4, 269-71. POIRIER, E., WATIER, L., ESPIE, E., WEILL, F. X., DE VALK, H. & DESENCLOS, J. C. 2008. Evaluation of the Impact On Human Salmonellosis of Control Measures Targeted to Salmonella Enteritidis and Typhimurium in Poultry Breeding Using Time-Series Analysis and intervention Models in France. Epidemiology & Infection, 136, 1217-24. PROMED-MAIL. 2012. E. coli EHEC - N America (02): ground beef, steak, O157, recall: 20121008.1328938 [Online]. Available: http://www.promedmail.org [Accessed December 11, 2012. PUBLIC HEALTH AGENCY OF CANADA. 2005. Notifiable Diseases On-Line [Online]. Available: http://dsolsmed.phac-aspc.gc.ca/dsol-smed/ndis/list-eng.php [Accessed December 13, 2012. PUBLIC HEALTH AGENCY OF CANADA 2011a. Canadian Integrated Program for Antimicrobial Resistance Surveillance (CIPARS) 2008. Guelph, ON. PUBLIC HEALTH AGENCY OF CANADA 2011b. National Enteric Surveillance Program (NESP): Annual Summary 2010. Winnipeg, MB. PUBLIC HEALTH AGENCY OF CANADA. 2012a. C-EnterNet [Online]. Available: http://www.phac-aspc.gc.ca/centernet/index-eng.php [Accessed October 9, 2012. PUBLIC HEALTH AGENCY OF CANADA. 2012b. National Enteric Surveillance Program (NESP) [Online]. Available: http://www.nml-lnm.gc.ca/NESP-PNSME/index-eng.htm [Accessed October 9, 2012. PUBLIC HEALTH AGENCY OF CANADA. 2012c. Public Health Notice: E. coli O157 illness related to beef [Online]. Available: http://www.phac-aspc.gc.ca/fs-sa/phn-asp/ecoli-1012-eng.php [Accessed December 11, 2012. PULSENET INTERNATIONAL. 2012. PulseNet International: The International Molecular Subtyping Network for Foodborne Disease Surveillance [Online]. Available: www.pulsenetinternational.org [Accessed October 9, 2012. RABINOWITZ, P. M. Linking Global Survellance for Human and Animal Diseases: Progress and Pitfalls. Princeton Seminar on Biosecurity, Biotechnology and Global Health, October 3, 2008 2008. RAVEL, A., SMOLINA, E., SARGEANT, J. M., COOK, A., MARSHALL, B., FLEURY, M. D. & POLLARI, F. 2010. Seasonality in Human Salmonellosis: Assessment of Human Activities and Chicken Contamination as Driving Factors. Foodborne Pathogens and Disease, 7, 785-94. RIDLEY, R. M. & BAKER, H. F. 1999. Big decisions based on small numbers: lessons from BSE. Veterinary Quarterly, 21, 86-92. ROBERTSON, C., SAWFORD, K., DANIEL, S. L., NELSON, T. A. & STEPHEN, C. 2010. Mobile Phone-Based Infectious Disease Surveillance System, Sri Lanka. Emerging Infectious Diseases, 16, 1524-31. SAKAI, T., SUZUKI, H., SASAKI, A., SAITO, R., TANABE, N. & TANIGUCHI, K. 2004. Geographic and Temporal Trends in Influenzalike Illness, Japan, 1992-1999. Emerging Infectious Diseases, 10, 1822-6. SARWARI, A. R., MAGDER, L. S., LEVINE, P., MCNAMARA, A. M., KNOWER, S., ARMSTRONG, G. L., ETZEL, R., HOLLINGSWORTH, J. & MORRIS, J. G., JR. 2001. Serotype Distribution of Salmonella Isolates from Food Animals After Slaughter Differs from That of Isolates Found in Humans. Journal of Infectious Diseases, 183, 1295-9. SAWFORD, K. 2011. Animal health surveillance for early detection of emerging infectious disease risks. University of Calgary. SAWFORD, K., ROBERTSON, C., GUNAWARDENA, S. & STEPHEN, C. 2011. Development and Application of a Framework for Emerging Infectious Disease Intelligence in Lower Resource Settings. Journal of Bioterrorism & Biodefense, S4:001. 206 SCALLAN, E., HOEKSTRA, R. M., ANGULO, F. J., TAUXE, R. V., WIDDOWSON, M., ROY, S. L., JONES, J. L. & GRIFFIN, P. M. 2011. Foodborne Illness Acquired in the United States—Major Pathogens. Emerging Infectious Diseases, 17, 7-15. SCHOLTISSEK, C. & NAYLOR, E. 1988. Fish Farming and Influenza Pandemics. Nature, 331, 215. SCOTCH, M., ODOFIN, L. & RABINOWITZ, P. 2009. Linkages Between Animal and Human Health Sentinel Data. BMC Veterinary Research, 5, 15. SHAFFER, L. E. 2007. Using Pre-diagnostic Data from Veterinary Laboratories to Detect Disease Outbreaks in Companion Animals. PhD, The Ohio State University. SHAFFER, L. E., FUNK, J. A., RAJALA-SCHULTZ, P., WAGNER, M. M., WITTUM, T. E. & SAVILLE, W. J. A. 2008. Evaluation of Microbiology Orders from a Veterinary Diagnostic Laboratory as a Potential Data Source for Early Outbreak Detection. Advances in Disease Surveillance, 6, 7. SHORTRIDGE, K. F., PEIRIS, J. S. M. & GUAN, Y. 2003. The next influenza pandemic: lessons from Hong Kong. Journal of Applied Microbiology, 94, 70S-79S. SMART, J. 2007. The Impact of BSE on Farmers' Crisis Management Capacity in Alberta. Prion Risk Management Workshop. Edinburgh, UK. SMITH, G. J. D., VIJAYKRISHNA, D., BAHL, J., LYCETT, S. J., WOROBEY, M., PYBUS, O. G., MA, S. K., CHEUNG, C. L., RAGHWANI, J., BHATT, S., PEIRIS, J. S. M., GYAN, Y. & RAMBAUT, A. 2009. Origins and evolutionary genomics of the 2009 swine-origin H1N1 influenza A epidemic. Nature, 459, 1122-1126. SMOLINSKI, M. S., HAMBURG, M. A. & LEDERBERG, J. E. 2003. Microbial threats to health: Emergence, detection, and response, Washington, DC, The National Academies Press. SOCIETY FOR RISK ANALYSIS. 2008. Glossary of Risk Analysis Terms [Online]. Available: http://www.sra.org/resources_glossary [Accessed November 4, 2012. SONESSON, C. & BOCK, D. 2003. A Review and Discussion of Prospective Statistical Surveillance in Public Health. Journal of the Royal Statistical Society. Series A (Statistics in Society), 166, 5-21. SOSIN, D. M. 2003. Draft Framework for Evaluating Syndromic Surveillance Systems. Journal of Urban Health, 80, i8-13. STARK, K. D., REGULA, G., HERNANDEZ, J., KNOPF, L., FUCHS, K., MORRIS, R. S. & DAVIES, P. 2006. Concepts for risk-based surveillance in the field of veterinary medicine and veterinary public health: review of current approaches. BMC Health Services Research, 6, 20. STATISTICS CANADA 1996. 1996 Census of Agriculture. Ottawa. STATISTICS CANADA 2001. 2001 Census of Agriculture. Ottawa. STATISTICS CANADA 2006. 2006 Census of Agriculture. Ottawa. STATISTICS CANADA. 2007. 2006 Census of Agriculture: Concepts, Methodology and Data Quality [Online]. Ottawa. Available: http://www.statscan.gc.ca/pub/95-629-x/2007000/4123850-eng.htm [Accessed December 5, 2011. STATISTICS CANADA. 2009. 2006 Census of Agriculture: Questionnaires [Online]. Ottawa. Available: http://www.statscan.gc.ca/ca-ra2006/q-eng.htm [Accessed 5 December 2011. STATISTICS CANADA. 