Improving rabies control in free-roaming dogs Michelle Kimberley Morters This dissertation is submitted for the degree of Doctor of Philosophy Wolfson College and Department of Veterinary Medicine, University of Cambridge Date of submission: 31st March, 2014 ii Abstract Canine-mediated rabies is a serious zoonosis responsible for at least 55,000 human deaths every year, primarily in less developed communities in Asia and Africa where domestic dogs are free-roaming. The disease can be effectively controlled through vaccinating at least 45% of the dogs in a population; however the impact of vaccinations on disease incidence may be affected by dynamic demographic and immunological processes. Specifically, the contribution of these processes, and their regulatory factors, to vaccination coverage and rabies transmission has not been comprehensively estimated. To improve rabies control, through field interventions and epidemiological modelling, more information regarding the effect of these processes, and their regulatory factors, on population and disease dynamics and vaccinal responses was needed. This required a multifaceted approach, using techniques from the fields of population ecology, vaccine-immunology, social science and epidemiological modelling. Demographic data were collected from four populations of free- roaming domestic dogs, two in South Africa and two in Indonesia where rabies is endemic. Longitudinal, individual-level data were obtained by direct observation and surveys, and community-level data by participatory methods. Longitudinal, serological data were collected from three cohorts within the populations. Epidemiological models were based on epidemic theory and empirical data from this current study and previous studies. A wide array of data were generated relevant to planning rabies control programmes, however of particular importance was evidence regarding positive and negative the impacts of human factors on population and disease dynamics. Nearly all of the dogs were owned, despite being free-roaming, and were accessible for vaccination through their owners; and population size was regulated through human demand for dogs and a substantial fraction of dogs was acquired from outside the communities. These translocated dogs may contribute to the spread of rabies, necessitating widespread and sustained vaccination programmes. Considerable differences in the handleability of dogs between locations and, thus ease of vaccine delivery, may also be attributable to differences in human-dog interactions. Finally, a critical review of the literature, and evaluation of epidemiological models, suggests that human interference in the transmission processes may reduce the incidence of rabies and vaccination threshold. This study has provided specific evidence that human behaviours are likely to be critically important in relation to the transmission and control of canine-mediated rabies – and is the first to properly identify this. Further detailed studies are required to explore these behaviours iii and how they vary culturally and geographically. In addition, the results highlight the critical role that demographic processes more generally, as well as immunological decay, play in influencing the long term success of rabies vaccination programmes. Overall, this research has provided valuable support for planning rabies control programmes and for parameterisation of epidemiological models of infectious diseases, including rabies. iv Acknowledgements This thesis is dedicated to my field teams and the communities. To my teams, thank you for working the long hours beside me – without your loyalty and dedication this thesis wouldn’t exist. Special thanks to David Khoza, Ni Nyoman Ari Puspa Dewi (Puspa), Putu Astri Wulandari Ernawan (Astri), I Dewa Gede Suyadnya (Dewa) and Ida Ayu Suarini Pramini (Dayu) – you have been my backbone in the field and I can’t thank you enough. To Cora Bailey, Sherry Grant and Luh Putu Listriani Wistawan (Lis), thank you for your incredible generosity, trusting me and throwing me the ball – I’m indebted to you for this opportunity. To the community leaders and residents of Zenzele, Braamfischerville, Kelusa and Antiga – I asked so much of you – thank you for being so gracious and generous and for embracing this project. Particular thanks to Andrew Conlan, TJ McKinley and Olivier Restif for your endless patience answering my steady stream of questions – without your technical support I’d still be in the starting blocks. Sincere thanks to James Wood, Sarah Cleaveland and Becky Whay for your tireless guidance and support throughout this very long process. Thank you very much to I Made Darmiyasa, Pak Adi, Johan Schoeman, Amelia Goddard, Louis Nel and Volker Schwann for your in-country support. Thanks to Tony Fooks, Dan Horton, Katie Hampson, Sunny Townsend, Sowmyaa Bharadwaj, Helen Dagut, Angela Wood, Darryn Knobel, Alex Wandeler, Sinead Sheerin and Meg Staff - your input has been invaluable. To all the crew at the labs including Trudy Goddard, Dave Seldon, Emma Wise, Hooman Goharriz, Cheryl Booth, Carien Muller, Gertie Pretorius - thank you for going above and beyond the call of duty. To Sheila, Ellen, Peter and Jan, the little, everyday things you did made a big difference especially when the work load was heavy – thank you. Immense thanks to my funders the International Fund for Animal Welfare and the World Society for the Protection of Animals, particularly Kate Atema and Elly Hiby for your continued support through thick and thin – I’m extremely grateful. To the countless others who have helped me along with way including Ray, Niela, Alf, Shannon, Lorraine, Kevin, Janice, Rai, Kadek, Dayu (vet), Martine, Nana, Wayan, Arya, Ketut, McDonald, Toby, Elijah, Sam, Nthabiseng and Pancho – thank you. Finally, to my family, I thank God for you and for Skype! - thank you for being there. v Table of Collaborators Individual providing assistance Organization Contribution Prof James Wood University of Cambridge PhD supervisor Prof Sarah Cleaveland University of Glasgow PhD advisor Dr Helen Whay University of Bristol PhD advisor Prof Johan Schoeman University of Pretoria Collaborator for studies in South Africa Prof I Made Damriyasa Udayana University (Bali) Collaborator for studies in Indonesia Sherry Grant, Yudisthira Foundation and Legal entity for research permits Luh Putu Listriani Wistawan Indonesian Animal Welfare (InAW) Data collection under MMs supervision CLAW employees Community Led Animal Welfare Translation, enumeration, dog handling (NGO, Johannesburg) InAW employees Indonesian Animal Welfare Translation, enumeration, dog handling, (NGO, Bali) vaccination, facilitation of the PRA BAWA employees Bali Animal Welfare Association Assistance with facilitation of PRA (NGO, Bali) Yudisthira employees Yudisthira Foundation Assistance with facilitation of PRA (NGO, Bali) Sowmyaa Bharadwaj Praxis Institute for Participatory Development of PRA exercises, team Practices (India) training and facilitation of PRA Laboratory analysis Prof Amelia Goddard, University of Pretoria Serum sample storage, complete cell Cheryl Booth, Gertie Pretorius, counts and microscopy Carien Muller Dr Volker Schwann University of Pretoria Faecal analysis Prof Anthony Fooks, Weybridge Animal Health Dr Daniel Horton, Veterinary Laboratory Agency FAVN assays (pipetting, plate Trudy Goddard, David Seldon, (UK) preparation and reading), LBV assays Hooman Goharriz Data analysis Dr TJ McKinley University of Cambridge Provided code for Monte Carlo estimates of early pregnancy, developed and fitted ordinal regression models (Chapter 2) Dr Andrew Conlan University of Cambridge Modification of formula to estimate R0 (Chapters 1 & 5 equations 1-3); provided the model code used in Chapter 5 Co-authors on papers (chapters) Prof James Wood; Dr TJ McKinley; see above Thoroughly reviewed the manuscripts; Dr Andrew Conlan; Dr Olivier Restif; provided extensive technical advice and Prof Sarah Cleaveland; guidance specific to each manuscript Dr Katie Hampson; Dr Helen Whay; [in particular OR & AC Chapter 1; Prof Anthony Fooks; Dr Daniel Horton; TJM Chapters 2 & 4; SW Chapter 3; Prof Johan Schoeman; AC Chapter 5] Prof Amelia Goddard; Prof I Made Damriyasa; Sowmyaa Bharadwaj vi Table of Contents ABSTRACT ii ACKNOWLEDGEMENTS iv TABLE OF COLLABORATORS v TABLE OF CONTENTS vi LIST OF TABLES ix LIST OF FIGURES ix LIST OF APPENDCIES x LIST OF ABBREVIATIONS xiv LIST OF SYMBOLS xv PREFACE xvi CHAPTER 1 Evidence-based control of canine rabies: a critical review of population density reduction 1.1 Introduction 1 1.2 Evidence for density-dependent transmission of rabies 3 1.2.1 Cycles in disease incidence 4 1.2.2 The relationship between R0 and host density 6 1.2.3 Thresholds for invasion and increasing incidence with population density 8 1.2.4 Impacts of density reduction 9 1.3 Conclusion 12 CHAPTER 2 The demography of free-roaming dog populations and applications to disease and population control 2.1 Introduction 15 2.2 Methods and materials 2.2.1 Research sites 17 2.2.2 Method and type of data collected 18 2.2.3 Analytical methods 21 vii 2.3 Results 2.3.1 Study population 22 2.3.2 Population size 22 2.3.3 Demographic processes 24 2.3.4 Declines in the starting cohorts 27 2.3.5 Ownership 29 2.3.6 Food sources 30 2.4 Discussion 30 2.5 Conclusion 33 CHAPTER 3 Participatory methods for the assessment of the ownership status of free-roaming dogs in Bali, Indonesia, for disease control and animal welfare 3.1 Introduction 34 3.2 Methods and materials 37 3.3 Results 39 3.4 Discussion 41 3.5 Conclusion 43 CHAPTER 4 Achieving population-level immunity to rabies in free-roaming dogs in Africa and Asia 4.1 Introduction 44 4.2 Methods and materials 4.2.1 Study populations 47 4.2.2 Vaccination and sampling 49 4.2.3 Covariates 51 4.2.4 Analytical methods 52 4.3 Results 4.3.1 General description of the study populations 56 4.3.2 Vaccination coverage 56 4.3.3 Assessment of prior vaccinations 56 4.3.4 Evaluation of antibody titres 57 4.4 Discussion 62 4.5 Conclusion 64 viii CHAPTER 5 Modelling human interference and rabies transmission in free-roaming dogs 5.1 Introduction 65 5.2 Methods and materials 5.2.1 Demographic data 69 5.2.2 Model structure and parameters 71 5.2.3 Model scenarios 74 5.3 Results 5.3.1 Population characteristics 76 5.3.2 Comparison of outbreak dynamics, R0 , and transmission rates 77 5.3.3 Probability of an outbreak with vaccination coverage 79 5.3.4 Comparison of outbreak dynamics, human-mediated incursions and 81 vaccination coverage 5.4 Discussion 83 5.5 Conclusion 86 CHAPTER 6 General discussion 87 REFERENCES 100 APPENDICES Chapter 2 121 Chapter 3 156 Chapter 4 164 Chapter 5 197 Chapter 6 203 Chapter 7 205 ix List of Tables Table 2.1 Sources of the dogs registered in Zenzele, Braamfischerville, Kelusa and Antiga. Table 2.2 Outcomes of the dogs registered in Zenzele, Braamfischerville, Kelusa and Antiga. Table 2.3 Mean estimates of vaccination coverage with time in Zenzele, Braamfischerville, Kelusa and Antiga. Table 2.4 Declines in the number of dogs in the starting cohorts in Zenzele, Braamfischerville, Kelusa and Antiga by age class and gender. Table 3.1 Summary of the three highest ranked food sources for Antiga. Table 3.2 Summary of the three highest ranked food sources for Kelusa. Table 4.1 The number of dogs in Zenzele, Kelusa and Antiga that were blood sampled during the study period. Table 5.1 Transitions and their probabilities in the individual-based and compartmental models. Table 5.2 Demographic events and their probabilities in the compartmental model. Table 5.3 Demographic parameter estimates for Zenzele, Braamfischerville, Kelusa and Antiga. Table 5.4 Population size and density in Zenzele, Braamfischerville, Kelusa and Antiga. List of Figures Figure 2.1 Variations in the size of the registered population in Zenzele, Braamfischerville, Kelusa and Antiga for the study period. Figure 2.2 Declines in the starting cohorts in Zenzele, Braamfischerville, Kelusa and Antiga for the study period. Figure 4.1 Rabies vaccine-induced virus neutralizing antibody (VNA) titres of all the dogs in the Zenzele research cohort that were blood sampled at four time points during the study period. Figure 4.2 Rabies vaccine-induced VNA titres of all the dogs in the Zenzele research cohort blood sampled during the study period. Figure 4.3 Rabies vaccine-induced VNA titres of all the dogs in the Kelusa and Antiga research cohorts blood sampled during the study period. Figure 5.1 Outbreak size in Zenzele, Braamfischerville, Kelusa, Kaler and Ketug for a range of R0 and transmission rates for the individual-based model. Figure 5.2 Probability of an outbreak in Kelusa at the time of vaccination and 12 and 24 months after vaccination for a range of vaccination coverage and transmission rates. Figure 5.3 Probability of outbreaks with human-mediated incursions for a range of vaccination coverage and transmission rates in Kelusa. Figure 5.4 Outbreak size with human-mediated incursions for a range of vaccination coverage and transmission rates in Kelusa. x List of Appendices Chapter 2 The demography of free-roaming dogs and applications to disease and population control Appendix 2.1 Development and implementation of the questionnaire used to collect individual-dog and household data in the study areas. Appendix 2.2 Characteristics of the households in the study areas. Appendix 2.3 Sex ratios of the registered populations in the study areas. Appendix 2.4 Owner reported litter sizes for the study populations, and litter sizes determined during sterilization of pregnant bitches elsewhere in Johannesburg and Bali. Appendix 2.5 Age structure of the study populations. Appendix 2.6 Confinement status of the registered populations. Appendix 2.7 Sterilization status of the registered populations. Appendix 2.8 Size and density of the registered populations. Appendix 2.9 Age class at registration. Appendix 2.10 Number of dogs in the registered populations gained per month. Appendix 2.11 Number of dogs lost from the registered populations per month. Appendix 2.12 Proportion of reproductively capable females pregnant per month in the registered populations. Appendix 2.13 Number of litters per month in the registered populations. Appendix 2.14 Mortality per month for the registered populations. Appendix 2.15 Mortality per month of dogs ≤12 months of age in the registered populations. Appendix 2.16 Causes of mortality of the registered dogs. Appendix 2.17 Reasons for sourcing a dog from outside the Johannesburg sites. Appendix 2.18 Reasons for sourcing a dog from outside the Bali sites. Appendix 2.19 Assessment of whether the acquisition of the dog was planned and if the owner looked in the research site for a dog. Appendix 2.20 Number of registered dogs acquired from outside the study area per month. Appendix 2.21 Outcomes of the pups born in the study areas. Appendix 2.22 Average time to loss or censoring of the starting cohorts. Appendix 2.23 Risk of loss from the starting cohorts by age class at the start of the study and gender. Appendix 2.24 Outcomes of the dogs in the starting cohorts. Appendix 2.25 Marginal positive predictive distributions for the explanatory variables in the ordinal regression model against the observed maximum body condition scores for Zenzele. Appendix 2.26 Variables included in the ordinal regression models used to investigate trends between body condition and increased caloric requirements from growth and lactation Appendix 2.27 Zenzele ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios. Appendix 2.28 Braamfischerville ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios. xi Appendix 2.29 Kelusa ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios. Appendix 2.30 Antiga ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios. Chapter 3 Participatory methods for the assessment of the ownership status of free- roaming dogs in Bali, Indonesia, for disease control and animal welfare Appendix 3.1 Summary of the study methodology. Appendix 3.2 Approximate number of participants at the PRA sessions. Appendix 3.3 Health status in relation to body condition for dogs determined by the Antiga participants. Appendix 3.4 Health status in relation to body condition for dogs determined by the Kelusa participants. Appendix 3.5 Rank of food sources of healthy dogs in order of importance. Appendix 3.6 Rank of food sources of owned dogs in order of importance. Appendix 3.7 Rank of food sources of unhealthy dogs in order of importance. Appendix 3.8 Rank of food sources of unowned dogs in order of importance. Chapter 4 Achieving population-level immunity to rabies in free-roaming dogs in Africa and Asia Appendix 4.1 Summary of the study methodology. Appendix 4.2 Summary of vaccination coverage in the study areas. Appendix 4.3 The number of dogs in the research cohorts and the number of unvaccinated controls in Bali that were blood sampled at each time point. Appendix 4.4 Sample selection and sampling technique for deep skin scrapes and sample selection for faecal analysis. Appendix 4.5 Life expectancies for the entire (ecological study) research population derived from the observed ages at the last time point of the study period. Appendix 4.6 Characteristics of the dogs in Bali (in January 2010) that were not vaccinated. Appendix 4.7 Description of the dogs in Zenzele in with baseline (day 0) VNA titres ≥0.5 IU/ml. Appendix 4.8 Summary of titres from the dogs in Zenzele necessary to assess the inadvertent inclusion of dogs vaccinated by the DoA in October 2009 in the research cohort. Appendix 4.9 Summary of the VNA titres of the dogs in the Zenzele research cohort present May 2006 and those that arrived into the population after May 2006. Appendix 4.10 Characteristics of the unvaccinated controls in Bali with titres ≥0.5 IU/ml. Appendix 4.11 Summary of the virus-neutralizing antibody (VNA) titres of the unvaccinated controls in Bali. Appendix 4.12 Characteristics (at vaccination) of the dogs in the Zenzele research and DoA cohorts with peak (day 30) titres <0.5 IU/ml. xii Appendix 4.13 Summary of the day 180 and 360 titres in the research cohorts [vaccinated dogs]. Appendix 4.14 Characteristics (at vaccination) of the dogs in the research cohorts with day 360 titres ≤0.1 IU/ml. Appendix 4.15 Correlation coefficients for the dogs in the Zenzele research cohort that were blood sampled at every time points. Appendix 4.16 Linear mixed effects model outputs for the research cohorts restricted to natural log of the VNA titres as the response variable and time as the covariate. Appendix 4.17 Observed and predicted geometric mean titres for each time point for the research cohorts. Appendix 4.18 Description of the covariates in the linear mixed effects and analysis of variance models (Appendices 4.19-4.22). Appendix 4.19 Linear mixed effects and analysis of variance model outputs for Zenzele which include natural log VNA titres as the response variable and the range of covariates (see Appendix 4.18). Appendix 4.19 Analysis of variance model outputs for Zenzele which include natural log of lymphocyte counts and the range of covariates (Appendix 4.18). Appendix 4.20 Linear mixed effects model outputs for Kelusa which include natural log of VNA titres as the response variable and the range of covariates (see Appendix 4.18). Appendix 4.21 Linear mixed effects model outputs for Antiga which include natural log of VNA titres as the response variable and the range of covariates (see Appendix 4.18). Appendix 4.22 Linear mixed effects model outputs for Kelusa and Antiga combined which include natural log of the VNA titres as the response variable and the range of covariates (see Appendix 4.18). Appendix 4.23 Contingency tables of the covariates in the linear mixed effects and analysis of variance models (see Appendices 4.19-4.22). Appendix 4.24 VNA titres of the dogs vaccinated by the Department of Livestock in Kelusa with Rabivet Supra 92. Appendix 4.25 The number of dogs in Zenzele with intestinal parasites at the start of the study period. Chapter 5 Modelling human interference and rabies transmission in free-roaming dogs Appendix 5.1 Outbreak size in Zenzele, Braamfischerville, Kelusa, Kaler and Ketug for a range of R0 and transmission rates for the compartmental model. Appendix 5.2 Probability of an outbreak in Zenzele, Braamfischerville, Kelusa, Kaler and Ketug at the time of vaccination and 12 and 24 months after vaccination for a range of vaccination coverage and transmission rates. Appendix 5.3 Mean estimates of vaccination coverage with time in the study populations. Appendix 5.4 Probability of outbreaks with human-mediated incursions for a range of vaccination coverage and transmission rates for Kaler. xiii Appendix 5.5 Probability of outbreaks with human-mediated incursions for a range of vaccination coverage and transmission rates for Ketug. Appendix 5.6 Outbreak size with human-mediated incursions for a range of vaccination coverage and transmission rates in Kelusa, Kaler and Ketug. Additional appendices Appendix 6.1 Rabies vaccine-induced virus neutralizing antibody (VNA) titres in pups in Zenzele. Appendix 7.1 Summary of the transcription errors in the questionnaires (see Chapter 2). Appendix 7.2 Analysis of the repeatability of the questionnaires. xiv List of Abbreviations Abbreviation Meaning AIC Akaike’s information criterion BCS Body condition score CBC Complete blood count CI Confidence interval CLAW Community Led Animal Welfare CM Compartmental model cm Centimetre DA2PPv Canine Distemper, Adenovirus Type 1 & 2, Parainfluenza and Parvovirus DoA Department of Agriculture DoL Department of Livestock DSS Deep skin scrape EDTA Ethylene diamine-tetraacetic acid FAVN Fluorescent antibody virus neutralization assay GMT Geometric mean titre GPS Global positioning system HIPD Hunting indicator of population density IBM Individual based model ICAM International Companion Animal Management Coalition IFAW International Fund for Animal Welfare InAW Indonesia Animal Welfare IU International Unit km Kilometre LBV Lagos Bat Virus ml Millilitre NGO Non-governmental organization nlme Non-linear mixed effects NPO Non-proportional odds OIE Office International des Epizooties / World Organisation for Animal Health OPL Oestrus, pregnancy, lactation PO Proportional odds PPA Posterior probability of association PRA Participatory rural appraisal RABV Rabies virus SD Standard deviation SE Standard error SEI Susceptible-Exposed-Infectious VNA Virus neutralizing antibody VP Vaccination point WHO World Health Organization WSPA World Society for the Protection of Animals xv List of Symbols Symbol Meaning o C Degrees Celsius cm 2 Area in centimetres km 2 Area in kilometres o E Degrees latitude o S Degrees longitude In Natural log K Carrying capacity % Percentage p p-value R0 Basic reproductive number Chapter 4 𝜷 Regression coefficient n Sample size 𝑵 Normal distribution 𝒓 Correlation coefficient 𝑿 Regression covariate 𝒀 Response variable 𝜽 Individual-level random effect term 𝝈𝟐 Variance 𝜺 Error term ∑ . Sum Chapter 5 𝜶 Removal rate that scales with population density 𝑨 Area 𝒃 Average gain rate 𝜷 Transmission rate 𝜹t Time step 𝜸 Removal rate 𝟏/𝜸 Infectious period 𝒎 Removal rate without human interference 𝝁 Average loss rate 𝒏 Population density 𝑵 Population size 𝟏/𝝈 Latent period xvi Preface Canine-mediated rabies claims tens of thousands of human lives every year, primarily in low- income communities in Asia and Africa. Although culling (i.e. the widespread killing of dogs regardless of infection status) is often used as the primary means to control the disease, mass vaccination of dog populations should be the main stay of effective control (WHO 2013). Planning vaccination campaigns depends on knowledge of the size and accessibility of dog populations and of rates of decline in vaccination coverage to below protective levels. This in turn requires an understanding of the demography and immune responses to rabies vaccine of domestic dogs. Understanding the factors that drive disease transmission, facilitating development of epidemiological models of local and spatial dynamics and for movement controls, can also support effective rabies control. Demographic data are also needed to plan canine population controls that are often used to address animal welfare and other public health concerns. However, there is a paucity of demographic and immunological data in the literature necessary for rabies and population control, and apparently an incomplete understanding of the drivers of local transmission of rabies, reflected in frequent culling dogs for rabies control. The existing literature includes various aspects of dog ecology that contribute substantially to rabies and population control. However, except for a limited number of more comprehensive studies of dog population dynamics (Boitani et al. 1995; Pal 2001), generally only facets of dog demography are reported in various studies of canine rabies, from a broad range of different study types. Any demographic data had generally been collected in cross-sectional studies, relying on owner recall for essential information, such as reproductive histories; most of these studies included only a sub-set or sample of households or dogs, often from large geographical areas. A further complication is that conflicting hypotheses regarding the factors that drive population and disease dynamics are presented. This may be, in part, a consequence of extrapolations of assumptions and observations of population and disease dynamics from wildlife to domestic dogs. For example, the WHO/WSPA Guidelines for Dog Population Management (WHO & WSPA 1990), frequently referred to by those implementing rabies and population controls, policy makers and academics, states that free- roaming dog population size is regulated by environmental resource constraints similar to wildlife (i.e. the dogs are effectively unowned), but also that free-roaming dogs generally have a reference household (i.e. the dogs are owned). Furthermore, serological studies of rabies vaccine-induced immune responses are generally limited to laboratory or pet dogs and xvii not to the free-roaming dogs that frequently transmit the disease. Finally, there is paucity in the literature regarding the factors that drive rabies transmission. This study aimed to address these limitations by generating individual-level, demographic and serological data, necessary for rabies control, from four entire populations through longitudinal, direct observations and owner questionnaires restricted to the recent past (described in Chapters 2 and 4). Community-level participatory exercises were also used as a novel approach to assess key aspects of ownership (described in Chapter 3). A thorough review of the role of population density in disease transmission was also undertaken in the context of population density-reduction through culling for rabies control (described in Chapter 1). My involvement in this research project began as Program Manager of the global Companion Animals Program for the International Fund for Animal Welfare (IFAW), faced with writing organizational policy on dog population management but also the aforementioned fundamental gaps and incongruities in the literature. I was also travelling extensively to observe population control measures in free-roaming dog populations and increasingly questioned the efficacy of these interventions. For similar reasons, the International Companion Animals Management Coalition (ICAM) http://www.icam-coalition.org/ was founded by Dr Elly Hiby of the World Society for the Protection of Animals (WSPA), myself (IFAW) and colleagues from several other organizations leading in dog population management. Collectively, these factors were the catalyst for my decision to undertake my PhD studies, to generate data directly relevant to my funders’ programmatic work on population and rabies control. The research was undertaken in collaboration with Pretoria and Udayana Universities and the support of several researchers in the field of rabies control, disease dynamics and animal welfare based at Glasgow, Cambridge and Bristol Universities. A part-time PhD allowed me to continue to engage with my funders and ICAM directly as a conduit of current information between the field and round table. It also provided the massive benefit of a sufficient period to collect a more extensive data set than would have been possible with a full-time PhD, with real, practical advantages for rabies and population control. Likewise, the luxury of time in the field afforded an intimidate knowledge of the communities by myself and the field teams, which was fairly unique and hugely advantageous to this type of research. The research was initially designed to evaluate demographic processes, and their regulatory factors, at the local level, for the purpose of improving population control. However, given xviii that most of the literature relevant dog ecology was in the context of rabies control, the broad application of these demographic data to rabies control became rapidly apparent. Similarly, the need for, and opportunity to incorporate, a serological sub-study to assess rabies vaccine- induced immune responses into the project became clear early after inception. The inclusion of a serological study was also relevant to IFAW and WSPA given their increasing involvement in canine-mediated rabies control, especially issues surrounding culling but also the implementation of vaccination programmes. Soon after initiating the study, a rabies outbreak occurred in Bali. One of the initial responses of the provincial government was extensive culling of the free-roaming dogs in an attempt to control the disease. In collaboration with world experts in rabies control and endorsements by the ICAM members, this precipitated my writing of recommendations for the provincial government for rabies control based on mass vaccination. In turn, this initiated my detailed review of the literature regarding the role population density in rabies transmission and the efficacy of culling for disease control. The extension of the cull to an area where I was conducting the research (thankfully at the tail end of data collection) added fuel to fire with regards publishing my review on culling. It also initiated a follow-up study given the unique opportunity to monitor the impact of culling on population dynamics, particularly focusing on human behaviour, where population dynamics had been previously monitored (thus providing historic control data). This follow-up was subsequently extended to monitor the effects of population sterilization and other veterinary interventions on population dynamics and animal welfare in the other research sites. Demographic, immunological and epidemiological data are integral to disease and population dynamics and controls. Therefore, this dissertation was deliberately written as a series of inter-related papers, with the view to publication in the hope that the information will be readily accessible to and helpful for those currently in the field implementing rabies controls, policy makers and academics. Chapter 1 (published) examines the factors that drive the transmission of rabies, specifically host population density and the reasons why density reduction, particularly through culling, fails to control rabies. The discussion introduces the role of human factors in the transmission and control of rabies. [M.K Morters, O Restif, K Hampson, S Cleaveland, J.L.N Wood, A.J.K Conlan. 2012. Evidence-based control of canine rabies: a critical review of population density reduction. Journal of Animal Ecology, 82, 6-14.] xix Chapter 2 (provisionally accepted 26 March, 2014) presents demographic data from four free-roaming dog populations, and includes evaluation of the regulatory effect of human behaviour on dog population dynamics and declines in rabies vaccination coverage following a pulse vaccination campaign. [M.K Morters, T.J McKinley, O Restif, A.J.K Conlan, K Hampson, H.R Whay, I Made Damriyasa, J.L.N Wood. The demography of free-roaming dog populations and applications to disease and population control. Journal of Applied Ecology.] Chapter 3 (provisionally accepted 27 February, 2014) describes community-level participatory methods applied to the study of dog ownership in Bali. This is a novel approach to assess the fraction of unowned dogs in populations, which complements the ecological study described in Chapter 2. [M.K Morters, S Bharadwaj, H.R Whay, S Cleaveland, I Made Damriyasa, J.L.N Wood. Participatory methods for the assessment of the ownership status of free-roaming dogs in Bali, Indonesia, for disease control and animal welfare. Preventive Veterinary Medicine.] Chapter 4 (under review) evaluates rabies vaccine-induced immune responses in free- roaming dogs, including the effects of waning immunity and demographic processes on vaccination coverage. [M.K Morters, T.J McKinley, D.L Horton, S Cleaveland, J.P Schoeman, O Restif, H.R Whay, A Goddard, A.R Fooks, I Made Damriyasa, J.L.N Wood. Achieving population-level immunity to rabies in free-roaming dogs in Africa and Asia. PLoS Neglected Tropical Diseases.] Chapter 5 explores the transmission dynamics of canine rabies using models that account for human interference in transmission as proposed in Chapter 1, demographic data from Chapter 2 and previously reported epidemiological parameter estimates. This is being prepared for publication and, thus far, has received input from A.J.K Conlan, O Restif and J.L.N Wood. Finally, Chapter 6 summarizes the results from the study and draws conclusions on the implications of the research and areas where further research is required. Cross-referencing in this thesis uses chapter number. Supporting documents, tables and figures from the papers are provided in appendices. The appendices have been submitted to the journals as supporting materials to the papers listed above. This dissertation is the result of my own work and includes nothing which is the outcome of work done in collaboration except where specifically indicated in the text and in the Table of Collaborators (see page v). xx This research was approved by the Ethics Committee, University of Cambridge, and the Animal Ethics Committee, University of Pretoria. Permits to collect demographic data were granted by the Ministry for Research and Technology (RISTEK), Indonesia. Blood samples were collected under the auspices of the Faculty of Veterinary Medicine, Udayana University, Bali, and permits for vaccination and blood collection were granted by the Balinese provincial and regencies Departments of Livestock, the districts Centres of Animal Health (UPT), and Kesbang, Pol and Linmas (the combined Agencies for National Unity, Politics and Protection). In all of the sites, informed consent was obtained prior to each survey and blood test from the community leaders and owners, who were kept fully informed of the purpose, approach and progress of the study. Survey respondents under 16 years of age were always interviewed with an adult present. Vaccination and blood sampling were only carried out with the owner, or responsible adult delegated by the owner, present and their express consent. The dissertation does not exceed the required word limit. Word count: 39,819 1 Chapter 1 Evidence-based control of canine rabies: a critical review of population density reduction Summary Control measures for canine rabies include vaccination and reducing population density through culling or sterilization. Despite evidence that culling fails to control canine rabies, efforts to reduce canine population density continue in many parts of the world. The rationale for reducing population density is that rabies transmission is density-dependent, with disease incidence increasing directly with host density. This may be based, in part, on an incomplete interpretation of historical field data for wildlife, with important implications for disease control in dog populations. Here we examine historical and more recent field data, in the context of host ecology and epidemic theory, to understand better the role of density in rabies transmission and the reasons why culling fails to control rabies. We conclude that the relationship between host density, disease incidence and other factors is complex and may differ between species. This highlights the difficulties of interpreting field data and the constraints of extrapolations between species, particularly in terms of control policies. We also propose that the complex interactions between dogs and people may render culling of free-roaming dogs ineffective irrespective of the relationship between host density and disease incidence. We conclude that vaccination is the most effective means to control rabies in all species. 1.1 Introduction Canine-mediated rabies is a serious zoonosis causing an estimated 55,000 human deaths per year (Knobel et al. 2005). Mortality from rabies is highest in developing communities in Africa and Asia where domestic dogs are predominately free-roaming (Ezeokoli & Umoh 1987; Butler & Bingham 2000; Kitala et al. 2002; Kayali et al. 2003a; Windiyaningsih et al. 2004; Kasempimolporn, Jitapunkul & Sitprija 2008). Social, economic and political factors contribute to the inadequate control of rabies in domestic dog populations (WHO 2004), accentuated by an incomplete understanding of disease dynamics. Knowledge of the factors that drive the transmission of rabies is needed for development of effective, sustainable disease control measures. 2 Two main methods are used to control canine rabies: vaccination (Cleaveland et al. 2003; WHO 2004; Schneider et al. 2005; Cleaveland et al. 2006) and measures aiming to reduce dog population density, usually by culling (i.e. the widespread killing of dogs regardless of infection status) (Beran & Frith 1988; Windiyaningsih et al. 2004) but also by sterilization (WHO 2004; Reece & Chawla 2006). Dog vaccinations are often undertaken as annual campaigns that aim to achieve 70% coverage (WHO 2004). This target coverage is supported by empirical evidence and theory, which indicates that a 70% coverage achieved during campaigns should maintain population immunity above the critical levels (20-45%) required to interrupt rabies transmission (Coleman & Dye 1996; Cleaveland et al. 2003; Hampson et al. 2009). This additional coverage above the critical level compensates for the loss in coverage arising from an increase in susceptible and loss of immune dogs through demographic and immunological processes (Hampson et al. 2009). Culling of dogs is also used, alone or with vaccination (Kaplan, Goor & Tierkel 1954; Larghi et al. 1988), based on the assumption that a physical reduction in the number of dogs must reduce the incidence of rabies despite evidence suggesting that it is ineffective (Beran & Frith 1988; WHO 2004; Windiyaningsih et al. 2004). Culling is still used, partly as a visible response to public concerns about rabies. It is also perceived to be easier to implement than annual vaccination of 70% of dogs, particularly if many are free-roaming and poorly socialized, and in areas where veterinarians and animal health workers have relatively little experience or confidence in handling dogs. In some areas, sterilizations are carried out together with vaccinations, on the basis that this is a more humane and culturally acceptable approach to reducing dog population density. The theoretical basis for rabies control measures involving culling or sterilization is the assumption that rates of transmission are density-dependent (Anderson et al. 1981; Wandeler et al. 1988; Cleaveland 1998; Hampson et al. 2007). This scaling of transmission rates occurs if the rate of encounters between susceptible and infectious individuals increases with host population density. Under this assumption, we expect that disease incidence will also increase with host density, as will the basic reproductive number (R0) that characterises the maximum reproductive potential of a pathogen. R0 is defined as the average number of secondary infections produced when one infected individual is introduced into a wholly susceptible population (Anderson & May 1991). For an epidemic to spread, R0 must, by definition, be >1. Hence, under density-dependent transmission there will exist a threshold density below which disease cannot invade a population. This contrasts with frequency-dependent disease transmission where the rate of contact and subsequent rates of transmission are assumed to be 3 independent of host density and a threshold density for invasion does not exist (Begon et al. 2002; Lloyd-Smith et al. 2005). Under either frequency- or density-dependent transmission, vaccination equally reduces both the number and proportion of susceptible individuals in a host population, and thus, the opportunities for transmission to occur. Therefore, the assumption that rabies transmission is density-dependent has little consequence for the efficiency of vaccination programmes. Conversely, the assumption is of critical importance with regards to control measures that aim to reduce dog population density. The net impact of culling and sterilization on subsequent rates of rabies transmission depends on the degree to which transmission scales with population density. Under the assumption of frequency dependence, density reduction will have no impact on the rate of transmission. Conversely, when transmission is density- dependent, there will be a threshold for disease invasion, and density reduction alone has the potential to achieve disease eradication. However, stochastic effects and antagonistic biological processes may complicate these simple relationships. Establishing the relationships between host density, disease incidence and other processes is therefore not only important for refinement of epidemiological models for rabies transmission, but has serious practical implications for the utility of density reduction in controlling rabies. In this study, we review current understanding of the role of density and other factors in rabies transmission in dogs in order to encourage reappraisal of the most appropriate and effective means of rabies control. Within the literature, and during the development of policy, extrapolations are often made between species, in particular between wildlife and domestic dogs. We therefore extend our review to rabies transmission in wildlife and highlight the differences and similarities with dog populations. We also compare the utility of various lines of evidence between species. This discussion will focus on fox rabies in particular, as empirical data on the local transmission of wildlife rabies are largely confined to this host species. 1.2 Evidence for density-dependent transmission of rabies It is difficult to determine the direct relationship between disease incidence, host density and transmission under field conditions, particularly for wildlife given their inaccessibility (Wandeler et al. 1974b; Macdonald & Voigt 1985; Beyer et al. 2010). Consequently, we are left with interpreting indirect and somewhat conflicting evidence regarding the role of density 4 in rabies transmission in wildlife and dogs. In this section we examine four key lines of evidence about the functional forms of rabies transmission. 