2012. Population by year, by province and territory [Online]. Available: http://www.statcan.gc.ca/tables-tableaux/sum-som/l01/cst01/demo02a-eng.htm [Accessed October 19, 2012. STEPHEN, C. 2010. RE: Personal communication. Type to VRBOVA, L. STEPHEN, C. 2011. RE: Personal communication. Type to VRBOVA, L. STEPHEN, C., ARTSOB, H., BOWIE, W. R., DREBOT, M., FRASER, E., LEIGHTON, T., MORSHED, M., ONG, C. & PATRICK, D. 2004. Perspectives On Emerging Zoonotic Disease Research and Capacity Building in Canada. Canadian Journal of Infectious Diseases & Medical Microbiology, 15, 339-44. STITT, T. 2010. RE: Personal communication. Type to VRBOVA, L. STITT, T., MOUNTIFIELD, J. & STEPHEN, C. 2007. Opportunities and obstacles to collecting wildlife disease data for public health purposes: Results of a pilot study on Vancouver Island, British Columbia. Canadian Veterinary Journal, 48, 83-90. 207 SZKLO, M. & NIETO, F. J. 2007. Epidemiology Beyond the Basics, 2nd ed., Sudbury, Massachussets, USA, Jones and Bartlett Publishers, Inc. TAM, J. S. 2002. Influenza A (H5N1) in Hong Kong: an overview. Vaccine, 20, S77-S81. TAYLOR, L. H., LATHAM, S. M. & WOOLHOUSE, M. E. 2001. Risk Factors for Human Disease Emergence. Philosophical Transactions of the Royal Society B: Biological Sciences, 356, 983-9. TAYLOR, M. 2011. RE: Personal communication. Type to VRBOVA, L. THACKER, S. B. 2000. Historical Development. In: TEUTSCH, S. M. & CHURCHILL, R. (eds.) Principles and Practice of Public Health Surveillance. 2nd ed. ed. New York, NY: Oxford University Press. THOMAS, M. K., MAJOWICZ, S. E., SOCKETT, P. N., FAZIL, A., POLLARI, F., DORE, K., FLINT, J. A. & EDGE, V. L. 2006. Estimated Numbers of Community Cases of Illness Due to Salmonella, Campylobacter and Verotoxigenic Escherichia Coli: Pathogen-specific Community Rates. Canadian Journal of Infectious Diseases & Medical Microbiology, 17, 229-34. TOUBIANA, L. & FLAHAULT, A. 1998. A Space-Time Criterion for Early Detection of Epidemics of influenzaLike-Illness. European journal of epidemiology, 14, 465. TSUI, F. C., ESPINO, J. U., DATO, V. M., GESTELAND, P. H., HUTMAN, J. & WAGNER, M. M. 2003. Technical description of RODS: a real-time public health surveillance system. Journal of the American Medical Informatics Association, 10, 399. UNAIDS, J. U. N. P. O. H. A. 2011. UNAIDS World AIDS Day Report 2011. Geneva. VALLERON, A. J. & VIDAL, J. J. Timely redistribution of information for epidemiological surveillance and alert: the experience from the French communicable diseases network. Proceedings / AMIA ...Annual Symposium.:830-4, 2002, Nov 9-13, 2002 2002 San Antonio, TX, USA. Philadelphia: Hanley & Belfus, Inc, 830. VAN METRE, D. C., BARKEY, D. Q., SALMAN, M. D. & MORLEY, P. S. 2009. Development of a Syndromic Surveillance System for Detection of Disease Among Livestock Entering an Auction Market. Journal of the American Veterinary Medical Association, 234, 658-64. VELIKINA, R., DATO, V. & WAGNER, M. M. 2006. Chapter 5: Governmental Public Health. In: WAGNER, M. M., MOORE, A. W. & ARYEL, R. M. (eds.) Handbook of Biosurveillance. Burlington, MA, USA: Elsevier Academic Press. VOURC'H, G., BRIDGES, V. E., GIBBENS, J., DE GROOT, B. D., MCINTYRE, L., POLAND, R. & BARNOUIN, J. 2006. Detecting Emerging Diseases in Farm Animals Through Clinical Observations. Emerging Infectious Diseases, 12, 204-10. WAGNER, M. M., GRESHAM, L. S. & DATO, V. 2006a. Chapter 3: Case Detection, Outbreak Detection, and Outbreak Characterization. In: WAGNER, M. M., MOORE, A. W. & ARYEL, R. M. (eds.) Handbook of Biosurveillance. Burlington, MA, USA: Elsevier Academic Press. WAGNER, M. M., HOGAN, W. R. & ARYEL, R. M. 2006b. Chapter 6: The Healthcare System. In: WAGNER, M. M., MOORE, A. W. & ARYEL, R. M. (eds.) Handbook of Biosurveillance. Burlington, MA, USA: Elsevier Academic Press. WAGNER, M. M., MOORE, A. W. & ARYEL, R. M. 2006c. Handbook of Biosurveillance, Amsterdam ; Boston, Elsevier Academic Press. WAGNER, M. M., TSUI, F. C., ESPINO, J. U., DATO, V. M., SITTIG, D. F., CARUANA, R. A., MCGINNIS, L. F., DEERFIELD, D. W., DRUZDZEL, M. J. & FRIDSMA, D. B. 2001. The emerging science of very early detection of disease outbreaks. Journal of Public Health Management & Practice, 7, 51. WHO, W. H. O. 2005. Combating Emerging Infectious Diseases in South-East Asia Region. New Dehli. WHO, W. H. O. 2012. Fact Sheet No. 104: Tuberculosis. Geneva. WHO, W. H. O., FAO, F. A. A. O. O. T. U. N. & OIE, O. I. D. E. 2004. Report of the WHO/FAO/OIE joint consultation on emerging zoonotic diseases. Geneva, Switzerland: Food and Agriculture Organization of the United Nations (FAO), World Health Organization (WHO), and World Organisation for Animal Health (OIE). WILESMITH, J. W., RYAN, J. B. & ATKINSON, M. J. 1991. Bovine Spongiform Encephalopathy: Epidemiological Studies on the Origin. Veterinary Record, 128, 199-203. 208 WOLFE, N. 2011. The viral storm : the dawn of a new pandemic age, New York, Times Books. WOLFE, N. D., C., P. D. & J., D. 2007. Origins of major human infectious diseases. Nature, 447, 279-283. WOLFE, N. D., SWITZER, W. M., CARR, J. K., BHULLAR, V. B., SHANMUGAM, V., TAMOUFE, U., PROSSER, A. T., TORIMIRO, J. N., WRIGHT, A., MPOUDI-NGOLE, E., MCCUTCHAN, F. E., BIRX, D. L., FOLKS, T. M., BURKE, D. S. & HENEINE, W. 2004. Naturally Acquired Simian Retrovirus Infections in Central African Hunters. Lancet, 363, 932-7. WOOLHOUSE, M. 2011. How to Make Predictions About Future Infectious Disease Risks. Philosophical Transactions of the Royal Society B: Biological Sciences, 366, 2045-54. WOOLHOUSE, M. E. & GOWTAGE-SEQUERIA, S. 2005. Host Range and Emerging and Reemerging Pathogens. Emerging Infectious Diseases, 11, 1842-7. WORD HEALTH ORGANIZATION 2002. Future Trends in Veterinary Public Health. WHO Technical Report Series. Geneva: World Health Organzation. WORD HEALTH ORGANIZATION 2005. Combating Emerging Infectious Diseases in South-East Asia Region. New Dehli: World Health Organzation. WORD HEALTH ORGANIZATION. 2007. International Statistical Classification of Diseases and Related Health Problems (10th revision, version for 2007) [Online]. Geneva: World Health Organzation. Available: http://apps.who.int/classifications/apps/icd/icd10online2007/ [Accessed May 18, 2010. WORD HEALTH ORGANIZATION. 