1.2.1 Cycles in disease incidence Cycles in disease incidence have motivated some of the most effective applications of population modelling in ecology (Anderson & May 1991; Begon, Harper & Townsend 1996). Mathematical models can explore how different biological hypotheses relate to the expected amplitude and period of cycles, providing insights into the drivers of transmission. Perhaps, the most successful examples of this have been in the study of childhood infectious diseases (Earn et al. 2000; Altizer et al. 2006) where detailed historical records have allowed the application of sophisticated methods of statistical inference (Bjornstad, Finkenstadt & Grenfell 2002; Grenfell, Bjornstad & Finkenstadt 2002). However, even in the absence of detailed data, models can provide useful insights simply through the ability of a given mechanism to generate periodic dynamics. Cycles have been observed for wildlife (Friend 1968; Bogel et al. 1974; Childs et al. 2000; Courtin et al. 2000; MacInnes et al. 2001) and canine rabies (Ernst & Fabrega 1989; Bingham et al. 1999a; Widdowson et al. 2002; Hampson et al. 2007), although periodicity in incidence is not a consistent finding (Macdonald & Voigt 1985; Zinsstag et al. 2009). The mechanistic driver of these cycles is widely assumed to be the interaction of density- dependent transmission, rabies-induced mortality and other demographic processes (Bogel et al. 1974; Steck & Wandeler 1980; Anderson et al. 1981; Childs et al. 2000; Hampson et al. 2007). However, it is important to determine if this assumption is correct given its implications for culling. Deterministic compartmental models have been used to describe rabies dynamics in wildlife (Anderson et al. 1981; Coyne, Smith & McAllister 1989) and domestic dogs (Cleaveland & Dye 1995; Coleman & Dye 1996; Kitala et al. 2002; Hampson et al. 2007; Carroll et al. 2010). These models assume random mixing, neglecting the spatial and social heterogeneity that exists in real populations. Within such “well-mixed” models, frequency-dependent transmission of fatal diseases inevitably leads to rapid die-out of the host population (Keeling & Rohani 2008). Under frequency dependence, the average reproductive potential of the pathogen is unchanged during the spread of an epidemic. With no mechanism to arrest the spread of disease, transmission continues and the host and parasite populations go extinct. In contrast, under the assumption of density-dependent transmission, epidemics will subside 5 when the host density falls below the invasion threshold (where R0 = 1). The time delay between epidemic peak and replenishment of the host population generates damped epidemic cycles through delayed density dependence. The assumption of density-dependent transmission is therefore the most parsimonious mechanism by which stable epidemic cycles for rabies can be supported within deterministic random mixing models. However, in structured populations, epidemic cycles may be generated by alternative mechanisms even when the transmission rate is frequency-dependent. Age structure is one such potential mechanism. Attack rates for rabies appear to vary considerably with age, with reported incidence in foxes in Europe (Wandeler et al. 1974b) and raccoons in Ontario (Rosatte et al. 2006) concentrated within adult age classes. Within an age-structured model, the net reproductive ratio of rabies will not only depend on the rates of transmission, but also on the age-distribution in the population (Anderson & May 1991). If the basic reproductive ratio is only above unity for a core-group of high-risk individuals, the epidemic can recede when this core-group is exhausted. The delay between depletion of the core group and replenishment through births can generate cycles in incidence that may be sustained by seasonal birth pulses (Davis & Wood 1959; Lloyd et al. 1976). Deterministic thresholds are not the only possible mechanism by which endemic co-existence of rabies could be maintained within frequency-dependent transmission models. An important limitation of deterministic models is that they do not account for the probability of local extinction of disease following an epidemic. In areas where rabies in foxes is not actively controlled, 3-4 yearly cycles in incidence are observed at regional levels [around 1000 km 2 in Europe and at the county level in Canada] (Johnston & Beauregard 1969; Bogel et al. 1974), and are out of phase between regions (Johnston & Beauregard 1969; Bogel et al. 1974; Macdonald & Voigt 1985). Epidemics have been associated with considerable reductions in host populations by up to 50% (Bogel et al. 1974). This reduction in the density of the host species within a region and the corresponding reduction in the instantaneous numbers of infective individuals will increase the chances of rabies becoming locally extinct before the host population is exhausted. Stochastic population thresholds for persistence of rabies can exist irrespective of the mode of transmission (Lloyd-Smith et al. 2005). Stochastic extinction and re-introduction of rabies following the local re-structuring of host populations (Steck & Wandeler 1980; Anderson et al. 1981; Macdonald & Voigt 1985), consistent with metapopulation dynamics, are also viable alternative mechanisms to generate these dynamics. 6 In conclusion, cycles in rabies incidence observed in wildlife could be supported by density- or frequency-dependent transmission when stochasticity and the heterogeneous structure of real populations are accounted for. Although deterministic density-dependent models have been used to describe rabies dynamics in domestic dogs, reactive vaccination can also drive cycles in incidence (Hampson et al. 2007). For example, in Zimbabwe between 1950 and 1995, the amplitude and interval of peaks in rabies varied (from 75 to 350 cases per year and inter-epidemic periods from 4 to 20 years) with the level of vaccination delivered during national vaccination campaigns (Bingham et al. 1999a). These observations provide little insight into the processes driving local disease dynamics for dogs. Rather, other evidence for the functional forms of transmission of canine rabies will be considered in the next sections. 1.2.2 The relationship between R0 and host density As discussed above, R0 is expected to increase with density for density-dependent transmission and remain constant irrespective of density for frequency-dependent transmission. R0 may be estimated from the (exponential) rate of growth early in an epidemic prior to significant susceptible depletion or implementation of control measures (Heffernan, Smith & Wahl 2005; Wallinga & Lipsitch 2007). Using this method, Hampson et al (2009) obtained estimates of R0 for canine rabies, across a wide geographical range, of between 1.05 and 1.72. The range of these estimates is similar to the statistical uncertainty in simulated epidemics when the biting behaviour of rabid dogs is accounted for. Dog population densities were reported for only four of these locations, ranging from 1.36 dogs km -2 in rural Tanzania to 110 unrestricted dogs per km 2 in urban Mexico. However, other locations cited in the study are likely to represent even higher densities, with the highest reported density in the general literature being 2388 dogs km -2 in Guayaquil, Ecuador (Beran & Frith 1988). The absence of any correlation between R0 and host density across such a large range of densities is consistent with earlier studies (Coleman & Dye 1996; Kitala et al. 2002) and suggests that if a relationship between transmission and dog density does exist, it must be quite weak. Equivalent data are not available for wildlife. Compared to canine rabies, incidence records generally have a lower temporal resolution (typically quarterly or annually) (Macdonald & Voigt 1985; Rhodes et al. 1998; Bingham et al. 1999b; Rosatte et al. 2006), and the ranges of host densities are narrower: 0.8-1.2 jackals km -2 during the breeding season on commercial farmland in Zimbabwe (Rhodes et al. 1998), 5.4-9.1 racoons km -2 (averaged over a four year 7 period) for rural Ontario (Rosatte et al. 2007) and 0.5-1.8 adult foxes km -2 in central Europe (Lloyd et al. 1976). This apparent lack of relationship between R0 and host density is most consistent with frequency-dependent transmission. However, as previously discussed, random mixing models with frequency-dependent transmission of rabies predict host extinction as soon as R0 exceeds unity. This prediction is inconsistent with the very low attack rates reported for canine rabies compared to wildlife rabies and with the absence of large declines in population densities from rabies-induced mortality (Hampson et al. 2007). Estimates of the incidence, or average monthly attack rates, are typically below 0.5% and rarely exceed 2% (Waltner- Toews et al. 1990; Windiyaningsih et al. 2004; Zinsstag et al. 2009; Tenzin et al. 2010; Putra et al. 2011; Tenzin et al. 2011). This incongruity between attack rates and the apparent scaling of R0 may be resolved by considering a more complex relationship between rabies dynamics in dogs and anthropogenic factors than has previously been assumed. Suspect rabid and in-contact dogs are often identified and killed swiftly by the community (Hampson et al. 2007; Hampson et al. 2009), a practice hereafter referred to as ‘selective removal’. This reduces the effective infectious period in dogs (Hampson et al. 2009) and could contribute to the relatively lower incidence as compared to wildlife. The selective removal of infectious and in-contact dogs was thought to have contributed to the control of rabies in eastern Bhutan (Tenzin et al. 2011) and the United Kingdom (Pastoret & Brochier 1998). Indeed, euthanasia (WSPA 2012) of infected dogs is advocated to control rabies (WHO 2004). Such behavioural responses to the spread of epidemics are rarely considered in epidemiological models (Ferguson 2007; Funk et al. 2009) but are likely to play a particularly important role in disease transmission within owned, and managed, populations. Selective removal may conceal the existence of density-dependent transmission processes if the rate of intervention also scales with density. We thus hypothesize that selective removal itself might be density-dependent for several reasons. First, rabid dogs may be more quickly spotted and selectively removed from areas with more people present. Second, given that most dogs are owned (WHO & WSPA 1990; Cleaveland & Dye 1995; Butler & Bingham 2000; Windiyaningsih et al. 2004), dog and human population densities are expected to correlate (Oboegbulem & Nwakonobi 1989; Matter et al. 1998; Butler & Bingham 2000). Finally, other anthropogenic factors that may interfere with contact processes, such as traffic or urban infrastructure, are also likely to scale with human and dog density. Therefore, the effective infectious period, as reduced by 8 selective removal, could scale inversely with human, and thus dog, population density. The estimates of R0 discussed above are conditional on the assumption of a fixed infectious period. Any systematic variation in the infectious period with population density could counteract the impact of density-dependent contact rates and result in R0 appearing density- independent. Under this hypothesis, density-dependent transmission could not be ruled out unequivocally for canine rabies. As a final consideration, stochastic fade-out is expected with low attack rates. However, rabies often appears to persist in dog populations. This may be because selective removal and stochastic processes are off-set by the continual translocation of dogs (some of them infected) by people (Beran & Frith 1988; Denduangboripant et al. 2005; Coetzee & Nel 2007; Kasempimolporn, Jitapunkul & Sitprija 2008; Zinsstag et al. 2009) consistent with metapopulation dynamics (Hanski & Gaggiotti 2004; Beyer et al. 2010). In conclusion, more intensive study of the mechanisms underlying rabies transmission and persistence in domestic dog populations is warranted to understand these empirical patterns. 1.2.3 Thresholds for invasion and increasing incidence with population density The existence of a threshold in host population density below which infection cannot spread (i.e. where R0 < 1) would be direct evidence in support of density-dependent transmission. Such invasion thresholds in wildlife and domestic dog populations have been proposed based on a limited number of studies that compared disease incidence between different geographic locations with different host densities (Steck & Wandeler 1980; Beran & Frith 1988; Cleaveland & Dye 1995). However, as discussed below, it is not possible to establish the relationship between host density and disease incidence based on these data. Threshold densities for invasion have been suggested to occur where canine rabies is observed to change from sporadic disease at lower densities to persistence at higher densities (Beran & Frith 1988; Cleaveland & Dye 1995). However, these observations could also be explained by increased stochastic fade-out of disease at lower densities where there are lower numbers of infected dogs. In general, the probability of stochastic fade-out will decrease with an increase in R0 or in the number of infected individuals (Lloyd-Smith et al. 2005). This effect may be particularly relevant to dogs where more infected individuals may be introduced into larger or more dense populations by people (Denduangboripant et al. 2005; Kasempimolporn, Jitapunkul & Sitprija 2008; Zinsstag et al. 2009). Consequently, the probability of stochastic fade-out is predicted to decrease with an increase in population size 9 or density. Even when R0 is invariant between populations of different sizes or densities, stochastic effects may give the impression of a deterministic threshold for invasion where one does not exist. This is particularly likely when R0 is low. Should a deterministic threshold for invasion exist, it may be obscured by these processes and be lower than estimated empirically. The key data used to support the existence of a threshold density in foxes are expressed in terms of the hunting indicator of population density (HIPD) (Steck & Wandeler 1980). HIPD is an indirect estimate of density, with well-known biases (Wandeler 1980; Macdonald & Voigt 1985). However, there are two specific issues with the use of these data to support a threshold density for fox rabies. First, HIPD estimates below the purported threshold density for invasion were not recorded, thus precluding any conclusion of an invasion threshold. Second, the observed positive correlation between the annual number of animal rabies cases per km 2 per year and the HIPD has been wrongly interpreted as evidence for density- dependent transmission. Assuming the HIPD correlates with host density, such a relationship would be expected whether transmission depends on fox density or not. Determining the mode of transmission would require an evaluation of disease incidence as a proportion of the total population size or density (Rothman, Greenland & Lash 2008), which cannot be inferred from HIPD. 1.2.4 Impacts of density reduction Density reduction, particularly culling (i.e. the widespread killing of hosts regardless of infection status), has been undertaken to reduce the incidence of rabies and therefore eliminate the disease on the basis that transmission is density-dependent. As previously discussed, the assumption of density dependence originates from the interpretation of cycles in wildlife rabies and thresholds for invasion for foxes and dogs. However, the fact that culling has failed to achieve sustained control of rabies in wildlife and dogs (Kaplan, Goor & Tierkel 1954; Anderson et al. 1981; Macdonald & Voigt 1985; Anderson 1986; Beran & Frith 1988; WHO 2004; Windiyaningsih et al. 2004; Cleaveland et al. 2006) may be the best evidence that a simple relationship between disease incidence and host population density does not exist for rabies. We now discuss evidence from culling programmes (dogs and wildlife) followed by more limited evidence on sterilization campaigns. 10 1.2.4.1 Culling Culling has been shown to be ineffective in controlling rabies in all host species. Rabies persisted in foxes in New York State despite ‘concentrated reduction campaigns’ following an outbreak in 1945, while simultaneous vaccination of dogs in the State eliminated rabies from this species (Friend 1968). Similarly, in Denmark in 1964, culling did not prevent rabies outbreaks in foxes; however, rabies did not occur where dogs in the same region had been vaccinated (Muller 1966; Muller 1971). In response to a rabies outbreak in 1997 nearly 300,000 dogs, approximately half of the population estimated at the start of the outbreak, were culled in Flores, Indonesia over a period of 4 years. However, in 2004 rabies was still endemic although the total dog population was still considerably reduced (Windiyaningsih et al. 2004). Culling failed to control canine rabies in Korea (Lee et al. 2001) and Israel (Kaplan, Goor & Tierkel 1954), whereas subsequent vaccination in both countries controlled the disease. Culling has been used to control ongoing outbreaks and to prevent the invasion of rabies in foxes. Declines in rabies cases have followed outbreaks irrespective of active culling (Bogel et al. 1974), with stochastic extinction expected (Anderson et al. 1981) particularly where disease-induced mortality is substantial (Bogel et al. 1974). Within a given area, culling might be expected to amplify these processes, increasing the probability of stochastic extinction regardless of density dependence. Indeed, rabies appeared to die-out in some areas where fox dens were gassed (Wandeler et al. 1974b). However, the limited data available are unclear regarding how culling interacts with disease-induced mortality during an epidemic and how it may change disease dynamics (Wandeler et al. 1974b). Other processes may also counter the effect of density reduction on disease incidence. Examples include social perturbations, as demonstrated in badger populations (Woodroofe et al. 2006a; Woodroofe et al. 2006b), and interactions between the level of culling, age structure (Bolzoni, Real & De Leo 2007) and demographic processes (Choisy & Rohani 2006). Culling has also failed to prevent outbreaks of rabies in foxes in previously unaffected areas or the recurrence of the disease in areas where it had died-out, as observed in southern Denmark (Muller 1971). Where density-dependent transmission has been assumed, invasion thresholds are reported to vary and to be low (i.e. <1 fox km -2 in Europe and <0.4 foxes km -2 in Ontario). Thus, even if transmission were density-dependent, reductions in density to below an invasion threshold may not be achievable practically or be sustainable (Wandeler et al. 1974a; Anderson et al. 1981). 11 Culling has generally failed to eliminate outbreaks of rabies in dogs. In our review of the scaling of rabies transmission rates with density (in the previous sections) we have found no conclusive evidence to support either the frequency-dependent or density-dependent assumption for canine rabies. We are therefore unable to unequivocally conclude that the ineffectiveness of culling is because transmission is frequency-dependent. An alternative explanation is that reductions in densities to below invasion thresholds are not achievable practically. Canine rabies can circulate where densities are as low as 1.36 dogs km -2 (Hampson 2009), which is substantially lower than the densities reported for most free- roaming dog populations. Under the assumption of density-dependent contact rates, culling and vaccination should have similar impacts on disease incidence. Thus, given estimated values of R0 <2, control should be achieved by culling at most half the population. Yet, in Flores, Indonesia, rabies persisted after this level of culling was achieved (Windiyaningsih et al. 2004). More generally, the stochastic persistence of canine rabies despite low attack rates and considerable density reduction is interesting irrespective of the mode of transmission. The fact that rabies often persists despite culling may be a function of human factors. The continual translocation of dogs (some infected) with people (Beran & Frith 1988; Denduangboripant et al. 2005; Coetzee & Nel 2007; Kasempimolporn, Jitapunkul & Sitprija 2008; Zinsstag et al. 2009) may off-set the selective removal of infectious and in-contact dogs and stochastic extinctions. Where culling occurs simultaneously, translocation may also off-set any reductions in the incidence of rabies. In addition, translocation may be exacerbated in response to culling campaigns. For example, within a few days of a village- wide cull in Kelusa, Bali, where rabies had not occurred previously, two residents brought in unvaccinated, potentially infected puppies from outside the village to replace their culled, vaccinated adult dogs. As attack rates are typically very low, culling predominately removes healthy dogs, and some of these may be vaccinated and hence unlikely to become infected. Other compensatory mechanisms may also off-set reductions in host density. These include concomitant reductions in mortality from reduced competition for food (although the actual intensity of competition in free-roaming dogs is unknown), reductions in the dumping of surplus puppies/unwanted dogs and improved care of dogs. To address these issues, we are currently investigating the effects of human behaviour in response to culling on dog population dynamics and disease transmission in Kelusa. The ethics of culling healthy, free-roaming animals in conjunction with vaccination programmes are also debatable. Raccoons have been culled on Wolfe Island, Ontario, as a means to reduce the number of animals that needed to be trapped and vaccinated (Rosatte et 12 al. 2007). The same justification may be extended to dogs, and a variable degree of culling of free-roaming dogs, historically regarded as ‘strays’, has often been undertaken alongside mass vaccination programmes (Wells 1954; Cheuk 1969; Larghi et al. 1988; Ernst & Fabrega 1989). However, despite appearances, the vast majority of free-roaming dogs in most societies globally are owned (WHO & WSPA 1990; Cleaveland & Dye 1995; Butler & Bingham 2000; Windiyaningsih et al. 2004) and in reasonable health. Not only are these dogs more accessible to vaccination than commonly recognised, but culling healthy animals can result in unintended negative consequences on both animal welfare and disease control. 1.2.4.2 Sterilization The use of immunological and chemical sterilization has been modelled for the control of rabies in wildlife and in dogs (Suppo et al. 2000; Smith & Cheeseman 2002; Carroll et al. 2010). However, only surgical sterilization has been used in dogs under field conditions. Sterilizations are usually carried out by nongovernmental organizations and local authorities, which aim to vaccinate and simultaneously sterilize at least 70% of the dog population (Totton 2009). Limited data suggest that these programmes reduce the incidence of rabies and may stabilize or gradually reduce population density over time-scales of several years (Reece & Chawla 2006; Totton 2009; Totton et al. 2010). However, the respective impacts of vaccination and sterilizations have not been assessed. Reductions in population density may plausibly reduce the number of dogs that require vaccination, although timely reductions in density may be constrained by resources and population dynamics (Hemachudha 2005). As with culling, the demand for dogs by communities may result in an increase in dog importation where local supply has been reduced by sterilization. Thus, we are studying the effect of human behaviour in response to sterilization on dog population dynamics and disease transmission in Antiga, Bali. 1.3 Conclusion There is still considerable uncertainty surrounding the role of density in the transmission of rabies in animal host species. Density has been assumed to be the key factor that drives transmission, with important implications for the use of population reduction as a means to control rabies. However, it is evident that the relationship between host density, disease incidence and other factors is complex and varies between species. Further research to determine the factors that drive rabies transmission would not only enhance development of 13 epidemiological models but also inform the development of effective, sustainable disease control measures. Determining the effect of density in the transmission of rabies in wildlife hosts is constrained by the lack of high-resolution data exhibiting sufficient variability in both disease incidence and host densities. We have discussed how cycles in the incidence of rabies in foxes and raccoons can occur under either frequency- or density-dependent transmission, and how both model structures could account for the failure of culling to control rabies. Although still limited, better quality data for dogs suggests a more complicated relationship between contact rates and host density. The evidence indicates that not only is reducing dog density ineffective at controlling rabies, but culling in particular often has unintended negative consequences. We advocate more systematic investigation of the human factors that could affect the dynamics of rabies in dogs, to understand possible contrasts with the situation in wildlife. In contrast to culling, vaccination programmes against rabies in dogs (Cleaveland et al. 2003; WHO 2004; Schneider et al. 2005; Cleaveland et al. 2006; Davlin & VonVille 2012) and wildlife (Wandeler et al. 1988; Brochier et al. 1991; MacInnes et al. 2001; Rosatte et al. 2007) have proven efficacy and feasibility across a wide range of settings and raise far fewer ethical or welfare issues. 14 Chapter 2 The demography of free-roaming dog populations and applications to disease and population control Summary Understanding the demography of domestic dog populations is essential for effective disease control, particularly of canine-mediated rabies. Demographic data are also needed to plan effective population management. However, no study has comprehensively evaluated the contribution of demographic processes (i.e. births, deaths and movement) to variations in dog population size or density, or determined the factors that regulate these processes, including human factors. We report the results of a three-year cohort study of domestic dogs, which is the first to generate detailed data on the temporal variation of these demographic characteristics. The study was undertaken in two communities in each of Bali, Indonesia and Johannesburg, South Africa, in rabies endemic areas and where the majority of dogs were free-roaming. None of the four communities had been engaged in any dog population management interventions by local authorities or animal welfare organizations. All identified dogs in the four communities were monitored individually throughout the study. We observed either no population growth or a progressive decline in population size during the study period. There was no clear evidence that population size was regulated through environmental resource constraints. Rather, almost all of the identified dogs were owned and fed regularly by their owners, consistent with population size regulated by human demand. Finally, a substantial fraction of the dogs originated from outside the population, entirely through the translocation of dogs by people, rather than from local births. These findings demonstrate that previously reported growth of dog populations is not a general phenomenon and challenge the widely held view that free-roaming dogs are unowned and form closed populations. These observations have broad implications for disease and population control. The accessibility of dogs for vaccination and evaluation through owners and the movement of dogs (some of them infected) by people will determine the viable options for disease control strategies. The impact of human factors on population dynamics will also influence the feasibility of annual vaccination campaigns to control rabies and population control through culling or sterilization. The complex relationship between dogs and people is critically important in the transmission and control of canine-mediated rabies. For effective 15 management, human factors must be considered in the development of disease and population control programmes. 2.1 Introduction Understanding the demography of domestic dog populations in the developing world is critical for planning effective population management and disease control, particularly for rabies which causes around 55,000 human deaths per year (Knobel et al. 2005), as well as for other dog-mediated zoonoses prevalent in developing countries (Macpherson, Meslin & Wandeler 2012). Most human rabies cases are caused by bites from dogs infected with this fatal encephalitic disease. Rabies is directly transmitted by bites between dogs and can be readily controlled by canine vaccination (Jackson 2013). Demographic processes (i.e. births, deaths and the movement of dogs into and out of populations) contribute to variations in population size, density, vaccination coverage and disease transmission. A number of studies, encompassing a range of geographic locations, have assessed aspects of canine demography, including longitudinal studies where population management was not systematic (Chomel et al. 1987; Kitala et al. 2001; Pal 2001; Hampson et al. 2009). However, the contributions of demographic processes to population and disease dynamics, and the factors that regulate these processes, have not been comprehensively investigated. Several studies have estimated variations in the size of dog populations, where most dogs are free-roaming and where there has been no population control (Brooks 1990; Butler & Bingham 2000; Kitala et al. 2001; Pal 2001; Hampson et al. 2009; Acosta-Jamett et al. 2010; Gsell et al. 2012). These all report population growth, but with the exception of Pal (2001), growth was determined indirectly from estimates of births and deaths and age structure in a sub-set of dogs, or extrapolated from human census data and dog to human ratios. Rapid declines in vaccination coverage, necessitating at least annual vaccination campaigns, were also determined from similar data (Brooks 1990; Kitala et al. 2001; Hampson et al. 2009; Acosta-Jamett et al. 2010; Gsell et al. 2012). The effect of movement on these variations was not considered. There is increasing evidence that most free-roaming dogs are owned (Cleaveland & Dye 1995; Butler & Bingham 2000; Windiyaningsih et al. 2004; Gsell et al. 2012). Movement of dogs by people may therefore contribute to population dynamics (Chomel et al. 1987; Beran & Frith 1988), the spread of rabies (Denduangboripant et al. 2005; Zinsstag et al. 2009; Talbi et al. 2010; Townsend et al. 2013b), and off-set the impact of population control programmes 16 that aim to reduce population density through culling (Beran & Frith 1988; Windiyaningsih et al. 2004) or sterilization (Totton et al. 2010). In canine epidemiological and ecological models, it is often implicitly assumed that free- roaming dog populations are regulated through environmental resource constraints on births and deaths (Kitala et al. 2002; Hampson et al. 2007; Zinsstag et al. 2009; Totton et al. 2010). This has been described in rabies control policy by the logistic growth model (Wandeler 1985; WHO & WSPA 1990); a simple model that assumes all individuals have equal access to resources at the population level and births and deaths vary linearly and uniformly with population density, a surrogate for environmental resource availability (Sibly & Hone 2002; Vandermeer & Goldberg 2003). While the overall contribution of resource availability to variations in vital rates is controversial, empirical evidence suggests that it is important in a range of feral ungulate and wildlife populations (Choquenot 1991; Albon et al. 2000; Coulson, Milner-Gulland & Clutton-Brock 2000; Coulson et al. 2001; Bonenfant et al. 2002; Coulson et al. 2004). The relationship between density and births and deaths is demonstrable in these populations because of large fluctuations in population density about a carrying capacity, K. These fluctuations are either intrinsic or the result of deliberate perturbations, i.e. harvesting. Simple models assume that as populations approach or exceed K, all individuals in the population uniformly experience the effects of resource depletion. In reality, the relationship between density and births and deaths is complex and may be non-linear with greatest change in births and deaths at or near K (Fowler 1981), and resource depletion primarily affecting those individuals with the highest nutritional requirements, e.g. the survival of rapidly growing juveniles (Clutton-Brock, Major & Guinness 1985; Coulson et al. 2001; Bonenfant et al. 2002). The assumption that environmental resource constraints regulate dog populations has never been properly investigated. In accordance with the logistic growth model, it implies the primary food source of free-roaming dog populations is available to all dogs and most likely to be environmental refuse, and that dog populations are self-sustaining and population size and vital rates are self-regulating; effectively the dogs are unowned. This contradicts increasing evidence that most free-roaming dogs are owned (Cleaveland & Dye 1995; Butler & Bingham 2000; Windiyaningsih et al. 2004; Gsell et al. 2012) with serious implications for disease and population control measures and for animal welfare. However, determining the relationship between dog population density and births and deaths is hampered by the lack of longitudinal data from dog populations with substantial variations in density. Although dog populations are frequently culled, culls are often non-systematic and difficult to monitor, 17 which complicates assessment of the role of environmental resource constraints in regulating free-roaming dog populations. Furthermore, similar to ecological studies of feral ungulates and wildlife, quantifying variations in the distribution, volume, nutritional content and, more critically, uptake of environmental resources is generally not practicable. However, the unique relationship between dogs and humans affords a multifaceted approach, combining community-based methods with direct observations to infer the mode of ownership and food sources for individuals. Using this multifaceted approach, we directly measured temporal variations in population size and the contributions of births, deaths and human-mediated movement on these variations. We also investigated the effect of human and other factors, including environmental resource constraints, associated with these demographic processes. To facilitate comparisons between different environments and cultures, two populations of free- roaming dogs were selected in Bali, Indonesia, and two in Gauteng Province, South Africa, where rabies is endemic and where free-roaming dogs have been assumed to be unowned. We found constant or declining population sizes with no clear evidence of population regulation by environmental resource constraints; that almost all identified dogs in the communities were owned and fed regularly by their owners, consistent with population size regulated by human demand; that a substantial fraction of dogs originated from outside the population; and, that high levels of vaccination coverage may afford protection from rabies for up to two years. 2.2 Methods and materials 2.2.1 Research sites Data were collected from four communities, two in South Africa and two in Indonesia. The sites were selected during preliminary visits to Johannesburg and Bali in 2007 based on criteria including community support for the study, the absence of previous dog population management interventions by local animal welfare non-governmental organizations (NGOs) or authorities, geographic accessibility, operator safety and the availability of NGO support for data collection and translation. Rabies outbreaks occurred in Bali in 2008 and Gauteng Province in 2010. The two sites in Gauteng Province, South Africa comprised the informal settlement Zenzele west of Johannesburg (26.15 o S and 27.41 o E) and Braamfischerville in Soweto (26.12 o S and 27.52 o E). The study area encompassed the entire Zenzele township, whereas approximately 18 one third of Braamfischerville was included in the study area to include a comparable number of dogs to Zenzele. In Indonesia, the two sites included the villages of Kelusa (8.26 o S and 115.15 o E) and Antiga (8.30 o S and 115.29 o E) on the island of Bali. In Kelusa, the study area encompassed the entire village with the exception of Banjar Yehtengeh, which is separated from the rest of the village by rice fields and jungle, the southern half of Banjar Kelikikawan and the entrances (i.e. a typical compound housing extended families) scattered along the main road leading into the village. In Antiga, the study area encompassed all of the main residential area (Banjars Kaler and Kelod). An additional area within Banjar Ketug included entrances scattered along a 2.7 km stretch of road winding through the jungle north of Kaler and Kelod. All households in the study areas were included in the sampling frame. In Zenzele the sample unit, or household, is a systematically numbered yard with usually one or two shacks and poor fencing. In Braamfischerville, a household is a systematically numbered yard with a small, fixed structure, and a variable number of shacks and variable quality fencing. In Bali, a household is equivalent to an “entrance” associated with variable quality fencing. Given the lack of street names and house numbers, entrances were identified by photograph. With the exception of Zenzele, the study areas were established with no new households built during the study period. In May 2009 one new street of shacks was erected at the north end of Zenzele. Antiga and Zenzele are of comparatively lower socioeconomic status, and Banjar Ketug and Zenzele are without household sewage systems or water supply. 2.2.2 Method and type of data collected Individual-level data for every identified dog in the study area were collected longitudinally by direct observation and questionnaire from March 2008 until April 2011. The study population comprised of every owned dog (i.e. dog belonging to a household in the study area). Each owned dog was included in the study population immediately upon identification at its household, and recorded by photograph (standardised dorsal and lateral views) and owner questionnaire and visually assessed (see below for more details). Pups were recorded but not photographed until their third month of life. Dogs were not photographed consistently during 2010 because handling that occurred due to rabies vaccination campaigns in that year (as part of the same research project) caused them to become flighty and difficult to photograph and because by that time the primary researcher and enumerators were familiar with the majority of dogs. Each dog in the study population was individually recognisable and monitored at its household through direct observation, by the primary researcher and 19 enumerators, and questionnaire for the remainder of the study period or until it was lost from the study area. Households were visited during door-to-door censuses undertaken every 3-4 months (3-5 months in Braamfischerville) during the study period. The first inter-census period was longer (Zenzele: ~5 months, Braamfischerville: ~8 months and Kelusa: ~ 4.5 months). Eleven censuses were undertaken in Zenzele and ten in Braamfischerville, ranging from 12-23 days and 16-31 days to complete respectively. Nine censuses were undertaken in each of Kelusa and Antiga, ranging from 8-16 days and 11-19 days to complete respectively. All households with female dogs were revisited by the enumerators between all the censuses. Therefore, every household in the study area was visited frequently during the study period and most owned dogs were observed directly by the primary researcher and enumerators. However, a proportion of dogs were owned transiently between household visits and were not observed directly by the research team; these dogs were also recorded by owner questionnaire. These were generally young dogs that were acquired and then died between successive visits. Households were visited on foot during daylight hours and in approximately the same order. In addition to owned dogs being monitored at their household, every dog encountered in a yard not their own or on the street during each census was identified by the primary researcher and enumerators as either belonging to a household in the study area or not. Each dog identified as not belonging to a household in the study area was classified as unowned. A description of the unowned dog was recorded and, whenever possible, the dog was photographed. Only two dogs in Johannesburg and eighteen dogs in Bali were identified as unowned; these dogs were not included in the study population but are reported separately (see Results section 2.3.5). Focus groups with community leaders, members and enumerators, including participatory rural appraisal (PRA) techniques (Kumar 2007), were undertaken in Zenzele during February 2008 to a. investigate the ecology of the dog population in the study area, and b. develop the questionnaires. See Appendix 2.1 for a description of questionnaire development and implementation. Information collected, by direct observation and questionnaire, for every identified owned dog during each census and revisit included: house number, dog’s name, age, gender, source, outcome (e.g. died, relocated), reason for ownership, physiological, clinical and (for females) reproductive status, nutrition (type and source), body condition score (BCS) and level of confinement. Body condition score was a surrogate for food volume given the practical limitations of quantifying food uptake from all possible sources, including 20 owners. The dogs in Bali could not be readily handled. Therefore, a standard 9-point scoring system (emaciated score 1 to obese score 9) validated in adults (Laflamme 1997) and modified to assess body condition score without palpation was used. The modified system had been validated using dual energy x-ray absorptiometry in 71 dogs, including a small number of growing dogs (German & Holden 2006; German et al. 2006). Each dog, generally in its third month of life or older, received two independent body condition scores from the primary researcher and enumerators during each census or revisit. Clinical examinations were undertaken during each census by the primary researcher, a qualified veterinarian. The date and age of dogs at acquisition was reported by owners and/or visually assessed, including from the dentition of pups and juveniles (Dyce, Sack & Wensing 1987) in Johannesburg. For most dogs, these data were re-recorded at least once during the study period. The date a dog was lost from the study area was generally only recorded once. The month of loss was reported by the owner for 68% of the lost dogs in Zenzele, 78% in Braamfischerville, 51% in Kelusa and 55% in Antiga. Except where stated (see Table 2.4), the remainder were assumed to be lost uniformly between the census or revisit in which they were last recorded and the subsequent one. Apart from the primary researcher, all enumerators were local residents employed by NGOs. Data collection was standardised through detailed enumerator training at the start of the study and repeated on the first day of each census. All known refuse in the study areas was evaluated non-systematically by the primary researcher. Refuse was photographed periodically, and its distribution and the presence of edible organic matter assessed subjectively at each visit. Participatory approaches were preferred to mark-recapture or more technically demanding surveillance techniques, such as monitoring the movement of dogs with GPS collars, to further investigate the presence of a resident population of healthy, unowned dogs (i.e. dogs not belonging to households in the study area or outside the study area) in the Bali villages. Measurement error and statistical variation, violations of mark-recapture model assumptions and the need for repeat photographic mark-recapture preclude the use of these techniques to identify a real number of unowned, healthy dogs resident in the population, particularly where this sub-group is likely to be small. On this basis, participatory exercises were undertaken from April 2011 – April 2012 and are reported separately (Chapter 3) . 21 2.2.3 Analytical methods For the analysis pups are defined being in their 1st – 3rd month of life (i.e. 0 – ~13 weeks of age), juveniles 4th – 12th month of life (i.e. ~14 – 52 weeks of age), and adults older than their 12th month of life. Owned dogs were included in the analysis once they reached their third month of life (i.e. approximately 8 weeks of age) (here after referred to as “registered” dogs). Pups born in households in the study area but lost before their third month of life and unowned dogs are reported separately. Non-parametric regression was used to explore trends in population size, mortality and pregnancy. Visual inspection of plots and autocorrelation and partial-autocorrelation were used to assess periodicity, particularly seasonality, for these variables. An extended data set (i.e. by an additional 5 months) for Antiga was available. The proportion of reproductively mature females pregnant per month was estimated to avoid variation in pregnancy confounded by any seasonal variation in population size, most likely from disease-induced mortality such as babesiosis which has a reported seasonal distribution (Collett 2000). Bitches continually confined to a dog proof yard were excluded from these analyses. Monte Carlo estimates of the proportion of females in early (i.e. not visible) pregnancy when lost from the study populations were obtained by sampling from the observed distributions of age at first pregnancy and interval between the first and subsequent litters. Mortality was also estimated as the proportion of dogs dying per month and in terms of total mortality, specific disease-induced mortality and “other” (i.e. disease-induced mortality and dogs found dead, missing entries and unknown causes). We used Cox proportional hazard models to evaluate the risk of loss from the starting cohorts by age class at the start of the study and by gender. To model declines in vaccination coverage, estimates of vaccination coverage were obtained by assigning a random sample of dogs from the starting cohort equal in size to the proportion assumed to be vaccinated, and determining those still present at 12 and 24 months. This process was repeated 1000 times to produce Monte Carlo estimates of vaccination drop-off. A Bayesian ordinal regression framework (McKinley, Morters & Wood 2014) was used to investigate whether there were clear trends between body condition and increased caloric requirements from growth and lactation (National Research Council 2006). All observations with complete information for the variables of interest (Appendix 2.26) were included in the analyses. To account for observer variability, we fitted four versions of the model using minimum or maximum BCS (between observers) as response variables, and two definitions of gestation and lactation 22 (estimated [63 days gestation and 12 weeks lactation] and observed). Analysis was repeated without the first time point (i.e. censuses March-May 2008) to allow for owner reported clinical signs for the previous three months (see Appendix 2.1). We tested for an association between population size and births and deaths. Data analyses were conducted using R (R Core Team 2014) and C. 2.3 Results 2.3.1 Study population Throughout the study there was a very high level of compliance, with a low rate of partial- and non-respondents (Zenzele 1.5%, Braamfischerville 1.2%, Kelusa 2.5% and Antiga 2.2%). While some respondents declined to complete the full questionnaire, all provided partial information, such as the number, source and outcome, and permitted visual assessment of their dogs periodically throughout the study period. A total of 3240 owned dogs were registered during the study period: 1022 in Zenzele, 882 in Braamfischerville, 707 in Kelusa and 629 in Antiga (Appendix 2.2). Unless stated otherwise, all results pertain to these dogs. The sex ratio was approximately even in Johannesburg but skewed (male: female 75:25) in Bali through killing of unwanted female puppies (Appendices 2.3 and 2.4), and most dogs were adult (Appendix 2.5). The majority (>90%) of dogs were free to roam intermittently or continuously in Zenzele, Kelusa and Antiga, whereas in Braamfischerville approximately 40% of dogs were confined most of the time (Appendix 2.6). Most dogs were not sterilized except for 14.1% and 26.9% of male dogs in Kelusa and Antiga respectively. These dogs were “traditionally” castrated by a community member at about 6 months of age (Appendix 2.7). 2.3.2 Population size Variations in population sizes are shown in Figure 2.1 and Appendix 2.8 (and age class at registration Appendix 2.9). Overall there was a decline in population size in Zenzele (linear regression p<0.001) and Antiga (p<0.001), while the population remained constant in Braamfischerville (p=0.6) and Kelusa (p=0.5), with no seasonal variation evident. The population decline in Antiga from March 2010 may be attributed to fewer dogs being gained during this period than prior to March 2010 (Mann-Whitney test p=0.01), whereas a similar number of dogs were lost during both periods (p=0.5) [the mean number of dogs gained per month before March 2010 was 12.4 and from March 2010 was 7.3; the mean number of dogs 23 lost before March 2010 was 10.6 and from March 2010 was 12.5]. Population size (and density) varied overall by a maximum of 22.1% (ranging from 3.8%) from the mean. The mean number of dogs gained and lost per month ranged between 10.3-18.7 and 11.4-20.3 respectively (Appendices 2.10 and 2.11). With the exception of Antiga, the percentage of dog-owning households was constant (Zenzele 12.0% at the start of the study and 10.0% at the end, Braamfischerville 7.5% and 7.8%, Kelusa 71.9% and 72.9%, and Antiga 49.0% declining to 41.6%) and the number of dogs per dog-owning household was unchanged at approximately 1.3 in Johannesburg and Antiga and 1.7 in Kelusa (Appendix 2.2). Figure 2.1 Variations in population size for the study period. Each large dot represents the number of dogs in the registered population at the end of that month after accounting for the dogs gained and lost during that month. The non-parametric regression line (-) shows the average variation in population size during the study period, and the small dotted lines (.) the 95% confidence intervals for the mean. 24 In Zenzele, Braamfischerville, Kelusa and Antiga 7.9%, 26.1%, 2.0% and 4.5% of the registered dogs respectively were reported as present during the previous inter-census or - revisit period by respondents but were not observed directly by the primary researcher and enumerators. Excluding the non-observed dogs from the data set for Zenzele, Kelusa and Antiga did not change the trends in population size. However, removal of non-observed dogs from Braamfischerville resulted in an overall increase in population size during 2008 and 2009 when the interval between surveys was intermittently longer than the other sites. 2.3.3 Demographic processes There was no overall increase or decrease in the proportion of dogs pregnant and dying per month, and there was no seasonal variation (Appendices 2.12 - 2.16). In Kelusa total and “other” mortality for the entire study population increased significantly with time (p<0.001), and there was an overall increase in juvenile mortality (p<0.01). In Zenzele, juvenile mortality during the first inter-census period was lower than during the rest of the study period. Exclusion of this period (or even the first two months of the study) resulted in constant juvenile mortality (p=0.2). At least one third of the population was sourced from outside the study area (Zenzele 40.8%, Braamfischerville 59.5%, Kelusa 36.5%, Antiga 43.0%) (Table 2.1). Owners reported a variety of reasons for obtaining a dog from outside the research site (see Methods and Materials), although the most common reason was opportunism (Appendices 2.17 and 2.18). Most owners had planned to get a dog (Appendix 2.19), including a proportion (Zenzele 56%, Braamfischerville 74%, Kelusa 16% and Antiga 11% minimum) of those who found their dog by chance outside the research site. There was no overall increase or decrease in the proportion of dogs acquired from outside the study area per month (Appendix 2.20). Less than one third of the dogs were born in the household, and 50% or less were born in the study area (Zenzele 50.2%, Braamfischerville 16.0%, Kelusa 37.8% and Antiga 31.3%) (Table 2.1). A substantial proportion (15-20%) of dogs disappeared, were stolen or unaccounted for (Table 2.2). Appendix 2.21 shows the outcomes of pups born in study households. 25 Table 2.1 Sources of the registered dogs. Zenzele Braamfischerville Kelusa Antiga sourced as pups born at address 128 (19.0%) 60 (9.9%) 128 (28.1%) 87 (24.5%) elsewhere in study area 210 (31.2%) 37 (6.1%) 44 (9.7%) 24 (6.8%) (address not reported) (98) (4) (20) (6) non-study area of the research site NA 12 (2.0%) 12 (2.6%) 10 (2.8%) research site but area not known NA 78 (12.8%) 73 (16.0%) 47 (13.2%) outside research site 186 (27.6%) 221 (36.3%) 120 (26.4%) 112 (31.5%) not known 25 (3.7%) 25 (4.1%) 28 (6.2%) 20 (5.6%) sourced as juveniles or adults a non-study area of the research site NA 5 (0.8%) 0 3 (0.8%) inside study area 11 (1.6%) 0 0 0 research site but area not known NA 21 (3.4%) 10 (2.2%) 7 (2.0%) outside research site 89 (13.2%) 124 (20.4%) 34 (7.5%) 28 (7.9%) not known 24 (3.6%) 26 (4.3%) 6 (1.3%) 17 (4.8%) total acquired 673 609 455 355 a includes a small number of dogs where the age at acquisition was not reported but was most likely juvenile or adult (see Appendix 2.9) 26 Table 2.2 Outcomes of the registered dogs. Zenzele Braamfischerville Kelusa Antiga died 513 (70.2%) 355 (59.5%) 322 (72.0%) 264 (68.4%) disappeared 58 (7.9%) 30 (5.0%) 58 (13.0%) 39 (10.1%) stolen 23 (3.1%) 38 (6.4%) 0 0 given away in the non-study area of the research site NA 13 (2.1%) 3 (0.7%) 4 (1.0%) given away outside research site 40 (5.5%) 66 (11.1%) 14 (3.1%) 26 (6.7%) given to meat trader NA NA 18 (4.0%) 14 (3.6%) relocated outside research site with owner 39 (5.3%) 34 (5.7%) 11 (2.5%) 7 (1.8%) dumped 0 0 0 5 (1.3%) other 1 (0.1%) 0 0 2 (0.5%) unaccounted for 44 (6.0%) 47 (7.9%) 17 (3.8%) 19 (4.9%) given away in study area but not found by enumerators 13 (1.8%) 2 (0.3%) 0 0 given away in research site but area not known NA 12 (2.0%) 4 (0.9%) 6 (1.6%) total lost 731 597 447 386 27 2.3.4 Declines in the starting cohorts At least 20% (Zenzele 28.2%, Braamfischerville 23.9%, Kelusa 39.6% and Antiga 51.7%) of the dogs registered during the starting censuses were present three years later at the end of the study (Figure 2.2). Assuming 60% vaccination coverage of the starting cohort, coverage was estimated to decline to between 22-33% after 24 months, and assuming an initial coverage of 80%, coverage would be expected to decline to between 29-43% after 24 months (Table 2.3). The yearly relative decrease of the starting cohorts varied across age classes, with greatest declines recorded in younger age classes (Table 2.4). In Bali, dogs that were adults at the start of the study remained in the starting cohorts on average significantly longer than those that were juveniles and pups (p<0.001). Similar trends were observed in Johannesburg (Appendices 2.22 and 2.23). Appendix 2.24 shows the outcomes of the dogs in the starting cohorts. Figure 2.2 Declines in the starting cohorts for the study period, generated by plotting the number of dogs remaining from each starting cohort that month as a percentage of population size that month. The beige region indicates the vaccination coverage (of 20-45%) required to interrupt rabies transmission in a population (Hampson et al. 2009). 28 Table 2.3 Mean estimates of vaccination coverage with time (95% confidence intervals). % of the population vaccinated at month 0 % vaccination coverage after month 12 % vaccination coverage after month 24 Zenzele 60 40.0 (40.0-40.1) 26.1 (26.0-26.1) 80 53.4 (53.4-53.5) 34.9 (34.9-35.0) Braamfischerville 60 37.0 (36.9-37.1) 22.0 (21.9-22.1) 80 49.3 (49.2-49.3) 29.2 (29.1-29.2) Kelusa 60 40.2 (40.1-40.3) 27.4 (27.3-27.4) 80 55.8 (55.8-55.9) 36.6 (36.5-36.7) Antiga 60 43.2 (43.1-43.3) 32.6 (32.5-32.7) 80 57.5 (57.4-57.6) 43.3 (43.2-43.4) Table 2.4 Declines in number of dogs in the starting cohorts by age class and gender (males: females). The numbers of dogs are based on individual-level mid-point data which is a close approximation to the population-level averaged data. initial cohort of dogs declines in cohort 0-12 months declines in cohort 0-24 months declines in cohort 0-36 months Zenzele adults 276 (147:128) c 41.3% (65:48) 61.6% (90:79) 75.4% (109:98) juveniles 73 (37:36) 43.8% (19:13) 76.7% (28:28) 84.9% (31:31) pups 20 (14:6) 65.0% (7:6) 75.0% (8:7) 85.0% (10:7) Braamfischerville adults 221 (113:107) c 41.6% (43:48) 60.6% (63:70) 72.9% (75:85) juveniles 52 (33:19) 65.4% (22:12) 82.7% (27:16) 88.5% (30:16) pups 15 (6:9) 53.3% (4:4) 66.7% (5:5) 86.7% (6:7) Kelusa adults 209 (158:51) 27.3% (38:19) 44.5% (62:31) 58.4% (82:40) a juveniles 43 (33:10) 55.8% (17:7) 60.5% (19:7) 72.1% (23:8) a pups 27 (17:9) c 51.9% (10:3) 74.1% (13:6) 85.2% (13:9) a Antiga adults 217 (172:43) d 24.4% (38:14) 38.2% (59:23) 53.9% (86:30) b juveniles 32 (20:12) 43.8% (8:6) 68.8% (14:8) 81.3% (17:9) b pups 19 (12:7) 52.6% (7:3) 73.7% (10:4) 89.5% (12:5) b a 0-33 months; b 0-37 months; c the gender was not reported for 1 dog; d the gender was not reported for 2 dogs 29 2.3.5 Ownership The participants of the Zenzele focus group agreed unanimously that all of the dogs residing in the study area were owned. Similarly, almost all of the dogs in Johannesburg identified by the primary researcher and enumerators were owned by households in the study areas. One healthy dog in Zenzele and one in Braamfischerville did not belong to a household in the study area (0.1% of all observed dogs in their third month of life or older in each study area); these dogs were each observed on only one occasion. One other dog owned by a household in Zenzele was abandoned. The dog was subsequently observed during successive survey periods around the rubbish heaps on the outskirts of the study area in worsening body condition (BCS 1-2), until it was eventually adopted by a neighbouring household. Almost all of the dogs identified in the Bali sites were also owned by households in the study areas. Eight dogs in Kelusa and ten in Antiga did not belong to households in the study areas (1.1% and 1.5% of all observed dogs in their third month of life or older in each study area respectively). All of these dogs were observed on only one occasion, and almost all (12/16) were emaciated (BCS ≤2) and had severe generalised dermatitis (16/18). Although residents in the vicinity reported that at least six of these dogs were unowned, it was not verified if these dogs were owned by households outside the study area or were indeed unowned. In Kelusa 3.1% and in Antiga 5.3% of dogs owned by households in the study areas had BCS ≤2 and suffered from generalised dermatitis either consistently or transiently during the study period. The majority of respondents in Bali (Kelusa 84% and 90% and Antiga 85% and 88% during the April 2008 and January 2009 censuses respectively) reported that there were no unowned dogs in either their banjar (April 2008) or village (January 2009). Most remaining respondents (71%) reported ≤10 unowned dogs in the banjar and village. These respondents assumed that dogs were unowned on the basis that they roamed (38%) and were often in poor condition or “uncared for” (48%) rather than known ownership status. Although limited to dog-owning respondents, given that most households have experience of owning a dog, these data may reflect community opinion overall. 30 2.3.6 Food sources Almost all owned dogs had been reportedly fed by their owner on the day of the interview or the previous day. In Zenzele 2.6%, in Braamfischerville 0.6%, in Kelusa 0.1% and in Antiga 1.1% of dogs were last fed at least two days prior to the interview on one occasion and in Antiga 0.7% on more than one occasion. Only one juvenile in Antiga was reported as “never fed” by the owner, however long-term feeding patterns for this dog could not be established because it died two months after registration. Most of the rubbish in the study areas consisted of inedible, inorganic matter. The remainder was poor quality organic matter, including food waste. Rubbish heaps around the periphery of Zenzele were removed by the community during the winters of 2009 and again 2010 after re-accumulation. The local authorities regularly collected household waste in Braamfischerville and very occasionally in Zenzele. In Kelusa and Antiga, organic matter was often incinerated or fed to pigs and occasionally used as compost. None of the sites were within 5 km of alternative food sources such as municipal rubbish dumps or commercial abattoirs. Body condition scores were generally unimodal including individuals expected to have increased energy requirements from growth and lactation (Appendix 2.25). Although on average lactating individuals were thinner than non-lactating females and males, most were not underweight and body condition ranges were similar to non-lactating females and males. Overall, there was no tendency for growing individuals (i.e. pups and juveniles) to be thinner than adults (Appendices 2.26-2.30). Rather, on average young adults (13-36 months) had less subcutaneous fat than the other age classes probably consistent with normal anatomical variation (Lund et al. 2006). There was no interaction between age and lactation on body condition. Furthermore, even though sterilized dogs were on average fatter than unsterilized dogs, the relationship between body condition and age was similar for all four sites including Johannesburg where most dogs were not sterilized. There was no association between body condition and the number of dogs in the household. 2.4 Discussion The longitudinal, individual-level data in this study provides the most detailed demographic data currently available for domestic dogs in low-income communities in Asia and Africa, and provides valuable support for planning disease and population control programmes and 31 parameterisation of epidemiological models of infectious diseases, including rabies, in these settings. A key finding was that almost all of the identified dogs were owned by households in the study area, despite the vast majority being free-roaming, and were accessible for evaluation and vaccination. Our results are consistent with an interpretation that dog population size in these communities is regulated by human demand for dogs, not environmental resource constraints (i.e. food from refuse). We observed a range of body condition scores for these dogs regardless of energy requirements dependent on age or reproductive status, including lactation when energy requirements can be more than double (National Research Council 2006). This is consistent with owners reporting that most dogs were fed individually at the household on a regular basis. The low nutritional value of the refuse, and extremely poor body condition of most dogs not owned by households in the study area, also suggests that environmental resources were probably inadequate to meet the energy requirements of those dogs not fed properly by an owner, and that these dogs would not be able to survive in these environments without provisioning. Consistent with studies of environmental resource constraints in feral ungulate and wildlife populations (Clutton-Brock, Major & Guinness 1985; Coulson et al. 2001; Bonenfant et al. 2002), where mammalian population sizes fluctuate around K (Sibly & Hone 2002), if dogs are competing for environmental resources at the population level, then thin individuals may be those with the highest energy requirements. Although this association could be obscured by behaviours, such as hunting, domestic dogs are predominately scavengers (Bradshaw 2006) and, where refuse is the main environmental resource available to scavenge, such as in this study, this is unlikely. Thus, our data suggest that free-roaming, domestic dogs are not “wildlife”, competing for environmental resources to survive; rather humans are responsible for providing adequate care for this domesticated species. Furthermore, while our observations were intermittent and limited to daylight hours, we did not observe a resident population of dogs in reasonable or good body condition that were not fed daily by an owner. Overall, these observations were consistent with community opinion expressed during this current study, and participatory exercises undertaken in Kelusa and Antiga during 2011 and 2012 that utilised systematic ranking methods to obtain consensuses regarding the food sources of free-roaming dogs (Chapter 3). While our findings may not be universally applicable, they agree with previous studies, primarily in sub-Saharan Africa, that report the majority of free-roaming dogs as owned and fed regularly by their owners (Brooks 1990; de Balogh, Wandeler & Meslin 1993; Butler & Bingham 2000). 32 Data from this study demonstrate the contribution of different demographic processes, including human-mediated movement, to variations in population size and support the view that dog population size is primarily a function of human factors. However, the cultural, economic and social factors driving the rates of acquisition and disposal of dogs (and thus ownership) are still poorly understood and warrant further investigation. This includes responses by community members to fear of rabies or liability arising from dog bites to people in rabies-endemic areas. Indeed a reluctance to acquire dogs because of rabies may have driven the declines in the Bali populations from January 2010 in the wake of a rabies epidemic. Population growth may, in part, be limited in established communities where geographical expansion is minimal and the number of household units relatively stable, such as in this study. Previous estimates of growth have generally been at the national (Brooks 1990) or district (Butler & Bingham 2000; Kitala et al. 2001; Hampson et al. 2009) level and may be reliable when considering the ecological heterogeneity within, and limited movement of dogs into and out of, large geographical areas. However, human-mediated movement of dogs, including over large geographical distances, can seed incursions of rabies and make endemic transmission more difficult to interrupt (Denduangboripant et al. 2005; Talbi et al. 2010; Townsend et al. 2013b). Therefore, the level or scale of interventions for control programmes and policy need to be considered carefully. This study also has important implications for the design of vaccination campaigns, as the frequency of campaigns required to maintain vaccination coverage above the critical threshold of 20-45% (Hampson et al. 2009) depends on the introductions of susceptible individuals into the population by people through the acquisition of dogs born locally or from outside the population and the loss of vaccinated individuals through deaths and the relocation of dogs by people. We observed variable declines in the starting cohorts with time and age class. If 100% of these dogs had been vaccinated against rabies, and given the rates of birth, death and human-mediated movement recorded in this population, coverage sufficient to disrupt rabies transmission would have been maintained throughout the three year study period when using vaccine with a three-year duration of immunity. More realistically, vaccination coverage tend to fall in the range of 60-80%, with 80% achieved in the study areas during 2010, including the Bali sites where the dogs were less easy to handle. In these situations, vaccination coverage following a single campaign would decline to threshold levels after two years. Similar levels of vaccination coverage have been readily achieved in Africa (Kaare et al. 2009) and Asia (Bogel & Joshi 1990), and these results emphasise that the benefits of vaccination campaigns can be long-lived. 33 Our findings have practical consequences in terms of dog population control. Mass sterilization programmes, often used as an adjunct to vaccination and advocated as a necessary component of dog rabies control, will have a limited effect where population growth is limited and a large proportion of the population originates from outside the area. Furthermore, given the ongoing demand for dogs, any reduction in the local supply of puppies from sterilization, or following a cull, might result in further movement of dogs by people into communities from outside, compounding the risk of disease introduction. 2.5 Conclusion Our results demonstrate the importance of human factors in the design and implementation of disease and population control programmes and epidemiological models. Although owners generally facilitate vaccination of their dogs against rabies, movement of dogs by people can increase the spread of rabies and, necessitating widespread and sustained vaccination programmes in rabies endemic areas. Human factors are therefore critical factors that must be considered in the development of disease and population control programmes. 34 Chapter 3 Participatory methods for the assessment of the ownership status of free-roaming dogs in Bali, Indonesia, for disease control and animal welfare Summary The existence of unowned, free-roaming dogs capable of maintaining adequate body condition without direct human oversight has serious implications for disease control and animal welfare, including reducing effective vaccination coverage against rabies through limiting access for vaccination, and absolving humans from the responsibility of providing adequate care for a domesticated species. Mark-recapture methods previously used to estimate the fraction of unowned dogs in free-roaming populations have limitations, particularly when most of the dogs are owned. We used participatory methods, described as Participatory Rural Appraisal (PRA), as a novel alternative to mark-recapture methods in two villages in Bali, Indonesia. PRA was implemented at the banjar (or sub-village) level to obtain consensus on the food sources of the free-roaming dogs. Specific methods included semi-structured discussion, visualisation tools and ranking. The PRA results agreed with the preceding household surveys and direct observations, designed to evaluate the same variables, and confirmed that a population of unowned, free-roaming dogs in sufficiently good condition to be sustained independently of direct human support was unlikely to exist. 3.1 Introduction Understanding the characteristics of free-roaming dog populations is essential for the design of effective interventions to control canine diseases, such as rabies, and improve animal welfare. A critical issue relates to the possible existence of unowned, free-roaming dogs that are in sufficiently good condition to be sustained without direct human oversight. Ownership issues are critical for the design of rabies vaccination campaigns. Owners generally facilitate vaccination of their dogs against rabies (Lembo et al. 2010; Knobel et al. 2013), whereas unowned dogs are likely to be more difficult to identify and access for vaccination, potentially reducing effective vaccination coverage (Hampson et al. 2009), particularly if the fraction of unowned dogs is large. There is increasing evidence that most free-roaming dogs are owned and accessible for prophylaxis (Childs et al. 1998; Matter et al. 1998; Butler & Bingham 2000; Estrada et al. 2001; Kayali et al. 2003b; Windiyaningsih et al. 2004; Kaare et al. 2009; Lembo et al. 2010; Gsell et al. 2012; Putra et al. 2013). Previous studies, using mark-recapture techniques to evaluate vaccination coverage, generally indicate only a small 35 proportion (<10%) of free-roaming dogs are unowned in a range of urban and rural locations (Fishbein et al. 1992; Matter & Fico 1998; Matter et al. 1998; Cleaveland et al. 2003; Kayali et al. 2003b; Durr et al. 2009; Kaare et al. 2009; Gsell et al. 2012), although estimates with an upper confidence limit as high as 37% have been reported (Vos & Turan 1998; Matter et al. 2000; Kayali et al. 2003b). None of these studies reported the health status of unowned dogs, which remains an important gap in our understanding of these populations. However, there is a perception, implied by the implementation of interventions to reduce reproductive potential of unowned dogs, that these dogs are in sufficiently good condition for the population to be sustained without direct human oversight. An important corollary of this assumption is that it absolves humans from the responsibility of providing adequate care for a domesticated species. During an intensive three-year study [April 2008 – December 2010] in the villages of Antiga and Kelusa, Bali, Indonesia, all identified, free-roaming dogs in the study area were monitored individually by direct observation and household questionnaire every 6-12 weeks (average of 250-300 dogs in each village) (Chapter 2). The study area encompassed most of the village and included every household in the main residential area. Almost all of the identified dogs were owned (i.e. belonged to a household in the study area) and fed regularly by their owner. Consistent with this finding was the observation that the vast majority of the owned dogs were in reasonable or good body condition, and only a small proportion (i.e. Antiga 5.3% and Kelusa 3.1%) “unhealthy” (i.e. with ribs clearly visible and concomitant generalised dermatitis). Only eight of the identified dogs in Kelusa and ten in Antiga did not belong to households in the study areas. All of these dogs were observed on only one occasion over the three year period, and almost all were emaciated (12/16) with severe generalised dermatitis (16/18). The poor condition of these dogs is consistent with the lack of edible refuse in the environment, based on subjective assessment, and householders reportedly rarely feeding dogs other than their own. Therefore, all of the healthy dogs resident in the study areas were identified as owned and fed by their owner, and there was no evidence for a resident population of dogs in reasonable or good body condition not fed daily by an owner. Similarly, during household surveys the majority (~80%) of householders reported that there were no unowned dogs, with the remainder reporting generally ≤10 unowned dogs at any one time in the community. Householders generally assumed dogs to be unowned based on their health and confinement status (i.e. ”thin with bad skin” and “on the street”) rather than specific knowledge of an owner. Overall, these results suggested that a sub-group of 36 unowned dogs, in sufficiently good body condition to be sustained independently of direct human oversight, did not exist in these two villages. However, given the implications for rabies control and animal welfare, this study aimed to generate additional evidence relating to the ecology and health of free-roaming dog populations using an alternative approach, specifically community-level participatory exercises. Community-based participatory methods, termed Participatory Rural Appraisal (PRA), have been used extensively for research purposes by those from outside the community (Chambers 1994a; Chambers 2007) including for veterinary epidemiology (Catley, Alders & Wood 2012) and rabies control (Okell et al. 2013). These methods facilitate the sharing of local knowledge, and typically involve visualisation tools and ranking or scoring, but may also include group discussion or semi-structured interviews (Chambers 1994a; Chambers 2007; Upjohn et al. 2013). Triangulation, or the comparison of PRA outputs with results generated by gold standard methods evaluating the same variables, is necessary to validate PRA outputs (Catley 1999; Catley, Alders & Wood 2012). Historically participatory approaches were developed to address discrepancies between perceived community-level issues determined through conventional surveys and by the community themselves (Catley 1999). While neither are gold standard methods, PRA outputs have been shown to agree with, and thus verify, key findings from a limited number of conventional surveys designed to assess the same variables (Chambers 1994b; Upjohn et al. 2013). From previous studies (Putra et al. 2013), we assume that if a fraction of the free-roaming dog populations in Antiga and Kelusa was indeed unowned and in reasonable body condition, these individuals would comprise <10% of the population. Therefore, we preferred community-based participatory exercises to mark-recapture approaches given that it may be difficult to differentiate a real number of unowned dogs in reasonable body condition from measurement error and statistical variation, which may be large and encompass zero (Matter et al. 2000; Kayali et al. 2003b; Totton et al. 2010; Belsare & Gompper 2013). This may be compounded by violations of mark-recapture model assumptions, such as closed and stable populations. We determined a priori that population size was unlikely to remain constant between marking and recapture through frequent gains and losses of dogs (Chapter 2). Furthermore, the study populations were not closed and were confluent with the other populations in the non-study areas and neighbouring villages. Free-roaming dogs may travel substantial distances (Garde et al. 2012), therefore owned, unconfined dogs from the neighbouring villages may wander into the research villages. 37 We used PRA in Kelusa and Antiga, as a novel approach in dog ecology studies, to draw on local knowledge to obtain community-level consensus regarding the food sources of free- roaming dogs according to health and ownership status to infer the existence (or not) of unowned dogs in adequate body condition. Specifically, the PRA aimed to generate additional information about the health and ownership status of free-roaming dogs for triangulation of data from direct observations and household surveys. 3.2 Methods and materials Participatory exercises were used to generate banjar-level discussion and consensus on the food sources of four independent categories of free-roaming dogs (owned, unowned, healthy and unhealthy). The exercises were designed to avoid associations between the categories so that ownership and health status were not confounded; thus, ownership status was not defined by health status. Antiga included three banjars; and Kelusa six banjars, including one banjar not involved in the preceding surveys (Yehtengah). The banjar heads extended an open invitation to every person in their banjar to attend the PRA session. Sessions occurred from April 2011 to April 2012. With the exception of one session in Kelusa with female only participants, all sessions were held in the evening to maximise attendance. Sessions were mixed (male and female) for the Antiga banjars and for one banjar in Kelusa; however, because of cultural differences, sessions were divided into male and female for the remaining Kelusa banjars. Female sessions were run as part of the women’s community groups due to a reluctance of women to attend mixed-gender banjar sessions. One banjar (Triwangsa) declined to host a female session, probably because of the caste divide unique to that banjar. The one banjar in Kelusa not involved in the preceding surveys was included in the PRA as a means to determine whether the preceding survey work may have confounded the PRA outcomes. Each session ran for approximately 3 hours and included a short video on rabies prevention at the end. The participatory exercises were developed and implemented by a fully trained, experienced external facilitator from Praxis – Institute for Participatory Practices (India). Exercises were first developed and piloted with a team of seventeen Balinese who worked for three local animal welfare organizations including the one involved in the preceding surveys. All were trained to facilitate the planned PRA exercises as a team by the external facilitator during two sessions the week before starting in the banjars. A team was trained in anticipation of large numbers of participants for each PRA session. 38 The PRA sessions in the banjars were implemented by the Balinese team, with the external facilitator overseeing implementation by the team in Antiga. The external facilitator was not present in Kelusa. The exercises were carried out in Bahasa and Balinese, and all verbal and written outputs were recorded in English during the sessions. All outputs were drawn by the banjar participants to accommodate the less-literate; outputs were also written by literate participants as desired. Drawings were done on paper (A1 for body mapping and ~15cm 2 sheets for food source ranking) using coloured pens. The exercises were in three sequential parts (i) semi-structured discussion at the banjar level, (ii) visualisation exercises at the group level with feedback at the banjar level, and (iii) ranking exercises at the banjar level (Appendix 3.1). The semi-structured discussion regarding dog ownership aimed to prepare the participants for the visualisation and ranking exercises and for open discussion throughout the session. The visualisation exercises involved group-level drawings of healthy and unhealthy dogs, followed by banjar-level discussion of the drawings to establish the body condition of healthy and unhealthy dogs. Finally, food sources for four independent categories of dogs – healthy, unhealthy, owned and unowned were discussed and ranked in order of importance at the banjar level. For each the participants were asked to ignore the other classification. For example, when ranking food sources for a healthy dog, participants ignored whether the dog was owned or unowned. Ranking was iterative, with rankings re-ordered based on discussion and debate, until consensus on the final rankings was reached. All possible food sources for each health or ownership category were listed. No attempt was made to quantify average volume of each food source in the diet of dogs in each category. With one exception (see Appendix 3.3), banjar attendees were divided into at least four groups for the group-level activities, with women in a separate group to ensure their involvement. The exercises were considered culturally appropriate given that Balinese are generally familiar with banjar-level meetings and are artistic. A summary of the three highest ranked food sources for each health and ownership category and for each village are shown in Tables 3.1 and 3.2. The summary is derived from Appendices 3.5-3.8. For example, food purchased from a pet shop by the owner and food prepared by the owner for the dog were ranked as the most important food sources for healthy dogs in Kelusa. These two food sources are stipulated in Table 3.2 as rank 1 for healthy dogs. 39 3.3 Results The sessions were well attended, generally ranging from approximately 80-110 participants per banjar in Antiga and 35-95 in Kelusa. Two male sessions in Kelusa were poorly attended - in Peliatan because of torrential rain and in Triwangsa because of the caste divide. In the mixed sessions, the majority of participants were male, with approximately 8-30 female participants and a small number of children or young teenagers (Appendix 3.2). Contrary to the preceding household surveys, during the semi-structured discussions ownership was generally based on behaviour, e.g. with owned dogs identified through being responsive to a specific person, and skin condition, rather than body condition. Only one banjar (Ayah male group) differentiated owned from unowned dogs based on body condition. At the group level, body condition was not consistently listed in relation to health status. However, with one exception, at the banjar level dogs with ribs clearly visible were deemed unhealthy (Appendices 3.3 and 3.4). At the group level the number of characteristics related to health ranged from 3-13 (mode 9) and were diverse, including skin condition, behaviour, reproductive health, appetite and body condition, and a range of clinical signs. At the banjar- level, dogs with a runny nose, watering eyes, lameness, ticks and bad skin were classified as unhealthy, except for lameness in Kaler (where opinion was split), runny nose in Roban, and eye discharge in Peliatan (where two participants disagreed with the consensus that these dogs were unhealthy). Healthy and owned dogs had similar food sources, with food prepared by an owner ranked as the most important for all banjars. Unhealthy and unowned dogs had similar food sources, with rubbish and faeces ranked as the most important overall (Tables 3.1 and 3.2). Only two banjars (Kelikikawan and Roban) inferred that unhealthy dogs may be owned and probably neglected (Appendix 3.7). The results were similar for the banjar in Kelusa not included in the conventional surveys, and for male and female groups, although the women tended to suggest fewer food sources for owned and healthy dogs, restricted to food provided by an owner (Appendices 3.5 and 3.6). 40 Table 3.1 Summary of the three highest ranked food sources for Antiga (derived from Appendices 3.5-3.8 which show all the food sources and their ranks for each banjar). status rank food source healthy 1 prepared by owner / owner leftovers 2 prepared by owner / offerings 3 neighbour leftovers / rubbish owned 1 prepared by owner / owner leftovers 2 pig food / rubbish 3 offerings / dead animals unhealthy 1 rubbish / faeces / dead animals 2 rubbish / dead animals 3 rubbish / pig food unowned 1 rubbish / faeces / pig food 2 rubbish / pig food 3 rubbish / offerings Table 3.2 Summary of the three highest ranked food sources for Kelusa (derived from Appendices 3.5-3.8 which show all the food sources and their ranks for each banjar, including by participants’ gender). status rank food source healthy 1 pet shop a / prepared by owner 2 pet shop / prepared by owner / owner leftovers 3 owner leftovers / stolen from the neighbour / stealing poultry owned 1 pet shop / prepared by owner / owner leftovers 2 pet shop / prepared by owner / owner leftovers / rubbish 3 owner leftovers / stolen from the neighbour / rubbish unhealthy 1 rubbish / faeces / dead animals / vermin 2 rubbish / faeces / pig food / dead animals 3 rubbish / faeces / pig food / dead animals / vermin unowned 1 rubbish / faeces / stolen from householders / dead animals 2 rubbish / faeces / pig food / offerings / dead animals 3 rubbish /pig food / offerings / stealing poultry / vermin a the purchase of pet food by owners in Kelusa, but not in Antiga, is consistent with the socioeconomic differences between the villages 41 3.4 Discussion This study demonstrates the benefit of drawing on local knowledge through community- based approaches in dog ecology studies. However, our research also highlights the challenge of definitively identifying resident unowned, healthy dogs, particularly where most of the dogs are owned. The results from this study were consistent with the preceding household surveys and direct observations which suggested that a resident population of healthy, unowned dogs was unlikely to exist. It had been determined a priori by these household surveys and direct observations that a. all of the healthy dogs resident in the study areas belonged to households and were fed regularly by the householders, b. a minority of dogs belonging to households in the study areas were underweight, c. the majority of the dogs not belonging to a household in the study area were emaciated (although residents in the study area reported that at least six of these dogs were unowned, it was not verified if these dogs were actually owned outside the study areas) and, d. a resident population of dogs in reasonable or good body condition not fed daily by an owner was not apparent. This suggests that healthy dogs were owned and fed by their owners and, consistent with an apparent lack of edible refuse in the environment, dogs not fed adequately by an owner were unable to find sufficient environmental resources to meet their energy requirements. These a priori observations are supported by the PRA results. Firstly, there was no perception at the banjar level that the most important food source for healthy dogs was anything other than an owner. Rather, food from an owner was ranked as the most important food source for healthy and owned dogs. Secondly, similar food sources were listed for unowned and unhealthy dogs, suggesting that unowned dogs are indeed unhealthy. The food sources ranked as most important for these dogs included rubbish, faeces and dead animals. Consistent with a priori observations, these food sources probably provide insufficient nutrition for free- roaming dogs and, therefore, unowned dogs are unhealthy through poor nutrition. Taken together, these results imply that it was unlikely that a resident population of unowned, free- roaming dogs in reasonable or good body condition, existed in Kelusa and Antiga that were effectively “invisible” amongst the owned, free-roaming dogs. A key implication of this finding is that, to maintain a reasonable health status, dogs are dependent upon direct provisioning by people; free-roaming dogs should not be considered as “feral” populations and people cannot be absolved from the responsibility of providing adequate care for this 42 species. Results from this study indicate that almost all of the dogs in Kelusa and Antiga are owned and are, therefore, likely to be accessible for vaccination. Although the key findings from the PRA and direct observations and surveys were consistent and similar to previous studies (Chambers 1994b; Upjohn et al. 2013), there were also important discrepancies. A high ranking score indicated that some food sources other than from the owner were perceived to be important (Tables 3.1 and 3.2 and Appendix 3.6), which did not accord with the preceding direct observations or survey results. In these surveys generally <2% of dogs were reported to have eaten rubbish or other food outside the household, and owned dogs were infrequently observed scavenging (Chapter 2). The discrepancy may be attributable to the PRA methodology with the participants encouraged to “brain storm” all possible food sources for each category of dog, and no attempt was made to determine the volume or frequency eaten of a particular food type. It may also reflect the contrast between the mixed and male groups and the women’s groups. The women generally restricted food sources for healthy and owned dogs to the owner (Appendices 3.5 and 3.6). These results are more consistent with the preceding survey results and may be more reliable given that generally the women organise the food and feed the dogs in this society. This study was designed to identify the existence of a resident population of unowned, healthy dogs, if there was one, by inference from local knowledge. An important consideration is that, for ecological studies such as this, PRA is limited to verification of data collected by conventional methods and is not optimal when used as a sole modality. For example, had rubbish been consistently ranked equal to or higher than an owner for healthy dogs then, from this result alone, it would not be possible to differentiate between a. a resident population of genuinely unowned healthy dogs in the study area, b. some or all of the owned dogs in the study area obtaining a proportion of their nutritional requirements from rubbish, or c. the community misidentifying dogs owned outside the study area and that wander into the study area from unowned dogs. However, owner derived food was consistently associated with both health and ownership; this does not directly address the question whether unowned and healthy dogs existed, but failed to provide any evidence for their existence in any banjar. This is also supported by the observation that nutritional sources for unowned and unhealthy dogs were of poor quality and the same as those that would have been available to unowned, healthy dogs had they existed. Had PRA results diverged from results generated by our conventional methods, or had PRA been used as a sole modality, then additional objective methods would have been invaluable to triangulate the PRA results. For the example above, this would involve approaches that might have included further, 43 intensive focus groups, or more technically demanding methods such as monitoring the movement of dogs with GPS collars. 3.5 Conclusion This study has demonstrated the value of alternative approaches to mark-recapture to establish the presence of unowned dogs in adequate body condition in free-roaming populations, particularly where the fraction is expected to be small. While questionnaires have been used previously for this purpose (Butler & Bingham 2000), PRA is a novel approach that can generate additional data to complement conventional surveys designed to evaluate the same variables. This study provides further evidence that there is unlikely to be a population of free-roaming dogs in Bali that is capable of maintaining adequate health without any direct human oversight, with fundamental implications for disease control and animal welfare. 