2012a. Cumulative number of confirmed human cases of avian influenza A(H5N1) reported to WHO [Online]. Available: http://www.who.int/influenza/human_animal_interface/EN_GIP_20120810CumulativeNumberH5N 1cases.pdf. WORD HEALTH ORGANIZATION. 2012b. Global Foodborne Infections Network (GFN) [Online]. Available: http://www.who.int/gfn/en [Accessed October 9, 2012. WORD HEALTH ORGANIZATION. 2012c. Variant Creutzfeldt-Jakob disease [Online]. Available: http://www.who.int/mediacentre/factsheets/fs180/en/index.html. WORD HEALTH ORGANIZATION. 2012d. Zoonoses [Online]. Geneva: World Health Organzation. Available: http://www.who.int/topics/zoonoses/en/ [Accessed March 13, 2012. WORD HEALTH ORGANIZATION, FOOD AND AGRICULTURE ORGANIZATION OF THE UNITED NATIONS & WORLD ORGANISATION FOR ANIMAL HEALTH (OIE) 2004. Report of the WHO/FAO/OIE Joint Consultation On Emerging Zoonotic Diseases. Geneva: Food and Agriculture Organization of the United Nations (FAO), World Health Organization (WHO), and World Organisation for Animal Health (OIE). WORLD BANK 2005. East Asia Update: Countering Global Shocks. Washington, DC. WORLD HEALTH ORGANIZATION. 2012. Available: http://www.who.int/hia/evidence/doh/en/ [Accessed December 12, 2012. WORLD ORGANISATION FOR ANIMAL HEALTH (OIE) 2011. Notification of animal and human diseases: Global legal basis. Paris: World Organisation for Animal Health. WORLD ORGANISATION FOR ANIMAL HEALTH (OIE). 2012a. Bovine Spongiform Encephalopathy (BSE) Geographical Distribution of Countries that reported BSE Confirmed Cases since 1989 [Online]. Available: http://www.oie.int/animal-health-in-the-world/bse-specific-data. WORLD ORGANISATION FOR ANIMAL HEALTH (OIE). 2012b. OIE Listed Diseases [Online]. Available: http://www.oie.int/animal-health-in-the-world/oie-listed-diseases-2012/ [Accessed October 19, 2012. WORLD ORGANISATION FOR ANIMAL HEALTH (OIE). 2012c. Outbreaks of Highly Pathogenic Avian Influenza (subtype H5N1) in poultry from the end of 2003 to 18 September 2012 [Online]. Available: http://www.oie.int/fileadmin/Home/eng/Animal_Health_in_the_World/docs/pdf/graph_avian_infl uenza/graphs_HPAI_18_09_2012.pdf. WU WIEN INSTITUTE FOR STATISTICS AND MATHEMATICS. 2012. The R Project for Statistical Computing [Online]. Available: www.r-project-org [Accessed October 9, 2012. 209 WURTZ, R. & POPOVICH, M. 2002. Animal disease surveillance: A framework for supporting disease detection in public health. Available: http://www.stchome.com/media/white_papers/WHP027A.pdf. WYOMING DEPARTMENT OF HEALTH. 2010. Wyoming Regional Veterinary Public Health Coordinators [Online]. Available: http://www.health.wyo.gov/sho/hazards/wrvc.html [Accessed December 5, 2011. YOB, M. Y., FIELD, H., SOHAYATI, A. R., MARIA, J., AZMIN, M. R. & C., M. Preliminary Investigation of Probable Reservoir Host of Nipah Virus. National Congress on Animal Health Production, 1999 Alor Gajah, Malaysia. Veterinary Association of Malaysia, 2. 210 Appendices 211 Appendix A: Appendices for Chapter 2 Appendix A.1: Emerging and Re-emerging Zoonoses Listed by Agent Viruses and prions (N=51) Andes Murray Valley encephalitis Australian bat lyssavirus Nipah Bagaza O'nyong-nyong Banna Oropouche Picobirnavirus Barmah Forest Puumala California encephalitis Rabies Cercopithecine herpes Reston Ebola Chikungunya Rift Valley fever Crimean-Congo Ross River hemorrhagic fever Dengue Sabia Eastern equine Salehabad encephalitis Tickborne encephalitis Sandfly fever Naples Guama Severe acute respiratory syndrome coronavirus Guanarito Hantaan Seoul Hendra Sin Nombre Influenza A* Sindbis Japanese encephalitis St. Louis encephalitis Junin Venezuelan equine encephalitis Laguna Negra Wesselsbron Lassa West Nile Machupo Western equine encephalitis Marburg Yellow fever Mayaro Zaire Ebola Menangle Zika Monkeypox Bovine spongiform encephalopathy agent Bacteria & rickettsia (N=29) Aeromonas caviae A. hydrophila A. veronii (var. sobria) Anaplasma phagocytophila Bacillus anthracis Borrelia burgdorferi Brucella melitensis Campylobacter fetus C. jejuni Helminths (N=9) Anisakis simplex Echinococcus granulosus Loa loa Metorchis conjunctus Onchocerca volvulus Strongyloides stercoralis Taenia solium Trichinella spiralis Wuchereria bancrofti Clostridium botulinum Ehrlichia chaffeensis Protozoa (N=11) Babesia microti E. ewingii Escherichia coli Cryptosporidium hominis C. parvum Francisella tularensis Leptospira interrogans Listeria monocytogenes Mycobacterium avium M. bovis M. marinum Giardia duodenalis Leishmania donovani L. infantum Plasmodium falciparum P. vivax Toxoplasma gondii Rickettsia prowazekii Salmonella enteritidis S. typhi Trypanosoma brucei T. cruzi S. typhimurium Shigella dysenteriae Vibrio cholerae V. parahaemolyticus Fungi (N=9) Histoplasma capsulatum Malassezia pachydermatis Penicillium marneffei Encephalitozoon cuniculi V. vulnificus Yersinia enterocolitica Y. pestis E. hellem E. intestinalis Enterocytozoon bieneusi Nosema connori Trachipleistophora hominis * Only Avian Influenza or other “animal” Influenzas were included; systems looking only at “human-human” Influenzas (i.e. those including both Influenza A and Influenza B) not designed to pick up potential “animal” influenzas (including avian) were not included. 