44 Chapter 4 Achieving population-level immunity to rabies in free-roaming dogs in Africa and Asia Summary Canine rabies can be effectively controlled by vaccination with readily available, high-quality vaccines. These vaccines should provide protection from challenge in healthy dogs for the claimed period for duration of immunity, which is often two or three years. It has been suggested that, in free-roaming dog populations where rabies is endemic, vaccine-induced protection may be compromised by immuno-suppression through malnutrition, infection and other stressors. This may reduce the proportion of dogs that seroconvert to the vaccine during vaccination campaigns and the duration of immunity of those dogs that seroconvert. Vaccination coverage may also be limited through insufficient vaccine delivery during vaccination campaigns and the loss of vaccinated individuals from populations through demographic processes. This is the first longitudinal study to evaluate temporal variations in rabies vaccine-induced serological responses, and factors associated with these variations, at the individual level in previously unvaccinated free-roaming dog populations. Individual- level serological and health-based data were collected from three cohorts of dogs in regions where rabies is endemic, one in South Africa and two in Indonesia. We found that the vast majority of dogs seroconverted to the vaccine; however there was considerable variation in titres, partly attributable to illness and lactation at the time of vaccination. Furthermore, >70% of the dogs were vaccinated through community engagement and door-to-door vaccine delivery, even in Indonesia where the majority of the dogs needed to be caught by net on successive occasions for repeat blood sampling and vaccination. This demonstrates the feasibility of achieving population-level immunity in free-roaming dog populations in rabies- endemic regions. However, attrition of immune individuals through demographic processes and waning immunity necessitates repeat vaccination of populations within at least two years to ensure communities are protected from rabies. These findings support annual mass vaccination campaigns as the most effective means to control canine rabies. 4.1 Introduction Canine-mediated rabies is a viral zoonosis, causing at least 55,000 human deaths every year (Knobel et al. 2005). Mortality from rabies is highest in less developed communities in Asia and Africa, where domestic dogs are free-roaming (Ezeokoli & Umoh 1987; Butler & 45 Bingham 2000; Kitala et al. 2002; Kayali et al. 2003a; Windiyaningsih et al. 2004; Reece & Chawla 2006; Kasempimolporn, Jitapunkul & Sitprija 2008); with increasing evidence that the majority are owned (Ezeokoli & Umoh 1987; Butler & Bingham 2000; Estrada et al. 2001; Windiyaningsih et al. 2004; Gsell et al. 2012) and, thus, generally accessible for vaccination (Lembo et al. 2010; Knobel et al. 2013). Canine rabies can be effectively controlled by vaccination (Cleaveland et al. 2003; Schneider et al. 2005; Cleaveland et al. 2006; WHO 2013) using readily available, high potency (antigenic value ≥1 IU/ml), inactivated cell-culture vaccines. These vaccines should provide protection from challenge in healthy dogs for the claimed period for duration of immunity (Council of Europe 2008), which is often two or three years. In free-roaming dog populations, vaccine-induced protection from rabies may be compromised for several reasons. These include: (a) insufficient vaccine delivery during vaccination campaigns (Lembo et al. 2010), (b) lack of repeat vaccination campaigns, with loss of vaccinated individuals from populations through demographic processes (Jackson 2013) (Chapter 2), and a substantial proportion of dogs probably vaccinated only once in their lifetime (Mitmoonpitak, Tepsumethanon & Wilde 1998), despite them often living beyond three years of age (Chapter 2); and, (c) the possibility of immuno-suppression through malnutrition, infection or other stressors (MSD Animal Health; Dionigi et al. 1977; Roitt, Brostoff & Male 2001), which may reduce the proportion of dogs that seroconvert or the duration of immunity of those dogs that seroconvert. These constraints may result in a decline in the vaccination coverage between campaigns to below 20-45%, the threshold necessary to control rabies (Hampson et al. 2009). Consequently, investigating the effectiveness of vaccination campaigns under field conditions is critical. The adaptive (B-cell humoral and T-cell cell-mediated) immune response to vaccination is complex. The humoral response generates virus neutralizing antibody (VNA), the primary correlate of protection induced by viral vaccines (Siegrist 2008; Johnson, Cunningham & Fooks 2010; Moore & Hanlon 2010; Jackson 2013). Cell mediated immunity (CMI) is also important for the development of vaccine-induced immunity (Thraenhart et al. 1994; Arya & Agarawal 2006; Corradi, Ferrari & Borghetti 2007) and acts in synergy with the humoral response (Siegrist 2008). Ongoing protection from challenge depends on the persistence of long-lived plasma cells, continuing to generate antigen-specific antibody, and B- and T- memory cells. The primary antibody response following vaccination generally correlates with the strength of the memory response (B- and T-cell) and, thus, the ability to induce secondary responses to subsequent challenge (Sikes et al. 1971; Brown, Merry & Beckenhauer 1973; 46 Bahloul et al. 2006; Lodmell et al. 2006; Hu et al. 2008; Siegrist 2008). In healthy dogs the quality of the primary immune response to vaccination depends on several factors, including the type of vaccine, with modified-live vaccines generally inducing superior responses, the route of administration, and the dose of vaccine antigen (Brown, Merry & Beckenhauer 1973; Aubert 1992; Lodmell et al. 2006; Hu et al. 2008; Siegrist 2008; Minke et al. 2009; Johnson, Cunningham & Fooks 2010). Laboratory challenge studies in healthy dogs support these observations. Following seroconversion, protection from rabies virus challenge correlates with peak VNA titre and final titre prior to challenge for inactivated, DNA and modified-live vaccines, with increased susceptibility to challenge once titres drop to near negligible levels (VNA titres <0.1 IU/ml or mouse serum neutralizing antibody titres <1:2 dilution) (Sikes et al. 1971; Brown, Merry & Beckenhauer 1973; Bunn, Ridpath & Beard 1984; Precausta et al. 1985; Sharpee, Nelson & Beckenhauer 1985; Bunn 1991; Aubert 1992; Bahloul et al. 2006; Lodmell et al. 2006; Hu et al. 2008). These studies used comparable antibody assays (Cliquet, Aubert & Sagne 1998; Ondrejkova et al. 2002) and virus challenge doses. Titres measured repeatedly over 3-4 years initially peaked and then declined rapidly, followed by a more gradual decline (Sikes et al. 1971; Brown, Merry & Beckenhauer 1973; Sharpee, Nelson & Beckenhauer 1985; Bahloul et al. 2006). While a titre of 0.5 IU/ml demonstrates seroconversion following vaccination (Kennedy 1998), the approximate threshold for protection following seroconversion may be 0.1 IU/ml (Precausta et al. 1985; Aubert 1992; Bahloul et al. 2006; CDC 2008). However, in the aforementioned experimental studies, only a proportion (<40%) of dogs with measureable titres following vaccination, but with negligible titres at the time of challenge succumbed to challenge, highlighting the importance of previously activated B- or T- cells allowing rapid response to challenge. Although the same relationship between VNA titre and protection from challenge is expected in immuno-suppressed dogs as in healthy dogs (MSD Animal Health; Roitt, Brostoff & Male 2001), no systematic comparison has been published to date. Reduced humoral immune responses have been shown in malnourished experimental dogs (Dionigi et al. 1977) and Gambian children vaccinated with human diploid-cell rabies vaccine (Moore et al. 2003), and pet dogs with anaemia or intestinal parasites vaccinated against rabies (Tepsumethanon et al. 1991; Aubert 1992). Several studies have evaluated the immune response in previously unvaccinated, mostly healthy pet dogs to high potency, inactivated rabies vaccine under field conditions (Tepsumethanon et al. 1991; Wilde et al. 1991; Sage et al. 1993; Sihvonen et al. 1996; Cliquet et al. 2003; Jakel et al. 2008; Berndtsson et al. 2011). All of these studies 47 report variable VNA titres up to 12 months following vaccination, including a proportion of dogs with titres ≤0.1 IU/ml (and generally a larger [17% to >42%] proportion with titres <0.5 IU/ml). These observations have serious implications for free-roaming dogs where their health status is more likely to be compromised. However, with the exception of one study in Peru (Chomel et al. 1987), no study has evaluated variations in vaccine-induced VNA in previously unvaccinated free-roaming dogs where rabies is endemic. Furthermore, no study has properly evaluated the factors associated with these variations. Cell-mediated immunity is technically difficult to measure under field conditions (Tizard & Ni 1998; Corradi, Ferrari & Borghetti 2007), however peripheral blood lymphocyte counts, which are predominately T-cells (Weiss & Wardrop 2010), may provide a straightforward, indirect assessment of CMI. Together with cytokine assays and measures of blastogenic responses of lymphocytes to mitogen, lymphocyte counts were used to assess immunomodulation in healthy dogs in response to vaccination (Phillips et al. 1989; Miyamoto et al. 1992; Strasser et al. 2003) and protein-calorie malnutrition (Dionigi et al. 1977), and in humans in response to protein-calorie malnutrition (Baron 1986). In dogs, malnutrition induced declines in immunoglobulin and lymphocyte function and counts. Therefore, lymphocyte counts together with rabies vaccine-induced titres and nutritional status may correspond to the overall immune status of an individual and susceptibility to infection. This study focused on evaluating temporal variations in vaccine-induced VNA, and factors associated with these variations, in three previously unvaccinated, owned free-roaming dog populations in South Africa and Indonesia, to better understand their effect on vaccination coverage. In addition, the efficiency of vaccine delivery and loss of vaccinated individuals from the cohorts were also assessed. 4.2 Methods and materials 4.2.1 Study populations See Appendix 4.1 for a summary of the methodology. Data were collected from three cohorts of dogs, one in South Africa, and two in Indonesia. The cohorts were part of a larger ecological study that commenced in March 2008 (Chapter 2). The South African cohort was located in Zenzele, an informal settlement 10 km west of Johannesburg (26.15 o S and 27.41 O E). In Indonesia the cohorts were located in the study areas of Kelusa (8.26 o S and 115.15 o E) and Antiga (8.30 o S and 115.29 o E), two villages on the island of Bali. Kelusa, 48 composed of six banjars (sub-villages), is inland. The study area encompassed the entire village except for Banjar Yehtengeh, separated from the rest of the village by rice fields and jungle, the southern half of Banjar Kelikikawan and the households scattered along the main road leading into the village. Antiga, a large village of six banjars, is located on the east coast. The bulk of the households are clustered into two banjars (Kaler and Kelod). The study area encompassed all of Kaler and Kelod. An additional area (Banjar Ketug) included households scattered along a 2.7 km stretch of road winding through the jungle north of Kaler and Kelod. Rabies is endemic in Indonesia and South Africa, with outbreaks occurring in Bali in 2008 and Gauteng Province in 2010. The Zenzele research cohort included every available dog in the entire township (which was the study area) in February 2010 that had not been previously vaccinated by the Department of Agriculture (DoA) during a vaccination point (VP) on the outskirts of the township in October 2009 (Appendix 4.2). All the dogs vaccinated by the DoA were identified within one week of the one day VP through a rapid door-to-door search, with verification by owners and inspection of certificates. The DoA had also set-up a VP on the outskirts of Zenzele in May 2006, thus vaccination history and certification were checked with each owner at the start of the study. VNA titres were also evaluated for anamnestic responses to vaccination consistent with previous vaccination. The Bali research cohorts included every available dog in the study areas of Kelusa and Antiga in January 2010 that had not been previously vaccinated by the Department of Livestock (DoL) as described below (Appendix 4.2). Prior to a rabies outbreak in 2008, vaccination against rabies was illegal in Bali and there had been no systematic vaccination programmes in either village prior to commencement of the study. Vaccination points were set up by the DoL in two banjars in Kelusa in December 2009 and in one banjar outside of the study area in Antiga in February 2010. The VPs were poorly attended because of community awareness of the research vaccination programme and because the owners could not readily handle their dogs. In Kelusa, 16 dogs from the study area attended the vaccination points. In Antiga only three dogs from the study area attended the vaccination point. All of the dogs resident in the study area were owned and had been previously identified by household, name and appearance through intensive monitoring by direct observation and survey since March 2008. Intensive monitoring of all of the dogs in the study area continued until April 2011. Therefore, all of the dogs in the study population were readily identified at the individual level during the study period. There was no evidence for a resident population 49 of unowned dogs (Chapters 2 and 3). All dogs in their third month of life or older were photographed (standardised dorsal and lateral views). Pups in their first or second month of life were recorded but not photographed. The same enumerators had tracked the majority of the cohorts at the individual level since March 2008 and were familiar with the dogs. 4.2.2 Vaccination and sampling Vaccine delivery was door-to-door for the research cohorts, and households were revisited repeatedly until the dog was caught for vaccination and blood sampling, or it was apparent that the dog could not be caught or the owner would not be available to give consent. A dog was also excluded from the study if the owner declined consent, the dog did not remain calm during restraint, there was a high index of suspicion that the dog may bite, or it was apparent the dog had a clinical condition that might have deteriorated as a result of restraint. All the dogs were carefully restrained by experienced personnel using the correct equipment and under the direct supervision of a veterinarian. In Zenzele, dogs were gently restrained with a leash and soft muzzle. In Bali most dogs could not be safely restrained by leash and muzzle and required restraint by net. Vaccinations and blood sampling were undertaken by experienced veterinarians. High-quality, sterile consumables (i.e. needle, syringe and blood tubes) were used for each vaccination and blood sample. Dogs in the research cohorts were vaccinated with 1ml of Rabisin (Merial Animal Health Limited), an inactivated rabies vaccine containing at least 1 IU/ml of rabies virus glycoprotein (GS57 Wistar strain) with an aluminium hydroxide adjuvant. Vaccine was administered subcutaneously into the neck or shoulder region. The vaccine cold chain was carefully preserved. Rabisin and Galaxy DA2PPv, a polyvalent vaccine against common infectious pathogens, was administered by the DoA during the October 2009 VP in Zenzele. Some dogs vaccinated at the VP may have received ivermectin. The DoL administered Rabisin during the February 2010 VP in Antiga, and Rabivet Supra 92, a locally produced cell-culture vaccine, during the December 2009 VP in Kelusa. Vaccine administration and storage by the local authorities were not observed. Different blood sampling schedules were required for Zenzele and Bali given the different methods of restraint and because the rabies outbreak in Bali escalated during 2009, forcing vaccination to be undertaken 6 months earlier than planned. Every dog in each research cohort, including neonates, was vaccinated at the start of the study (day 0) (Zenzele n=259, Kelusa n=284 and Antiga n=259 vaccinated [Appendix 4.2]), and every available dog from 50 about 6-8 weeks of age was blood sampled (see Table 4.1 and Appendix 4.3 for the number of dogs blood sampled at each time point). Blood was collected from the Zenzele research cohort on day 0 (immediately prior to vaccination) and then approximately 30, 90, 180 and 360 days following vaccination. The dogs vaccinated by the DoA were also blood sampled 8-10 days after the VP. Samples were then collected approximately 30, 90, 180 and 360 days following the VP. In Zenzele, only those dogs that had been vaccinated were blood sampled. Rabies vaccine-induced VNA was measured at each time point. Complete blood counts (CBCs) measured on days 0, 180 and 360 for the research cohort. In Bali, samples were collected on approximately day 180 and 360 following vaccination. Every available dog, whether vaccinated or not, was blood sampled at both time points and analysed for rabies vaccine-induced VNA. Unvaccinated dogs constituted the control group, and included those dogs not caught for vaccination on day 0 and those that arrived into the study populations after day 0. The sixteen dogs in Kelusa and three dogs in Antiga vaccinated by the DoL, in December 2009 and February 2010 respectively, were blood sampled at the same time as the research cohort. In all the sites, households were visited in approximately the same order at each time point, so the number of days between samples were similar for each dog. Table 4.1 The number of dogs in the research cohorts and the number of unvaccinated controls in Bali that were blood sampled at each time point (this table is reproduced with additional information in Appendix 4.3). day 0 a day 30 day 90 day 180 day 360 Zenzele vaccinated dogs 190 183 148 134 103 Kelusa vaccinated dogs _ _ _ 168 124 Kelusa unvaccinated dogs _ _ _ 70 79 Antiga vaccinated dogs _ _ _ 163 126 Antiga unvaccinated dogs _ _ _ 35 49 a day 0 immediately prior to vaccination for the research cohort 51 For each sample, 5-7ml of blood was collected from the jugular or cephalic vein and divided into plain and ethylene diamine-tetraacetic acid (EDTA) containing blood tubes. The blood tubes were immediately coded by date, house number and dog identification and placed in cool boxes with ice packs. Serum was separated by centrifugation within 8 hours of collection and refrigerated at 4-6 o C for up to 48 hours prior to freezing. All the sera were transported frozen in dry shippers to the Weybridge Animal Health Veterinary Laboratory Agency in the United Kingdom for fluorescent antibody virus neutralization (FAVN) assays. EDTA whole blood samples were refrigerated and then tested within 48 hours of collection for CBCs. Approximately 10 grams of faeces was collected manually on day 0 from 107 dogs randomly selected from the Zenzele cohort for routine analysis. Upon collection, the faecal sample pots were similarly coded and kept in the cool boxes, then refrigerated until being tested. Complete blood counts and faecal analysis were undertaken by the Faculty of Veterinary Science, University of Pretoria. Suitable laboratory facilities were not accessible in Bali for these tests. Finally, 32 dogs from Kelusa and Antiga combined were selected on day 180 from those dogs diagnosed with generalised dermatitis during the preceding survey for deep skin scrapes (DSS) from affected areas of skin to determine the prevalence of Demodex spp. See Appendix 4.4 for an explanation of sample selection for the DSS and faecal analysis. 4.2.3 Covariates Factors that may influence the immune response to rabies vaccine were selected on their measurability under field conditions, particularly by vaccinators. These factors had been previously quantified at the individual level as part of the larger ecological study that commenced in March 2008, and the methods used to quantify the factors are described elsewhere (Chapter 2). In summary, the factors were categorical and measured by direct observation and questionnaire at the time of vaccination (gender, age class , pregnancy, lactation, sterilization status [Bali only], intestinal parasites [Zenzele only]) or within 6 weeks of vaccination (body condition, clinical signs associated with serious illness, protein intake [Bali only], and generalised dermatitis [Bali only]) (Dionigi et al. 1977; Tepsumethanon et al. 1991; Aubert 1992; Roitt, Brostoff & Male 2001; Moore et al. 2003; Mansfield et al. 2004; Kennedy et al. 2007; Guaguere, Prelaud & Craig 2008; Miller, Griffin & Campbell 2013). See Appendix 4.18 for a detailed description of the covariates. Time (points) was treated as a continuous variable. 52 4.2.4 Analytical methods 4.2.4.1 Laboratory tests VNA was measured by fluorescent antibody virus neutralization (FAVN), a method prescribed by the Office International des Epizooties (OIE) (Cliquet, Aubert & Sagne 1998). In order to evaluate the variability in titres, including ≤0.1 IU/ml, the assay was modified to include a two-fold dilution with reciprocal dilutions ranging from 2 to 4096. Fifty percent endpoint titres, estimated by the Spearman-Karber method (WHO 1996), were converted into international units (IU/ml) by comparison with a standard serum. All samples were tested within two weeks of thawing and re-frozen within three weeks of testing. Except during assay preparation, all thawed samples were refrigerated. All of the samples from the same dog were tested within the same batch. Consequently, samples from each dog were frozen for a variable amount of time between collection and testing and a proportion of the samples were stored for over 12 months. To evaluate the effect of storage time and freeze-thaw cycles on titres, 25 samples were randomly selected from the first batch tested. These samples had been frozen (-20 o C) for over 2 years between the initial and repeat tests. To rule out cross-reaction with Lyssaviruses other than Rabies Virus (RABV), 30 samples were randomly selected from the Zenzele research cohort (day 0) and 60 from the Bali research cohorts and controls (day 180 and 360) combined and tested against Lagos Bat Virus (LBV), an antigenically divergent virus from Phylogroup II Lyssaviruses (Horton et al. 2010). Complete blood counts were determined by an automated cell counter (ADVIA 2120 Siemens) using impedence counting, flow cytochemistry, laser light scattering and validated vet package software. The differential leukocyte counts were confirmed by manual counting. Deep skin scrapes and faecal samples were evaluated using standard protocols (Soulsby 1982; Miller, Griffin & Campbell 2013) (Appendix 4.4). 4.2.4.2 Statistical methods A range of models were used to explore the relationship between time after vaccination and physiological and health status at the time of vaccination on titre. Correlation coefficients for titres and log titres were determined for a combination of time points (i.e. day 30, 90, 180 and 53 360) for the vaccinated dogs in Zenzele. This suggested that dogs with higher peak VNA titres also had higher titres towards the end of the study period (further described in Results section 4.3.4.2). Linear mixed effects models were fitted to the longitudinal data from the vaccinated dogs in the research cohorts using the nlme package in R (3.0.1) (Pinheiro et al. 2013; R Core Team 2014). Dogs vaccinated by the local authorities in Zenzele in October 2009, Kelusa in December 2009 and Antiga in February 2010 were excluded from these analysis because the administration of a standardised dose of Rabisin was not observed. The response variable, of VNA titre (here after referred to as “titre”) following vaccination, was modelled as the natural log of the titre (determined by Box-Cox transformation) expressed in IU. Therefore, baseline (day 0) titres were dropped from the Zenzele models and the unvaccinated (control) dogs were excluded from the Bali models. Explanatory variables included time (points) and the covariates described in section 4.2.3 (and Appendix 4.18) as fixed effects, and dog as a random effect. All individuals with complete information for the variables of interest were included in the models. Forward and backwards stepwise regression compared the full range of covariates and their biologically plausible interactions to the null model. The models with the lowest Akaike’s Information Criteria (AIC) for the highest number of observations were retained. Models were first fitted to each cohort separately. The Bali cohorts were then combined and the model refitted with dog nested within study area (i.e. village) as a random effect. Finally, all the research cohorts were combined and the models refitted. Each of these models were fitted with and without upper outliers (i.e. day 30 titres ≥128 IU/ml for 7 dogs in Zenzele, and day 180 titres ≥11.3 IU/ml for 4 dogs in Kelusa and 15 dogs Antiga) in order to exclude dogs from the analysis that may have been previously vaccinated by the DoA in Zenzele in May 2006, as part of vaccination campaigns outside of Kelusa and Antiga, or privately by their owners. Upper outliers were defined according to vaccination history, breed, age, source, geographical location and post-vaccinal titres (further described in Results section 4.3.3). The models take the form:   ijiijppijppijij XXXY    2 1110ln  , where ijY is titre and ijkX ( pk ,,1 ) are the covariates for observation inj ,,1 on individual mi ,,1 , where the final covariate ( ijpX ) is time. Hence, time is modelled as a 54 quadratic curve (Figure 4.1). The vector   10 ,,  p  is a vector of regression coefficients, and the vector  m ,,1  corresponds to a set of individual-level random effect terms, such that    m i i 1 0 . Finally the error terms  2,0~  Nij . This model was fitted to the full data set for Zenzele, which included all the time points (i.e. day 30, 90, 180 and 360). The data set included one to four data points for each individual depending on the availability of the individual for blood sampling during the study period. Times were adjusted by 30 days to allow the model intercepts to correspond to day 30 (peak) titres. A model using an exponential decay (rather than quadratic) over time was also fitted, however the quadratic model provided a marginally better fit to the data across this range, and so only the results from the quadratic model are reported here. Figure 4.1 Titres of all the dogs (n=82) in the Zenzele research cohort that were blood sampled all at four time points (30, 90, 180 and 360 days after vaccination). Upper outliers (i.e. the seven dogs with day 30 titres ≥128 IU/ml) are excluded. The geometric mean titre is shown in red. 55 The Bali data contained only one or two data points for each individual (i.e. day 180 and 360), and so instead a linear relationship to time was used (instead of quadratic). To facilitate comparisons with the Bali cohorts, linear models were fitted to truncated data sets for Zenzele (i.e. day 90 or 180 to 360), and the time (points) were adjusted to allow the intercepts to correspond to titres on day 90 or 180 respectively. Experimental studies report a spike in titre immediately following vaccination, followed by a prolonged, slow decline in titre (Bahloul et al. 2006). Although a quadratic relationship with time fits the Zenzele data set well over the observed range of the data (Appendix 4.17), it does not monotonically decrease over time and hence is a poor choice for predictions beyond the range of the data. Exponential decay models do decease monotonically, but do not have heavy enough tails given how we would expect the titres to decay outside the range of the data, based on previous studies (Bahloul et al. 2006). Fitting alternative models to a skewed distribution with heavy tail for predictions is challenging given that there is insufficient data in the extremes in order to robustly estimate the tail. Therefore, linear models, fitted to Zenzele data sets that exclude peak (day 30) titres, were selected to approximate prolonged, slow declines in titre for predictions in GMT beyond the last time point (Appendix 4.16). To explore the relationship between the natural log of the day 30 (peak) titres and the covariates described in Section 4.2.3. and lymphocyte and eosinophil counts on day 0, 180 and 360 for Zenzele, linear models were fitted to these data and model selection performed using stepAIC with the MASS package in R (3.0.1) (Venables & Ripley 2002). These models were equivalent to an analysis of variance. An association between eosinophil counts and antibody titres at each time point was also assessed. The final models were checked for violation of constant variance and normal error distribution assumptions. Mann-Whitney tests were used to compare titres between (a) vaccinated dogs in Zenzele, Kelusa and Antiga for the same time points, (b) unvaccinated dogs in Kelusa and Antiga for the same time points, and (c) dogs present in Zenzele in May 2006 and those that arrived into the population after May 2006. The natural log of the titre was used for these comparisons. The Mann-Whitney test was also used to compare peak (day 30) titres between dogs with day 360 titres <0.5 IU/ml and ≥0.5 IU/ml in Zenzele, and the cube root transformation of the titre was used to stabilise the variance according to a Box-Cox transformation. Population structures were stable (Chapter 2), therefore life expectancies were estimated, by standard 56 analysis of vertical life tables (Caughley 1977; Pianka 1999; Gsell et al. 2012), from the observed ages of the entire study population at the end of the study period. 4.3 Results 4.3.1 General description of the study populations Almost all of the dogs in the study populations were owned but free-roaming, with <10% confined continuously or frequently during the study period March 2008 – April 2011. There was an approximately even ratio of male to female dogs in Zenzele, but the ratio was skewed towards males (approximately 75%) in Bali. Less than 2% of dogs were sterilized in Zenzele, but castration of juvenile male dogs by community members was common in Bali (approximately 14% in Kelusa and 27% in Antiga) (Chapter 2). Life expectancy was at least 3 years for the majority of dogs in the study populations (Appendices 2.5 and 4.5). 4.3.2 Vaccination coverage High vaccination coverage was achieved through door-to-door vaccine delivery: 82% (259/315) in Zenzele, 81% (284/351) in Kelusa and 79% (259/327) in Antiga. Similar coverage (75-86%) was achieved in Bali for blood sampling at day 180 and 360, despite many of the dogs having been caught on at least one previous occasion (Appendix 4.2). The characteristics of dogs that avoided capture are described in Appendix 4.6. The sex ratio and age distribution of these dogs were similar to the overall population (Appendix 2.5). Attrition of the cohorts occurred during the study period through mortality, particularly of neonates, but also through the relocation and disappearance of dogs (Chapter 2). Of the 259 dogs vaccinated in Zenzele at the start of the study, 103 (40%) were sampled at the last time point. Similar proportions were recorded in Kelusa (44%, n=124) and Antiga (49%, n=126) (Appendices 4.2 and 4.3). 4.3.3 Assessment of prior vaccinations In the Zenzele research cohort, upper outliers were defined as dogs with peak titres (on day 30) of 128 IU/ml or greater (n=7). Some of these dogs were either in the study area in May 2006 or may have been previously independently vaccinated by their owner. Baseline titres of the upper outliers were ≤0.25 IU/ml, most with a titre of ≤0.09 IU/ml. The history of those individuals with the next highest titre (91 IU/ml) varied, and included seven dogs that were born in Zenzele after October 2009. 57 It is unlikely that any of the dogs vaccinated by the DoA four months prior to initiation of vaccination of the research cohort were inadvertently included in the research cohort. The day 0 titres of the research cohort (including upper outliers ranged from 0.06 - 1 IU/ml with a GMT of 0.1 IU/ml) were substantially lower than the day 90 titres of the DoA cohort (including upper outliers ranged from 0.06 -128 IU/ml with a GMT of 2.8 IU/ml). Thirteen (20%) of the dogs vaccinated by the DoA had titres ≤1 IU/ml 90 days after vaccination, of which 6 had titres <0.5 IU/ml and four of these were non-responders (i.e. day 30 titre of <0.5 IU/ml). Only five dogs in the research cohort had day 0 titres ≥0.5 IU/ml, and of these none appeared to have an anamnestic response to the vaccine (day 30 titres ranged from 1.4 - 45 IU/ml) (Appendices 4.7 and 4.8). There were no differences in the distributions of titres for dogs in Zenzele probably present in May 2006, when the DoA vaccinated, and those that arrived into the population after May 2006 (Appendix 4.9). In the Bali research cohorts, upper outliers were defined as dogs with day 180 titres of 11.3 IU/ml or greater (n=4 in Kelusa; n=15 in Antiga). For some of these dogs, information provided by owners, breed, source and geographical location was suggestive of vaccination undertaken independently by their owner or as part of vaccination campaigns outside of Kelusa and Antiga. Several (n=15) unvaccinated controls had titres ≥0.5 IU/ml (Appendix 4.10). The titres of the unvaccinated controls are summarised in Appendix 4.11. 4.3.4 Evaluation of antibody titres 4.3.4.1 Titre variations in the vaccinated dogs The quality of the serum samples was excellent, with only a few samples with slight to moderate haemolysis. Most dogs in Zenzele seroconverted (97% of the research and 92% of the DoA cohorts had titres ≥0.5 IU/ml at day 30), however there was considerable variability in titres at each time point (Figure 4.2). The estimated dog-dog variation (random effect) in peak titres (quadratic model intercept) was large (+/- 2SD 1.8 – 99 IU/ml) (Appendix 4.16, model 1). Excluding upper outliers, the observed geometric mean titres (GMT) at day 30 for the research cohort (of 15 IU/ml, Appendix 4.17) was comparable to experimental (Bahloul et al. 2006; Hu et al. 2008) and field (Tepsumethanon et al. 1991) studies of previously unvaccinated dogs. The maximum peak titre was more than double the upper limit of the other studies (40-50 IU/ml), however those dogs with peak titres >40 IU/ml included seven dogs born in Zenzele after October 2009 which were unlikely to have been vaccinated prior to commencement of the study. There was similar variability in the titres at each time point 58 for the Bali cohorts (Figures 4.3; Appendix 4.16, models 3-5). See Appendix 4.12 for details of the dogs in Zenzele that did not seroconvert to the vaccine. Figure 4.2 Titres of all the dogs in the Zenzele research cohort. Upper outliers (i.e. the seven dogs with day 30 titres ≥128 IU/ml) are excluded. The median titre (thick, horizontal line), 25th and 75th percentiles (thin horizontal lines), and either minimum and maximum titres or 1.5x the interquartile range (dashed vertical lines) are shown for each time point after vaccination (at day 30, 90, 180 and 360). Day 0 shows the distribution of titres immediately prior to vaccination. 59 Figure 4.3 Titres of all the dogs in the Kelusa and Antiga research cohorts. Upper outliers (i.e. the four dogs in Kelusa and fifteen dogs in Antiga with day 180 titres ≥11.3 IU/ml) are excluded. The median titre (thick, horizontal line), 25th and 75th percentiles (thin horizontal lines), and either minimum and maximum titres or 1.5x the interquartile range (dashed vertical lines) are shown for each time point after vaccination (at day 180 and 360). Although the GMTs for the Bali cohorts were statistically significantly higher than Zenzele (Mann-Whitney test p≤0.05) for day 180 and 360, the value of the means and modes were comparable between cohorts at each time point (Appendix 4.13). Less than 10% of each cohort had titres of ≤0.1 IU/ml at day 360. For Zenzele, peak titres did not exceed 5.7 IU/ml for these dogs, and three were non-responders. See Appendix 4.14 for details of the dogs in each cohort with titres ≤0.1 IU/ml 360 days after vaccination. Between 20-40% of dogs overall had titres <0.5 IU/ml at the last time point. Excluding upper outliers, dogs in Zenzele with day 360 titres <0.5 IU/ml had a statistically significantly (Mann-Whitney test p<0.001) lower day 30 GMT (6.6 IU/ml, n=38) compared to dogs with day 360 titres ≥0.5 IU/ml (23.6 IU/ml, n=57); this is consistent with the correlations between time points discussed below. 4.3.4.2 Kinetics of titres in the vaccinated dogs In Zenzele, titres declined rapidly between day 30 and 90, then gradually from day 90 (Figures 4.1 and 4.2). Log titres were closely correlated across all the time points, including between day 30 and 360 (excluding outliers correlation coefficient r = 0.55) and day 30 and the mean log titre for day 90, 180 and 360 (r = 0.72) (Appendix 4.15). Consequently, dogs 60 with higher peak titres tended to have higher titres at the final time point. The model assuming a quadratic relationship between titre and time was a good fit, with predicted GMT for the day 30 (peak) and 360 titres congruent with the observed means (Appendix 4.16, model 1; Appendix 4.17). Peak (day 30) titres could not be extrapolated from the linear models for Bali, but the decline in titres between 180 and 360 in Kelusa was similar to Zenzele (Appendix 4.16, models 2-4). The GMT for Antiga declined only marginally with time (slope p=0.4; Appendix 4.16, model 5), with a rate of decline less than quarter that of Zenzele and Kelusa. Overall, the predicted GMTs were comparable to the observed titres (Appendix 4.16, models 1-5; Appendix 4.17). The magnitude of any decline in titre as a consequence of extended storage time or freeze- thaw cycles was not great compared to normal background variation. This agrees with other studies evaluating the effect of storage time and freeze-thaw cycles on blood proteins (Thoresen et al. 1995; Reynolds et al. 2006). The decline in titre for 22 (88%) of the samples did not exceed normal inter-assay variation of two-fold or less (Gilbert et al. 2013). 4.3.4.3 Factors associated with variations in titre When comparing the research cohorts, all models with time (points) had lower AICs than the null models except for Antiga when upper outliers were excluded, indicating that time after vaccination had an effect on titre. While there were no clear patterns across the cohorts between variations in titre and the covariates described in Methods and materials section 4.2.3., lactation and health status emerged as significant covariates. For Zenzele, apart from lactation at vaccination, there were no statistically significant (p<0.05) associations of titre with age, gender, reproductive and health status, and body condition when accounting for all the time points (i.e. for models of the longitudinal data). Time and lactation were the only covariates retained in the quadratic model with the lowest AIC, where the negative effect of lactation was statistically significant (p≤0.02) (Appendix 4.19, model 1). When the response variables was restricted to peak (day 30) titres, titres again varied significantly with lactation (p≤0.02) (Appendix 4.19, model 2). Overall, the GMT of lactating dogs (~6 IU/ml) was less than half that of males and non-lactating females. Clinical signs at the time of vaccination was also significant (p=0.04) when the response variable was restricted to peak (day 30) titres but only when upper outliers were included in the model (Appendix 4.19, model 3). Those dogs with clinical signs at the time of vaccination had a GMT of 11 IU/ml, approximately half that of dogs without clinical signs (21 IU/ml). When 61 body condition was dropped from the model, clinical signs at the time of vaccination was marginally statistically significant (p=0.06) (Appendix 4.19, Note 4); the differences in factor levels (i.e. with and without clinical signs at vaccination) between the two models were comparable (9.4 IU/ml including and 8.0 IU/ml excluding body condition). Lactation and health status at vaccination were also the main covariates of significance in the Bali villages. Lactation and pregnancy were excluded from the models for Kelusa (Appendix 4.20) given their small group sizes (Appendix 4.23) and these covariates were not retained in any of the models for Antiga (Appendix 4.21). However, a positive effect of lactation at vaccination was marginally statistically significant (p=0.07) to statistically significant (p=0.05) for Kelusa and Antiga combined when upper outliers were included in the model (Appendix 4.22, model 1 and Note 2). When generalised dermatitis at the time of vaccination was included in the combined model, lactation was no longer significant (Appendix 4.22, model 2). Generalised dermatitis was statistically significant (p≤0.02), although the GMT was only approximately 0.5 IU/ml less than the baseline (i.e. dogs not lactating without dermatitis GMT ~2 IU/ml). Generalised dermatitis was retained in models for Kelusa, but it was not statistically significant (p=0.15) (Appendix 4.20, model 1) even though 44 (37%) of dogs at the time of vaccination were affected (Appendix 4.23). However, generalised dermatitis at the time of vaccination was generally highly statistically significant for the Antiga cohort (p≤0.01) reducing the GMT by up to half (Appendix 4.21, models 2 and 3). Convergence errors, regardless of fitting method, precluded full evaluation of the data set combining all three research cohorts. 4.3.4.4 Rabivet Supra 92 Of the 16 dogs vaccinated in December 2009 with Rabivet Supra 92 all had titres <0.5 IU/ml except for one dog at day 180 and two at day 360 sampling (Appendix 4.24). 4.3.4.5 Lagos Bat Virus assays The samples were negative for the Zenzele cohort. All the samples were negative for the Bali cohorts except for one dog in Kelusa, with a 50% end-point titre of 1/64 upon initial testing and 1/32 upon re-testing. 62 4.3.4.6 Other diagnostics Lymphocyte counts were significantly associated with body condition at the time of vaccination (p=0.03) (Appendix 4.19, model 4), however there was no association with peak (day 30) titres (p>0.05). There were no associations between eosinophil counts and titres at any time point. Almost all of the dogs had intestinal parasites, primarily Ancylostoma spp. (Appendix 4.25) (Minnaar, Krecke & Rajput 1999; Minnaar & Krecke 2001). Consequently, there was insufficient variability to determine the effect of intestinal parasites on immune response to vaccination. One dog was positive for Demodex spp. on deep skin scrape. 4.4 Discussion The longitudinal, individual-level data from this study provides the most detailed serological data currently available for domestic dogs in rabies endemic areas, and provides valuable support for planning rabies vaccination programmes. This study reinforces the importance of frequent and regular vaccination campaigns to ensure effective vaccination coverage is maintained. Dogs with lower peak titres had correspondingly lower titres at the end of the study, with titres <0.5 IU/ml at the last time point (day 360) for 20-40% of the dogs and <0.1 IU/ml for 3-8% of the dogs (Appendix 4.13); the implication being an increased susceptibility to natural exposure with time in the dogs with low titres (Precausta et al. 1985; Aubert 1992; Bahloul et al. 2006; CDC 2008; Siegrist 2008). Robust demographic data from these study populations indicates, two years after a pulse campaign which achieved 80% vaccination coverage, at least 20-45% vaccination coverage would remain (Chapter 2), which is the critical threshold necessary to prevent rabies (Hampson et al. 2009). However, from our model predictions (Appendix 4.16), we speculate that a substantial proportion of the dogs remaining in Zenzele two years after vaccination may have titres <0.1 IU/ml, potentially dropping effective vaccination coverage to below the critical threshold. Models were constrained to two time points for the Bali cohorts, but predicted similar declines in the GMT for Kelusa. The vast majority of the dogs seroconverted following vaccination (with a peak titre of ≥0.5 IU/ml), regardless of health status. However, there was considerable variation in titres at each time point for all the cohorts. Peak titres were not measured for the Bali cohorts, however day 180 titres were comparable to Zenzele, therefore it is likely that a similar proportion of dogs 63 to Zenzele seroconverted following vaccination. Identification of risk factors associated with lower titres may promote targeted boostering to maintain vaccination coverage. Clinical conditions around the time of vaccination reduced the immune response to the vaccine in all the cohorts; in particular, generalised dermatitis provided a ‘visible marker’ for a reduced immune response, with practical implications for rabies control. While demodicosis was assumed to be an important cause of generalised dermatitis associated with immuno- suppression in Bali, the mostly negative skin scrapes suggests that dermatophytosis may be more likely, consistent with both the tropical climate and immuno-suppression (Guaguere, Prelaud & Craig 2008; Miller, Griffin & Campbell 2013). This warrants further investigation given that a substantial proportion of the dogs (37%-46%; Appendix 4.23) were affected, potentially reducing the effectiveness of vaccination. Lactation at the time of vaccination in Zenzele and the Bali cohorts combined was marginally significant statistically, however its biological significance is unclear. Lactation is associated with loss of body condition in all the research sites (Chapter 2), consistent with immuno-suppression observed in Zenzele. The reason for the opposite effect in Bali cannot be readily explained (Lloyd 1983; Lloyd, Amerasinghe & Soulsby 1983). While this incongruity may warrant further investigation in larger study populations, on balance lactating bitches should be vaccinated, with re- vaccination following weaning. Our study demonstrated an advantage of community engagement and door-to-door vaccine programmes over the use of simple vaccination points. We consistently achieved vaccination coverage above 70% through door-to-door vaccine delivery, even in Bali where the majority of the dogs needed to be caught by net on successive occasions. Similar coverage was achieved across the rest of the island through door-to-door vaccine delivery in 2010 and 2011 (Putra et al. 2013). This compares to a vaccination coverage of only 27% through the vaccination point in Zenzele and a very low vaccine uptake (5%) in Kelusa. The utility of vaccination points is likely to differ between locations according to local circumstances. Similar to other communities in Africa, Europe and central Asia where free-roaming dogs are handleable (Bogel & Joshi 1990; Cleaveland et al. 2003; Kaare et al. 2009; Lembo et al. 2010), it is likely that the majority of the dogs in Zenzele could have been delivered to the vaccination point by their owners, and the low vaccination coverage was probably the result of inadequate advertising (Durr et al. 2009) and limited operating hours during a work / school day. Vaccine uptake in Kelusa was, in part, affected by community awareness of the research vaccination programme, however the majority of the dogs could not be handled by their owners or the vaccinators, thus necessitating restraint by net (Putra et al. 2013). The 64 reasons for the difference in handleability between locations are unclear. Restraint by net is more stressful to the dog, time consuming and costly than by leash and muzzle. In order to improve welfare, facilitate more cost-effective and efficient delivery of vaccines (and other prophylactics), and improve evaluation of the dogs in Bali and similar communities, extending our studies to evaluate the differences in husbandry, environment and other factors influencing the temperament of the dogs in the sites is warranted. This research has generated valuable data that may contribute to rabies control, including through improving epidemiological models. However, understanding variation between dogs in titres measured from field studies is challenging. Some covariates which may impact on titres, such as lactation and health status, are measurable, whereas others such as genetics and stress are harder to assess in real time. Further evaluation of factors associated with variation in immunity over time since vaccination, including both serological responses and direct assessment of CMI, and recording vaccine failures is warranted and may require larger populations studied and over longer time periods. 