212 Appendix A.2: Emerging and Re-emerging Zoonoses Listed by Transmission Route and Disease Zoonoses Transmitted by Direct Contact, Alimentary (Foodborne and Waterborne), or Aerogenic (Airborne) Routes Aeromonas infection: Aeromonas hydrophila Anisakiasis: Anisakis simplex Anthrax: Bacillus anthracis Argentine hemorrhagic fever: Junin virus (Arenavirus) Avian tuberculosis: Mycobacterium avium Bat Lyssavirus (formerly known as Pteropid bat virus): Australian bat lyssavirus Bovine tuberculosis: Mycobacterium bovis Bolivian hemorrhagic fever (also known as black typhus or Machupo virus): Machupo virus (Arenavirus) Botulism: Clostridium botulinum Bovine spongiform encephalopathy (BSE): BSE prion Brazilian hemorrhagic fever: Sabia virus (Arenavirus) Brucellosis (also called Undulant fever, Malta fever): Brucella melitensis Bubonic plague (also known as Black Death, Great Plague): Yersinia pestis Campylobacterosis: Campylobacter coli, Campylobacter jejuni, Campylobacter spp Cercopithecine herpesvirus 1 (or B virus) infection: Cercopithecine herpesvirus 1 Cholera: Vibrio cholerae Cryptosporidiosis : Cryptosporidium hominis, C. Parvum Ebola hemorrhagic fever: Reston ebolavirus, Zaire ebolavirus (previously Reston Ebola virus , Zaire Ebola virus) Echinococcosis (also known as hydatid disease or hydatid cyst): Echinococcus granulosus Giardiasis: Giardia duodenalis (formerly also Lamblia intestinalis and also known as Giardia duodenalis and Giardia intestinalis) Hantavirus pulmonary syndrome (HPS): Sin Nombre virus, Andes virus, Laguna Negra virus (Hantavirus) Hendra hemorrhagic bronchopneumonia: Hendra virus (Henipavirus) Hemorrhagic colitis: Escherichia coli (E. coli) Hemorrhagic fever with renal syndrome: Hantaan virus, Seoul virus, Puumala virus (Hantavirus) Hemolytic uremic syndrome (HUS): Escherichia coli (E. coli) Histoplasmosis: Histoplasma capsulate, Ajellomyces capsulatus (telomorph) Human monkeypox: Monkeypox virus (Orthopoxvirus) Influenza: Influenza A virus* Lassa hemorrhagic fever: Lassa virus (Arenavirus) Listeriosis: Listeria monocytogenes Malassezia pachydermatis infection (seborrhoeic dermatitis and otitis externa in dogs): Malassezia pachydermatis Marburg hemorrhagic fever: Lake Victoria marburgvirus (previously Marburg virus) Menangle: Menangle virus (Family Paramyxoviridae, genus not yet assigned) Metorchiasis: Metorchis conjunctus Microsporidiosis (can also exhibit as Encephalitozoonosis , Cerebral Microsporidiosis) : Encephalitozoon cuniculi, Encephalitozoon hellem, Encephalitozoon intestinalis, Enterocytozoon bieneusi, Nosema connori, Trachipleistophora hominis Nipah hemorrhagic bronchopneumonia: Nipah virus (Henipavirus) Penicilliosis: Penicillium marneffei Picobirnavirus: Picobirnavirus Pork tapeworm: Taenia solium Rabies: Rabies virus (Lyssavirus) Salmonellosis: Salmonella, S. enterica, S. enteritidis, S. typhi, S. typhimurium SARS (Severe Acute Respiratory Syndrome): Severe acute respiratory syndrome virus (Coronavirus) Strongyloidiasis: Strongyloides stercoralis Swimming Pool Granuloma: Mycobacterium marinum Toxoplasmosis: Toxoplasma gondii 213 Trichinellosis (also called trichinosis, or trichiniasis): Trichinella spiralis Tularemia : Francisella tularensis Venezuelan hemorrhagic fever (VHF): Guanarito virus (Arenavirus) Vibrio infections: Vibrio parahaemolyticus, Vibrio vulnificus Yersiniosis: Yersinia enterocolitica Zoonoses Transmitted by Hematophagous Arthropods Hard ticks: African tick typhus (also called African tick-bite fever): Rickettsia africae Babesiosis: Babesia microti Crimean-Congo hemorrhagic fever: Crimean-Congo hemorrhagic fever virus (Nairovirus) Human granulocytotropic anaplasmosis (HGA) (also known as Human Granulocytic Ehrlichiosis (HGE) and Sennetsu Fever): Anaplasma phagocytophilia Human ehrlichiosis: Ehrlichia ewingii Human monocytic ehrlichiosis (HME): Ehrlichia chaffeensis Lyme disease (European version called Tickborne encephalitis): Borrelia burdorferi Soft Ticks: Kyasanur Forest disease (also known as Monkey disease): Kyasanur Forest disease virus (Flavivirus) Lice: Trench fever (also called Wolhynia fever, shin bone fever, quintan fever, five-day fever, Meuse fever, His disease and His-Werner disease): Bartonella quintana Mosquitoes: Bagaza virus: Bagaza virus (Flavivirus) Banna virus: Banna virus (Seadomavirus) Barmah Forest: Barmah Forest virus (Alphavirus) California encephalitis (viral encephalitis): California encephalitis virus (Orthobunyavirus) Chikungunya fever: Chikungunya virus (Alphavirus) Dengue fever (also called Dengue hemorrhagic fever, Dengue shock syndrome): Dengue virus (Flavivirus) Eastern Equine Encephalitis (EEE): Eastern equine encephalitis virus (Alphavirus) Filariasis (also called lymphatic filariasis or elephantiasis): Wuchereria bancrofti Guama virus: Guama virus (Othobunyavirus) Japanese encephalitis (also known as Japanese B encephalitis): Japanese encephalitis virus (Flavivirus) Leptospirosis: Leptospira interrogans Malaria: Plasmodium sp. – Plasmodium falciparum and Plasmodium vivax Mayaro virus fever: Mayaro virus (Alphavirus) Murray Valley encephalitis (formerly known as Australian encephalitis): Murray Valley encephalitis virus (Flavivirus) O’nyong nyong fever: O'nyong-nyong virus (Alphavirus) Oropouche Fever: Oropouche virus (Orthobunyavirus) Rift Valley fever: Rift Valley fever virus (Phlebovirus) Ross River epidemic polyarthritis (also known as Ross River fever): Ross River virus (Alphavirus) Shigellosis: Shigella dysenteriae Sindbis fever: Sindbis virus (Aplhavirus) St. Louis Encephalitis (SLE): St. Louis encephalitis virus (Flavivirus) Venezuelan Equine Encephalitis: Venezuelan equine encephalitis virus (Alphavirus) Western Equine Encephalitis (WEE): Western equine encephalitis virus (Alphavirus) Wesselsbron: Wesselsbron virus (Flavivirus) West Nile illness, West Nile fever, West Nile neurological illness: West Nile virus [WNV] (Flavivirus) Yellow fever (also called yellow jack, black vomit or vomito negro in Spanish, or sometimes American Plague): Yellow fever virus (Flavivirus) Zika fever: Zika virus (Flavivirus) Sandflies: Sandfly fever : Salehabad virus; Sandfly fever Naples virus (Phlebovirus) 214 Leishmaniasis (visceral, cutaneous, and mucocutaneous leishmaniasis, kala-azar, dumdum fever): Leishmania donovani, Leishmania donovani infantum, Leishmania donovani chagasi Tse Tse Fly/Reduviid Bugs: Sleeping sickness (also known as African trypanosomiasis, Nagana): Trypanosoma brucei Chagas disease (also known as American trypanosomiasis): Trypanosoma cruzi Deer Flies/Black Flies: Loa loa filariasis (also loiasis and African eyeworm): Loa loa River Blindness (also known as onchocerciasis): Onchocerca volvulus Fleas: Cat-Scratch Fever (also called Cat-scratch disease, Cat-Scratch Adenitis, Cat-Scratch-Oculoglandular Syndrome, Debre's Syndrome, Debre-Mollaret Syndrome, Foshay-Mollaret Cat-Scratch Fever, Foshay-Mollaret syndrome, Foshay-Mollaret Cat-Scratch Fever Syndrome, Lymphadenitis-Regional Nonbacterial, Lymphoreticulosis-Benign Inoculation, maladie des griffes du chat, Parinaud oculoglandular disease, and Petzetakis' disease): Bartonella henselae, Bartonella clarridgeiae Murine typhus: Rickettsia typhi, Rickettsia felis, Rickettsia prowazekii * Only Avian Influenza or other “animal” Influenzas were included; systems looking only at “human-human” Influenzas (i.e. those including both Influenza A and Influenza B) not designed to pick up potential “animal” influenzas (including avian) were not included. 215 Appendix A.3: Data Extracted from Articles No. 1 2 3 Field Name System Name Purpose Location 4 5 6 7 Population Year Started Organizations Involved Agent: Known and/or Unknown System Type 8 9 10 11 12 13 14 15 16 17 18 Syndromes/Diseases Under Surveillance Type of Data Collected Data Category: Human/Animal/Other Method of Data Collection and Analysis Evaluation: Timeliness Evaluation: Sensitivity/Specificity Evaluation: Other Evaluated Category: Yes/No Role of Public Health Inspectors References Field Description the name of the system the purpose of the system the location of the system, systems entered into the database at their highest form of aggregation, i.e. if the system was both local and national, it was entered at the national level (Continent, “International” was used for systems that spanned two or more countries, and Country) a description of the population the system covers the year the system started operating the organizations involved in the operation of the system the nature of the infectious disease agent the system could identify, whether the agent is known or defined, or unknown and undefined the type of system, such as whether it is a true surveillance system, monitoring system, or research project the types of diseases or syndromes that the system indentifies the type of data the system collects, such as laboratory diagnoses or administrative health records the type of data collected category, defining whether the data collected was human, and/or animal, and/or other a description of the methods employed by the system to collect the necessary data, and how the data was analyzed whether the system was evaluated for timeliness whether the sensitivity or specificity of the system was determined whether any other evaluation of the system was explicitly performed, see methods for more details a categorical assessment of whether an evaluation was conducted, see methods for more details detailed account of the role of public health inspectors or analogous personnel in the system a list of all references associate with the system, as often more than one reference was associated with one system 216 Appendix A.4: Evaluation Criteria Used in the Canadian Field Epidemiology Program’s Surveillance System Evaluation Reports (N=7). Evaluation Criteria Timeliness Acceptability Utility/Relevance Flexibility Sensitivity/Specificity/Positive Predictive Value Data Quality Representativeness Simplicity Sustainability Number of Reports Using Criteria 5 4 4 3 3 2 2 2 1 217 Appendix A.5: MEDLINE Search Terms Used for the Review Pilot Search Database: Ovid MEDLINE(R) 1950 to Present with Daily Update Search Strategy: --------------------------------------------------------1 population surveillance/ (27146) 2 surveillance.mp. (72696) 3 or/1-2 (72696) 4 zoonoses/ (7640) 5 3 and 4 (384) 6 limit 5 to yr="1985 - 2007" (355) 7 limit 6 to english language (286) 8 6 not 7 (69) Final Search Database: Ovid MEDLINE(R) 1950 to Present with Daily Update Search Strategy: --------------------------------------------------------1 artificial intelligence/ or expert systems/ or fuzzy logic/ or knowledge bases/ or natural language processing/ or "neural networks (computer)"/ (18972) 2 medical informatics/ or medical informatics applications/ (5026) 3 Public Health Informatics/ (571) 4 decision making, computerassisted/ or diagnosis, computer-assisted/ (14602) 5 information systems/ or clinical laboratory information systems/ (16796) 6 decision support systems, clinical/ or geographic information systems/ or hospital information systems/ or integrated advanced information management systems/ or knowledge bases/ or management information systems/ or ambulatory care information systems/ or clinical pharmacy information systems/ or database management systems/ or decision support systems, management/ or medical records systems, computerized/ or reminder systems/ (29392) 7 databases/ or databases, factual/ (25911) 8 computer simulation/ or computer systems/ or computer communication networks/ (73001) 9 decision support techniques/ or data interpretation, statistical/ or decision trees/ (35950) 10 systems analysis/ or operations research/ or systems integration/ (9622) 11 data collection/ or death certificates/ or hospital records/ or medical records/ or medical records systems, computerized/ (93920) 12 vital statistics/ (3604) 13 morbidity/ or incidence/ or prevalence/ or mortality/ or "cause of death"/ or child mortality/ or fatal outcome/ or hospital mortality/ or infant mortality/ or maternal mortality/ or survival rate/ (369637) 14 decision$.mp. (142670) 15 expert$.mp. (52741) 16 computer$.mp. (334061) 17 informatic$.mp. (8533) 18 information system$.mp. (37270) 19 or/1-18 (990362) 20 Disease Outbreaks/ (42573) 21 Disease Reservoirs/ (9659) 22 Disease Transmission/ (1165) 23 Environmental Medicine/ (265) 24 Environmental Microbiology/ (2738) 25 Environmental Monitoring/ (29649) 26 Inhalation Exposure/ (2600) 27 Food Contamination/ (18165) 28 Communicable Disease Control/ (12815) 29 Mandatory Reporting/ (1190) 30 disease management/ (4987) 31 disease notification/ (2052) 32 population surveillance/ or sentinel surveillance/ (28870) 33 epidemiologic methods/ (20117) 34 health care surveys/ or interviews/ or questionnaires/ or incidence/ or prevalence/ (353762) 35 community health planning/ (3176) 36 disaster planning/ (6213) 37 Health Plan Implementation/ (1599) 38 public health practice/ or communicable disease control/ (14760) 39 disease notification/ (2052) 40 sanitation/ or food inspection/ (6229) 41 universal precautions/ or environmental monitoring/ (30852) 42 primary prevention/ (8896) 43 veterinary medicine/ (15450) 44 control$.mp. (1835160) 45 