4.5 Conclusion This study demonstrates that the vast majority of free-roaming dogs, in two regions of Africa and Asia where rabies is endemic, seroconverted to rabies vaccine regardless of health status producing titres that exceeded 0.5 IU/ml, the level considered necessary to protect against rabies. Declines in vaccination coverage following a vaccination campaign occur through mortality / emigration of vaccinated dogs and birth / immigration of unvaccinated, susceptible dogs. Robust demographic data from the study populations show that two years after vaccinating at least 70% of dogs during a pulse vaccination campaign, vaccination coverage remained within 20-45% (Chapter 2), the range necessary to control rabies (Hampson 2009). However, our serological data indicates that dogs with lower peak (day 30) titres had correspondingly lower end point (day 360) titres. We speculate that a proportion of vaccinated dogs remaining in the study populations after two years will probably have titres below the approximate threshold for protection (<0.1 IU/ml) thus dropping effective vaccination coverage to below the critical threshold (of 20-45%). This emphasizes the importance of re-vaccinating within two years. Vaccination of all dogs during annual campaigns is therefore recommended as the most effective means of ensuring that individual immunity and population coverage are both maintained at sufficient levels to control rabies. 65 Chapter 5 Modelling human interference and rabies transmission in free-roaming dogs Summary Deterministic, compartmental models have been used to explore local transmission dynamics of canine rabies, particularly for the development and implementation of control measures. These models have assumed density-dependent transmission and included demographic parameters limited to births and deaths derived from samples of owned dogs at the district or city level. More recent demographic and epidemiological evidence from the field justifies re- evaluation of local transmission processes of canine rabies, including the possible effects of human interference (i.e. through the killing or translocation of diseased dogs) on transmission. We apply a stochastic compartmental model to explore the transmission dynamics of rabies in a range of realistic, free-roaming dog populations. We model the effect of human interference through assuming that the removal rate of infected dogs scales with human, and thus dog, population density. We also investigated the value of incorporating individual-level demographic data into the model. We show that empirical attack rates suggest that the basic reproductive number (R0) in local populations (e.g. at the village level) may be <1 and lower than the previously estimated global R0 (e.g. at the provincial level). We find that even a low rate of humans bringing infected dogs into a local population may off-set the benefits of high vaccination coverage. These findings suggest that human factors should be considered in epidemiological modelling and disease control for canine rabies; and, regular, thorough vaccination campaigns at the local level are essential, especially in the presence of interventions that may increase human-mediated movement of dogs, such as mass sterilization programmes. Finally, we found no clear advantage of individual-based models over simpler compartmental models with exponential age distributions. 5.1 Introduction Canine-mediated rabies is a serious zoonosis, responsible for at least 55,000 human deaths per year (Knobel et al. 2005). Understanding the transmission dynamics of the disease is essential for its control. Epidemiological models, based on observational data, have been used to explore spatial dynamics (Beyer et al. 2010; Beyer et al. 2012; Townsend et al. 2013b) and local transmission processes (Cleaveland & Dye 1995; Kitala et al. 2002; Hampson et al. 66 2007; Zinsstag et al. 2009), particularly to inform the development and implementation of control measures. The local transmission processes have been evaluated using deterministic, compartmental models, including demographic parameters (births and deaths only) derived from samples of households at the district or city level, and transmission parameters from observations of natural infections. These models assume density-dependent transmission, whereby the population is “well-mixed” and encounters between susceptible and infectious individuals increase with population density (Keeling & Rohani 2008). Under this assumption, disease incidence is expected to increase with host density, as will the basic reproductive number R0, an important epidemiological parameter that indicates whether an outbreak will spread. R0 is defined as the average number of secondary infections produced when one infected individual is introduced into a wholly susceptible population (Anderson & May 1991). For a disease to spread epidemically, R0 must exceed the threshold value of 1. For simple, “well-mixed” models R0 depends on the rate of transmission and the infectious period, and the expected outbreak size depends on R0. However, as discussed below, more recent empirical evidence (Hampson et al. 2009) (Chapter 2) justifies re-evaluation of local transmission processes, including the possible effects of human interference on transmission, which relates to complex interactions between people and dogs (Chapter 1). Currently these effects are poorly understood; however exploratory models based on realistic free-roaming dog populations, that account for human interference, may provide new insights into local disease dynamics and, thus, direct future field research and enhance rabies control policy. Canine rabies can be controlled effectively by vaccination (Cleaveland et al. 2003; Belotto et al. 2005; Schneider et al. 2005; Cleaveland et al. 2006; WHO 2013). Vaccination campaigns are often undertaken annually and aim to achieve 70% coverage of the dog population (WHO 2013). Empirical evidence and theory suggests that a vaccination coverage of 70% should maintain population immunity above the critical levels (20-45%) required to interrupt rabies transmission (Coleman & Dye 1996; Hampson et al. 2009). A vaccination coverage of 20- 45% is consistent with a low R0 (<2), estimated from empirical data of early exponential growth of epidemics in dog populations (Heffernan, Smith & Wahl 2005; Wallinga & Lipsitch 2007) across a wide geographic range (Hampson et al. 2009; Townsend et al. 2013b). In these studies, R0 was generally estimated at the global level (i.e. over large geographic areas such as cities or districts) and was consistently <2 (ranging from 1.05-1.72) irrespective of location, suggesting that transmission may be independent of population density. An R0 <2 is also consistent with a low incidence, or average monthly attack rate, of canine rabies compared to other species (Bogel et al. 1974). Attack rates of canine rabies 67 have also generally been estimated at the global level and are typically <0.5% and rarely exceeds 2% (Waltner-Toews et al. 1990; Windiyaningsih et al. 2004; Rothman, Greenland & Lash 2008; Zinsstag et al. 2009; Tenzin et al. 2010; Putra et al. 2011; Tenzin et al. 2011; Townsend et al. 2013a; Townsend et al. 2013b). Incidence data in local populations (e.g. at the village level) is limited, although the outbreak size distribution has been reported for the village level in the Serengeti, Tanzania, with a median outbreak size <5 cases and upper limit of approximately 35 cases (Hampson et al. 2009). While an R0 of 1.2 was estimated from outbreak data at the district level for the Serengeti, R0 at the village level was not reported. The apparent lack of relationship between R0 and population density is most consistent with frequency-dependent transmission, where the rate of transmission is assumed to be independent of host density. In this case, “well-mixed” models with frequency-dependent transmission predict host extinction for fatal diseases, such as rabies, as soon as R0 exceeds unity (Keeling & Rohani 2008). However, this prediction is inconsistent with the very low attack rates reported for dogs. Alternatively, a consistently low R0 (<2) irrespective of dog population density may be explained, in part, by selective removal, or the swift identification and killing of infectious and in-contact dogs by local communities (Beran 1982; Ogunkoya, Will & Ezeokoli 1984; Ezeokoli & Umoh 1987; Pastoret & Brochier 1998; Belotto et al. 2005; Hampson et al. 2009; Tenzin et al. 2011; Townsend et al. 2013b). Selective removal reduces the effective infectious period in dogs (Hampson et al. 2009) which may limit outbreak size (Pastoret & Brochier 1998; Tenzin et al. 2011), reducing disease incidence, R0 , and the vaccination threshold, and exacerbate stochastic fade-out, particularly in small populations (Lloyd-Smith et al. 2005). We have hypothesized that selective removal may scale with density concealing the existence of density-dependent transmission (Chapter 1). The basis of this hypothesis is that rabid dogs may be spotted and selectively removed from areas with more people present and that human and dog densities are expected to correlate (Oboegbulem & Nwakonobi 1989; Matter et al. 1998; Butler & Bingham 2000). Therefore, if the effective infectious period, as reduced by selective removal, scales inversely with human, and thus dog, population density R0 will be relatively invariant irrespective of population density. This is consistent with simple epidemic model predictions of exponential growth of diseased individuals (𝐼) early in an epidemic: )))(1exp(()0()( 0 tRItI   . 68 The exponential growth rate is therefore ))(1( 0  R and not only depends on the value of R0 but the removal rate (𝛾) and turnover rate (𝜇) for the population (Keeling & Rohani 2008); thus, the number of diseased individuals in a population may scale with R0 or the removal rate. Given the variable but often protracted latent period of rabies (Jackson 2013), demographic processes may also contribute to disease control through the loss of individuals incubating the disease from local populations. The positive effect of human interference, and stochastic and demographic processes, on the control of transmission in local populations, may be off-set by the influx of infected dogs mediated by people. Human-mediated movement of dogs (some of them infected) may be substantial (Hill 1985; Chomel et al. 1987; Beran & Frith 1988; Matter & Fico 1998; Estrada et al. 2001; Kongkaew et al. 2004) and contribute to the spatial spread of rabies and local persistence through reintroductions (Wells 1954; Shone 1962; Beran 1982; Robinson et al. 1996; Denduangboripant et al. 2005; Coetzee & Nel 2007; Kasempimolporn, Jitapunkul & Sitprija 2008; Zinsstag et al. 2009; Beyer et al. 2010; Talbi et al. 2010; Townsend et al. 2013b). This may be exacerbated by negative community reactions to culling (i.e. the widespread killing of dogs regardless of infection status) (Beran & Frith 1988; Windiyaningsih et al. 2004) and sterilization (WHO 2004; Reece & Chawla 2006; Totton 2009), as discussed below. Both culling and sterilization aim to reduce host density in order to reduce disease incidence, even though there is no evidence that the incidence of rabies increases with host density in dog populations (Coleman & Dye 1996; Kitala et al. 2002; Hampson et al. 2009; Townsend et al. 2013b). Sterilizations have been assumed to extend vaccination coverage by reducing the number of new, susceptible dogs entering the population through reducing local births and increasing the longevity of sterilized dogs; although the empirical evidence for this is equivocal (Totton 2009; Totton et al. 2010; Jackson 2013). Sterilizations have also been advocated to reduce the proportion of dogs under 12 months of age given that a proportionally higher incidence of rabies is often reported in this age class (Belcher, Wurapa & Atuora 1976; Malaga, Lopez Nieto & Gambirazio 1979; Beran 1991; Mitmoonpitak, Tepsumethanon & Wilde 1998; Widdowson et al. 2002); even though it may be simpler to increase vaccination coverage in young dogs (Appendix 6.1), which may be comparatively lower (Cleaveland et al. 2003; Flores-Ibarra & Estrella- Valenzuela 2004; Awoyomi, Adeyemi & Awoyomi 2007). However, given increasing evidence that the vast majority of free-roaming dogs in most societies globally are owned 69 (WHO & WSPA 1990; Cleaveland & Dye 1995; Robinson et al. 1996; Butler & Bingham 2000; Estrada et al. 2001; Windiyaningsih et al. 2004; Gsell et al. 2012) and that population size is regulated by the demand for dogs (Chapter 2), killing owned dogs and reducing the local supply of puppies available to meet the demand for dogs may perversely precipitate an influx of dogs into local populations. For example, within a few days of a village-wide cull in Kelusa, Bali Province, Indonesia, two residents brought in unvaccinated, potentially infected puppies from outside the village to replace their culled, vaccinated adult dogs. Community responses to culling and sterilization, in particular their effect on the movement of dogs, is currently being monitored in Kelusa and Antiga, Bali. This study aims to explore the local dynamics of canine rabies in realistic free-roaming dog populations using simple epidemic models that account for human interference (both the selective removal and movement of dogs) in transmission. We use robust demographic data (Chapter 2) to build an individual-based model for rabies transmission in dog populations. We model the effect of human interference through assuming that the removal rate of diseased dogs scales with human, and thus dog, population density; and investigate whether outbreak dynamics in local populations are consistent with reported estimates of R0. We also evaluate the effect of a variable influx of infected dogs by people, under different vaccination coverages, on outbreak size and frequency with a view to the evaluation of culling and sterilization in Bali villages on the implications of dog movement on rabies control. Comparisons are made with a simpler stochastic, compartmental SEI (Susceptible-Exposed- Infectious) model to assess the value of incorporating individual-level demographic data into the model. 5.2 Methods and materials 5.2.1 Demographic data Detailed demographic data were collected from four research sites, two in Gauteng Province, South Africa and two in Bali Province, Indonesia from March 2008 until April 2011. The research sites were part of a larger ecological study (Chapter 2), and were selected based on specific criteria, which included the absence of previous dog population management interventions by local animal welfare non-governmental organizations (NGOs) or authorities. Rabies outbreaks occurred in Bali Province in 2008 and Gauteng Province in 2010. The two sites in Gauteng Province, South Africa comprised the informal settlement Zenzele west of Johannesburg (26.15 o S and 27.41 o E) and Braamfischerville in Soweto (26.12 o S and 70 27.52 o E). The study area encompassed the entire Zenzele township, whereas approximately one third of Braamfischerville was included in the study area to include a comparable number of dogs to Zenzele. In Indonesia, the two sites included the villages of Kelusa (8.26 o S and 115.15 o E) and Antiga (8.30 o S and 115.29 o E) on the island of Bali. In Kelusa, the study area encompassed the entire village with the exception of Banjar (sub-village) Yehtengeh, which is separated from the rest of the village by rice fields and jungle, the southern half of Banjar Kelikikawan and households scattered along the main road leading into the village. In Antiga, the study area encompassed all of the main residential area (Banjars Kaler and Kelod). An additional area within Banjar Ketug included entrances scattered along a 2.7 km stretch of road winding through the jungle north of Kaler and Kelod. All households in the study areas were included in the sampling frame. Individual-level data for every identified dog in the study area were collected longitudinally by direct observation and questionnaire through door-to-door surveys every 6-12 weeks during the study period (of 36 months in Zenzele and Braamfischerville, 34 months in Kelusa and 38 months in Antiga). The study population comprised of every owned dog (i.e. dog belonging to a household in the study area). Pups were defined being in their 1st – 3rd month of life (i.e. 0 – ~13 weeks of age), juveniles 4th – 12th month of life (i.e. ~14 – 52 weeks of age), and adults older than their 12th month of life. The demographic data included in the models comprised of all owned dogs in the third month of their life or older in the study areas. The majority (>90%) of dogs were free to roam intermittently or continuously in Zenzele, Kelusa and Antiga, whereas in Braamfischerville approximately 40% of dogs were confined most of the time. Therefore, in addition to owned dogs being monitored at their household, every dog encountered in a yard not their own or on the street during each census was identified as either belonging to a household in the study area or not. Each dog identified as not belonging to a household in the study area was classified as unowned. Only two dogs in Johannesburg and eighteen dogs in Bali were identified as unowned; these dogs were not included in the study population. The date each dog was acquired was reported by owners, and for most dogs the date was re- recorded at least once during the study period. The date each dog was lost from the study area was generally only recorded once. The month of loss was reported by the owner for 68% of the dogs in Zenzele, 78% in Braamfischerville, 51% in Kelusa and 55% in Antiga. The remainder were randomly allocated to a month between the census or revisit in which they were last recorded and the subsequent one. The origin and outcome of each dog was also recorded. 71 Human demographic data were collected in Zenzele and Braamfischerville during eight and six of the door-to-door surveys respectively, and in Kelusa and Antiga during seven of the door-to-door surveys (including the first and last surveys for all of the sites). The same data were also collected during one survey in 2009 from a random sample of 80 and 141 non-dog owning households in the Zenzele and Braamfischerville respectively; and from all non- and ex-dog owning households during one survey in 2009 in the Bali sites (a total of 50 households in Kelusa and 99 households in Antiga). 5.2.2 Model structure and parameters Rabies is a fatal disease, generally with a long latent period (Jackson 2013) between infection and infectiousness. We therefore consider models with susceptible (S), exposed (E) and infectious (I) states. The models depend on three key epidemiological parameters for rabies: the transmission rate (𝛽), the latent period (1/𝜎) and the infectious period (1/𝛾) (Tables 5.1 and 5.2). Table 5.1 Transitions and their probabilities in the individual-based and compartmental models. description transition probability of transition in time 𝛿t a dog becomes exposed to the virus S → E (𝛽SI)𝛿𝑡 a dog becomes infectious E → I 𝜎E𝛿𝑡 a dog dies from the virus I → ∅ 𝛾I𝛿𝑡 N is the total number of dogs in the population at any one time (i.e. N = S+E+I) 72 Table 5.2 Demographic events and their probabilities in the compartmental model. description event probability of an event in time 𝛿𝑡 a susceptible dog is gained into the population ∅ → S bN a susceptible dog is lost from the population S → ∅ 𝜇S an exposed dog is lost from the population E → ∅ 𝜇E an infectious dog is lost from the population I → ∅ 𝜇I a vaccinated dog is lost from the population V → ∅ 𝜇V b = the mean number of dogs gained per capita per day, which is assumed to = 𝜇 [the mean number of dogs lost per capita per day] The exact number of dogs gained (from all sources) and lost (for all reasons) for each study population were known (see section 5.2.1.). The study populations are small (average of <350 dogs in each community; Table 5.3) therefore the model is stochastic, which is the preferred method for modelling outbreak dynamics in small populations (Keeling & Rohani 2008). The majority of the dogs in the populations were free-roaming consistent with the model assumption of “well-mixed” populations. For the individual-based model population size was not fixed and the demography was deterministic, updating the population by removing and adding dogs lost and gained from the study population at each (monthly) time step (𝛿𝑡). For the compartmental model, the loss rates were fixed at average values derived from the demographic data. The population size therefore fluctuates around the initial value. Table 5.3 Demographic parameter estimates for the study populations. location area 𝐴𝐿 (km 2 ) mean population size N population size at time 0 mean number of dogs lost per capita per day 𝜇 Zenzele 0.72 323 349 0.00208 Braamfischerville 1.26 283 288 0.00195 Kelusa 0.66 287 252 0.00156 Kaler (Antiga) 0.30 175 166 0.00148 Ketug (Antiga) 0.63 82 81 0.00140 73 Both models assume constant latency and removal rates, with the latent and infectious periods exponentially distributed. Empirical studies of these distributions indicate that they are less dispersed than predicted by exponential distributions and potentially better described by gamma distributions (Hampson et al. 2009). However, we have no information or intuition on how the dispersion of infectious period distributions may scale with population density. Therefore, for simplicity we only explore the constant rate (exponential) model here. The compartmental model is simulated by the (Gillespie) Stochastic Simulation Algorithm. The individual-based model uses a modified form of the Gillespie algorithm, where a demographic event is defined before the Markov simulation. At each time step (𝛿𝑡) a Markov simulation is attempted and carried out only if no demographic (non-Markov) events occur during the time step. When the average latent period is small compared to the average life expectancy, as for canine rabies, then the expression for R0 for an SEI model reduces to that for an SI model (Keeling & Rohani 2008). Therefore, we used equation 1 to estimate R0. R0 = (N𝛽𝐿 + 𝑖)/(𝛾[𝑛] + 𝜇) equation 1 Transmission is density-dependent however, to account for a relatively invariant R0 irrespective of local population density consistent with empirical data (Hampson et al. 2009), the removal rate (𝛾) was scaled according to the average dog density (i.e. dogs km-2) for each study population (𝑛) as proxy for human density. The overall removal rate (𝛾) in the model is composed of the removal rate without human interference (𝑚), fixed as the reciprocal of the average infectious period of 3.7 days without selective removal (Hampson et al. 2009), and the removal rate that scales with the average study population density (𝛼) (equation 2). This generates a lower bound in population density below which selective removal does not occur. 𝛾 = 𝑚 + 𝛼[𝑛] equation 2 The transmission rate (𝛽0) was initially calculated in units of population per units of area (i.e. dogs per km 2 per day) (see below). In order to parameterize the stochastic model with a transmission rate in per capita units (dog.days) and average study population size (N) (rather than average study population density), 𝛽0 is adjusted by the local study area (A𝐿) (i.e. km 2 ) (equation 3). The average loss rate per capita per day and the number of infected dogs entering the population (from outside the population) per day (𝑖) are small and, thus, ignored when calculating removal rates. 74 𝛽𝐿 = 𝛽0 / A𝐿 equation 3 A baseline transmission rate (𝛽0) of 0.0469 dogs per km 2 per day was derived from the most robust empirical data reported, obtained through contact tracing, in the Serengeti and Ngorongoro Districts in Tanzania (Hampson et al. 2009). This study estimated a global R0 of 1.19 (from the initial epidemic growth curve) for the Serengeti and an infectious period of 3.1 days, accounting for selective removal. An average global dog population density (of 8.19 km -2 ) was estimated for these those two districts combined, based on the distribution of reported outbreaks for the districts. A map of the region indicates heterogeneous dog population density (Lembo et al. 2010), and the average dog density at the village level may be substantially higher than the global density. Therefore, transmission rates of 0.0076 dogs per km 2 per day and 0.0038 dogs per km 2 per day were recalculated on the basis that the average local dog density across the region was arbitrarily 50 and 100 dogs km -2 respectively. Although the latent period is highly variable (Jackson 2013), the average latent period was fixed at 25 days consistent with the empirical data (Hampson et al. 2007; Hampson et al. 2009). An R0 of 1.2 was also estimated for Bali from the initial growth curve of the 2008 epidemic (Townsend et al. 2013b). 5.2.3 Model scenarios An outbreak is defined as at least two cases not interrupted by an interval of more than one month, as per Hampson et al (2009). Single cases that arise from seeding with an exposed dog and exposed seeds that do not transition to the infectious state (i.e. infected individuals that are lost from the population through demographic processes) are defined as index cases. The probability of an outbreak and median outbreak size and the quantiles were estimated from distributions of 1000 simulations for each of the following scenarios: 5.2.3.1 Comparison of outbreak dynamics, R0 , and transmission rates We investigate whether outbreak dynamics in the study populations are consistent with reported estimates of R0 for canine rabies. We also explore the sensitivity of the model for the range of R0 and transmission (and removal) rates. We use a range of R0 (from 0.5 to 1.2) and transmission rates (0.0469, 0.0076 and 0.0038 dogs per km 2 per day area adjusted [𝛽𝐿]) for each study population, with models seeded with a single exposed dog at the start of the time series. Models are seeded with exposed dogs on the assumption that most local incursions 75 occur through human-mediated movement of an infected (not infectious) dog, and a proportion of these dogs will be lost from the population before becoming infectious. 5.2.3.2 Probability of an outbreak with vaccination coverage In order to benchmark how well simple demographic models describe the rate of decline in vaccination coverage we simulate the probability of an outbreak for a range of vaccination coverage (from 0% to 80%) for an R0 of 0.7 and 1.2 and the maximum and minimum transmission rates (0.0469 and 0.0038 dogs per km 2 per day area adjusted [𝛽𝐿]) for each population. Vaccination coverage was calculated for the total number of dogs present in the population at the start of the study period for the individual-based model and for the mean population size for the compartmental model (Table 5.3). Models are seeded with a single exposed dog at the time of vaccination (at the start of the time series) and 12 and 24 months following vaccination. Outbreak size was not estimated because truncation of the time series precluded proper comparisons between the time points. Estimates of vaccination coverage in the study populations were obtained by assigning a random sample of dogs from the starting cohorts equal in size to the proportion assumed to be vaccinated, and determining those still present at 12 and 24 months. This process was repeated 1000 times to produce Monte Carlo estimates of the drop-off in vaccination coverage. 5.2.3.3 Comparison of outbreak dynamics, human-mediated incursions and vaccination coverage We evaluate the effect of human-mediated incursions of infected dogs on outbreak dynamics in the Bali populations for a range of vaccination coverage (from 0% to 80%) at the start of the time series using the individual-based model. We use parameter values that are biologically plausible for local transmission derived from the initial analysis described under 5.2.3.1 Comparison of outbreak dynamics, R0 , and transmission rates. These values include an R0 of 0.7 and area adjusted transmission rates of 0.0076 and 0.0038 dogs per km 2 per day [𝛽𝐿]. The probability of a human-mediated incursion was estimated from recent outbreak (Townsend et al. 2013b) and geographic (Putra et al. 2011) data for Bali and demographic data for the study populations (Chapter 2). Townsend et al (2013) estimated the probability of human-mediated transport of dogs across the island to be 0.05-0.09. This estimate agrees with the observed proportion of dogs relocated by people out of Kelusa and Antiga during the study period. The probability of human-mediated transport of dogs of 0.05 was combined 76 with the outbreak (incidence) and geographical data for Bali and the observed distribution of dogs gained into each study population. This was to estimate the baseline probability that a dog gained into the study population at some point during the study period was exposed. This baseline probability applies to a study population subject to rabies but not population controls, such as sterilizations. Increases in the probability of human-mediated incursions of canine rabies may occur in populations subject to management interventions. Therefore, we also modelled the effect on outbreak dynamics of incremental increases in the probability (from the baseline described above) that a dog gained into the study population is exposed. In this scenario, we assume the total number of dogs gained during the study period (and population size) remained the same but the proportion of dogs gained from outside the research site increased by 10%, 20% and 50% from the baseline. This is on the basis that the observed rates of ownership in the study populations were constant during the study period (see Chapter 2) and would probably not change in response to population management. All simulations were truncated at month 34 and any truncated outbreaks (a maximum of 6.4%) were included in the analysis. Although the entry of free-roaming rabid dogs from communities adjacent to the study populations cannot be excluded, when considering the local terrain surrounding the study areas and the observation that all dogs that moved into the study populations were brought in by people, incursions into the study populations are more likely to be mediated through human translocation of exposed dogs. Therefore, for the purposes of the exploratory analysis, entry of free-roaming rabid dogs into the study populations were ignored. Any transient dip in population size from culling was also ignored. All analysis were done using R (R Core Team 2014). 5.3 Results 5.3.1 Population characteristics Although limited to five sites, consistent with other studies, there was a higher density of dogs where there was a higher density of people (Table 5.4). However, the relationship between human and dog population density is not directly comparable between countries. For example, the average dog population densities for Zenzele and Kelusa are similar (~440 km -2 ) but the human population density in Zenzele (14410 km -2 ) is four times that of Kelusa (3436 km -2 ). The proportion of children younger than 7 years of age was higher in Johannesburg 77 compared to Bali (Zenzele 20.4%, Braamfischerville 15.4%, Kelusa 6.8% and Antiga 10.6%). There was considerable inward and outward movement of dogs by people, with at least one third of the dog population sourced from outside the research sites, and 5-10% of pups and 10-20% of juveniles and adults leaving the research sites (Chapter 2). Table 5.4 Study population size and density. Zenzele Braamfischerville Kelusa Kaler (Antiga) c Ketug (Antiga) c number of households in the study area 2170 a 2844 203 287 78 2212 b mean dog population density (dogs km -2 ) 449.29 225.19 435.61 582.28 130.41 mean number of people per household 4.78 5.27 11.17 6.78 5.86 mean human population size 10375.75 14987.88 2267.51 1945.86 457.08 mean human population density (people km -2 ) 14410.76 11895.14 3435.62 6486.2 725.52 mean human: dog ratio 32.07: 1 52.82: 1 7.89: 1 11.14: 1 5.56: 1 a before May 2009; b from May 2009 following the erection of an additional street of shacks May 2009; c includes data until May 2011 5.3.2 Comparison of outbreak dynamics, R0 , and transmission rates Outbreak size increased with R0 and population size (Figure 5.1 and Appendix 5.1). Excluding index cases, for all of the study populations an R0 of approximately 0.7 generated outbreak distributions with a median outbreak size of 3 and upper 95% quantile of approximately 20 cases, comparable to observational data from Tanzania (Hampson et al. 2009). The median outbreak size was similar for all the populations, increasing from three cases for an R0 of 0.5 to ten cases for an R0 of 1.2. There was a disproportionate increase in the 95% quantile with R0 and population size, with a maximum outbreak of ~300 cases in Zenzele (average population size of 323 dogs), corresponding to a maximum outbreak size of 63 cases in Ketug (average population size of 82 dogs) (i.e. for an R0 of 1.2). 78 Figure 5.1 Outbreak size for a range of R0 and transmission rates (𝜷) for the individual-based model (IBM). Each vertical line represents the distribution in outbreak size generated from 1000 simulations. The dot on the vertical line represents the estimated median outbreak size. The top and bottom of the vertical line are equivalent to the estimated 95% and 5% quantile respectively. The top horizontal line indicates the upper limit in outbreak size (of approximately 35 cases) and the bottom horizontal line indicates the median outbreak size (of approximately 3 cases) observed in villages in the Serengeti District (Hampson et al. 2009). N = population size. 79 Median outbreak sizes were comparable for the individual-based and compartmental models; however the 95% quantiles were generally lower for the compartmental models (Figure 5.1 and Appendix 5.1). For a given R0 there were no substantial differences in outbreak size with transmission and removal rates, although there was a tendency for slightly larger outbreaks with longer infectious periods in Zenzele, Kelusa and Kaler. However, the outbreak dynamics were highly stochastic, with any differences in outbreak size with the transmission and removal rates much smaller than the intrinsic variability in outbreak size. The largest difference occurred in Kelusa for an R0 of 1.2, with an increase in the 95% quantile from 160 with an average infectious period of 1.4 hours to 192 with an infectious period of 17.4 hours. The probability of an outbreak increased with R0 but not population size. For an R0 of 0.7 the probability of an outbreak was ~40% (36-43%), a small fraction of exposed dogs (<6%) did not transition to the infectious state (i.e. were lost through demographic processes), and 54- 62% of exposed dogs transitioned to the infectious state but did not instigate an outbreak, rather remained as isolated (index) cases. 5.3.3 Probability of an outbreak with vaccination coverage The probability of an outbreak following vaccination of a proportion of the population at the start of the time series was similar for the individual-based and compartmental models (Figure 5.2 for a summary of the results for Kelusa and Appendix 5.2 for the results for all of the populations). 80 Figure 5.2 Probability of an outbreak in Kelusa at the time of vaccination (t_0) and 12 and 24 months after vaccination for a range of vaccination coverage and transmission rates (beta). The top plots were generated using the individual-based model (IBM) and the bottom plots using the compartmental model (CM). Vaccination coverage of 80% generally reduced the probability of an outbreak by two thirds immediately following pulse vaccination undertaken at the start of the time series. Although the probability of an outbreak approximately doubled by 12 months after the pulse vaccination (Appendix 5.2). For example, in Kelusa (for an R0 of 0.7, transmission rate of 0.0058 dogs per km 2 per day and infectious period of 10.1 hours) the probability of an outbreak decreased from 39% prior to vaccination to 11% immediately following 80% 81 vaccination, but then increased to 24% by 12 months following vaccination. In general, for vaccination coverage of <60% the probability of an outbreak at 24 months was similar to the unvaccinated study population. Appendix 5.3 shows the estimates of the proportion of the study population still immune 12 and 24 months after vaccination of 60% and 80% of the starting cohort. Declines in vaccination coverage estimated by the SEI models agree with those estimated from the starting cohorts. For example, in Kelusa vaccination coverage declined to 35% in the study population 24 months after a pulse vaccination campaign that achieved 80% coverage. This agrees with the probability of an outbreak of 30% estimated at 24 months after achieving 80% vaccination and vaccination coverage of between 20-40% at the time of vaccination (Appendix 5.2). 5.3.4 Comparison of outbreak dynamics, human-mediated incursions and vaccination coverage Human-mediated incursions of exposed dogs appear to off-set reductions in the probability of an outbreak achieved through vaccination even with R0 <1 (Figure 5.3 and Appendices 5.4 and 5.5). For example, in Kelusa (for a transmission rate of 0.0058 dogs per km 2 per day), 80% vaccination coverage at the start of the time series caused a ~25% reduction in the probability of at least one outbreak (from 34% to 26%) as a consequence of normal background (i.e. baseline) movement of dogs during the study period. However, a 20% increase in the probability of human-mediated incursions increased the probability of an outbreak to 33%, comparable to that in unvaccinated populations. For vaccination coverage <80%, a 20% increase in human-mediated incursions increased the probability of an outbreak to above the unvaccinated baseline population. Vaccination coverage <60% at the start of the time series for Kelusa had no effect on the probability of at least one outbreak during the study period or median outbreak size. Overall, median outbreak size reduced slightly (from 3.5 to 2.5) with vaccination coverage ≥60%. Maximum outbreak size reduced substantially with vaccination; in Kelusa by more than two thirds with 80% vaccination coverage (Figure 5.4. and Appendix 5.6). 82 Figure 5.3 Probability of outbreaks with human-mediated incursions for a range of vaccination coverage at the start of the time series and transmission rates (𝜷) in Kelusa. The plots were generated using the individual-based model (IBM). The probability of an outbreak during the 34 month study period is shown on the y-axis. The average number of exposed dogs entering the study population during the study period are shown on the x-axis; and were estimated from empirical data described in section 5.2.3.3. The baseline probability that a dog gained into the study population during the study period was exposed was obtained by dividing 1.16 (number of exposed dogs) by the total number of dogs gained into the study population during the study period (n=478). This probability was included in the model. The number (and the probability) of exposed dogs gained during the study period was then increased incrementally by 10%, 20% and 50% from the baseline as shown. The horizontal line provides a reference to allow for comparisons. 83 Figure 5.4 Outbreak size with human-mediated incursions for a range of vaccination coverage at the start of the time series and transmission rates (𝜷) in Kelusa (this figure is reproduced in Appendix 5.6 with additional information). The plots were generated using the individual-based model (IBM). The number of exposed dogs gained into the study population during the 34 month study period are shown on the x-axis; see Figure legend 5.3 for the derivation of the number (and probability) of exposed dogs gained into the study population during the study period. Each vertical line represents the distribution in outbreak size generated from 1000 simulations. The dot on the vertical line represents the estimated median outbreak size. The top and bottom of the vertical line are equivalent to the estimated 95% and 5% quantile respectively. 