response.mp. (1118156) 46 prevent$.mp. (566536) 47 early warning.mp. (1143) 48 threat$.mp. (59068) 49 agrobioterrorism.mp. (1) 50 (bio-surveillance or biosurveillance).mp. (29) 51 outbreak$.mp. (55709) 52 monitor$.mp. (350590) 53 detect$.mp. (971613) 54 surveillance$.mp. (72819) 55 alert$.mp. (15727) 56 contaminat$.mp. (94199) 57 exposure$.mp. (381229) 58 emergenc$.mp. (158314) 59 diagnos$.mp. (1252022) 60 notification.mp. (5509) 61 or/20-60 (5438486) 62 Communication/ (42166) 63 dialogue.mp. (4248) 64 Communication Barriers/ (2659) 65 Cooperative Behavior/ (10892) 66 (data adj3 shar$).mp. (1050)67 (ownership adj3 data).mp. (65) 68 Program Development/ (11917) 69 consensus/ (1704) 70 Decision Making/ (41957) 71 dynamic environment$.mp. (291) 72 Information Dissemination/ (3665) 73 "diffusion of innovation"/ or technology transfer/ (7708) 74 interdisciplinary communication/ (2484) 75 Interprofessional Relations/ (32153) 76 International Cooperation/ (27287) 77 Internationality/ (6771) 78 cross-disciplinary.mp. (220) 79 (interstate or inter-state).mp. (528) 80 Public Health Administration/ (11818) 81 systems integration/ (4513) 82 multi-institutional systems/ or hospital shared services/ (7981) 83 "Decision Support Systems, Management"/ (763) 84 Management Information Systems/ (3318) 85 infrastructure.mp. (4971) 86 ((corporate or organization$ or health unit$) adj10 plan$).mp. (10476) 87 "Organization and Administration"/ (13938) 88 ((polic$ or decision) adj5 maker$).mp. (6565) 89 network$.mp. (106344) 90 hierarchy.mp. (6907) 91 authority.mp. (8296) 92 formali?ation.mp. (356) 93 codification.mp. (325) 94 jurisdiction.mp. (958) 95 (coordination adj5 activit$).mp. (868) 96 (coordination adj10 system$).mp. (1124) 97 (coordination or coordination).mp. (25076) 98 government/ or federal government/ or "united states department of agriculture"/ or "united states dept. of health and human services"/ or "united states centers for medicare and medicaid 218 services"/ or united states public health service/ or "centers for disease control and prevention (u.s.)"/ or "national institute for occupational safety and health"/ or national center for health care technology/ or "national center for health statistics (u.s.)"/ or "national institutes of health (u.s.)"/ or "united states agency for healthcare research and quality"/ or "united states food and drug administration"/ or exp "united states health resources and services administration"/ or united states indian health service/ or "united states office of research integrity"/ or united states environmental protection agency/ or united states government agencies/ or "united states occupational safety and health administration"/ or local government/ or state government/ or government programs/ (76066) 99 Confidentiality/ (15579) 100 (cross-disciplinar$ or crossdisciplinar$).mp. (228) 101 (interdisciplinar$ or interdisciplinar$).mp. (12712) 102 ((law$ or regulation$ or rule$) adj20 (observance$ or adherence$ or enforce$)).mp. (4313) 103 or/62-102 (457139) 104 19 and 61 and 103 (52172) 105 aeromonas/ (2619) 106 caviae.mp. and aeromonas/ (382) 107 aeromonas caviae.mp. (206) 108 or/105-107 (2644) 109 Aeromonas hydrophila/ (622) 110 hydrophila.mp. and aeromonas/ (944) 111 or/109-110 (1506) 112 veronii.mp. and aeromonas/ (170) 113 sobria.mp. and aeromonas/ (426) 114 (aeromonas veronii or aeromonas sobria).mp. (288) 115 or/112-114 (512) 116 Anisakiasis/ (273) 117 Anisakis/ (298) 118 simplex.mp. and (Anisakis/ or Anisakiasis/) (228) 119 Anisakis simplex.mp. (297) 120 or/116-119 (473) 121 Anthrax/ (2592) 122 Bacillus anthracis/ (2215) 123 or/121-122 (3897) 124 Argentine hemorrhagic fever.mp. (148) 125 Hemorrhagic Fever, American/ (340) 126 Junin virus/ (69) 127 Arenaviruses, New World/ (415) 128 or/124-127 (597) 129 Tuberculosis, Avian/ (393) 130 Mycobacterium avium/ (1927) 131 or/129-130 (2240) 132 lyssavirus/ (161) 133 ((pteropid or bat$) adj5 (virus$ or lyssavirus$)).mp. (556) 134 Rhabdoviridae Infections/ (515) 135 136 137 138 139 or/132-134 (1019) Tuberculosis, Bovine/ (1707) Mycobacterium bovis/ (6465) or/136-137 (7575) Hemorrhagic Fever, American/ (340) 140 Hemorrhagic Fevers, Viral/ (1022) 141 Arenaviruses, New World/ (415) 142 Machupo virus$.mp. (50) 143 (bolivian adj5 fever).mp. (43) 144 or/139-143 (1506) 145 Botulism/ (2145) 146 Clostridium botulinum/ (1996) 147 or/145-146 (3602) 148 Encephalopathy, Bovine Spongiform/ (1876) 149 prions/ or prpc proteins/ or prpsc proteins/ or prp 27-30 protein/ (6537) 150 or/148-149 (7809) 151 arenavirus/ or lassa virus/ or lymphocytic choriomeningitis virus/ or arenaviruses, new world/ or junin virus/ or pichinde virus/ (2518) 152 (sabia adj5 virus$).mp. (14) 153 Arenaviridae Infections/ (197) 154 Hemorrhagic Fevers, Viral/ (1022) 155 or/151-154 (3527) 156 Brucellosis/ (7007) 157 Brucellosis, Bovine/ (1598) 158 or/156-157 (8199) 159 Brucella melitensis/ (616) 160 Malta fever.mp. (85) 161 Undulant fever.mp. (54) 162 or/159-161 (743) 163 or/158,162 (8353) 164 Plague/ (3176) 165 Yersinia pestis/ (2203) 166 Yersinia Infections/ (2647) 167 black death.mp. (107) 168 bubonic plague.mp. (240) 169 or/164-168 (7181) 170 campylobacter/ or campylobacter coli/ or campylobacter fetus/ or campylobacter hyointestinalis/ or campylobacter jejuni/ or campylobacter lari/ or campylobacter rectus/ or campylobacter sputorum/ or campylobacter upsaliensis/ (7058) 171 Campylobacterosis.mp. (6) 172 Campylobacter Infections/ (4585) 173 or/170-172 (8184) 174 Herpesvirus 1, Cercopithecine/ (211) 175 exp Herpesviridae Infections/ (78911) 176 174 and 175 (145) 177 or/174,176 (211) 178 Cholera/ (5706) 179 vibrio cholerae/ or vibrio cholerae non-o1/ or vibrio cholerae o1/ or vibrio cholerae o139/ (5464) 180 or/178-179 (9410) 181 Cryptosporidiosis/ (3130) 182 Cryptosporidium hominis.mp. (56) 183 cryptosporidium/ or cryptosporidium parvum/ (3144) 184 or/181-183 (4315) 185 Hemorrhagic Fever, Ebola/ (464) 186 Ebolavirus/ (616) 187 or/185-186 (826) 188 echinococcosis/ or echinococcosis, hepatic/ or echinococcosis, pulmonary/ (12900) 189 echinococcus/ or echinococcus granulosus/ or echinococcus multilocularis/ (2459) 190 or/188-189 (13480) 191 Giardiasis/ (3659) 192 giardia/ or giardia lamblia/ (2668) 193 Giardiavirus/ (15) 194 Lamblia intestinalis.mp. (54) 195 (Giardia adj3 (duodenalis or intestinalis)).mp. (692) 196 or/191-195 (5258) 197 Hantavirus Pulmonary Syndrome/ (317) 198 Sin Nombre virus/ (43) 199 ((andes or sin nuombre or laguna negra) adj5 (virus or hantavirus)).mp. (69) 200 Hantavirus/ (1473) 201 197 and 200 (163) 202 or/197-199,201 (362) 203 Hendra Virus/ (30) 204 Henipavirus Infections/ (60) 205 Paramyxoviridae Infections/ (1831) 206 or/203-205 (1907) 207 206 and (hendra or hemorrhagic or bronchopneumonia).mp. (104) 208 or/203-204,207 (146) 209 escherichia coli/ or escherichia coli o157/ (183629) 210 Hemorrhagic colitis.mp. (478) 211 209 and 210 (287) 212 Escherichia coli Infections/ (18672) 213 210 and 212 (203) 214 or/211,213 (318) 215 Hemorrhagic Fever with Renal Syndrome/ (1753) 216 Hantaan virus/ (227) 217 Seoul virus/ (25) 218 Puumala virus/ (86) 219 or/215-218 (1857) 220 Hemolytic-Uremic Syndrome/ (3412) 221 Escherichia coli/ (180844) 222 220 and 221 (378) 223 or/220,222 (3412) 224 Histoplasmosis/ (4264) 225 Histoplasma/ (1667) 226 Histoplasma capsulatum.mp. (1711) 227 Ajellomyces capsulatus.mp. (9) 228 or/225-227 (2355) 229 224 and 228 (1386) 230 or/224,229 (4264) 231 Monkeypox/ (68) 232 Monkeypox virus/ (215) 233 or/231-232 (248) 219 234 Influenza, Human/ (17375) 235 influenza a virus/ or influenza a virus, h1n1 subtype/ or influenza a virus, h2n2 subtype/ or influenza a virus, h3n2 subtype/ or influenza a virus, h3n8 subtype/ or influenza a virus, h5n1 subtype/ or influenza a virus, h5n2 subtype/ or influenza a virus, h7n7 subtype/ or influenza a virus, h9n2 subtype/ (13120) 236 or/234-235 (26099) 237 Lassa Fever/ (352) 238 Lassa virus/ (340) 239 or/237-238 (551) 240 Listeria monocytogenes/ (6853) 241 listeria infections/ or meningitis, listeria/ (5296) 242 Listeriosis.mp. (2027) 243 240 and 242 (946) 244 241 and 242 (1763) 245 or/243-244 (1915) 246 Otitis Externa/ (1721) 247 Dogs/ (248573) 248 246 and 247 (200) 249 pachydermatis.mp. (187) 250 Dermatitis, Seborrheic/ (1905) 251 or/248-250 (2257) 252 Malassezia/ (1045) 253 251 and 252 (337) 254 Marburg Virus Disease/ (234) 255 Marburgvirus/ (243) 256 or/254-255 (371) 257 Paramyxoviridae/ (874) 258 Paramyxoviridae Infections/ (1831) 259 measles/ or subacute sclerosing panencephalitis/ or mumps/ (13395) 260 258 not 259 (1773) 261 menangle.mp. (17) 262 or/260-261 (1786) 263 or/257,262 (2621) 264 Trematode Infections/ (2605) 265 Opisthorchidae/ (27) 266 Metorchis.mp. (39) 267 Metorchiasis.mp. (5) 268 or/264-267 (2640) 269 microsporidiosis/ or encephalitozoonosis/ (868) 270 Cerebral Microsporidiosis.mp. (2) 271 Encephalitozoon cuniculi/ (275) 272 Encephalitozoon hellem.mp. (112) 273 Encephalitozoon intestinalis.mp. (139) 274 Enterocytozoon bieneusi.mp. (296) 275 Nosema connori.mp. (5) 276 Trachipleistophora hominis.mp. (15) 277 Encephalitozoon/ (238) 278 or/270-277 (775) 279 or/269,278 (1181) 280 Nipah Virus/ (84) 281 Henipavirus Infections/ (60) 282 or/280-281 (92) 283 Penicilliosis.mp. (148) 284 Penicillium marneffei.mp. (284) 285 Penicilliosis.mp. and Penicillium/ (113) 286 or/284-285 (326) 287 or/283,286 (356) 288 Picobirnavirus/ (23) 289 Picobirnavirus$.mp. (41) 290 or/288-289 (41) 291 RNA Virus Infections/ (211) 292 290 and 291 (10) 293 or/290,292 (41) 294 Taenia solium/ (259) 295 taeniasis/ or cysticercosis/ or neurocysticercosis/ (4931) 296 or/294-295 (4971) 297 Rabies/ (6533) 298 Rabies virus/ (2845) 299 or/297-298 (7743) 300 salmonella infections/ or paratyphoid fever/ or salmonella food poisoning/ or typhoid fever/ (18322) 301 salmonella enteritidis/ or salmonella typhi/ or salmonella typhimurium/ (27788) 302 or/300-301 (41155) 303 SARS Virus/ (1604) 304 Severe Acute Respiratory Syndrome/ (3056) 305 or/303-304 (3624) 306 Strongyloidiasis/ (2545) 307 Strongyloides stercoralis/ (519) 308 or/306-307 (2596) 309 Granuloma/ (15917) 310 swim$ pool$.mp. (1521) 311 Swimming Pools/ (1079) 312 or/310-311 (1521) 313 309 and 312 (35) 314 Mycobacterium marinum/ (211) 315 or/313-314 (244) 316 toxoplasmosis/ or toxoplasmosis, animal/ or toxoplasmosis, cerebral/ or toxoplasmosis, ocular/ (12109) 317 Toxoplasma/ (6978) 318 gondii.mp. (6814) 319 317 and 318 (4896) 320 or/316,319 (14138) 321 Trichinellosis.mp. (757) 322 Trichinosis/ (3755) 323 321 and 322 (706) 324 trichiniasis.mp. (29) 325 or/323-324 (735) 326 Trichinella spiralis/ (793) 327 or/325-326 (1395) 328 Tularemia/ (1907) 329 Francisella tularensis/ (1152) 330 or/328-329 (2462) 331 Hemorrhagic Fever, American/ (340) 332 Hemorrhagic Fevers, Viral/ (1022) 333 or/331-332 (1335) 334 Venezuela/ (2975) 335 venezuela$.mp. (5364) 336 or/334-335 (5364) 337 333 and 336 (17) 338 ((Venezuela$ adj3 hemorrhagic) and fever).mp. (16) 339 or/337-338 (22) 340 Arenaviruses, New World/ (415) 341 Guanarito$.mp. (23) 342 or/340-341 (425) 343 Vibrio parahaemolyticus/ (1103) 344 Vibrio Infections/ (1616) 345 parahaemolyticus.mp. (1728) 346 or/343,345 (1728) 347 344 and 346 (378) 348 or/343,347 (1183) 349 or/343,348 (1183) 350 Vibrio vulnificus/ (215) 351 Vibrio Infections/ (1616) 352 vulnificus.mp. (864) 353 or/350,352 (864) 354 351 and 353 (397) 355 or/350,354 (507) 356 Yersinia enterocolitica/ (2847) 357 Yersiniosis.mp. (428) 358 Yersinia Infections/ (2647) 359 356 and 357 (210) 360 357 and 358 (338) 361 or/359-360 (356) 362 or/108,111,115,120,122,128,131,135,138,1 44,147,150,155,163,169 (46590) 363 or/173,177,180,184,187,190,196,202,208,2 14,219,223 (46911) 364 or/230,233,236,239,245,253,256,263,268,2 79 (39759) 365 or/282,287,293,296,299,302,305,308,315,3 20 (74667) 366 or/325,327,330,333,337,339,342,349,355,3 61 (7323) 367 or/362-366 (208216) 368 Rickettsia/ (1733) 369 Ticks/ (9097) 370 or/368-369 (10530) 371 africae.mp. (74) 372 370 and 371 (60) 373 African tick typhus.mp. (4) 374 African tick-bite fever.mp. (63) 375 Rickettsia Infections/ (1970) 376 africae.mp. (74) 377 375 and 376 (45) 378 Tick-Borne Diseases/ (668) 379 Rickettsia/ (1733) 380 378 and 379 (78) 381 or/372-374,377,380 (154) 382 Babesiosis/ (2393) 383 Babesia microti/ (73) 384 Hemorrhagic Fever Virus, Crimean-Congo/ (244) 385 Crimean-Congo hemorrhagic fever.mp. (197) 386 or/384-385 (302) 387 Anaplasma