5.4 Discussion The models used in this study have facilitated exploration of the transmission of rabies in realistic free-roaming dog populations. The structure of the models was underpinned by assumptions regarding the nature of human interference in transmission processes and their application limited through parameterisation from data not from the study populations. Nonetheless, the main outputs, that R0 at the local (e.g. village) level may be <1 and lower than the global (e.g. provincial level) R0 , and that the local influx of infected dogs by people may off-set the benefits of vaccination, are biologically plausible and important for disease control policy, and hence epidemiological modelling, as discussed below. To model the hypothesis of human interference in transmission processes, specifically the selective removal of infectious and in-contact dogs scaling with dog population density, we scaled the infectious period, fixing R0 irrespective of local (i.e. study population) density. This hypothesis is supported by empirical evidence of selective removal (Beran 1982; Ogunkoya, Will & Ezeokoli 1984; Pastoret & Brochier 1998; Belotto et al. 2005; Hampson et 84 al. 2009; Tenzin et al. 2011; Townsend et al. 2013b) and a correlation between dog and human population densities (Oboegbulem & Nwakonobi 1989; Matter et al. 1998; Butler & Bingham 2000), also evident in the study populations (Table 5.4). While there are currently no published data regarding scaling of selective removal with density, given the possible importance of human interference in transmission processes, this warrants investigation. In light of this study, this investigation should consider two aspects. The first is how human demography may affect any scaling of selective removal with density. Although there was a higher density of dogs where there was a higher density of people within Johannesburg and Bali, the relationship between human and dog densities is not directly comparable between countries. However, when considering local human demography and culture, the difference between countries in the numbers of people available to selectively remove a rabid dog may be reduced. For example, there was a higher proportion of very young children in Johannesburg compared to Bali. Young children are unlikely to chase and kill rabid dogs, thus reducing the number and density of people likely to selectively remove a rabid dog in Johannesburg. Other factors, such as the proportion of householders at work during the day and the elderly and sick may also be important. The second aspect that warrants consideration is the degree to which the effective infectious period and transmission rate is reduced through any human interference. For a given R0 outbreak size was fairly insensitive to the values of the transmission and removal rates in this study. Therefore, from the perspective of modelling outbreak size it is only the value of R0, and not the particular combination of values selected for these parameters, that is important. However, the values of these parameters have important biological significance that may be important for predicting the impact of different interventions. To model the effect of human- mediated incursions in Kelusa a transmission rate of 4.2 dogs per km 2 per year and an infectious period of ~5 hours was used. These parameters were selected on the assumption that they may reflect what is biologically plausible. The parameter values are consistent with the reported incidence of rabies in Bali (Townsend et al. 2013a; Townsend et al. 2013b) and reports that community members chased and killed a rabid dog on two separate occasions in the non-study area of Kelusa. Given the rapidity with which the community members were reported to have killed the dogs, it is reasonable to assume that the rabid dogs were unlikely to have been left to roam for an extended period of time. However, the exact duration that the rabid dogs were in the village, presenting a public health risk, is unknown. In this respect, the degree with which any selective removal reduces the effective infectious period justifies further investigation. Finally, any other anthropogenic factors that may scale with human and 85 dog density and interfere with the contact process, such as traffic and urban infrastructure, should also be considered. A key prediction from this study is that an R0 of ~0.7 generates outbreak size distributions consistent with that observed at the village level in the Serengeti, with a median outbreak size <5 cases and an upper limit of approximately 35 cases (Hampson et al. 2009). Predictions of a small number of cases at the local level is also consistent with estimates of disease incidence typically <0.5% (Waltner-Toews et al. 1990; Windiyaningsih et al. 2004; Zinsstag et al. 2009; Tenzin et al. 2010; Tenzin et al. 2011) and the lack of empirical evidence of large fluctuations in dog population density during an outbreak (Hampson et al. 2007). This suggests that the R0 at the local (e.g. village) level is lower than the R0 estimated for larger geographic areas. This difference may be the result of human-mediated movement of infected dogs, such that exposed dogs are lost from the local population but not from the global population, and the roaming of rabid dogs out of the local population. Together with selective removal and demographic and stochastic processes, the loss of infected dogs from the local population through human-mediated movement may also contribute to reducing the local R0 <1 with important implications for rabies control, as discussed below. The difference between the local R0 and global R0 may also reflect the duration of outbreaks at the global level, which are generally protracted over months or years, compared to small clusters of cases over shorter time frame at the local level. The implication that human interference may have a substantial effect on local and spatial dynamics should be considered in epidemiological models and disease control policy. Culling is often instigated at the local level, for example by the banjar Head in Bali, as a means to curb a rabies outbreak. However, culling should not be advocated on the grounds that selective removal of diseased animals alone may contribute to the control of rabies (Pastoret & Brochier 1998; Tenzin et al. 2011) and because for an R0 below 1 the probability of an outbreak is low. For an R0 <1 any outbreak generally only precipitates a short chain of infection even in large populations (Lloyd-Smith et al. 2005) and are therefore “self-limiting” irrespective of any culling. However, even though human interference and demographic and stochastic processes may reduce R0 to <1, reducing the probability of large outbreaks, vaccination still has a protective effect at the local level and is essential. A global R0 of <2 corresponds to a vaccination threshold of 20-45% necessary to reduce the R0 to unity, below which deterministic models predict that outbreaks are curbed. Most vaccination programmes are implemented at regional levels and vaccination coverage above this threshold should be maintained through widespread annual vaccination campaigns achieving a vaccination 86 coverage of at least 70%. When considering stochastic processes, short, rather than long, chains of cases are more likely to occur when R0 <1 and in smaller populations (Lloyd-Smith et al. 2005), which may explain the uniformity in the outbreak distributions for an R0 <1. These short chains of infection still present a public health risk. As our results demonstrate, vaccination of at least 80% will substantially reduce the probability of an outbreak even in small populations. However, vaccination coverage declines through demographic and immunological processes, necessitating frequent, thorough local vaccination campaigns (Figure 5.2 and Appendix 5.2). Regular, comprehensive vaccination programmes at the local level are particularly important when considering human-mediated incursions of exposed dogs. Such incursions may limit the efficacy of vaccination even with high coverage (Figure 5.3 and Appendices 5.4 and 5.5). There was considerable inward and outward movement of dogs by people in the study populations (Chapter 2). Therefore, it could be useful to limit baseline movement, by encouraging the local acquisition of dogs, and to ensure that vaccinations are untaken simultaneously with interventions that may increase the influx of infected dogs, such as mass sterilization programmes. Sterilization programmes, implemented at the population-level, are often advocated as a necessary component of canine rabies control (WHO 2004; Reece & Chawla 2006), but are also undertaken to address animal welfare concerns. As with culling, the demand for dogs by communities may result in an increase in dog importation where local supply has been reduced by sterilization; and sterilizations programmes (and culling) may in fact promote the spread of rabies. Therefore, vaccination at the time of sterilization, but also repeatedly following sterilization programmes is strongly recommended. Evaluation of the effect of human behaviour in response to sterilization and culling on dog population dynamics and disease transmission requires far greater consideration in future research. 5.5 Conclusion We have used simple stochastic epidemic models to explore the local transmission of rabies in realistic free-roaming dog populations. Our modelling of human interference, with regards selective removal and an increase in the movement of infected dogs as an unintended consequence of population management, is currently based on assumption. Community responses to culling and sterilization, in particular their effect on the movement of dogs, is currently being monitored in Kelusa and Antiga, Bali. However, given the critical role that human interference may play in the transmission of canine rabies, this warrants more extensive research. 87 Chapter 6 General discussion Canine-rabies can be effectively controlled through vaccination (Cleaveland et al. 2003; Schneider et al. 2005; Cleaveland et al. 2006; WHO 2013), however the impact of vaccinations on disease incidence may be affected by dynamic demographic and immunological processes, particularly through their effect on vaccination coverage and rabies transmission. To improve rabies control, through field interventions and epidemiological modelling, the research presented in this thesis used a multifaceted approach to explore the effect of demographic and immunological processes, and their regulatory factors, on population and disease dynamics and vaccinal responses in two free-roaming dog populations in Johannesburg, South Africa, and two in Bali, Indonesia. The main conclusions from this work are: 1. Despite complex population dynamics, and geographical and cultural diversity, the ecology of the dog populations was similar. 2. Human factors are likely to be critically important in the transmission and control of canine-mediated rabies. 3. Many popular perceptions regarding the ecology of free-roaming dogs and the control of canine rabies may be wrong. 4. Widespread, sustained vaccination programmes are essential in rabies endemic areas. There are several important similarities between the study populations that affect canine rabies transmission and control and population control. These similarities are primarily a function of people and their close relationship with domestic dogs; this relationship is universally recognised. In contrast to other species affected by rabies, human behaviours may have important positive and negative effects on population and disease dynamics in domesticated dogs, which in turn impacts on planning rabies control programmes; and, this is the first study to properly identify this. Almost all of the identified dogs in the study population were owned and fed regularly by their owners, with important implications for vaccine delivery during vaccination campaigns. 88 Owners generally facilitate vaccination of their dogs (Lembo et al. 2010; Knobel et al. 2013) through identification and restraint of their animal. Unowned dogs may be more difficult to access for vaccination, potentially reducing effective vaccination coverage (Hampson et al. 2009), particularly if the fraction of unowned dogs is large. At least 70% of the dog population was vaccinated through door-to-door vaccine delivery and with the assistance of owners, even in Bali where the majority of the dogs needed to be caught by net. In Bali community engagement, including a simple ad hoc pictorial pamphlet delivered door-to-door, raised awareness of rabies and explained what a vaccine is and the reason for vaccinating dogs in order to protect people from rabies. The pamphlet was designed to cater for the illiterate and poorly educated. As a result, even though most owners could not restrain their dogs manually, many managed to confine their dogs to the kitchen to facilitate capture by net. This reduced the time and effort required to catch the dogs. Furthermore, all identified dogs resident in the study population belonged to households in the study area, therefore vaccination coverage was accurately measured during vaccine delivery. The mode of ownership of the dogs in the study population is consistent with increasing evidence that, globally, most free-roaming dogs are owned (WHO & WSPA 1990; Cleaveland & Dye 1995; Butler & Bingham 2000; Estrada et al. 2001; Windiyaningsih et al. 2004; Gsell et al. 2012) and fed regularly by their owner (Brooks 1990; de Balogh, Wandeler & Meslin 1993; Butler & Bingham 2000). However, historically it was assumed that the free- roaming dogs in the four communities were unowned or owned but neglected, whereby the dogs primarily scavenged for food to survive. This initial uncertainty regarding ownership status parallels conflicting reports in the literature regarding the mode of ownership and food sources of free-roaming dogs. Implicit in many canine epidemiological and ecological models is the assumption that dog populations are regulated through environmental resource constraints on births and deaths (Wandeler 1985; WHO & WSPA 1990; Kitala et al. 2002; Vandermeer & Goldberg 2003; Hampson et al. 2007; Zinsstag et al. 2009; Totton et al. 2010). This implies that all individuals in the population are unowned and compete for a common resource to survive, most likely environmental refuse, consistent with limited observations (Pal 2001). By contrast, other estimates of the unowned fraction of the population are variable, ranging from 0%, including in sub-Saharan Africa (Butler & Bingham 2000) and South East Asia (Estrada et al. 2001; Windiyaningsih et al. 2004), to 20- 30% (Vos & Turan 1998; Matter et al. 2000; Durr et al. 2009). A further complication is that a fraction of owned dogs may be neglected and left to scavenge a variable proportion of their diet (Ortega-Pacheco et al. 2007a; Ratsitorahina et al. 2007), or that a fraction of the 89 population “belongs” to the community and source variable proportions of their diet from refuse and community members but not a specific owner (Reece & Chawla 2006; Reece 2007). Although the health status of unowned or owned neglected dogs is not reported, it is a widely-held perception that they are in sufficiently good condition to be sustained without direct human oversight. This variability in mode of ownership may reflect true differences in the ecology of dogs globally. Alternatively, it may be a consequence of extrapolations of ecological principles from wildlife to a domesticated species, classification of dogs as unowned if owners are not immediately apparent, or a consequence of the practical difficulties definitively identifying unowned dogs in reasonable or good body condition, particularly where most dogs are owned. Definitive identification of the fraction of unowned, healthy dogs in the study population would have required intensive, individual-level surveillance of dogs identified as not belonging to households in the study area and quantification of variations in the availability, nutritional content and uptake of environmental resources; neither of which were practicable within the scope of this study. Measures of food provided by owners were also constrained through illiteracy and poor education and the inability to validate the measurements. Indirect assessment of population-level ownership, i.e. population regulation through environmental resource constraints, was hampered by the lack of variation in population size (or density) and births and deaths. Mark-recapture methods used previously to identify the unowned fraction of a population (Fishbein et al. 1992; Matter & Fico 1998; Matter et al. 1998; Vos & Turan 1998; Matter et al. 2000; Cleaveland et al. 2003; Kayali et al. 2003b; Durr et al. 2009; Kaare et al. 2009; Gsell et al. 2012) may have been limited by measurement error and statistical variations, violations of model assumptions and, more importantly, practical limitations of repeat measures using photographic mark-recapture to identify a resident population. Therefore, a multifaceted approach, combining all practicable methods, was used to assess mode of ownership and food sources of the dogs in the study population. Methods included repeat direct observations of identified dogs, household surveys and subjective assessment of environmental refuse, community-based participatory methods (Chambers 1994a; Chambers 2007) which took advantage of the close relationship between dogs and people and local knowledge of the enumerators as well as the community, and assessment of the relationship between body condition and energy requirements. From this approach, there was no evidence that population size was regulated through environmental resource constraints, nor was there any evidence that a fraction of the healthy population was unowned or owned but neglected. Rather, population size was regulated 90 through the demand for dogs. All dogs resident in the study population that were in reasonable or good body condition belonged to households in the study area and were fed regularly by the household. Most of the small number of dogs that did not belong to households in the study area were emaciated. Although it was not verified whether these dogs were owned outside the study area, their condition was consistent with community perceptions that they were unowned, not fed by non-specific community members, and that environmental refuse probably provided inadequate nutrition. A common assertion is that free-roaming dogs in impoverished communities are unowned or owned but neglected. This was not the case in the study populations; most householders provided adequate care and nutrition for their dogs despite their poverty. In addition to vaccine delivery, these observations have wide-reaching implications for animal welfare. Had a population of unowned, healthy dogs been identified it may absolve humans from the responsibility of providing adequate care for a domesticated species. Several other important commonalities between the four communities impact the planning of rabies control, particularly epidemiological modelling and vaccine delivery. No population growth, or a progressive decline in population size, occurred during the study period, consistent with similar rates of (susceptible) dogs gained into (means of 10-19) and (potentially vaccinated) dogs lost from (means of 11-20) the study population each month (Appendices 2.10 and 2.11). Reliable estimates of these demographic processes, including growth rates, are required for dynamical models, including compartmental models of local transmission (Cleaveland & Dye 1995; Kitala et al. 2002; Hampson et al. 2007; Zinsstag et al. 2009; Totton et al. 2010). As previously discussed, population size, and the acquisition of (susceptible) dogs, is regulated through human demand for dogs. Previous estimates of growth have generally been at the national (Brooks 1990) or district (Butler & Bingham 2000; Kitala et al. 2001; Hampson et al. 2009) level and may reflect the ecological heterogeneity within, and limited movement of dogs into and out of, large geographical areas. This study has generated reliable estimates of demographic parameters and has demonstrated that growth may be limited in established communities where geographical expansion is minimal. A common perception is that, in communities where the majority of dogs are free-roaming, there are “too many dogs”. Contrary to expectations, household rates of ownership were consistently low, with only a fraction of the households owning dogs (approximately 10% in Johannesburg, 70% in Kelusa and 40-50% in Antiga) and an average of 1.3-1.7 dogs per dog owning household, comparable to the rates of dog ownership in the UK (Murray et al. 2010). 91 Low rates of ownership may support vaccine delivery, such that fewer dogs need to be vaccinated to achieve the target vaccination coverage of at least 70% during annual vaccination campaigns (Coleman & Dye 1996; Cleaveland et al. 2003; Hampson et al. 2009; WHO 2013). Furthermore, a substantial fraction (20-40%) of the population was present for the entire study period. A large proportion of mature adults may live longer than previously recognised (Oboegbulem & Nwakonobi 1989; Brooks 1990; Rautenbach, Boomker & de Villiers 1991; Margawani & Robertson 1995; Robinson et al. 1996; Butler & Bingham 2000; Kitala et al. 2001). While every available dog should be vaccinated against rabies irrespective of age class, vaccinating adults may be particularly important in maintaining vaccination coverage. Vaccination coverage in the study population was estimated to remain within the critical threshold of 20-45% necessary to curb an outbreak (Hampson et al. 2009) two years after a pulse campaign that achieved 60% coverage (assuming that all vaccinated individuals were protected from challenge). This deterministic threshold is consistent with estimates of a global R0 of <2. Most control programmes are implemented at the regional level and vaccination coverage above this threshold should be maintained through widespread, annual vaccination campaigns achieving a vaccination coverage of at least 70%. However, dynamical models from this study, that account for human interference (i.e. selective removal, or the swift identification and killing of infectious and in-contact dogs by local communities, and the translocation of dogs by people) in transmission, predict an R0 of <1 in local populations (e.g. a Bali village) with outbreaks generally limited to short chains or clusters of cases consistent with epidemic theory (Lloyd-Smith et al. 2005). These short chains of infection present a significant public health risk. In light of these predictions, previous interpretations (in Chapters 1, 2 and 4) of the critical threshold (reducing R0to unity) in local populations may be insufficient to prevent human disease. However, high levels of vaccination coverage is still predicted to substantially reduce the probability of a cluster of cases even in small populations. Thus, the recommended vaccination coverage of 70% (WHO 2013) should also be achieved at the local level. Vaccination coverage declines in populations through time through demographic and immunological processes. Although almost all of the dogs in Johannesburg, and presumably Bali, seroconverted (virus neutralizing antibody titre ≥0.5 IU/ml) to high-quality, inactivated cell-culture rabies vaccine irrespective of health status, titres were predicted to decline to near negligible levels (<0.1 IU/ml) within two years for a proportion of the vaccinated population. There may be an increased susceptibility to natural exposure in dogs with low titres (Precausta et al. 1985; Aubert 1992; Bahloul et al. 2006; CDC 2008; Siegrist 2008), but also 92 in the remaining unvaccinated dogs in the population given the decline in effective vaccination coverage and herd immunity. This reinforces the importance of frequent and thorough vaccination campaigns. A substantial fraction of the study population was sourced from outside the research sites by the dog owners. Translocation of dogs may contribute to the spatial spread of rabies between populations (Denduangboripant et al. 2005; Coetzee & Nel 2007; Kasempimolporn, Jitapunkul & Sitprija 2008; Beyer et al. 2010; Talbi et al. 2010; Townsend et al. 2013b) and incursions of rabies into local populations (Chomel et al. 1987; Beran & Frith 1988; Robinson et al. 1996; Zinsstag et al. 2009). Human-mediated incursions of exposed dogs may off-set reductions in disease incidence, R0 , and vaccination threshold through human interference and demographic and stochastic processes, thus necessitating widespread and sustained vaccination campaigns. Complex population dynamics may also have practical consequences in terms of population control. Sterilization programmes, implemented at the population level, has been advocated as a necessary component of canine rabies control (WHO 2004; WHO 2013). These programmes are associated with a reduction in the incidence of rabies when sterilized dogs are vaccinated simultaneously (Reece & Chawla 2006), however it is uncertain to what extent sterilization itself is necessary for rabies control. Sterilization programmes may gradually stabilise or reduce population size (or density) over several years (Reece & Chawla 2006; Totton 2009; Totton et al. 2010). Sterilizations are also hypothesized to extend vaccination coverage by increasing the longevity of dogs and reducing the number of new, susceptible dogs entering the population through reducing local births. Although reductions in population density are unlikely to reduce the transmission of rabies (Coleman & Dye 1996; Kitala et al. 2002; Hampson et al. 2009), they may reduce the number of dogs that require vaccination. However, mass sterilizations may have a reduced effect where population growth is already limited and a large proportion of the population originates from outside the area. Furthermore, reducing the number of dogs that require vaccination may be unnecessary when rates of dog ownership are already low. From the available data, the effect of sterilizations on population structure (Totton 2009; Totton et al. 2010) and life expectancy is unclear. Increasing the adult fraction may only have a minor impact on vaccination coverage if the fraction prior to the intervention was already large (Totton et al. 2010). Most critically, the demand for dogs by communities may result in an increase in dog importation (some of them infected with rabies) where the local supply has been reduced by sterilization (or culling discussed below). Dynamical models from this study indicate that small (i.e. 20%) increases 93 in human-mediated incursions may off-set the advantages of high levels of vaccination, particularly the probability of an outbreak. Therefore, evaluation of the effect of human behaviour in response to sterilization (and culling) on dog population dynamics and disease transmission needs far more research attention. Sterilizations are also advocated to reduce the number of puppies and juveniles in a population on the basis that a proportionally higher incidence of rabies is often reported in these age classes (Belcher, Wurapa & Atuora 1976; Malaga, Lopez Nieto & Gambirazio 1979; Beran 1991; Mitmoonpitak, Tepsumethanon & Wilde 1998; Widdowson et al. 2002); with a serious risk to the public given the close relationship between humans and puppies (Mitmoonpitak, Wilde & Tepsumethanon 1997; Taiwo et al. 1998; WHO 1998; Widdowson et al. 2002; Awoyomi, Adeyemi & Awoyomi 2007). The higher incidence of rabies in young dogs may be the effect of low vaccination coverage in this age class, given that puppies are often excluded from vaccination programmes (Chomel et al. 1987; Beran & Frith 1988; Brooks 1990; Matter & Fico 1998; Mitmoonpitak, Tepsumethanon & Wilde 1998; Matter et al. 2000; Gunatilake, Wimalaratne & Perera 2003; Flores-Ibarra & Estrella-Valenzuela 2004; Kongkaew et al. 2004; Awoyomi, Adeyemi & Awoyomi 2007; Durr et al. 2009; Kaare et al. 2009; Touihri et al. 2011; Davlin et al. 2013) and population immunity declines following a vaccination campaign. Dogs less than 12-13 weeks of age are generally excluded from vaccination programmes on the assumption that they have maternal antibodies and immature immune systems which may limit the immune response to rabies vaccine. Evaluation of the effect of maternal antibodies and immune function of puppies on rabies vaccine-induced immune responses is limited. Maternal antibody may interfere with immune responses (Day & Schultz 2011; Siegrist 2012; Tizard 2013), particularly in puppies 8 weeks of age or younger vaccinated with modified live vaccine under field conditions (Aghomo, Oduye & Rupprecht 1990). However, at least under experimental conditions, maternal antibodies and immune function may not limit the immune response to inactivated vaccines which stimulate both B- and T- cell responses (Siegrist 2012), as demonstrated in puppies vaccinated with Rabisin at 2 weeks of age (Chappius 1998). The results from this study support these prior observations (Appendix 6.1). Therefore, rather than try to reduce the proportion of puppies in the population through sterilizations, it may be more efficient to simply vaccinate them with commercially available, high quality, inactivated vaccine. Culling is commonly instigated at the local (e.g. banjar or sub-village) level in response to rabies cases. Similar to sterilization programmes (discussed above), culling assumes that encounters between susceptible and infectious individuals and, thus, disease incidence 94 increases with host density (i.e. is density-dependent). Culling reduces host density in order to reduce the incidence of the disease, and reduce host density to below the deterministic threshold for invasion (R0 = 1) to halt the spread of an epidemic (chapter 1). Although the role of density (and other factors) in the transmission of rabies is currently poorly understood, the available evidence suggests that there is no clear relationship between dog population density and transmission (Coleman & Dye 1996; Kitala et al. 2002; Hampson et al. 2009), and reductions in density through the mass killing of predominately healthy dogs (or through sterilizations) are unlikely to reduce the transmission of rabies. While it cannot be ruled out that rabies transmission is independent of host density (i.e. is frequency-dependent), consistent with the low incidence of canine rabies (generally <0.5%) (Waltner-Toews et al. 1990; Windiyaningsih et al. 2004; Rothman, Greenland & Lash 2008; Zinsstag et al. 2009; Tenzin et al. 2010; Putra et al. 2011; Tenzin et al. 2011; Townsend et al. 2013a; Townsend et al. 2013b) a more plausible explanation is that selective removal “disrupts” density- dependent transmission. Selective removal reduces the effective infectious period (Hampson et al. 2009) and may reduce the incidence of rabies (Pastoret & Brochier 1998; Tenzin et al. 2011). Should selective removal scale with human, and thus dog, density as hypothesized (Chapter 1), then the effective infectious period will scale inversely with host density and transmission will appear to be independent of host density. Selective removal presents an alternative mechanism to density reduction to reduce R0 to below unity. Indeed, euthanasia of dogs showing clinical signs is advocated to control rabies (WHO 2013). As noted above, dynamical models from this study, that account for selective removal and stochastic and demographic processes, predict an R0 <1 in local populations, with a low probability of an outbreak. Any outbreaks that do occur are generally small and, thus, “self-limiting” irrespective of culling. Furthermore, reducing host density to below the deterministic threshold may not achievable. Even in Ketug, the smallest of the research populations with a density of ~130 dogs per km 2 and assuming the lowest transmission rate used in the dynamical models (of 0.0038 dogs per km 2 per day), density would have to be almost halved to reduce R0 to unity. Similar to sterilizations, although culling may reduce the number (but not proportion) of dogs requiring vaccination (Rosatte et al. 2007), density reduction may be off-set by the continual influx of dogs brought into the population by people. This influx may be exacerbated in response to culling, with a corresponding increase in the incursions of rabies especially into larger populations. However, more importantly, culling is ethically questionable given that most free-roaming dogs are probably owned and 95 accessible for vaccination, and vaccination has proven efficacy (Cleaveland et al. 2003; Schneider et al. 2005; Cleaveland et al. 2006; WHO 2013). One important difference between the communities was in the handleability of the dogs for vaccination. The majority of the dogs in Johannesburg were amenable to gentle restraint, with a soft muzzle and leash, for vaccination and blood sampling. In contrast, in Bali the majority of the dogs could not be safely restrained without a net. Although the reasons for this difference are unclear, it may be attributable to subtle cultural differences in human-dog interactions. In particular, preliminary data from the follow-up study and informal discussion with community members indicate that puppies are more likely to be tied up in Bali during the sensitive period critical for social development (Serpell & Jagoe 1995; Appleby, Bradshaw & Casey 2002). Restraint by net is more stressful to the dog, time consuming and costly, making vaccine delivery more difficult, potentially deterring those responsible for rabies control to undertake regular and thorough vaccination campaigns. Implications of this research The findings from this research have important implications for rabies control, particularly given the similarities between the four communities in this study and other communities where most free-roaming dogs are owned and fed by their owners. In contrast to other species affected by rabies, the complex relationship between dogs and people is critically important in the transmission and control of canine-mediated rabies and should not be ignored. Community involvement is essential for vaccine delivery and also to access essential information for rabies and population control. It is often assumed that free- roaming dogs are unowned, effectively wildlife, or that the dogs are neglected by owners and that community members are disinterested in their welfare. Rather, the starting point should be that most, if not all, free-roaming dogs in a community are owned and wanted, particularly if they are in reasonable or good body condition, most are accessible for vaccination through their owner, and that the community can provide essential information regarding the dogs in their community. Human factors should also be considered with regards the level or scale of interventions. Human-mediated movement of dogs may contribute to the spatial spread and incursions of rabies, therefore vaccination programmes need to be spatially ambitious and sustained. Vaccination coverage at regional levels should be maintained above the deterministic threshold of 20-45%, thus vaccination campaigns should achieve a uniform, widespread 96 coverage of 70% (Townsend et al. 2013b; WHO 2013). Importantly, while the deterministic threshold of 20-45% necessary to curb an outbreak may not be applicable at the local level, dynamical models from this study predict that regularly vaccinating 60-80% of the local population substantially reduces the probability and size of chains of infection. Given declines in vaccination coverage following campaigns through demographic and (probably) immunological processes, and predicted rebound in the probability of an outbreak following vaccination, vaccination campaigns should be repeated annually consistent with World Health Organization recommendations (WHO 2013). The level or scale of interventions intended to reduce population density (or size) should also be considered. Where mass sterilization programmes are implemented over a large geographical area, such as at the city level (Reece & Chawla 2006; Totton et al. 2010), the acquisition of dogs from outside the area may have a limited effect on declines in population density on the basis that the influx of dogs may be relatively small compared to number born in the population. Whereas, consistent with this study, a substantial proportion of the population may originate from outside smaller communities or more localised areas, off- setting any reductions in density through sterilization at this level. Indeed, simple assumptions that populations are closed and most dogs originate locally may not be valid, especially in smaller populations. Reductions in the number of puppies born locally may be off-set by an influx of dogs from outside the area. Sterilization programmes are undertaken for a variety of reasons (Jackson 2013). These programmes are generally expensive and labour intensive (Morters pers comm), therefore it is important that the aims of the programme are clearly defined and the limitations of the programme understood. For example, targeted sterilization, i.e. determined as appropriate for each dog individually, may positively impact animal welfare and nuisance behaviour (Reece, Chawla & Hiby 2013) but may have limited impacts on population density. Although it may be counterintuitive, it should not be assumed that fewer dogs in a population (achieved through density reduction, especially culling) equates to less rabies. There is still considerable uncertainty regarding the factors that drive rabies transmission, particularly the role of host density, however culling is not recommended as a means to control canine rabies (Chapter 1) (WHO 2013). Rather, rabies should be controlled through regular, widespread vaccination with high quality, inactivated vaccine, that achieves at least 70% coverage (WHO 2013), preferably in conjunction with community engagement. 97 The effect of human behaviour on rabies transmission, specifically human-mediated movement of exposed dogs, has been factored into spatial models of canine rabies (Beyer et al. 2010; Townsend et al. 2013b). However, human behaviour should also be considered for models of local disease dynamics. Amendments to these models may include no population growth or decline in population size, exclusion of density-dependent demographic processes specific to population regulation through environmental resource constraints, and demographic parameters that account for selective removal and human-mediated movement of dogs. Further studies required Priorities for future work include: 1. Determining the factors responsible for the differences in the handleability of dogs may reduce the need for restraint by net for vaccination, lessening the cost and time for vaccination delivery and improving animal welfare. Proper evaluation of human-animal interactions by an anthrozoologist and other factors, such as genetic differences in temperament, is warranted. To balance public health risks, simultaneous evaluation of the risk of infection of owners from zoonotic pathogens, such as dermatophytes and rabies, through increased handling of their dogs (WHO 1998; Widdowson et al. 2002) should be undertaken. 2. Assessment of human behaviour in response to culling and sterilization is currently underway in Kelusa and Antiga, Bali. Specifically, these interventions may increase the translocation of dogs, potentially introducing rabies into the community. Culling and mass sterilization programmes are carried out world-wide in response to rabies outbreaks, but also for population management. This study showed a substantial fraction of dogs is sourced from outside the village independent of culling and sterilization. Should these interventions drive an increase in human-mediated movement of dogs sufficient to increase the local incidence of rabies, this adds to the importance of controlling rabies by vaccinations and not density reduction. Where sterilization programmes are established, then it may be mandatory that all dogs in the location of the programme (whether they are sterilized or not) are vaccinated against rabies. 3. Evaluating the factors that determine the rates of ownership may be important for disease and population control, particularly the effect of rabies outbreaks on the acquisition and 98 disposal of dogs. Density reduction, through culling or mass sterilization programmes, is often undertaken in response to rabies outbreaks. However, the impact of these interventions on disease control may be further limited when declines in population size (or density) are already driven by the community, possibly through the fear of rabies or liability arising from dogs bites to people in rabies-endemic areas. Although the reasons for the low rate of ownership in the study areas are unclear, informal discussions with community members suggest that poverty, i.e. being able to afford to feed a dog, and work commitments are causal factors. If this is indeed the case, it reinforces the points that community members generally consider the provision of adequate care for their dogs a priority and the importance of community engagement in disease and population control. 4. Exploring evidence for selective removal of diseased dogs scaling with dog population density might improve dynamical models of local transmission, however obtaining such evidence may be unrealistic. The distinctive presentation of rabies has facilitated detailed contact tracing, which has contributed significantly to understanding local and global transmission dynamics and the effectiveness of vaccination strategies (Hampson et al. 2009). While collecting similar data across geographic locations with different dog densities may provide insights into local transmission dynamics it may prove impractical. 5. Evidence for resident populations of healthy, unowned dogs would have wide-reaching implications for rabies control and animal welfare. However, robust evidence of this sub- group would require intensive monitoring, possibly with GPS collars, for ownership and variations in food sources and health status of at least a small number of dogs identified by locals as not belonging to a reference household. Although there is increasing evidence that these dogs are unlikely to exist, given the heterogeneity in communities (dogs and people) globally, investigation is warranted, particularly India where edible refuse is reported to be a food source for free-roaming dogs (Pal 2001; Reece 2007). Nonetheless, reliable identification of unowned, healthy dogs may be impractical. 6. Evidence of Lagos Bat-like Virus in Bali, suggesting spill-over of a Lyssavirus from bats to dogs, presents a serious public health concern that justifies further investigation (Hayman et al. 2013; Peel et al. 2013). Exposure of the public to Lyssaviruses other than RBV is of particular concern in light of anecdotal reports of a bat-cave temple priest 99 dying from encephalitis unrelated to exposure to a dog and two dogs vaccinated with Rabisin at least twice over several months dying from rabies. 100 References Acosta-Jamett, G., Cleaveland, S., Cunningham, A.A. & deC. Bronsvoort, B.M. (2010) Demography of domestic dogs in rural and urban areas of the Coquimbo region of Chile and implications for disease transmission. Preventive Veterinary Medicine, 94, 272-281. Aghomo, H.O., Oduye, O.O. & Rupprecht, C.E. (1990) The serological response of young dogs to the Flury LEP strain of rabies virus vaccine. Veterinary Research Communications, 14, 415-425. Albon, S.D., Coulson, T.N., Brown, D., Guinness, F.E., Pemberton, J.M. & Clutton-Brock, T.H. (2000) Temporal changes in key factors and key age groups influencing the population dynamics of female red deer. Journal of Animal Ecology, 69, 1099-1110. Altizer, S., Dobson, A., Hosseini, P., Hudson, P., Pascual, M. & Rohani, P. (2006) Seasonality and the dynamics of infectious diseases. Ecology Letters, 9, 467-484. Anderson, R.M. (1986) Vaccination of wildlife reservoirs. Nature, 322, 304-305. Anderson, R.M., Jackson, H.C., May, R.M. & Smith, A.M. (1981) Population dynamics of fox rabies in Europe. Nature, 289, 765-771. Anderson, R.M. & May, R.M. (1991) Infectious diseases of humans - dynamics and control. Oxford University Press, New York, United States. Appleby, D.L., Bradshaw, J.W.S. & Casey, R.A. (2002) Relationship between agressive and avoidance behaviour by dogs and their experience in the first six months of life. The Veterinary Record, 150, 434-438. Arya, S.C. & Agarawal, N. (2006) Immunogenicity of newly introduced Vero cell rabies vaccine. Vaccine, 24, 847-848. Aubert, M.F.A. (1992) Practical significance of rabies antibodies in cats and dogs. Revue Scientifique et Technique-Office International, 11, 735-760. AusVet EpiTools http://epitools.ausvet.com.au/content.php?page=PrevalenceSS. Awoyomi, O.J., Adeyemi, I.G. & Awoyomi, F.S.O. (2007) Socioeconomic factors associated with non-vaccination of dogs against rabies in Ibadan, Nigeria. Nigerian Veterinary Journal, 28, 59-63. Bahloul, C., Taieb, D., Diouani, M.F., Ahmed, S.B.H., Chtourou, Y., B'chir, B.I., Kharmachi, H. & Dellagi, K. (2006) Field trials of a very potent rabies DNA vaccine which induced long lasting virus neutralizing antibodies and protection in dogs in experimental conditions. Vaccine, 24, 1063-1072. 101 Baron, R.B. (1986) Malnutrition in hospitalized patients - diagnosis and treatment. Western Journal of Medicine, 144, 63-67. Begon, M., Bennett, M., Bowers, R.G., French, N.P., Hazel, S.M. & Turner, J. (2002) A clarification of transmission terms in host-microparasite models: numbers, densities and areas. Epidemiology and Infection, 129, 147-153. Begon, M., Harper, J.L. & Townsend, C.R. (1996) Ecology, Third Edition, Third edn. Blackwell Science Limited, Oxford, United Kingdom. Belcher, D.W., Wurapa, F.K. & Atuora, D.O.C. (1976) Endemic rabies in Ghana. The American Journal of Tropical Medicine and Hygiene 25, 724-729. Belotto, A., Leanes, L.F., Schneider, M.C., Tamayo, H. & Correa, E. (2005) Overview of rabies in the Americas. Virus Research, 111, 5-12. Belsare, A.V. & Gompper, M.E. (2013) Assessing demographic and epidemioloigic parameters of rural dog populations in India during mass vaccination campaigns. Preventive Veterinary Medicine, 111, 139-146. Beran, G.W. (1982) Ecology of dogs in the central Philippines in relation to rabies control efforts. Comparative Immunology, Microbiology and Infectious Diseases, 5, 265-270. Beran, G.W. (1991) Urban rabies. The natural history of rabies, 2nd edition (ed. G.M. Baer), pp. 428-443. CRC Press, United States. Beran, G.W. & Frith, M. (1988) Domestic animal rabies control: an overview. Reviews of Infectious Diseases, 10, S672-S677. Berndtsson, L.T., Nyman, A.-K.J., Rivera, E. & Klingeborn, B. (2011) Factors associated with the success of rabies vaccination of dogs in Sweden. Acta Veterinaria Scandinavica, 53, 1-7. Beyer, H.L., Hampson, K., Lembo, T., Cleaveland, S., Kaare, M. & Haydon, D.T. (2010) Metapopulation dynamics of rabies and the efficacy of vaccination. Proceedings of the Royal Society of London. Series B, Biological sciences, 278(1715), 2182-2190. Beyer, H.L., Hampson, K., Lembo, T., Cleaveland, S., Kaare, M. & Haydon, D.T. (2012) The implications of metapopulations dynamics on the design of vaccination campaigns. Vaccine, 30, 1014-1022. Bingham, J., Foggin, C.M., Wandeler, A.I. & Hill, F.W.G. (1999a) The epidemiology of rabies in Zimbabwe. 1. Rabies in dogs (Canis familiaris). Onderstepoort Journal of Veterinary Research, 66, 1-10. Bingham, J., Foggin, C.M., Wandeler, A.I. & Hill, F.W.G. (1999b) The epidemiology of rabies in Zimbabwe. 2. Rabies in jackals (Canis adustus and Canis mesomelas). Onderstepoort Journal of Veterinary Research, 66, 11-23. 102 Bjornstad, O.N., Finkenstadt, B.F. & Grenfell, B.T. (2002) Dynamics of measles epidemics: estimating scaling of transmission rates using a time series SIR model. Ecological Monographs, 72, 169-184. Bogel, K., Arata, A.A., Moegle, H. & Knorpp, F. (1974) Recovery of reduced fox populations in rabies control. Zentralblatt fur Veterinarmedizin Reihe B, 21, 401-412. Bogel, K. & Joshi, D.D. (1990) Accessibility of dog populations for rabies control in Kathmandu, Nepal. Bulletin of the World Health Organization, 68, 611-617. Boitani, L., Francisci, F., Ciucci, P. & Andreoli, G. (1995) Population biology and ecology of feral dogs in central Italy. The domestic dog: its evolution, behaviour, and interactions with people (ed. J. Serpell), pp. 217-244. Cambridge University Press, Great Britain. Bolzoni, L., Real, L. & De Leo, G. (2007) Transmission heterogeneity and control strategies for infectious disease emergence. PLoS One, 1-5. Bonenfant, C., Gaillard, J.-M., Klein, F. & Loison, A. (2002) Sex- and age-dependent effects of population density on life history traits of red deer Cervus elaphus in a temperate forest. Ecography, 25, 446-458. Bradshaw, J.W.S. (2006) The Evolutionary Basis for the Feeding Behavior of Domestic Dogs (Canis familiaris) and Cats (Felis catus). The Journal of Nutrition, 136, 1927S-1931S. Brochier, B., Kieny, M.P., Costy, F., Coppens, P., Bauduin, B., Lecocq, J.P., Languet, B., Chappius, G., Desmettre, P., Afiademanyo, K., Libois, R. & Pastoret, P.-P. (1991) Large-scale eradication of rabies using recombinant vaccinia-rabies vaccine. Nature, 354, 520-522. Brooks, R. (1990) Survey of the dog population of Zimbabwe and its level of rabies vaccination. The Veterinary Record, 127, 592-596. Brown, A.L., Merry, D., L & Beckenhauer, W.H. (1973) Modified live-virus rabies vaccine produced from Flury high egg-passage virus grown on an established canine-kideny cell line: three-year duration of immunity study in dogs. American Journal of Veterinary Research, 34, 1427-1432. Bunn, T.O. (1991) Canine and feline vaccines, past and present. The natural history of rabies (ed. G.M. Baer), pp. 415-425. CRC Press, United States. Bunn, T.O., Ridpath, H.D. & Beard, P.D. (1984) The relationship between rabies antibody titres in dogs and cats and protection from challenge. Rabies information exchange (ed. G.M. Baer). Centres for Disease Control, Georgia, USA. Butler, J.R.A. & Bingham, J. (2000) Demography and dog-human relationships of the dog population in Zimbabwean communal lands. The Veterinary Record, 147, 442-446. 103 Carroll, M.J., Singer, A., Smith, G.C., Cowan, D.P. & Massei, G. (2010) The use of immunocontraception to improve rabies eradication in urban dog populations. Wildlife Research, 37, 676-687. Catley, A. (1999) Methods on the move. International Institute for Environment and Development, London. Catley, A., Alders, R.G. & Wood, J.L.N. (2012) Participatory epidemiology: approaches, methods and experiences. The Veterinary Journal, 191, 151-160. Caughley, G. (1977) Analysis of vertebrate populations. The Blackburn Press, United States. CDC (2008) Centers for Disease Control and Prevention: Recommendations of the Advisory Committee on Immunization Practices (ACIP). Morbidity and Mortality Weekly Report, pp. 1-36. United States. Chambers, R. (1994a) The origins and practice of participatory rural appraisal. World Development, 22, 953-969. Chambers, R. (1994b) Participatory rural appraisal (PRA): analysis of experience. World Development, 22, 1253-1268. Chambers, R. (2007) From PRA to PLA and pluralism: practice and theory. Working paper 286. Institute of Development Studies, United Kingdom. Chappius, G. (1998) Neonatal immunity and immunisiation in early age: lessons from veterinary medicine. Vaccine, 16, 1468-1472. Cheuk, H. (1969) A review of the history and control of rabies in Hong Kong. Agricultural science Hong Kong, 1, 141-147. Childs, J.E., Curns, A.T., Dey, M.E., Real, L.A., Feinstein, L., Bjornstad, O.N. & Krebs, J.W. (2000) Predicting the local dynamics of epizootic rabies among raccoons in the United States. Proceedings of the National Academy of Sciences, 97, 13666-13671. Childs, J.E., Robinson, L.E., Sadek, R., Madden, A., Miranda, M.E. & Miranda, N.L. (1998) Density estimates of rural dog populations and an assessment of marking methods during a rabies vaccination campaign in the Philippines. Preventive Veterinary Medicine, 33, 207-218. Choisy, M. & Rohani, P. (2006) Harvesting can increase severity of wildlife disease epidemics. Proceedings of the Royal Society of London. Series B, Biological sciences, 273, 2025-2034. Chomel, B., Chappuis, G., Bullon, F., Cardenas, E., David de Beublain, T., Maufrais, M.C. & Giambruno, E. (1987) Serological results of a dog vaccination campaign against rabies in Peru. Revue Scientifique et Technique-Office International, 6, 97-113. 104 Choquenot, D. (1991) Density-dependent growth, body condition, and demography in feral donkeys: testing the food hypothesis Ecology, 72, 805-813. Cleaveland, S. (1998) Epidemiology and control of rabies. The growing problem of rabies in Africa. Transactions of the Royal Society of Tropical Medicine and Hygiene, 92, 131- 134. Cleaveland, S. & Dye, C. (1995) Maintenance of a microparasite infecting several host species: rabies in the Serengeti. Parasitology, 111, S33-S47. Cleaveland, S., Kaare, M., Knobel, D. & Laurenson, M.K. (2006) Canine vaccination - providing broder benefits for disease control. Veterinary Microbiology, 117, 43-50. Cleaveland, S., Kaare, M., Tiringa, P., Mlengeya, T. & Barrat, J. (2003) A dog rabies vaccination campaign in rural Africa: impact on the incidence of dog rabies and human dog-bite injuries. Vaccine, 21, 1965-1973. Cliquet, F., Aubert, M. & Sagne, L. (1998) Development of a fluorescent antibody virus neutralisation test (FAVN test) for the quantificaiton of rabies-neutralising antibody. Journal of Immunological Methods, 212, 79-87. Cliquet, F., Verdier, Y., Sagne, L., Aubert, M., Schereffer, J.L., Selve, M., Wasniewski, M. & Servat, A. (2003) Neutralising antibody titration in 25,000 sera of dogs and cats vaccinated against rabies in France, in the framework of the new regulations that offer an alternative to quarantine. Revue Scientifique et Technique-Office International, 22, 857-866. Clutton-Brock, T.H., Major, M. & Guinness, F.E. (1985) Population regulation in male and female red deer. Journal of Animal Ecology, 54, 831-846. Coetzee, P. & Nel, L.H. (2007) Emerging epidemic dog rabies in coastal South Africa: a molecular epidemiological analysis. Virus Research, 126, 186-195. Coleman, P.G. & Dye, C. (1996) Immunization coverage required to prevent outbreaks of dog rabies. Vaccine, 14, 185-186. Collett, M.G. (2000) Survey of canine babesiosis in South Africa. Journal of the South African Veterinary Association, 71, 180-186. Corradi, A., Ferrari, L. & Borghetti, P. (2007) Parameters for evaluating the cell-mediated immune response during viral infection: diagnostic and prognostic applications. Veterinary Research Communications, 31, 103-107. Coulson, T., Catchpole, E.A., Albon, S.D., Morgan, B.J.T., Pemberton, J.M., Clutton-Brock, T.H., Crawley, M.J. & Grenfell, B.T. (2001) Age, sex, density, winter weather, and population crashes in Soay sheep. Science, 292, 1528-1531. 105 Coulson, T., Guinness, F., Pemberton, J. & Clutton-Brock, T. (2004) The demographic consequences of releasing a population of red deer from culling. Ecology, 85, 411- 422. Coulson, T., Milner-Gulland, E.J. & Clutton-Brock, T. (2000) The relative roles of density and climatic variation on population dynamics and fecundity rates in three contrasting ungulate species. Proceedings of the Royal Society of London. Series B, Biological sciences, 267, 1771-1779. Council of Europe (2008) European pharmacopoeia Rabies vaccine (inactivated) for veterinary use. Council of Europe, France. Courtin, F., Carpenter, T.E., Paskin, R.D. & Chomel, B.B. (2000) Temporal patterns of domestic and wildlife rabies in central Namibia stock-ranching area, 1986-1996. Preventive Veterinary Medicine, 43, 13-28. Coyne, M.J., Smith, G. & McAllister, F.E. (1989) Mathematic model for the population biology of rabies in raccoons in the mid-Atlantic. American Journal of Veterinary Research, 50, 2148-2154. Crawley, M.J. (2007) The R book. Wiley, Chichester, United Kingdom. Davis, D.E. & Wood, J.E. (1959) Ecology of foxes and rabies control. Public Health Reports, 74, 115-118. Davlin, S., Lapiz, S.M., Miranda, M.E. & Murray, K. (2013) Factors associated with dog rabies vaccination in Bohol, Philippines: results of a cross-sectional cluster survey conducted following the island-wide rabies elimination campaign. Zoonoses and public health, 60, 494-503. Davlin, S.L. & VonVille, H.M. (2012) Canine rabies vaccination and domestic dog population characteristics in the developing world: A systematic review. Vaccine, 30, 3492-3502. Day, M.J. & Schultz, R.D. (2011) Veterinary immunology principles and practice. Manson Publishing, London, United Kingdom. de Balogh, K.K.I.M., Wandeler, A.I. & Meslin, F.-X. (1993) A dog ecology study in an urban and a semi-rural area of Zambia. Onderstepoort Journal of Veterinary Research, 60, 437-443. Denduangboripant, J., Wacharapluesadee, S., Lumlertdacha, B., Ruankaew, N., Hoonsuwan, W., Puanghat, A. & Hemachudha, T. (2005) Transmission dynamics of rabies virus in Thailand: implications for disease control. BMC Infectious Diseases, 5, 1-11. 106 Dionigi, R., Zonta, A., Dominioni, L., Gnes, F. & Ballabio, A. (1977) The effects of total parenteral nutrition on immunodepression due to malnutrition. Annals of Surgery, 185, 467-474. Durr, S., Mindekem, R., Kaninga, Y., Moto, D.D., Meltzer, M.I., Vounatsou, P. & Zinsstag, J. (2009) Effectiveness of dog rabies vaccination programmes: comparison of owner- charged and free vaccination campaigns. Epidemiology and Infection, 137, 1558- 1567. Dyce, K.M., Sack, W.O. & Wensing, C.J.G. (1987) Textbook of Veterinary Anatomy. Saunders, Philadelphia, U.S.A. Earn, D.J.D., Rohani, P., Bolker, B.M. & Grenfell, B.T. (2000) A simple model for complex dynamical transitions in epidemics. Science, 287, 667-670. Ernst, S.N. & Fabrega, F. (1989) A time series analysis of the rabies control programme in Chile. Epidemiology and Infection, 103, 651-657. Estrada, R., Vos, A., De Leon, R. & Mueller, T. (2001) Field trial with oral vaccination of dogs against rabies in the Philippines. BMC Infectious Diseases, 1, 1-7. Ezeokoli, C.D. & Umoh, J.U. (1987) Epidemiology of rabies in northern Nigeria. Transactions of the Royal Society of Tropical Medicine and Hygiene, 81, 268-272. Ferguson, N. (2007) Capturing human behaviour. Nature, 446, 733. Fishbein, D.B., Frontini, M.G., Dobbins, J.G., Collins, E.F., Huerta, G.Q., de Jesus, J., Rodriquez, G., Woo-Ming, B., Ramos, J.G., Belotto, A.J., Torres, J.M.B., Yenne, K.M., Linhart, S.B. & Baer, G.M. (1992) Prevention of canine rabies in rural Mexico: an epidemiologic study of vaccination campaigns. American Journal of Tropical Medicine and Hygiene, 47, 317-327. Flores-Ibarra, M. & Estrella-Valenzuela, G. (2004) Canine ecology and socieconomic factors associated with dogs unvaccinated against rabies in a Mexican city across the US- Mexico border. Preventive Veterinary Medicine, 62, 79-87. Fondati, A., De Lucia, M., Furiani, N., Monaco, M., Ordeix, L. & Scarampella, F. (2009) Prevalence of Demodex canis - positive healthy dogs at trichoscopic examination. Veterinary Dermatology, 21, 146-151. Fowler, C.W. (1981) Density dependence as related to life history strategy. Ecology, 62, 602- 610. Friend, M. (1968) History and epidemiology of rabies in wildlife in New York. New York Fish and Game Journal, 15, 71-97. 107 Funk, S., Gilad, E., Watkins, C. & Jansen, V.A.A. (2009) The spread of awareness and its impact on epidemic outbreaks. Proceedings of the National Academy of Sciences, 106, 6872-8677. Garde, E., Serpell, J., Pérez, G., Vanderstichel, R., Zenteno, N. & Dalla Villa, P. (2012) Behaviour assessment of male dogs pre- and post surgical and non-surgical sterilization in Puerto Natales, Chile 1st International Conference on Dog Population Management. York, United Kingdom. German, A.J. & Holden, S.L. (2006) Subjective estimation of body condition can predict body fat mass as well as condition scoring with an established 9-point scale. BSAVA, pp. 508. Birmingham, United Kingdom. German, A.J., Holden, S.L., Moxham, G.L., Holmes, K.L., Hackett, R.M. & Rawlings, J.M. (2006) A simple, reliable tool for owners to assess the body condition of their dog or cat. The Journal of Nutrition, 136, 2031S-2033S. Gilbert, A., Fooks, A.R., Hayman, D.T.S., Horton, D.L., Muller, T., Plowright, R., Peel, A.J., Bowen, R., Wood, J.L.N., Cunningham, A.A. & Rupprecht, C.E. (2013) Deciphering serology to further understand the ecology of infectious diseases in wildlife. Ecohealth, 10, 298-313. Grenfell, B.T., Bjornstad, O.N. & Finkenstadt, B. (2002) Dynamics of measles epidemics. II. Scaling noise, determinism and predictability with the time series TSIR model. Ecological Monographs, 72, 185-202. Gsell, A., Knobel, D.L., Cleaveland, S., Kazwala, R.R., Vounatsou, P. & Zinsstag, J. (2012) Domestic dog demographic structure and dynamics relevant to rabies control in urban areas in Africa: the case of Iringa, Tanzania. BMC Veterinary Research, 8, 1-10. Guaguere, E., Prelaud, P. & Craig, M. (2008) A practical guide to canine dermatology. Merial, Italy. Gunatilake, M., Wimalaratne, O. & Perera, K.A.D.N. (2003) Persistence of antibody titres in adult dogs and puppies following anti-rabies immunization. The Ceylon Journal of Medical Science, 46, 67-72. Hampson, K., Dushoff, J., Bingham, J., Bruckner, G., Ali, Y.H. & Dobson, A. (2007) Synchronous cycles of domestic dog rabies in sub-Saharan Africa and the impact of control efforts. Proceedings of the National Academy of Sciences, 104(18), 7717- 7722. Hampson, K., Dushoff, J., Cleaveland, S., Haydon, D.T., Kaare, M., Packer, C. & Dobson, A. (2009) Transmission dynamics and prospects for the elimination of canine rabies. PLoS Biology, 7(3), 462-471. 108 Hanski, I. & Gaggiotti, O.E. (2004) Ecology, Genetics and Evolution of Metapopulations: Standard Methods for Inventory and Monitoring. Elsevier Academic Press, USA. Hayman, D.T.S., Bowen, R.A., Cryan, P.M., McCracken, G.F., O'Shea, T.J., Peel, A.J., Gilbert, A., Webb, C.T. & Wood, J.L.N. (2013) Ecology of zoonotic infectious diseases in bats: current knowledge and future direction. Zoonoses and public health, 60, 2-21. Heffernan, J.M., Smith, R.J. & Wahl, L.M. (2005) Perspectives on the basic reproductive ratio. Journal of the Royal Society Interface, 2, 281-293. Hemachudha, T. (2005) Rabies and dog population control in Thailand: success or failure? J Med Assoc Thai, 88, 120-123. Hill, F.W.G. (1985) A survey of the animal population of four high density suburbs in Harare. Zimbabwe Veterinary Journal, 16, 31-36. Horton, D.L., McElhinney, L.M., Marston, D.A., Wood, J.L.N., Russell, C.A., Lewis, N., Kuzmin, I.V., Fouchier, R.A.M., Osterhaus, A.D.M.E., Fooks, A.R. & Smith, D.J. (2010) Quantifying antigenic relationships among the Lyssaviruses. Journal of Virology, 84, 11841-11848. Hu, R.L., Fooks, A.R., Zhang, S.F., Liu, Y. & Zhang, F. (2008) Inferior rabies vaccine quality and low immunization coverage in dogs (Canis familiaris) in China. Epidemiology and Infection, 136, 1556-1563. Jackson, A.C. (2013) Rabies: Scientific basis of the disease and its management, 3rd edn. Academic Press, San Diego, United States. Jakel, V., Konig, M., Cussler, K., Hanschmann, K. & Thiel, H.-J. (2008) Factors influencing the antibody response to vaccination against rabies. Towards the elimination of rabies in Eurasia (eds B. Dodet, A.R. Fooks, T. Muller & N. Tordo), pp. 431-437. Scientific & Technical Department of the OIE. Jefferys, H. (1961) The Theory of Probability, Third Edition. Oxford, United Kingdom. Johnson, N., Cunningham, A.F. & Fooks, A.R. (2010) The immune response to rabies virus infection and vaccination. Vaccine, 28, 3896-3901. Johnston, D.H. & Beauregard, M. (1969) Rabies epidemiology in Ontario. Bulletin of the Wildlife Disease Association, 5, 357-370. Kaare, M., Lembo, T., Hampson, K., Ernest, E., Estes, A., Mentzel, C. & Cleaveland, S. (2009) Rabies control in rural Africa: evaluating strategies for effective domestic dog vaccination. Vaccine, 27, 152-160. 109 Kaplan, M.M., Goor, Y. & Tierkel, E.S. (1954) A field demostration of rabies control using chicken-embryo vaccine in dogs. Bulletin of the World Health Organization, 10, 743- 752. Kasempimolporn, S., Jitapunkul, S. & Sitprija, V. (2008) Moving towards the elimination of rabies in Thailand. Journal of the Medical Association of Thailand, 91, 433-437. Kasempimolporn, S., Mitmoonpitak, C., Chaiyabutr, N., Supakorn, K., Brahmasa, R. & Sitprija, V. (1996) Maternal antibodies against rabies in Thai puppies. A preliminary study. Journal of the Medical Association of Thailand, 79, 36-39. Kayali, U., Mindekem, R., Yemadji, N., Oussiguere, A., Naissengar, S., Ndoutamia, A.G. & Zinsstag, J. (2003a) Incidence of canine rabies in N'Djamena, Chad. Preventive Veterinary Medicine, 61, 227-233. Kayali, U., Mindekem, R., Yemadji, N., Vounatsou, P., Kaninga, Y., Ndoutamia, A.G. & Zinsstag, J. (2003b) Coverage of pilot parenteral vaccination campaign against canine rabies in N'Djamena, Chad. Bulletin of the World Health Organization, 81, 739-745. Keeling, M.J. & Rohani, P. (2008) Modeling Infectious Diseases in Humans and Animals. Princeton University Press, Princeton, New Jersey, USA. Kennedy, I. (1998) Quarantine and rabies: a reappraisal. Report by the advisory group on quarantine to the Rt Hon Nick Brown MP, Minister of Agriculture, Fisheries and Food. MAFF Publications. London. Kennedy, L.J., Lunt, M., Barnes, A., McElhinney, L., Fooks, A.R., Baxter, D.N. & Ollier, W.E.R. (2007) Factors influencing the antibody response of dogs vacccinated against rabies. Vaccine, 25, 8500-8507. Kitala, P., McDermott, J., Kyule, M., Gathuma, J., Perry, B. & Wandeler, A. (2001) Dog ecology and demography information to support the planning of rabies control in Machakos District, Kenya. Acta Tropica, 78, 217-230. Kitala, P.M., McDermott, J.J., Coleman, P.G. & Dye, C. (2002) Comparison of vaccination strategies for the control of dog rabies in Machakos District, Kenya. Epidemiology and Infection, 129, 215-222. Knobel, D.L., Cleaveland, S., Coleman, P.G., Fevre, E.M., Meltzer, M.I., Miranda, M.E.G., Shaw, A., Zinsstag, J. & Meslin, F.-X. (2005) Re-evaluating the burden of rabies in Africa and Asia. Bulletin of the World Health Organization, 83, 360-368. Knobel, D.L., Lembo, T., Morters, M., Townsend, S.E., Cleaveland, S. & Hampson, K. (2013) Dog rabies and its control. Rabies: scientific basis of the disease and its management, 3rd edition (ed. A.C. Jackson), pp. 591-615. Elsevier, United States. 110 Kongkaew, W., Coleman, P., Pfeiffer, D.U., Antarasena, C. & Thiptara, A. (2004) Vaccination coverage and epidemiological parameters of the owned-dog population in Thungsong District, Thailand. Preventive Veterinary Medicine, 65, 105-115. Kumar, S. (2007) Methods for community participation. A complete guide for practitioners. Intermediate Technology Publications Ltd, Warwickshire, United Kingdom. Laflamme, D. (1997) Development and validation of a body condition score system for dogs. Canine Practice, 22, 10-15. Larghi, O.P., Arrosi, J.C., Nakajata-A, J. & Villa-Nova, A. (1988) Control of urban rabies. Rabies (eds J.B. Campbell & K.M. Charlton), pp. 407-422. Kluwer Academic Press, Boston, USA. Lee, J.H., Lee, M.J., Lee, J.B., Kim, J.S., Bae, C.S. & Lee, W.C. (2001) Review of canine rabies prevalence under two different vaccination programmes in Korea. The Veterinary Record, 148, 511-512. Lembo, T., Hampson, K., Kaare, M.T., Ernest, E., Knobel, D., Kazwala, R.R., Haydon, D.T. & Cleaveland, S. (2010) The feasibility of canine rabies elimination in Africa: dispelling doubts with data. PLoS Neglected Tropical Diseases, 4, 1-9. Lloyd-Smith, J.O., Cross, P.C., Briggs, C.J., Daugherty, M., Getz, W.M., Latto, J., Sanchez, M.S., Smith, A.B. & Swei, A. (2005) Should we expect population thresholds for wildlife disease? Trends in Ecology and Evolution, 20(9), 511-519. Lloyd, H.G., Jensen, B., van Haaften, J.L., Niewold, F.J.J., Wandeler, A., Bogel, K. & Arata, A.A. (1976) Annual turnover of fox populations in Europe. Zentralblatt fur Veterinarmedizin Reihe B, 23, 580-589. Lloyd, S. (1983) Effects of pregnancy and lactation upon infection. Veterinary Immunology and Immunopathology, 4. Lloyd, S., Amerasinghe, P.H. & Soulsby, E.J.L. (1983) Periparturient immunosupression in the bitch and its influence on infection with Toxocara canis. Journal of Small Animal Practice, 24, 237-247. Lodmell, D.L., Ewalt, L.C., Parnell, M.J., Rupprecht, C.E. & Hanlon, C.A. (2006) One-time intradermal DNA vaccination in ear pinnae one year prior to infection protects dogs against rabies virus. Vaccine, 24, 412-416. Lund, E.M., Armstrong, P.J., Kirk, C.A. & Klauser, J.S. (2006) Prevalence and risk factors for obesity in adult dgos from private US veterinary practices. Journal of Applied Research in Veterinary Medicine, 4, 177-186. 111 Macdonald, D.W. & Voigt, D.R. (1985) The Biological Basis of Rabies Models. Population Dynamics of Rabies in Wildlife (ed. P.J. Bacon), pp. 71-108. Academic Press, London, UK. MacInnes, C.D., Smith, S.M., Tinline, R.R., Ayers, N.R., Bachmann, P., Ball, D.G.A., Calder, L., A, Crosgrey, S.J., Fielding, C., Hauschildt, P., Honig, J.M., Johnston, D.H., Lawson, K.F., Nunan, C.P., Pedde, M.A., Pond, B., Stewart, R.B. & Voigt, D.R. (2001) Elimination of rabies from red foxes in eastern Ontario. Journal of Wildlife Diseases, 37, 119-132. Macpherson, C.N.L., Meslin, F.-X. & Wandeler, A.I. (2012) Dogs, zoonoses and public health, 2nd edition. CABI, Wallingford, United Kingdom. Malaga, H., Lopez Nieto, E. & Gambirazio, C. (1979) Canine rabies seasonality. International Journal of Epidemiology, 8, 243-245. Mansfield, K.L., Burr, P.D., Snodgrass, D.R., Sayers, R. & Fooks, A.R. (2004) Factors affecting the serological response of dogs and cats to rabies vaccination. The Veterinary Record, 154, 423-426. Margawani, K.R. & Robertson, I.D. (1995) A survey of urban pet ownership in Bali. The Veterinary Record, 147, 486-488. Matter, H. & Fico, R. (1998) Study of the structure and density of a dog population in Tekirdag (Turkey). The Journal of Etlik Veterinary Microbiology, 9, 9-24. Matter, H.C., Schumacher, C.L., Kharmachi, H., Hammami, S., Tlatli, A., Jemli, J., Mrabet, L., Meslin, F.X., Aubert, M.F.A., Neuenschwander, B.E. & El Hicheri, K. (1998) Field evaluation of two bait delivery systems for the oral immunization of dogs against rabies in Tunisia. Vaccine, 16, 657-665. Matter, H.C., Wandeler, A.I., Neuenschwander, B.E., Harischandra, L.P.A. & Meslin, F.X. (2000) Study of the dog population and the rabies control activities in the Mirigama area of Sri Lanka. Acta Tropica, 75, 95-108. McKinley, T.J., Morters, M. & Wood, J.L.N. (2014) Bayesian model choice in cumulative link ordinal regression models: an application in a longitudinal study of risk factors affecting body condition score in a dog population in Zenzele, South Africa. Bayesian Analysis, 10, 1-30. Merial Animal Health Limited Rabisin data sheet http://www.noahcompendium.co.uk/Merial_Animal_Health_Ltd/Rabisin/-66604.html Miller, W.H., Griffin, C.E. & Campbell, K.L. (2013) Muller & Kirk's Small Animal Dermatology 7th Edition. Elsevier, St. Louis, United States. 112 Minke, J.M., Bouvet, J., Cliquet, F., Wasniewski, M., Guiot, A.L., Lemaitre, L., Cariou, C., Cozette, V., Vergne, L. & Guigal, P.M. (2009) Comparison of antibody responses after vaccination with two inactivated rabies vaccines. Veterinary Microbiology, 133, 283-286. Minnaar, W.N. & Krecke, R.C. (2001) Helminths in dogs belonging to people in a resource- limited urban community in Gauteng, South Africa. Onderstepoort Journal of Veterinary Research, 68, 111-117. Minnaar, W.N., Krecke, R.C. & Rajput, J.I. (1999) Helminth parasites of dogs from two resource-limited communities in South Africa. Journal of the South African Veterinary Association, 70, 92-94. Mitmoonpitak, C., Tepsumethanon, V. & Wilde, H. (1998) Rabies in Thailand. Epidemiology and Infection, 120, 165-169. Mitmoonpitak, C., Wilde, H. & Tepsumethanon, V. (1997) Current status of animal rabies in Thailand. Journal of Veterinary Medical Science, 59, 457-460. Miyamoto, T., Taura, Y., Une, S., Yoshitake, M., Nakama, S. & Watanabe, S. (1992) Changes in blastogenic responses of lymphocytes and delayed type hypersensivity responses after vaccination in dogs. Journal of Veterinary Medical Science, 54, 945- 950. Moore, S.E., Goldblatt, D., Bates, C.J. & Prentice, A.M. (2003) Impact of nutritional status on antibody responses to different vaccines in undernourished Gambian children. Acta Paediatrica, 92, 170-176. Moore, S.M. & Hanlon, C.A. (2010) Rabies-specific antibodies: measuring surrogates of protection against a fatal disease. PLoS Neglected Tropical Diseases, 4, 1-7. MSD Animal Health Nobivac Rabies data sheet http://www.msd-animal- health.co.uk/Products_Public/Nobivac_Rabies/090_Product_Datasheet.aspx Mueller, R.S. (2004) Treatment protocols for demodicosis: an evidence based review. Veterinary Dermatology, 14, 75-89. Muller, J. (1966) The reappearance of rabies in Denmark. Bulletin of the Wildlife Disease Association, 65, 21-29. Muller, J. (1971) The effect of fox reduction on the occurence of rabies. Observations from two outbreaks of rabies in Denmark. Bulletin of the World Health Organization, 75, 763-776. Murray, J.K., Browne, W.J., Roberts, M.A., Whitmarsh, A. & Gruffydd-Jones, T.J. (2010) Number and ownership profiles of cats and dogs in the UK. Veterinary Record, 166, 163-168. 113 National Research Council (2006) Nutrient Requirements of Dogs and Cats. The National Academies Press, Washington DC, United States. Nayak, D.C., Tripathy, S.B., Dey, P.C., Ray, S.K., Mohanty, D.N., Parida, G.S., Biswal, S. & Das, M. (1997) Prevalence of canine demodicosis in Orissa, India. Veterinary Parasitology, 73, 347-352. Noakes, D.E., Parkinson, T.J. & England, G.C.W. (2009) Veterinary reproduction and obstetrics, 9th edition. Saunders, United Kingdom. Oboegbulem, S.I. & Nwakonobi, I.E. (1989) Population density and ecology of dogs in Nigeria: a pilot stdy. Revue Scientifique et Technique-Office International, 8, 733- 745. Ogunkoya, A.B., Will, L.A. & Ezeokoli, C.D. (1984) Rabies in Oyo State, Nigeria. International Journal of Zoonoses, 11, 84-94. Okell, C.N., Pinchbeck, G.P., Stringer, A.P., Tefera, G. & Christley, R.M. (2013) A community-based participatory study investigating the epidemiology and effects of rabies to livestock owner in rural Ethiopia. Preventive Veterinary Medicine, 108, 1-9. Ondrejkova, A., Suli, J., Ondrejka, R., Benisek, Z., Franka, R., Svrcek, S., Madar, M. & Bugarsky, A. (2002) Comparison of the detection and quantification of rabies antibodies in canine sera. Veterinarni Medicina 47, 218-221. Ortega-Pacheco, A., Rodriguez-Buenfil, J.C., Bolio-Gonzalez, M.E., Sauri-Arceo, C., Jimenez-Coello, M. & Forsberg, C.L. (2007a) A survey of dog populations in urban and rural areas of Yucatan, Mexico. Anthrozoos, 20, 261-274. Ortega-Pacheco, A., Segura-Correa, J.C., Jimenez-Coello, M. & Forsberg, C.L. (2007b) Reproductive patterns and reproductive pathologies of stray bitches in the tropics. Theriogenology, 67, 382-390. Pal, S.K. (2001) Population ecology of free-ranging urban dogs in West Bengal, India. Acta Theriologica, 46, 69-78. Palmer, S.R., Soulsby, E.J.L., Torgerson, P.R. & Brown, D.W.G. (2011) Oxford textbook of zoonoses: biology, clinical practice and public health control, 2nd edition. Oxford University Press, United Kingdom. Pastoret, P.P. & Brochier, B. (1998) Epidemiology and elimination of rabies in Western Europe. The Veterinary Journal, 156, 83-90. Peel, A.J., Sargan, D.R., Baker, K.S., Hayman, D.T.S., Barr, J.A., Crmeri, G., Suu-Ire, R., Broder, C., Lembo, T., Lin-Fa, W., Fooks, A.R., Rossiter, S.J., Wood, J.L.N. & Cunningham, A.A. (2013) Continental-wide panmixia of an African fruit bat 114 facilitates transmission of potentially zoonotic viruses. Nature Communication, 4, 1- 14. Phillips, T.R., Jensen, J.L., Rubino, M.J., Yang, W.C. & Schultz, R. (1989) Effects of vaccines on the canine immune system. Canadian Journal of Veterinary Research, 53, 154-160. Pianka, E.R. (1999) Vital statistics of populations. United States. Pinheiro, J., Bates, D., DebRoy, S., Deepayan, S. & Team, R.D.C. (2013) nlme: Linear and nonlinear mixed effects models. Precausta, P., Soulebot, J.P., Chappius, G., Brun, A., Bugand, M. & Petermann, H.G. (1985) NIL2 cell inactivated tissue culture vaccine against rabies - immunization of carnivores. Springer-Verlag, Berlin. Putra, A.A.G., Hampson, K., Girardi, J., Hiby, E., Knobel, D., Mardiana, I.W., Townsend, S. & Scott-Orr, H. (2013) Responses to a rabies epidemic, Bali, Indonesia, 2008-2011. Emerging Infectious Diseases, 19, 648-651. Putra, A.A.G., Mardiana, I.W., Scott-Orr, H., Hiby, E., Hampson, K., Haydon, D., Girardi, J., Knobel, D. & Townsend, S. (2011) Progress of a rabies elimination program in Bali, Indonesia. Rabies in Asia Conference. Colombo, Sri Lanka. R Core Team (2014) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria URL http://www.R-project.org/. Ratsitorahina, M., Rasambainarivo, J.H., Raharimanana, S., Rakotonandrasana, H., Andriamiarisoa, M.-P., Rakalomanana, F.A. & Richard, V. (2007) Dog ecology and demography in Antananarivo-2007. BMC Veterinary Research, 5, 1-7. Rautenbach, G.H., Boomker, J. & de Villiers, I.L. (1991) A descriptive study of the canine population in a rural town in southern Africa. Journal of the South African Veterinary Association, 62, 158-162. Reece, J.F. (2007) Rabies in India: an ABC approach to combating the disease in street dogs. The Veterinary Record, 161, 292-293. Reece, J.F. & Chawla, S.K. (2006) Control of rabies in Jaipur, India, by the sterilisation and vaccination of neighbourhood dogs. The Veterinary Record, 159, 379-383. Reece, J.F., Chawla, S.K. & Hiby, A.R. (2013) Decline in human dog-bite cases during a street dog sterilisation programme in Jaipur, India. The Veterinary Record, 172, 18- 21. Reynolds, B., Taillade, B., Medaille, C., Palenche, F., Trumel, C. & Lefebvre, H.P. (2006) Effect of repeated freeze-thaw cycles on routine plasma biochemical constituents in canine plasma. Veterinary Clinical Pathology, 35, 339-340. 115 Rhodes, C.J., Atkinson, R.P.D., Anderson, R.M. & Macdonald, D.W. (1998) Rabies in Zimbabwe: reservoir dogs and the implications for disease control. Philosophical Transactions of the Royal Society of London (B), 353, 999-1010. Robinson, L.E., Miranda, M.E., Miranda, N.L. & Childs, J.E. (1996) Evaluation of a canine rabies vaccination campaign and characterization of owned-dog populations in the Philippines. Southeast Asian Journal of Tropical Medicine and Public Health, 27, 250-256. Rodriguez-Vivas, R.I., Ortega-Pacheco, A., Rosado-Aguilar, J.A. & E, B.G.M. (2003) Factors affecting the prevalence of mange-mite infestations in stray dogs of Yucatan, Mexico. Veterinary Parasitology, 115, 61-65. Roitt, I., Brostoff, J. & Male, D. (2001) Immunology 6th Edition. Mosby, United Kingdom. Rosatte, R., MacDonald, E., Sobey, K., Donovan, D., Bruce, L., Allan, M., Silver, A., Bennett, K., MacDonald, K., Gibson, M., Buchanan, T., Stevenson, B., Davies, C., Wandeler, A. & Muldoon, F. (2007) The elimination of raccoon rabies from Wolfe Island, Ontario: animal density and movements. Journal of Wildlife Diseases, 43, 242-250. Rosatte, R., Sobey, K., Donovan, D., Bruce, L., Allan, M., Silver, A., Bennett, K., Gibson, M., Simpson, H., Davies, C., Wandeler, A. & Muldoon, F. (2006) Behaviour, movements and demographics of rabid raccoons in Ontario, Canada: management implications. Journal of Wildlife Diseases, 42, 589-605. Rothman, K.J., Greenland, S. & Lash, T.L. (2008) Modern Epidemiology Third Edition, Third edn. Lippincott Williams & Wilkins, Philadelphia, USA. Sage, G., Khawplod, P., Wilde, H., Lobaugh, C., Hemachudha, T., Tepsumethanon, W. & Lumlertdaecha, B. (1993) Immune response to rabies vaccine in Alaskan dogs: failure to achieve a consistently protective antibody response. Transactions of the Royal Society of Tropical Medicine and Hygiene, 87, 593-595. Schneider, M.C., Belotto, A., Ade, M.P., Correa, E., Tamayo, H., Medina, G. & Rodrigues, M.J. (2005) Epidemiological situation of human rabies in Latin America in 2004. Epidemiological Bulletin / Pan-American Health Organization, 26, 2-4. Serpell, J. & Jagoe, J.A. (1995) Early experience and the development of behaviour. The domestic dog its evolution, behaviour and interactions with people (ed. J. Serpell), pp. 80-102. Cambridge University Press, Cambridge, United Kingdom. Sharpee, R.L., Nelson, L.D. & Beckenhauer, W.H. (1985) Inactivated tissue culture rabies vaccine with three years immunogenicity in dogs and cats. Rabies in the tropics (eds E. Kuwert, C. Merieux, H. Koprowski & K. Bogel). Springer-Verlag, Berlin. 116 Shone, D.K. (1962) Rabies in southern Rhodesia: 1900 to 1961. Journal of the South African Veterinary Medical Association, 33, 567-580. Sibly, R.M. & Hone, J. (2002) Population growth rate and its determinants: an overview. Philosophical Transactions of the Royal Society of London (B), 357, 1153-1170. Siegrist, C.-A. (2008) Vaccine immunology. Vaccines Fifth Edition (eds S. Plotkin, W. Orenstein & P. Offit), pp. 17-36. Saunders, USA. Siegrist, C.-A. (2012) Vaccine immunology. Vaccines Sixth Edition (eds S. Plotkin, W. Orenstein & P. Offit), pp. 16-32. Saunders, USA. Sihvonen, L., Kulonen, K., Neuvonen, E. & Pekkanen, K. (1996) Rabies antibodies in vaccinated dogs. Acta Veterinaria Scandinavica, 36, 87-91. Sikes, R.K., Peacock, G.V., Acha, P., Arko, R.J. & Dierks, R. (1971) Rabies vaccines: duration-of-immunity study in dogs. Journal of the American Veterinary Medical Association, 159, 1491-1499. Smith, G.C. & Cheeseman, C.L. (2002) A mathematical model for the control of diseases in wildlife populations: culling, vaccinations and fertility control. Ecological Modelling, 150, 45-53. Soulsby, E.J.L. (1982) Helminths, arthropods and protozoa of domestic animals, 7th edition. Bailliere Tindall, United States. Steck, F. & Wandeler, A. (1980) The epidemiology of fox rabies in Europe. Epidemiologic Reviews, 2, 71-96. Strasser, A., May, B., Teltscher, A., Wistrela, E. & Niedermuller, H. (2003) Immune modulation following immunization with polyvalent vaccines in dogs. Veterinary Immunology and Immunopathology, 94, 113-121. Suppo, C., Naulin, J.-M., Langlais, M. & Artois, M. (2000) A modelling approach to vaccination and contracpetion programmes for rabies control in fox populations. Proceedings of the Royal Society of London. Series B, Biological sciences, 267, 1575- 1582. Taiwo, V.O., Antia, R.E., Adeniran, G.A., Adeyemi, G., Alaka, O.O. & Ohore, O.G. (1998) Rabies in dogs and cats in southwestern Nigeria: laboratory reports (1985-1995). Tropical Veterinarian, 16, 9-13. Talbi, C., Lemey, P., Suchard, M.A., Abdelatif, E., Elharrak, M., Jalal, N., Faouzi, A., Echevarria, J.E., Moron, S.V., Rambaut, A., Campiz, N., Tatem, A.J., Holmes, E.C. & Bourhy, H. (2010) Phylodynamics and human-mediated dispersal of a zoonotic virus. PLoS Pathogens, 6, e1001166. 117 Tenzin, Dhand, N.K., Dorjee, J. & Ward, M.P. (2011) Re-emergence of rabies in dogs and other domestic animals in eastern Bhutan, 2005-2007. Epidemiology and Infection, 139, 220-225. Tenzin, Sharma, B., Dhand, N.K., Timsina, N. & Ward, M.P. (2010) Reemergence of rabies in Chhukha District, Bhutan, 2008. Emerging Infectious Diseases, 16, 1925-1930. Tepsumethanon, W., Polsuwan, C., Lumlertdaecha, B., Khawplod, P., Hemachudha, T., Chutivongse, S., Wilde, H., Chiewbamrungkiat, M. & Phanuphak, P. (1991) Immune response to rabies vaccine in Thai dogs: a preliminary report. Vaccine, 9, 627-630. Thoresen, S.I., Tverdal, A., Havre, G. & Morberg, H. (1995) Effects of storage time and freezing temperature on clinical chemical parameters from canine serum and heparinzed plasma. Veterinary Clinical Pathology, 24, 129-133. Thraenhart, O., Kreuzfelder, E., Hillebrandt, M., Marcus, I., Ramakrishnan, K., Fu, Z.F. & Dietzschold, B. (1994) Long-term humoral and cellular immunity after vaccination with cell-culture rabies vaccine in man. Clinical Immunology and Immunopathology, 71, 287-292. Thrusfield, M. (2005) Veterinary epidemiology, 3rd edition. Blackwell Publishing, United Kingdom. Tizard, I. & Ni, Y. (1998) Use of serological testing to assess immune status of companion animals. Journal of the American Veterinary Medical Association, 213, 54-60. Tizard, I.R. (2013) Veterinary Immunology Ninth Edition. Elsevier, St. Louis, United States. Totton, S.C. (2009) Stray dog population health and demographics in Jodhpur, India, following a spay/neuter/rabies vaccination program. PhD Thesis, University of Guelph, Canada. Totton, S.C., Wandeler, A.I., Zinsstag, J., Bauch, C.T., Ribblw, C.S., Rosatte, R.C. & McEwen, S.A. (2010) Stray dog population demographics in Jodhpur, India following a population control / rabies vaccination program. Preventive Veterinary Medicine, 97, 51-57. Touihri, L., Zaouia, I., Elhili, K., Dellagi, K. & Bahloul, C. (2011) Evaluation of mass vaccination campaign coverage against rabies in dogs in Tunisia. Zoonoses and public health, 58, 110-118. Townsend, S.E., Lembo, T., Cleaveland, S., Meslin, F.X., Miranda, M.E., Putra, A.A.G., Haydon, D.T. & Hampson, K. (2013a) Surveillance guidelines for disease elimination: a case study of canine rabies. Comparative Immunology, Microbiology and Infectious Diseases, 36, 249-261. 118 Townsend, S.E., Sumantra, I.P., Pudjiatmoko, Bagus, G.N., Brum, E., Cleaveland, S., Crafter, S., Dewi, A.P.M., Dharma, D.M.N., Dushoff, J., Girardi, J., Gunata, I.K., Hiby, E.F., Kalalo, C., Knobel, D.L., Mardiana, I.W., Putra, A.A.G., Schoonman, L., Scott-Orr, H., Shand, M., Sukanadi, I.W., Suseno, P.P., Haydon, D.T. & Hampson, K. (2013b) Designing programs for eliminating canine rabies from islands: Bali, Indonesia as a case study. PLoS Neglected Tropical Diseases, 7, 1-11. Upjohn, M.M., Attwood, G.A., Lerotholi, T., Pfeiffer, D.U. & Verheyen, K.L.P. (2013) Quantitative versus qualitative approaches: a comparison of two research methods applied to identification of key health issues for working horses in Lesotho. Preventive Veterinary Medicine, 108, 313-320. Vandermeer, J.H. & Goldberg, D.E. (2003) Population ecology: first principles. Princeton University Press, Oxfordshire, United Kingdom. Venables, W.N. & Ripley, B.D. (2002) Modern applied statistics with S. Springer, United States. Vos, A. & Turan, B. (1998) Study of the dog population in Istanbul, Turkey. Journal of Etlik Veterinary Microbiology, 9, 25-34. Wallinga, J. & Lipsitch, M. (2007) How generation intervals shape the relationship between growth rates and reproductive numbers. Proceedings of the Royal Society of London. Series B, Biological sciences, 274, 599-604. Waltner-Toews, D., Maryono, A., Akoso, B.T., Wisynu, S. & Unruh, D.H.A. (1990) An epidemic of canine rabies in central Java, Indonesia. Preventive Veterinary Medicine, 8, 295-303. Wandeler, A. (1980) Hunters and foxes. Habitat classification, fox populations and rabies spread. Institute of Terrestrial Ecology, Huntingdon, United Kingdom. Wandeler, A.I. (1985) Ecological and epidemiological data requirements for the planning of dog rabies control. International conference on rabies in the tropics (ed. E. Kuwert), pp. 657-661. Springer-Verlag, Berlin. Wandeler, A.I., Capt, S., Gerber, H., Kappeler, A. & Kipfer, R. (1988) Rabies epidemiology, natural barriers and fox vaccination. Parassitologia, 30, 53-57. Wandeler, A.I., Muller, J., Wachendorfer, G., Schale, W., Forster, U. & Steck, F. (1974a) Rabies in wild carnivores in central Europe. III. Ecology and biology of the fox in relation to control operations. Zentralblatt fur Veterinarmedizin Reihe B, 21, 765-773. Wandeler, A.I., Wachendorfer, G., Forster, U., Krekel, H., Schale, W., Muller, J. & Steck, F. (1974b) Rabies in wild carnivores in central Europe. I. Epidemiological studies. Zentralblatt fur Veterinarmedizin Reihe B, 21, 735-756. 119 Weiss, D.J. & Wardrop, J.K. (2010) Schalm's veterinary hematology 6th edition. Wiley Blackwell, United States. Wells, C.W. (1954) The control of rabies in Malaya through compulsory mass vaccination of dogs. Bulletin of the Wildlife Disease Association, 10, 731-742. WHO (1996) Laboratory techniques in rabies, 4th edition. (eds F.-X. Meslin, M.M. Kaplan & H. Koprowski). World Health Organization, Geneva. WHO (1998) Field application of oral rabies vaccines for dogs. WHO/EMC/ZDI/98.15. World Health Organization (WHO), Geneva. WHO (2004) World Health Organization Expert Consultation on Rabies First Report. WHO Technical Report Series 931. Geneva. WHO (2013) World Health Organization Expert Consultation on Rabies Second Report. WHO Technical Report Series 982. World Health Organization, Geneva. WHO & WSPA (1990) Guidelines for Dog Population Management. WHO/ZOON/90.166. World Health Organization (WHO) and World Society for the Protection of Animals (WSPA), Geneva. Widdowson, M.-A., Morales, G.J., Chaves, S. & McGrane, J. (2002) Epidemiology of urban canine rabies, Santa Cruz, Bolivia, 1972-1997. Emerging Infectious Diseases, 8, 458- 461. Wilde, H., Chutivongse, S., Tepsumethanon, W. & Choomkasien, P. (1991) Rabies in Thailand. Reviews of Infectious Diseases, 13, 644-652. Windiyaningsih, C., Wilde, H., Meslin, F.-X., Suroso, T. & Widarso, H.S. (2004) The rabies epidemic on Flores Island, Indonesia (1998-2003). Journal of the Medical Association of Thailand, 87, 1389-1393. Woodroofe, R., Donnelly, C.A., Cox, D.R., Bourne, F.J., Cheeseman, C.L., Delahay, R.J., Gettinby, G., McInerney, J.P. & Morrison, W.I. (2006a) Effects of culling on badger Meles meles spatial organization: implications for the control of bovine tuberculosis. Journal of Applied Ecology, 43, 1-10. Woodroofe, R., Donnelly, C.A., Jenkins, H.E., Johnston, W.T., Cox, D.R., Bourne, F.J., Cheeseman, C.L., Delahay, R.J., Clifton-Hadley, R.S., Gettinby, G., Gilks, P., Hewinson, R.G., McInerney, J.P. & Morrison, W.I. (2006b) Culling and cattle controls influence tuberculosis risk for badgers. Proceedings of the National Academy of Sciences, 103(40), 14713-14717. WSPA (2012) http://www.rabiesblueprint.com/IMG/pdf/Link72_Euthanasia_WSPA.pdf last accessed 17.5.12. 120 Zinsstag, J., Durr, S., Penny, M.A., Mindekem, R., Roth, F., Menendez Gonzalez, S., Naissengar, S. & Hattendorf, J. (2009) Transmission dynamics and economics of rabies control in dogs and humans in an African city. Proceedings of the National Academy of Sciences, 106(35), 14996-15001. 121 Appendix 2 Appendix 2.1 Questionnaire development and implementation. The questionnaires were piloted in a sub-set of households in a non-study area of Braamfischerville (n=~15). Questionnaires were subsequently modified for use in Bali through meetings with the enumerators in April 2008. The same questionnaire was used for each census and revisit in all four study areas, with the exception of a few culturally specific questions and, from the second census onwards, the addition of owner reported clinical signs for the previous three months. Specific questions related to either the previous seven days or the previous three months, i.e. since the last survey. Responses were categorical, for example: When did you last feed this dog? – today / yesterday /other (specify) /don’t know; or open, for example When did you get this dog? The format of the question Why did you not get a dog from inside the site (names)? was changed from open ended to categorical in November 2009 because of cultural difficulties interpreting the question. In Bali, the questionnaire was bilingual (English and Bahasa with the Bahasa back-translated). Several languages are spoken in Johannesburg, including English, therefore the questionnaire was written in English and the accuracy of the various translations checked regularly with the multi-lingual enumerators throughout the study period. The respondent was the person/s in the household that the householders collectively identified as most knowledgeable about the dog, which was not necessarily the owner. Respondents under 16 years of age were always interviewed with an adult present. The same respondent was generally interviewed at each time point. When a respondent was not available (primarily in the Johannesburg sites), the household was revisited at least once during the same survey period to locate a respondent. 122 Appendix 2.2 Household characteristics. Zenzele Braamfischerville Kelusa Antiga number of households in the study area 2170 a / 2212 b 2844 203 365 total number of registered dogs 1022 882 707 600 c / 629 d number of dog-owning households at first time point 261 (12.0%) 212 (7.5%) 146 (71.9%) 179 (49.0%) number of dog-owning households at the last time point 222 (10.0%) 222 (7.8%) 148 (72.9%) 153 (41.9%) c / 152 (41.6%) d number of dogs per dog-owning household at first time point 1 dog: 202 1 dog: 165 1 dog: 76 1 dog: 130 2 dogs: 43 2 dogs: 35 2 dogs: 47 2 dogs: 36 3 dogs: 11 3 dogs: 10 3 dogs: 15 3 dogs: 8 4+ dogs: 5 4 dogs: 2 4+ dogs: 8 4+ dogs: 5 number of dogs per dog-owning household at the last time-point 1 dog: 170 1 dog: 175 1 dog: 76 1 dog: 112 c / 107 d 2 dogs: 38 2 dogs: 39 2 dogs: 49 2 dogs: 32 / 34 3 dogs: 12 3 dogs: 4 3 dogs: 20 3 dogs: 7 / 10 4 dogs: 2 4+ dogs: 4 4+ dogs: 3 4 dogs: 2 / 1 a before May 2009; b from May 2009 following the erection of an additional street of shacks May 2009; c until December 2010; d until May 2011; the estimated mean human population density (km -2 ) for each study area: Zenzele 14410.76, Braamfischerville 11895.14, Kelusa 3435.62, Antiga 2558.95 (Kelod / Kaler 6486.20 and Ketug 725.52) 123 Appendix 2.3 and 2.4 Sex ratios of the registered population (see tables below). The marked skew towards males in Kelusa and Antiga and the disparity between litter size determined during routine surgical sterilization of dogs from other villages in Bali and reported litter sizes for the study populations is probably a function of under-reporting of dumped neonates, particularly females. Although not systematically reported, at least 12% and 44% of neonates were killed, i.e. typically drowned in rice fields or the river, in Kelusa and Antiga respectively. In Bali the majority of dogs were owned for security and males were perceived to be better guard dogs than females; whereas in Johannesburg dogs were owned for a variety of reasons, including companionship and security. Appendix 2.3 Sex ratios of the registered populations (males: females) b . first time point last time point Zenzele 0.54: 0.46 0.53: 0.46 Braamfischerville 0.52: 0.48 0.58: 0.42 Kelusa 0.75: 0.25 0.75: 25 Antiga 0.77: 0.23 0.81: 0.19 a a including 2011 data; b excludes missing information 124 Appendix 2.4 Owner reported litter sizes for the study populations, and litter sizes determined during sterilization of pregnant bitches from similar populations elsewhere in Johannesburg and Bali during the study period. number of litters mean median minimum maximum litter size at sterilization Johannesburg 22 7.0 6 4 11 reported litter size Zenzele 273 5.7 6 1 13 reported litter size Braamfischerville 117 6.0 6 1 12 litter size at sterilization Bali 8 5.6 5.5 3 9 reported litter size Kelusa 161 3.2 3 1 7 reported litter size Antiga 130 3.3 3 1 9 125 Appendix 2.5 Age structure of the study population (see graphs below). Owner reported ages were often inconsistent. Therefore, population age structures were determined from the age distributions at the last time point, which contain the largest fraction of adults with known ages. Most dogs in their 42nd month of life or less had been observed as a pup or juvenile during the study period, so their true age was known. Additionally, several (Zenzele n= 9, Braamfischerville n= 4, Kelusa n= 9 and Antiga n=3) litters were born during the last time point. Appendix 2.5 Age structures for the registered population were determined from the age distribution at the last time point, which consists of the largest fraction of adults with known ages. The x-axis shows the number of individuals in each age class and the y-axis shows the month of life. For example, dogs in age class 2-3 are in their second or third month of life. 126 Appendix 2.6 Confinement (see table below). The method of confinement was variable. Dogs were confined, i.e. prevented from free- roaming, by tethers, pens, dog proof fences or a combination of these. “Frequently confined” dogs were reported at each interview for the entire period they were in the study population to have been confined during the previous seven days, although the frequency and duration of release during these seven days was variable. Confinement was also observed by the enumerators and the continuity of confinement between censuses discussed with the respondents. Very few dogs were never released. 