phagocytophilum/ (271) 388 Ehrlichiosis/ (1221) 389 (phagocytophilum or phagocytophilia).mp. (353) 390 granulocytic.mp. (6195) 391 or/389-390 (6413) 392 388 and 391 (554) 393 (human adj5 anaplasmosis).mp. (76) 394 (granulocytic EHRLICHIOSIS and human).mp. (424) 395 Sennetsu Fever.mp. (2) 220 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 or/392-395 (667) or/387,396 (728) Ehrlichia/ (945) ewingii.mp. (53) 398 and 399 (41) Ehrlichia ewingii.mp. (34) or/399-401 (53) Ehrlichiosis/ (1221) Humans/ (9619411) human$.mp. (9824443) 404 or 405 (9824443) 403 and 406 (810) Human ehrlichiosis.mp. (146) or/407-408 (847) or/402,409 (869) Ehrlichia chaffeensis/ (274) monocytic ehrlichiosis.mp. (184) 413 humans/ or human$.mp. (9824443) 414 412 and 413 (118) 415 monocytic.mp. (10533) 416 Ehrlichiosis/ (1221) 417 415 and 416 (176) 418 or/414,417 (195) 419 or/411,418 (383) 420 Lyme Disease/ (6297) 421 Encephalitis, Tick-Borne/ (1904) 422 Borrelia burgdorferi/ (1198) 423 421 and 422 (12) 424 Encephalitis Viruses, TickBorne/ (1887) 425 420 and 421 (106) 426 Tickborne encephalitis.mp. (67) 427 420 and 422 (818) 428 or/420-427 (9499) 429 Hemorrhagic Fevers, Viral/ (1022) 430 Tick-Borne Diseases/ (668) 431 Flavivirus Infections/ (173) 432 or/429-431 (1852) 433 Kyasanur.mp. (129) 434 432 and 433 (20) 435 Kyasanur Forest Disease/ (38) 436 Monkey disease.mp. (21) 437 Kyasanur Forest disease.mp. (126) 438 or/434-437 (149) 439 (Kyasanur adj10 virus$).mp. (78) 440 Flavivirus/ (790) 441 Kyasanur.mp. (129) 442 440 and 441 (27) 443 or/439,442 (87) 444 or/438,443 (149) 445 Bartonella quintana/ (166) 446 Trench Fever/ (135) 447 (Wolhynia adj5 fever$).mp. (1) 448 (quintan adj5 fever$).mp. (2) 449 (trench adj5 fever$).mp. (182) 450 or/446-449 (183) 451 or/445,450 (269) 452 or/381,386,397,410,419,428,438,444,451 (11368) 453 bagaza.mp. (4) 454 Flavivirus/ (790) 455 ntaya.mp. (16) 456 454 and 455 (2) 457 or/455-456 (16) 458 or/453,457 (20) 459 coltivirus/ or colorado tick fever virus/ (57) 460 Reoviridae Infections/ (1462) 461 banna.mp. (42) 462 460 and 461 (4) 463 or/461-462 (42) 464 459 and 461 (7) 465 or/463-464 (42) 466 or/459,465 (92) 467 Alphavirus Infections/ (455) 468 barmah.mp. (70) 469 467 and 468 (35) 470 (Barmah and virus$).mp. (66) 471 Alphavirus/ (550) 472 barmah.mp. (70) 473 471 and 472 (38) 474 or/470,472 (70) 475 or/469-470,472-473 (70) 476 Encephalitis, California/ (256) 477 encephalitis virus, california/ or la crosse virus/ (492) 478 or/476-477 (545) 479 Chikungunya virus/ (466) 480 Alphavirus Infections/ (455) 481 Chikungunya.mp. (561) 482 480 and 481 (88) 483 481 or 482 (561) 484 or/479,483 (561) 485 dengue/ or dengue hemorrhagic fever/ (3574) 486 Dengue Virus/ (2609) 487 or/485-486 (4657) 488 Encephalomyelitis, Eastern Equine/ (29) 489 Encephalitis Virus, Eastern Equine/ (306) 490 or/488-489 (316) 491 filariasis/ or elephantiasis, filarial/ (6003) 492 Wuchereria bancrofti/ (1609) 493 or/491-492 (6165) 494 Orthobunyavirus/ (70) 495 guama.mp. (29) 496 Bunyaviridae Infections/ (381) 497 495 and 496 (2) 498 or/494-495 (92) 499 Encephalitis, Japanese/ (1605) 500 Encephalitis Virus, Japanese/ (1177) 501 or/499-500 (2216) 502 leptospirosis/ or weil disease/ (5295) 503 Leptospira interrogans/ (1226) 504 or/502-503 (5778) 505 malaria/ (24885) 506 malaria, avian/ (356) 507 malaria, cerebral/ (947) 508 malaria, falciparum/ (7894) 509 blackwater fever/ (68) 510 malaria, vivax/ (1513) 511 or/505-510 (33962) 512 plasmodium/ (5070) 513 plasmodium falciparum/ (15320) 514 plasmodium malariae/ (676) 515 plasmodium ovale/ (37) 516 plasmodium vivax/ (2470) 517 or/512-516 (21420) 518 or/511,517 (42008) 519 Mayaro virus fever.mp. (1) 520 Alphavirus/ (550) 521 Monkey Diseases/ (3517) 522 mayaro.mp. (65) 523 520 and 522 (37) 524 521 and 522 (2) 525 or/519,522-524 (65) 526 Encephalitis Virus, Murray Valley/ (74) 527 Encephalitis, Arbovirus/ (1615) 528 (murray adj5 valley).mp. (275) 529 527 and 528 (104) 530 Australian encephalitis.mp. (11) 531 Murray Valley encephalitis.mp. (238) 532 or/529-531 (257) 533 or/526,532 (261) 534 Alphavirus/ (550) 535 Alphavirus Infections/ (455) 536 O'Nyong-nyong.mp. (53) 537 534 and 536 (21) 538 535 and 536 (14) 539 or/536-538 (53) 540 Oropouche.mp. (39) 541 Bunyaviridae/ (781) 542 Bunyaviridae Infections/ (381) 543 540 and 541 (5) 544 540 and 542 (19) 545 or/540,543-544 (39) 546 Rift Valley Fever/ (454) 547 Rift Valley fever virus/ (350) 548 or/546-547 (586) 549 Ross River virus/ (263) 550 (ross river adj5 (virus$ or disease or fever or polyarthritis)).mp. (342) 551 ALPHAVIRUS INFECTIONS/ (455) 552 550 and 551 (113) 553 or/550,552 (342) 554 or/549,553 (342) 555 Dysentery, Bacillary/ (5986) 556 Shigella dysenteriae/ (1474) 557 or/555-556 (6883) 558 Sindbis Virus/ (1709) 559 Alphavirus Infections/ (455) 560 sindbis$.mp. (2226) 561 559 and 560 (119) 562 or/558,561 (1720) 563 "Encephalitis, St. Louis"/ (393) 564 "Encephalitis Virus, St. Louis"/ (280) 565 or/563-564 (543) 566 Encephalomyelitis, Venezuelan Equine/ (316) 567 Encephalitis Virus, Venezuelan Equine/ (787) 568 or/566-567 (893) 569 Encephalomyelitis, Western Equine/ (10) 570 Encephalitis Virus, Western Equine/ (391) 571 or/569-570 (395) 572 Flavivirus/ (790) 573 Flavivirus Infections/ (173) 574 Wesselsbron.mp. (71) 575 572 and 574 (19) 576 573 and 574 (6) 577 or/574-576 (71) 221 578 West Nile Fever/ (1605) 579 West Nile virus/ (1754) 580 or/578-579 (2232) 581 Yellow Fever/ (1852) 582 Yellow fever virus/ (763) 583 or/581-582 (2293) 584 Flavivirus/ (790) 585 Flavivirus Infections/ (173) 586 zika.mp. (53) 587 584 and 586 (15) 588 585 and 586 (1) 589 or/586-588 (53) 590 Phlebotomus Fever/ (156) 591 Phlebovirus/ (128) 592 (naples adj20 virus$).mp. (75) 593 (sandfly
UBC Theses and Dissertations
Use of animal data in public health surveillance for emerging zoonotic diseases Vrbova, Linda 2013
Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.
- 24-ubc_2013_spring_vrbova_linda.pdf [ 4.54MB ]
- JSON: 24-1.0073819.json
- JSON-LD: 24-1.0073819-ld.json
- RDF/XML (Pretty): 24-1.0073819-rdf.xml
- RDF/JSON: 24-1.0073819-rdf.json
- Turtle: 24-1.0073819-turtle.txt
- N-Triples: 24-1.0073819-rdf-ntriples.txt
- Original Record: 24-1.0073819-source.json
- Full Text