127 Appendix 2.6 Confinement status of the registered populations. Zenzele Braamfischerville Kelusa Antiga b confined at some point while in the study population a 308 (30.1%) 551 (62.5%) 80 (11.3%) 80 (12.7%) confined continuously or frequently while in the study population <10% ~ 40% <10% <10% never confined 548 (53.6%) 106 (12.0%) 568 (80.3%) 492 (78.2%) missing information 166 (16.2%) 225 (25.5%) 59 (8.3%) 57 (9.1%) a including dogs confined continuously or frequently while in the study population; b including 2011 data 128 Appendix 2.7 Sterilization status of the registered populations. Zenzele a Braamfischerville b Kelusa c Antiga d males sterilized 7 (0.7%) 50 (5.7%) 100 (14.1%) 169 (26.9%) females sterilized 9 (0.9%) 50 (5.7%) 1 (0.1%) 1 (0.2%) not sterilized 773 (75.6%) 593 (67.2%) 489 (69.2%) 359 (57.1%) “don’t know” 9 (0.9%) 19 (2.2%) 0 0 missing information 224 (21.9%) 169 (19.2%) 117 (16.5%) 100 (15.9%) a most dogs were acquired sterilized from boss; b n=101 sterilized however the gender of one of these dogs was not known; c males castrated “traditionally” and the one female was acquired sterilized from boss; d including 2011 data 129 Appendix 2.8 Size and density of the registered populations. study area km 2 median population size population mean population minimum population maximum linear regression size density km -2 size density km -2 variation from mean size density km -2 variation from mean slope SE p-value Zenzele 0.72 321 323.03 448.65 285 395.83 11.8% 362 502.78 12.1% -1.524 0.232 <0.001 Braamfischerville 1.26 285 283.08 224.67 256 203.17 9.6% 294 233.33 3.8% 0.078 0.142 0.586 Kelusa 0.66 291 288.97 437.83 252 381.82 12.8% 312 472.73 8.0% 0.160 0.220 0.474 Antiga a 0.93 265 256.13 275.41 205 220.43 20.0% 286 307.53 11.7% -1.364 0.283 <0.001 Kelod / Kaler 0.30 180.5 174.50 581.67 136 453.33 22.1% 195 650.00 11.7% _ _ _ Ketug 0.63 84 81.50 129.37 66 104.76 19.0% 94 149.21 15.3% _ _ _ Linear regression models were fitted to assess the overall trend in the population size with time; the slope, error and p-value are shown a including the sub-villages (banjars) of Kelod / Kaler and Ketug; linear regression was applied at the village level only 130 Appendix 2.9 Age class at registration (excluding the starting cohorts). pups juveniles adults age not reported Zenzele 549 (81.6%) 42 (6.2%) 75 (11.1%) 7 (1.0%) Braamfischerville 433 (71.1%) 73 (12.0%) 93 (15.3%) 10 (1.6%) Kelusa 405 (89.0%) 28 (6.2%) 20 (4.4%) 2 (0.4%) Antiga a 323 (84.1%) 30 (7.8%) 30 (7.8%) 1 (0.3%) a 2011 data included 131 Appendix 2.10 Number of dogs in the registered populations gained per month (excluding the starting cohorts). mean median minimum maximum Zenzele 18.69 18 6 42 Braamfischerville 16.46 15 7 41 Kelusa 13.79 11 6 31 Antiga 10.32 9 2 28 132 Appendix 2.11 Number of dogs lost from the registered populations per month (excluding the starting cohorts). mean median minimum maximum Zenzele 20.31 19 7 34 Braamfischerville 16.14 16 3 38 Kelusa 13.55 13 3 25 Antiga 11.35 10 2 22 133 Appendix 2.12 and 2.13 Reproductively capable bitches (see tables below). Bitches were observed to be reproductively capable (i.e. in oestrus, which resulted in pregnancy) during their eighth and seventh month of life in the Johannesburg and Bali sites respectively. Reproductive efficiency was assessed in bitches across all four sites, by examining the monthly distribution of dogs pregnant as a proportion of reproductively capable females. The observed and estimated (i.e. which assumed a proportion of dogs lost the following month were early pregnant) distributions were similar, although pregnancy rates were higher in Bali. There was no overall increase or decrease in the proportion of dogs pregnant and the number of litters per month, and there was no clear evidence of seasonal variation. Interpretation of the time series analysis is constrained by small numbers of females pregnant or whelping per month, although plots for Kelusa and Antiga and autocorrelation and partial autocorrelation for Antiga indicates periodicity with an interval of two months. This may be a consequence of the enumerators tending to record bitches whelping the month of the survey rather than the correct month within the usual 6 week inter-survey period for households with female dogs. This noise was not evident in the plots for population size for Bali (Figure 2.1) probably because, compared to all sources of registered dogs, only a small proportion originated from within the study area (Table 2.1). In Zenzele 48.7%, Braamfischerville 63.6%, Kelusa 41.6% and Antiga 33% dogs categorised as reproductively capable did not reproduce during the study period, however these dogs encompass a range of ages, observational periods, reproductive histories, husbandry practices and health status. Consequently, differentiating the factors influencing fecundity, such as nutritional status, pathogens (Noakes, Parkinson & England 2009) and husbandry is not straight forward. In contrast to a previous study (Ortega-Pacheco et al. 2007b), on average pregnant dogs were more likely to be in a higher body condition category than non-pregnant dogs however this relationship does not vary across body condition class (Appendices 2.26- 2.30). Despite careful enumerator training in body condition scoring, this variation may be the result of incorrect body condition scores allocated to pregnant dogs with enlarged abdomens. Furthermore, this model does not account for longitudinal variations in body condition prior to pregnancy necessary to determine the effect on body condition on the probability of pregnancy. 134 Appendix 2.12 Proportion of reproductively capable females pregnant per month in the registered populations. Zenzele Braamfischerville Kelusa Antiga observed proportion pregnant (number) estimated proportion pregnant observed proportion pregnant (number) estimated proportion pregnant observed proportion pregnant (number) estimated proportion pregnant observed proportion pregnant (number) estimated proportion pregnant mean 0.090 (8.49) 0.097 0.046 (3.34) 0.049 0.117 (4.94) 0.123 0.111 (3.91) 0.114 median 0.087 (8) 0.091 0.045 (3) 0.049 0.107 (5) 0.120 0.083 (3) 0.091 minimum 0.011 (3) 0.028 0 (0) 0 0 (0) 0 0.024 (1) 0.026 maximum 0.227 (20) 0.231 0.107 (8) 0.108 0.306 (12) 0.316 0.250 (9) 0.250 slope 2.89E-05 3.49E-05 1.04E-05 1.29E-05 3.00E-04 6.00E-04 -6.13E-05 -5.66E-05 SE 2.23E-05 2.11E-05 1.31E-05 1.29E-05 1.40E-03 1.40E-03 3.95E-05 3.97E-05 p-value 0.202 0.107 0.433 0.327 0.828 0.702 0.131 0.164 The estimated proportion pregnant includes the observed proportion pregnant plus an estimate of the proportion of females lost from the study population the following month that were early pregnant (i.e. lost before pregnancy could be confirmed visually) Linear regression models were fitted to assess the overall trend in the proportion pregnant with time; the slope, error and p-value are shown 135 Appendix 2.13 Number of litters per month in the registered populations. Zenzele Braamfischerville Kelusa Antiga mean 8.38 3.42 5.35 4.18 median 8 3 5 3 minimum 3 0 0 1 maximum 20 8 12 9 slope 0.062 -0.004 0.053 -0.051 SE 0.061 0.031 0.056 0.042 p-value 0.320 0.910 0.355 0.241 Linear regression models were fitted to assess the overall trend in the number of litters per month; the slope, error and p-value are shown 136 Appendix 2.14 – 2.16 Mortality of registered dogs (see tables below). With regards mortality, the category of “other” combines disease (including injury) -induced mortality and dogs found dead or where the cause of mortality was not known; based on Appendix 2.16 the majority of the dogs in the latter categories probably died from disease or injury. There was no overall increase or decrease in the proportion of dogs dying per month, except for Kelusa where total mortality and combined reported and assumed (i.e. “other”) disease- induced mortality increased (p<0.001). There was a marginal increase in reported (i.e. “specific”) disease-induced mortality (p=0.08). There was no specific cause for the increase in total mortality in Kelusa; the number of dogs hit-by-cars (n=70) more than doubled in 2009 and 2010 (mean 2.5 deaths per month), the number of dogs killed by people (n=18) progressively increased from 0 per month in 2008 to 1 per month in 2010, and the number of dogs with missing information (n=36) doubled in 2010 to 1.7 per month. There is no clear evidence of periodicity, including seasonal variation, for overall mortality, specific disease- induced mortality and “other” mortality. Interpretation of the time series is constrained by the small numbers of dogs dying per month. Although the clinical signs at the time of death were reported, in most cases the aetiology could not be determined. Rather, the reported clinical signs primarily served to verify that mortality was from illness rather than some other cause (Appendix 2.16). Mortality due to starvation (12% pups) has only been reported once (Pal 2001). In this study, body condition at the time of death was not consistently reported given the practical difficulties of verification by the enumerators. On average animals suffering from clinical conditions likely to cause weight loss were indeed thinner than animals not suffering from these conditions (Appendices 2.25-2.30), and this relationship did not vary with body condition class. Thus, weight loss would be expected in dogs suffering from these conditions at the time of death. Loss of body condition in sick dogs may be a consequence of nausea and inappetence rather than increased metabolic requirements. Concomitant neglect may also occur, particularly for dogs with generalised dermatitis, often with malodourous skin. Almost all of the dogs with generalised dermatitis were in Bali, and were the majority of those dogs with observed clinical signs (Appendices 2.26, 2.29-2.30). Owners only occasionally reported mortality from starvation (Appendix 2.16). More dogs were observed to be severely underweight (body condition score ≤2) than reported to have died from neglect. 137 Appendix 2.14 Mortality per month for the registered populations (includes all age classes). Linear regression models were fitted to assess the overall trend in the number and proportion dying with time; the slope, error and p-value are shown Note: “other” combines disease (including injury)-induced mortality and dogs found dead or where the cause of mortality was not known on the basis that the majority of the dogs in the latter categories probably died from disease or injury (Appendix 2.16). <0.001 6.77E-06 2.79E-05 0.061 0.008 0.034 0.017 6 6.790.0190.034 0.005 2 14 4.50E-03 1.50E-03 0.082 4.63E-06 8.32E-06 0.038 0.008 2.60E-03 0.074 0.003 0.019 total total disease other prop.numberprop.numberprop.number 0.021 0.0235.58 3 0.343 2.30E-03 disease other prop.numberprop.numberprop.number 0.348 7.40E-06 -7.03E-06 0.079 0.010 0.033 0.034 8 8.76 0.545 5.40E-06 3.30E-06 0.045 0.041 0.026 12 total disease other prop.numberprop.numberprop.number 12.640.0280.044 0.269 8.34E-06 1.20E-031.90E-03 26 20 0.072 0.01035 9.36E-06 0.081 0.017 6.97E-06 6.70E-06 0.057 0.009 0.281 9.66E-06 1.05E-05 0.557 2.80E-03 1.70E-03 23 8.19E-03 1119 22 7.26E-06 -2.63E-06 0.072 0.007 0.0224.787.73 14 11 0.043 012 5.590.018 1.20E-03 5.07E-06 1.79E-06-9.60E-046.00E-04 0.7270.4240.701 8 5 0.022 0.8750.7260.290 6.35E-06 0.060 -1.51E-06 0.039 12 00.004 0.019 6 Zenzele Braamfischerville p-value SE mean median minimum maximum slope 9 5 13 0.029 0.022 2 0.5360.344 0.720 23 13 1.00E-03-2.00E-03 <0.001 1.40E-032.01E-03 0.0399.1914.25 9.760.0316.119.59 10 5 2.10E-03 1.50E-03 0.735 2.10E-03 -7.00E-04 21 3.20E-03 12 8 0.040 0.570 0.615 otherdisease total AntigaKelusa <0.001 1.70E-035.12E-06 1.42E-05 0.047 0.012 prop.numberprop.numberprop.number 5.92E-06 0.008 0.030 0.030 4.27E-06 1.40E-03 -2.00E-04 138 Appendix 2.15 Mortality per month of dogs ≤ 12 months (i.e. ~52 weeks) of age for the registered populations. a including March & April 2008 p-values for total = 0.057, disease = 0.048, other = 0.041 Linear regression models were fitted to assess the overall trend in the number and proportion dying with time; the slope, error and p-value are shown Note: “other” combines disease (including injury)-induced mortality and dogs found dead or where the cause of mortality was not known on the basis that the majority of the dogs in the latter categories probably died from disease or injury (Appendix 2.16). 0.442 0.694 1.07E-04 p-value 0.134 0.192 a 0.193 0.205 a 0.109 0.172 a 0.662 0.678 0.352 0.581 0.251 3.23E-02 1.10E-04 3.43E-02 1.19E-04 2.69E-02 9.93E-05 2.90E-02 0.0079 0.001990.439 0.409 0.567 0.575 0.00018 0.0002 0.00184 0.731 1.80E-04 3.66E-02 1.20E-04 4.64E-02 1.55E-04 4.46E-02 0.00765 slope 8.25E-02 2.40E-04 4.85E-02 1.50E-04 7.63E-02 1.32E-04 3.00E-01 3.86E-04 2.51E-02 9.58E-05 -2.49E-02 -8.63E-05 1.61E-01 5.55E-04 8.56E-02 1.57E-04 3.21E-02 1.15E-04 4.31E-02 1.50E-04 3.80E-02SE 5.38E-02 1.61E-02 1.25E-049.25E-05 1.07E-02 9.37E-053.74E-04 1.19E-022.90E-04 1.09E-012.17E-04 -1.97E-02 -6.57E-05 0.027 8 0.0270.048 7 0.025 13 0.045 10 0 0 maximum 12 0.038 10 0.032 10 0.034 14 0 0 1 0.004 0 0 10 0.035 8 0.028 9 0.0320.036 8 0.003 1 0.004 0 02 0.006 1 0.003 0 0minimum 2 0.006 1 0.003 3.70 0.014 2.46 0.001 median 6 0.019 4 0.013 5.5 3 0.010 4 0.010 32 0.007 3 1 2.97 0.012 0.018 4 0.014 0.009 3.24 0.0110.016 3.03 0.011 4.22 0.014 3 0.011 0.014 3 0.010 3 0.011 Zenzele Braamfischerville Kelusa 0.015 4.58 0.016 2.67 prop. number prop. number prop. number mean 6.58 0.021 4.19 0.013 6.03 0.019 4.49 number prop. Antiga total disease other total disease other total disease other total disease other prop. number prop.number prop. number prop.number prop. number prop. number prop. number prop. number 139 Appendix 2.16 Causes of mortality of the registered dogs. Zenzele Braamfischerville Kelusa Antiga sick a 328 (63.9%) 218 (61.4%) 176 (54.7%) 157 (59.5%) hit-by-car 37 (7.2%) 21 (5.9%) 70 (21.7%) 37 (14.0%) killed by owner 4 (0.8%) - 6 (1.9%) 5 (1.9%) killed by someone 5 (1.0%) 3 (0.8%) 12 (3.7%) 18 (6.8%) severe weather ˉ 2 (0.6%) ˉ ˉ neglect / starvation 3 (0.6%) ˉ 1 (0.3%) 4 (1.5%) b ceremonial sacrifice NA NA 6 (1.9%) 3 (1.1%) eaten by the owner NA NA ˉ 4 (1.5%) found dead 23 (4.5%) 19 (5.3%) 5 (1.4%) 2 (0.8%) dumped ˉ ˉ 3 (0.8%) 5 (1.9%) other 15(2.9%) 14 (3.9%) 7 (2.2%) 5 (1.9%) don't know 21 (4.1%) 17 (4.8%) 21 (6.5%) 2 (0.8%) missing entries 77 (15.0%) 61 (17.2%) 15 (4.7%) 22 (8.3%) 513 355 322 264 a includes infections and injuries; b includes 3 pups that did not nurse properly 140 Appendix 2.17 – 2.19 Reasons for sourcing of dogs outside of the research sites (see tables below). Across all communities, the vast majority of owners reported that getting a dog was a conscious decision. Many of the owners sought dogs from outside the community that they lived in; of those owners who planned to get a dog and sourced it from outside the research site, <50% (Zenzele 48%, Braamfischerville 38%, Kelusa 32%, Antiga 30%) tried to find a dog in the community that they lived in. Different cultural perspectives meant that determining why a dog was not sourced from inside the research site was difficult, hence the large amount of missing information. Owners who did not plan to get their dogs generally acquired them opportunistically, but the majority of these dogs were not unwanted; rather, at the time of acquisition owners deliberately decided that they wanted to keep their dogs for specific purposes, e.g. security. 141 Appendix 2.17 Reasons for sourcing a dog from outside the Johannesburg sites. a Question: why did you not the dog from inside the research site? Zenzele Braamfischerville % % chance acquisition 16.8 21.1 gift 16.3 16 relocated with owner 12.4 3.4 rescued 3.5 1.7 stray / found 0.5 0.8 concerned the dog will return to the original owner 1.5 0.0 a dog was not available 6.9 5.9 a pup was not available 9.4 14.8 an adult was not available 1.0 0.0 the right breed was not available 11.9 21.5 from a breeder 0.0 0.4 the available breed was preferred 1.5 1.7 a healthy dog was not available 6.4 8.0 dogs in the research site are unhealthy 2.5 1.3 dogs in the research site are unclean 1.5 0.0 a dog with the right temperament was not available 5.0 7.6 a male was not available 1.5 3.8 a female was not available 1.0 0.0 dog sitting 1.5 0.8 convenient to get from relative 3.5 11.8 convenient to get from friend 4.0 12.2 from boss or relative's / work / colleague 1.2 7.6 from stranger 0.0 0.4 did not want to ask friends / family / neighbours in the research site 0.5 4.2 did not know who to ask in the research site 0.0 8.9 brought home 1.0 0.4 free dog not available in BF / dog was free 0.0 1.7 no time to look in the research site / dog readily available 0.5 0.0 from a second house 0.0 0.8 other 0.5 2.1 don't know 3.0 0.0 a either could not find a dog or assumed that a dog could not be found Information was available for 73.5% (202/275) of the dogs in Zenzele and 68.7% (237/345) of the dogs in Braamfischerville. Most of the missing information is prior to November 2009 when the relevant questions were changed to improve interpretation. 142 Appendix 2.18 Reasons for sourcing a dog from outside the Bali sites. a Question: why did you not get the dog from inside the research site? Kelusa Antiga % % chance acquisition 39.3 49.0 gift 3.6 6.3 rescued 0.0 1.0 stray / found 2.7 0.0 a dog was not available 7.1 2.1 a pup was not available 5.4 2.1 the right breed was not available 11.6 12.5 the available breed was preferred 0.0 5.2 a healthy dog was not available 8.9 9.4 the available dog was healthy 0 1.0 a dog with the right temperament was not available 7.1 1.0 the available dog was the right temperament 0.0 2.1 a male was not available 3.6 1.0 a male dog was available 0.0 1.0 convenient to get from a relative 22.3 30.2 convenient to get from a friend 25.0 15.6 from boss or relative's / work / colleague 4.5 6.3 did not want to ask friends / family / neighbours in the research site 1.8 2.1 brought home 1.8 0 for ceremony 0.0 2.1 no time to look in the research site 0.0 5.2 from a second house 2.7 4.2 a either could not find a dog or assumed that a dog could not be found Information was available for 76.2% (112/147) of the dogs in Kelusa and 69.1% (96/139) of the dogs in Antiga. Most of the missing information is prior to November 2009 when the relevant questions were changed to improve interpretation. These data exclude 7 dogs in Kelusa and 1 dog in Antiga born in the dry field and categorized as from outside the research site in Table 2.1. 143 Appendix 2.19 Assessment of whether the acquisition of the dog was planned and if the owner looked in the research site for a dog. Question: did you plan to get a dog? Question: did you look for a dog in the research site? yes no missing data yes no missing data Zenzele 83 12 180 38 39 198 Braamfischerville 158 16 171 60 96 189 Kelusa 47 29 71 15 56 76 Antiga 33 31 75 10 54 75 Most of the missing information is prior to November 2009 when these questions were added to the survey to improve interpretation of the question: Why did you not get the dog from inside the research site? (see Appendices 2.17 and 2.18) 144 Appendix 2.20 Registered dogs acquired from outside the study areas per month. Zenzele Braamfischerville Kelusa Antiga number prop. number prop. number prop. number prop. mean 7.57 0.023 9.63 0.034 5.09 0.018 4.67 0.018 median 7 0.022 8 0.029 5 0.017 3 0.012 minimum 2 0.006 3 0.010 0 0 0 0 maximum 21 0.060 21 0.074 11 0.040 13 0.049 slope -6.90E-02 -1.13E-04 -1.55E-02 -7.06E-05 9.43E-02 3.25E-04 -0.01671 -6.62E-05 SE 6.32E-02 1.92E-04 6.45E-02 2.29E-04 4.70E-02 1.70E-04 0.06781 2.49E-04 p-value 0.282 0.561 0.811 0.760 0.054 0.064 0.807 0.792 Linear regression models were fitted to assess the overall trend in the number and proportion of registered dogs from outside the study area with time; the slope, error and p-value are shown 145 Appendix 2.21(i) Outcome of pups (see table below). The number of pups born in households in the Bali study areas may be substantially higher than shown given that female neonates were often dumped and, thus, litter size was probably deliberately under-reported. Where the complete litter size was not known by an owner, litter size was rounded up based on the average number of foetuses observed in gravid uteri during routine surgery (Appendix 2.4). The complete litter size was not known for 9% of the litters born in Zenzele during the study period, 7% in Braamfischerville 9% in Kelusa and 6% in Antiga. Appendix 2.21(ii) Unknown sources and outcomes of dogs (see table below and Tables 2.1-2.2). Tables 2.1 and 2.2 (and Appendix 2.21) show the contribution of unknown sources and outcomes of dogs to the dynamics of the populations. Based on the distribution of known outcomes, the majority of dogs with unknown outcomes probably died given that they were not observed by the enumerators. This sub-group should not be ignored in population models or control policies. In addition, a proportion of the registered dogs from the study area but where the household of origin was not identified a were probably from the pool of locally born pups given away in the study area or research site and where the recipient household was not identified b . The residuals from the second pool of dogs b may have died or subsequently moved out of the study area given that they were not observed by the enumerators (Zenzele 98 a : 138 b , Braamfischerville 4: 62, Kelusa 20: 55, Antiga 6: 88). 146 Appendix 2.21 Outcomes of the pups born in the study areas. Zenzele Braamfischerville Kelusa Antiga kept where born (registered) 128 (7.2%) 60 (7.8%) 128 (20.9%) 87 (18.1%) given away in study area address known (registered) 112 (6.3%) 33 (4.3%) 24 (3.9%) 18 (3.7%) total number pups born in study area registered 240 (13.5%) 93 (12.1%) 152 (24.8%) 105 (21.8%) given away in study area address not known / pup not found 132 (7.4%) 8 (1.0%) 1 (0.2%) 0 given away in research site but area not known NA 54 (7.0%) 54 (8.8%) 88 (18.3%) given away in non-study area of the research site NA 3 (0.4%) 8 (1.3%) 23 (4.8%) given away but location not known 6 (0.3%) 0 0 0 given away out outside research site 103 (5.8%) 65 (8.5%) 30 (4.9%) 18 (3.8%) died 1040 (58.6%) 438 (57.0%) 272 (44.4%) 179 (37.2%) disappeared 25 (1.4%) 7 (0.9%) 6 (1.0%) 10 (2.1%) stolen 27 (1.5%) 31 (4.0%) 0 0 relocated outside the research site with owner 13 (0.7%) 0 0 2 (0.4%) relocated to non-study area of the research site with owner NA 7 (0.9%) 0 0 unaccounted for 190 (10.7%) 63 (8.2%) 90 (14.7%) 56 (11.6%) total number pups born in study area 1776 769 613 481 Includes pups born 1 or 2 months prior to the start of the study period that were present at the start of the study period 147 Appendix 2.22 Average time to loss or censoring of the starting cohorts. factor level average time (months) Zenzele age adult 19.8 juvenile 16.7 pup 13.0 gender male 18.6 female 19.0 Braamfischerville age adult 20.1 juvenile 13.2 pup 17.2 gender male 19.4 female 18.0 Kelusa age adult 23.3 juvenile 16.9 pup 16.8 gender male 22.8 female 18.4 Antiga age adult 26.7 juvenile 18.4 pup 16.2 gender male 26.2 female 21.0 148 Appendix 2.23 Risk of loss from the starting cohorts by age class at the start of the study and gender. observations study period (months) factors retained in the minimum adequate model factor level exp(coef) se(coef) 2.5%CI 97.5%CI p-value Zenzele 368 36 age adult - - - - - juvenile 1.315 0.145 0.990 1.748 0.059 pup 1.761 0.253 1.074 2.889 0.025 Braamfischerville 287 36 age adult - - - - - juvenile 1.862 0.168 1.339 2.589 <0.001 pup 1.396 0.289 0.793 2.458 0.248 Kelusa 278 33 age + gender adult - - - - - juvenile 1.777 0.201 1.197 2.637 <0.001 pup 1.831 0.235 1.156 2.901 <0.001 female - - - - - male 0.582 0.164 0.422 0.802 <0.001 Antiga 266 37 age adult - - - - - juvenile 2.203 0.218 1.437 3.376 <0.001 pup 2.749 0.262 1.646 4.590 <0.001 All the cox proportional hazards models included duration in the study, age class at the start of the study and gender. The model values vary slightly from the averages shown in Appendix 2.22 because of how the model handled censoring (Crawley 2007) 149 Appendix 2.24 Outcomes of the dogs in the starting cohorts. Zenzele Braamfischerville Kelusa Antiga total died 204 (71.1%) 145 (65.9%) 127 (72.2%) 116 (72.5%) died of disease 132 (64.7%) 83 (57.2%) 82 (64.6%) 72 (62.1%) disappeared 19 (6.6%) 5 (2.3%) 21 (11.9%) 13 (8.1%) stolen 8 (2.8%) 12 (5.5%) 0 0 given away in non-study area of the research site NA 1 (0.5%) 0 0 given away outside research site 16 (5.6%) 21 (9.5%) 5 (2.8%) 10 (6.3%) given to meat trader NA NA 11 (6.3%) 7 (4.4%) relocated outside research site with owner 15 (5.2%) 15 (6.8%) 3 (1.7%) 0 dumped 0 0 0 1 (0.6%) other 1 (0.3%) 0 0 1 (0.6%) unaccounted for 19 (6.6%) 14 (6.4%) 9 (5.1%) 10 (6.3%) given away in study area but not found by enumerators 5 (1.7%) 1 (0.5%) 0 0 given away in research site but area not known NA 6 (2.7%) 0 2 (1.3%) total lost 287 220 176 160 150 Appendix 2.25 Marginal positive predictive distributions for the explanatory variables, including observed pregnancy and lactation, against the observed maximum body condition scores for Zenzele. Bars represent the data, the points are the marginal predictive means and the error bars are the 95% prediction intervals. These distributions are representative of the four communities, including maximum and minimum body condition scores and arbitrary and observed pregnancy and lactation. OPL includes non-lactating females and males (N) and lactating (L) and pregnant (P) dogs. Age categories are in months (see Appendix 2.26). 151 Appendix 2.26 Variables included in the ordinal regression models. See Appendix 2.27 for application of the variable categories. variable categories body condition score 1&2 (emaciated & very underweight), 3 (underweight), 4 (slightly underweight), 5 (ideal), 6-9 (slightly overweight - obese); maximum body condition = largest of the two condition scores; minimum body condition = smallest of the two condition scores gender female vs. male lactating non-lactating female and male vs. lactating female; arbitrary lactation period = 12 weeks lactation period; observed lactation period = lactation observed by the enumerators pregnant non-pregnant female and male vs. pregnant female; arbitrary pregnancy period = 63 days gestation; observed pregnancy period = pregnancy observed by the enumerators sterilized not sterilized vs. sterilized confined not confined at some point since the previous census vs. confined at some point during the previous census age 1-6 months a vs. 7-12 months, 13-36 months, mature adult b number of dogs per household number of dogs per household observed minor clinical signs healthy vs. clinical conditions unlikely to cause weight loss observed by the enumerators observed major clinical signs healthy vs. clinical conditions likely to cause weight loss (and generalised dermatitis); includes dogs with a combination of minor and major clinical signs owner reported minor clinical signs healthy vs. only clinical conditions unlikely to cause weight loss for the previous 7 days and 3 months (i.e. since the last census) owner reported major clinical signs healthy vs. clinical conditions likely to cause weight loss (and generalised dermatitis) for the previous 7 days and 3 months (i.e. since the last census); includes dogs with a combination of minor and major clinical signs; short duration = one episode or one day duration, medium = 2-6 days duration, long = 7+ days duration a dogs in age class 1-6 months are between birth [beginning of their 1st month of life] and ~26 weeks of age [end of their 6th month of life], dogs in age class 7-12 months are between ~27 weeks of age [beginning of their 7th month of life] and ~52 weeks of age [end of their 12th month of life], and dogs in age class 13-36 months are between ~53 weeks of age [beginning of their 13th month of life] and ~156 weeks of age [end of their 36th month of life]; b mostly dogs first observed as adults at the start of the study period, but also dogs first observed as adults after the start of study period, therefore the exact age of these dogs could not be determined by direct observation [Note: dogs in age classes 1-6, 7-12 and 13-36 months were observed as pups or juveniles during the study period, so their true age was known] 152 Appendix 2.27 Zenzele ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios. The first three columns (Proportional Odds [PO], Non-Proportional Odds [NPO] and excluded) show the posterior probabilities of association (PPA) for each variable, averaged across all the competing models. Only those variables with a PPA ≥0.5 are included (Jefferys 1961). The averaged posterior means and standard deviations for each variable are also shown. Where the means and SDs are the same across the levels (1-4) only a single result is shown. In reference to the results for observed lactation and gestation and maximum and minimum body condition scores, males are on average ~1.2× more likely to have a lower body condition score (BCS) than females. Lactating females are, on average, ~ 2.2× more likely to have a lower BCS than equivalent males and non-pregnant females. Pregnant females are ~1.5× more likely to be have a higher BCS than non-pregnant females and males. For age, relative to the 1–6 month category, dogs aged 7–12 months are ~1.2× more likely to have a higher BCS; and, dogs aged 13–36 months are ~1-1.2× more likely to have a lower BCS. The mature adult age class has very strong evidence of an NPO structure. Mature adults are ~1–2.5× more likely to have a higher BCS, depending on the category level. For maximum BCS, mature dogs were clustered in the higher body condition categories; whereas, for the minimum BCS, these dogs were distributed across body condition categories but were, on average, fatter than dogs in the other age categories. A likely explanation is that this pattern reflects normal morphological variation - generally, as dogs become older their activity levels will decrease, resulting in a general increase in BCS. Dogs with major and minor clinical sings were ~2x and 1.1x more like to have a lower BCS than healthy dogs respectively. 0.15 0.210.47001.00 ¯¯¯ 0.06 0.06 (0.62, 0.65, 0.70, 0.73) 0.065 (0.28, 0.24, 0.25, 0.31) 0.15 0.170.300.06 1.00 0 0.0005 0.84 0.94 0.941.00 0.0005 0 0 (-0.003, 0.27, 0.75, 1.50) -0.18 0.13 1.00 1.00 minimum body condition score 0 0 0.0015 ¯ 0 ¯ 0 NPO 0.0005 0 0 ¯ 0 0 0.0005 SD minimum body condition scoremaximum body condition score SD 0.18 ¯ 0 excluded 0 (-0.56, -0.55, -0.55, -0.55) -0.19 SD ¯ (0.23, 0.18, 0.18, 0.21) 0.16 0.15 ¯ ¯ ¯ ¯ 0 1.00 1.00 0 0 SD ¯ 0.16 0.22 0.99 ¯ ¯ 0 ¯ (0.23, 0.18, 0.17, 0.21) 0.16 0.15 (-0.006, 0.27, 0.75, 1.50) mean 01.00 -0.28 excludedPOexcludedPO 01.00 0.005 0 0.009 ¯ 0 ¯ (0.25, 0.22, 0.22, 0.22)(0.46, 0.45, 0.45, 0.45) NPO 0 NPO 0.11 0.12 0.210.53 0.47 0.10 (0.13, 0.11, 0.11, 0.12) 0.21 PO mean PO 0 -1.30 -0.54 -0.15 1.00 0.45 ¯ -0.941.00 ¯ 0 meanmean -1.00 -0.14 0.31 -0.28 1.00 NPO 0.290.71 -1.10 ¯ 1.00 0 0 0.0015 Zenzele maximum body condition score 01.00 01.00 01.00 observed major clinical signs observed minor clinical signs confined 1.00 01.00 0.0020.99 -1.00 -0.14 (-0.53, -0.52, -0.52, -0.52) 0.270.73 excluded gender:male 0.69 -0.27 lactating: observed lactation period lactating: arbitrary lactation period pregnant: arbitrary gestation period 1.00 1.00 ¯ -0.89 ¯¯ 00.0021.00 0.88 1.00 -0.57 (0.14, 0.14, 0.15, 0.15) 0.11 (-0.16, -0.17, -0.16, -0.16) 0.12 0 ¯ 0 0.23 ¯¯ 1.00 ¯ ¯ 0.180 ¯ 0.001 0 0.005 0.120 1.00 1.00 pregnant: observed gestation ¯ mature adult 13-36 months 7-12 months 0 0 1.00 0 0 ¯ 0.15 (0.13, 0.12, 0.12, 0.13) 0.180 ¯ 0 (0.15, 0.15, 0.15, 0.16) 0.21 0.12 0.18 0 0 0 (0.68, 0.72, 0.78, 0.83) 0.09 0.34 (0.25, 0.18, 0.19, 0.27) 0.15 0 -1.30 0.16 ¯ ¯¯ 0.15 0.11 ¯ ¯ 0 ¯ 0 ¯ 0.096 0 -0.14 -1.30 153 Appendix 2.28 Braamfischerville ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios; see Appendix 2.27 for interpretation of the results. The first three columns (PO, NPO and excluded) show the posterior probabilities of association (PPA) for each variable, averaged across all the competing models. Only those variables with a PPA ≥0.5 are included (Jefferys 1961). The averaged posterior means and standard deviations for each variable are also shown. Where the means and SDs are the same across the levels (1-4) only a single result is shown. ¯ 0.0065 -0.240.22 0.22 0.35 0.99 0.520.30 0.0065 0 0 0 minimum body condition score 0 0 0.0005 ¯ ¯ 0.0005 0 0 ¯ 0.0005 ¯ 0.27 excluded meanPOexcluded meanPOexcluded meanPO minimum body condition score NPO NPOSD SD SDNPOPO ¯ 0.94 0.019 ¯ -1.200.98 ¯ ¯ Braamfischerville maximum body condition score maximum body condition score excluded mean observed major clinical signs observed minor clinical signs sterilised mature adult 13-36 months 7-12 months -1.401.00 -0.23 0.68 0.42 0.64 0.96 -0.24 0.240.96 1.00 0.96 0.0360 0 0 0.99 0 0 0.0005 ¯ ¯ ¯ 0.22 0.21 0.440.70 ¯¯ 0.41 0.330.59 0.670.99 ¯ 0.0065 0 0 ¯ ¯¯¯ 0.26 ¯¯ ¯ ¯ ¯ 0.28 0.35 0 0.32 0.21 0.21 0.22 0.24 -1.30 -0.38 0 00.15 0 0 1.00 1.00 1.00 0 0 -1.50 -0.24 1.00 0 0 -1.30 -0.36 1.00 1.00 0.21 0.15 ¯ 0.27 ¯ 0.47 ¯ 0.0005 0.51 ¯ 0.98 ¯ pregnant: observed gestation 0.019 ¯ ¯¯ ¯¯ 0 ¯ 0 0.001 ¯ 0 pregnant: arbitrary gestation period ¯ 0.37 0.42 lactating: observed lactation period lactating: arbitrary lactation period ¯ 0.25 ¯ 0.25 ¯ 0.16 ¯ -0.69 ¯ 0.75 ¯ 0.75 0.036 0.036 0.058 ¯ ¯ ¯ ¯ -1.00 ¯ 0.94 ¯ (0.0095, 0.0095, 0.009, 0.0095) 0.48 0.28 ¯ 0 0.15 0.33 ¯ ¯ ¯ (0.39, 0.39, 0.39, 0.40) 0.35 ¯ 0.004 1.00 0.38 0.15 0.31 ¯ ¯ ¯ 0 ¯ 0.241.00 SD ¯ ¯ 0.004 ¯ ¯ -1.50 ¯ ¯ NPO ¯ ¯ ¯¯ ¯ ¯ 0.058 154 Appendix 2.29 Kelusa ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios; see Appendix 2.27 for interpretation of the results. The first three columns (PO, NPO and excluded) show the posterior probabilities of association (PPA) for each variable, averaged across all the competing models. Only those variables with a PPA ≥0.5 are included (Jefferys 1961). The averaged posterior means and standard deviations for each variable are also shown. Where the means and SDs are the same across the levels (1-4) only a single result is shown. 0.0005 ¯ ¯ 0 mean 0.24 ¯ 0 0.20 0 1.00 0.16 00.27 0.13 0.16 0.13 0.96 ¯ 0.24 ¯ (0.23, 0.22, 0.23, 0.24) b 0 ¯ ¯ 0.001 ¯ ¯ 0.001 0.23 ¯ 0 ¯ ¯ 0.40 ¯ -0.87 1.00 ¯ 1.00 ¯ -1.20 ¯ ¯ 1.00 ¯ mean pregnant: observed gestation pregnant: arbitrary gestation period lactating: observed lactation period lactating: arbitrary lactation period 1.00 0.36 ¯ -0.92 ¯ 0 ¯ 0 1.700 1.00 1.700 0 -0.511.00 0 -1.501.00 0 0 0 0 0 -0.48 -1.40 0.002 0.21 0 0.19 PO NPO 0 0.001 ¯ ¯ ¯¯ 0 0 0 0 -0.931.00 0 0 -1.20 -1.100 ¯ 6-12 months of age confined 1.00 1.00 1.00 observed major clinical signs observed minor clinical signs sterilised 1.00 1.00 1.00 -0.88 -0.50 0.13 0.16 1.00 1.00 1.00 0.0015 0 0 -1.40 1.00 1.00 1.00 1.00-0.83 0 0 0 0 0.17 0.27 0 0 0 0 0.23 0 0 1.00 -0.46 -0.51 0.19 0 -0.88 mature adult 0 0 0.21 1.200 Kelusa excludedPO 13-36 months of age POexcluded maximum body condition score excluded mean maximum body condition score excluded meanSDNPO minimum body condition score NPO minimum body condition score SD SD SD 0 0 0 0 0 0 ¯ 0 ¯ NPO 0 0 00 ¯ 0 ¯ 0 0.23 1.00 1.00 ¯ ¯ ¯ ¯ 0.24 ¯ 1.20 1.70 1.00 ¯ PO 0.17 (0.23, 0.22, 0.22, 0.22) b 0.22 0.13 ¯ 0.17 0.35 -0.85 0 ¯ 0.001 0.19 0.17 0.36 ¯ 0.22 0.23 0.27 0 0 0 1.40 0 -0.551.00 0.18 1.00 0 0 ¯ 0.001 0 0 1.00 1.00 1.00 -1.10 -0.56 -0.941.00 1.40 1.10 1.00 1.00 1.00 1.70 -0.51 -1.50 1.000.26 0.22 155 Appendix 2.30 Antiga ordinal regression model averaged posterior means and standard deviations for the log cumulative odds ratios; see Appendix 2.27 for interpretation of the results. The first three columns (PO, NPO and excluded) show the posterior probabilities of association (PPA) for each variable, averaged across all the competing models. Only those variables with a PPA ≥0.5 are included (Jefferys 1961). The averaged posterior means and standard deviations for each variable are also shown. Where the means and SDs are the same across the levels (1-4) only a single result is shown. excludedSD minimum body condition score 0 0.34 0.27 0.26 1.00 1.00 0.95 0.96 ¯ ¯ PO -0.85 -0.32 0 0 0.049 0.27 0.14 0.19 0.35 0.27 -0.86 -0.33 0.94 0.76 0 -1.401.00-1.40 -0.34 0 0 1.00 0.13 0.19 0.26 0 0.001 0 0 0 1.000 -0.90 -0.25 1.100.007 0.96 1.000.13 0.19 (0.34, 0.36, 0.31, 0.33) b 0.80 0 0.001 0.035 0 0 0 0 0.041 0.041 0 0.004 0.031 0 0.13 0.19 1.00 1.00 0.95 0.97 1.00 ¯ ¯ ¯ number of dogs per household observed major clinical signs observed minor clinical signs confined sterilised mature adult 13-36 months of age 6-12 months of age 1.00 1.00 1.00 pregnant: observed gestation pregnant: arbitrary gestation period lactating: observed lactation period lactating: arbitrary lactation period ¯ ¯ ¯ ¯ ¯ -0.67 -1.500 0 0.970.028 -0.88 -0.25 1.10 -0.29 1.00 0 1.000.010.990.025 0 0 0.96 1.00 0.81 -0.36 -0.31 0 0.04 0 0 0 0 0 (0.33, 0.33, 0.30, 0.31) b 0.27 1.00 0.26 0.24 ¯ ¯ ¯ ¯ ¯ ¯ PO Antiga maximum body condition scoremaximum body condition score POexcluded mean ¯ ¯ ¯¯ 0.15 -0.380.85 NPO 0 ¯ 0.001 ¯ NPO 0 ¯ 0 0 ¯ ¯ 0.23 0.07 0.26 SD ¯ ¯ excluded mean 0 0 -0.37 -1.50 -0.701.00 1.00 1.00 0.27 0.25 0.23 0.072 ¯ ¯ NPO 0.85 0.83 ¯ SD ¯ 0.25 0.24 0.077 0.49 0 0 0 0 0 0.15 0 0.0005 0.0005 0 0 0 ¯ NPO 1.00 excluded ¯ ¯ (0.29, 0.28, 0.29, 0.29) b meanPO 0.076 0.450.97 minimum body condition score ¯ ¯¯ 0.92 -0.410.076 0.92 ¯ 0.28 ¯ 0.0005 ¯ 0.001 SD ¯ 0.24 0.069 0.96 ¯ ¯ 0 -0.35 ¯ ¯ ¯ 0 1.00 -0.750 -0.331.00 -0.320 0.045 0.91 0 0.76 ¯ ¯ ¯ ¯ -0.72 ¯ ¯ ¯¯ 1.00 ¯ ¯ ¯ ¯ ¯ ¯ mean 156 Appendix 3 Appendix 3.1 Study methodology. order activity level process 1 introduction banjar ▪ introduce facilitators ▪ state purpose of the session - to understand their perspective ▪ reiterate no right or wrong answers 2 semi- structured discussion - general banjar ▪ feedback on the following questions: ▫ who has a dog? or, do you have a dog? ▫ how do you keep your dogs? (confined or unconfined) ▫ how do you identify an owned and unowned dog? ▫ what do you know about rabies? or, do you have any questions about rabies? 3 visualisation tool - body mapping groups ▪ each group randomly allocated one of the following: ▫ healthy male dog, healthy female dog, unhealthy male dogs, unhealthy female dog ▪ each group asked to "Draw what you think is a healthy / unhealthy male / female dog" ▪ affected parts of the body are marked, or characteristics listed banjar ▪ each group shares what they have drawn with the rest of the banjar ▪ after each group, the banjar is asked if anything is missing from the drawings ▪ finally, while referring to the healthy dog picture, ask if the dog is still healthy if it has: i) a runny nose, ii) watering eyes, iii) ribs showing, iv) limps, v) ticks, vi) bad skin 4 semi- structured discussion - food sources banjar ▪ ask what are the various things which affect a dog's health ▪ as soon as "food" is suggested, conversation is directed to sources ▪ four volunteers asked to draw each food source on a separate piece of paper as they come up in conversation 5 ranking - food sources banjar ▪ first in reference to the healthy dog (picture hanging on a flipchart board) identify which food sources are relevant and irrelevant; and then rank the relevant food sources in order of importance (food source pictures semi-permanently tacked onto the picture of the healthy dog so they can be re-arranged as necessary) ▪ repeat the process for unhealthy, owned and unowned dogs 6 conclusion banjar ▪ thank the banjar for their participation ▪ explain the (research) application of the information they provided ▪ ask if the banjar has any questions for the facilitators 157 Appendix 3.2 Approximate number of participants. village banjar total number of participants men women children Antiga Kaler (mixed) 110 95 a ~15 0 Ketug (mixed) 90 67 8 b ~15 (<14y) Kelod (mixed) 84 32 32 20 Kelusa Kelikikawan (mixed) 60 50 ~ 10 c – Roban (male) 35 33 2 0 Roban (female) 54 0 50 4 Ayah (male) 41 41 0 0 Ayah (female) 44 0 41 3 Peliatan (male) 12 12 0 0 Peliatan (female) 74 0 54 d 4 Triwangsa (male) 16 16 0 0 Yehtengah (male) 92 92 0 0 Yehtengah (female) 45 0 45 0 a includes 15 male youths; b includes 2-3 female youths; c including children; d includes ~8 male and 8 female young teenagers Note: A small number of participants came and went during each session. Most children were ≤10 years of age and did not participate in the PRA exercises. 158 Appendix 3.3 Health status in relation to body condition for Antiga. Kaler (mixed) Ketug (mixed) Kelod (mixed) group-level description of: healthy male dog fat fat fat healthy female dog fat strong, eats well not done a unhealthy male dog skinny not listed skinny, weak unhealthy female dog missing entry ribs showing not done a banjar-level response to the question: Is a dog still healthy if it looks like this (the healthy healthy unhealthy unhealthy dog picture) but has ribs showing ? b a low attendance due to Hindu ceremonies therefore the banjar was divided into only two groups; b see Table 3.1 for full details of this question 159 Appendix 3.4 Health status in relation to body condition for Kelusa. Kelikikawan (mixed) Roban (male) Roban (female) Ayah (male) Ayah (female) group-level description of: healthy male dog fat fat fat not listed fat healthy female dog fat fat fat, strong not listed fat, good appetite unhealthy male dog skinny skinny not listed skinny not listed unhealthy female dog not listed not listed skinny, no appetite skinny skinny banjar-level response to the question: Is a dog still healthy if it looks like this (the healthy unhealthy unhealthy a unhealthy unhealthy unhealthy dog picture) but has ribs showing ? b Peliatan (male) Peliatan (female) Triwangsa (male) Yehtengah (male) Yehtengah (female) group-level description of: healthy male dog fat not listed fat, good appetite fat not listed healthy female dog not listed not listed good appetite fat not listed unhealthy male dog not listed not listed not listed not listed not listed unhealthy female dog not listed no appetite not listed skinny skinny banjar-level response to the question: Is a dog still healthy if it looks like this (the healthy unhealthy unhealthy unhealthy unhealthy unhealthy dog picture) but has ribs showing ? b a the majority of the banjar said unhealthy, however two banjar members disagreed and said healthy; b refer Appendix 3.1 for full details of this question 160 Appendix 3.5 Rank of food sources of healthy dogs in order of importance (1 = most important). a fresh offerings; b equal with food prepared by owner; c no women's community group in Banjar Triwangsa food prepared by owner for the dog 1 1 2 2 1 1 2 2 2 2 2 1 1 leftovers from the owner 4 1 2 b 2 3 3 3 3 3 3 3 3 food purchased from pet shop by owner 1 2 1 1 1 1 1 2 2 leftovers from the neighbour 3 4 4 5 stealing food from the neighbour 3 school children 9 offerings 2 2 a 6 5 4 4 4 rubbish 5 spoiled food in the ditch 8 school rubbish 5 4 5 waste from the local stalls 3 8 5 6 4 4 pig food from the sty 6 6 4 stealing food from the local stalls 6 stealing poultry 3 5 6 5 dead animals 7 7 fish from the river 7 cassava in dry fields 7 Yehtengah (male) Triwangsa c (male) Peliatan (female) Peliatan (male) Ayah (female) Antiga Kelusa Kaler (mixed) Ketug (mixed) Kelod (mixed) Kelikikawan (mixed) Roban (male) Roban (female) Ayah (male) Yehtengah (female) 161 Appendix 3.6 Rank of food sources of owned dogs in order of importance (1 = most important). food prepared by owner for the dog 4 1 4 2 1 1 2 1 2 1 1 1 leftovers from the owner 1 1 3 1 3 3 3 5 3 3 2 4 food purchased from pet shop by owner 1 2 2 1 2 1 2 4 2 leftovers from the neighbour 6 7 7 stealing food from the neighbour 3 school children 4 offerings 3 6 5 4 4 rubbish 10 school rubbish 2 4 8 waste from the local stalls 5 2 6 3 3 3 pig food from the sty 2 2 9 6 stealing food from the local stalls 4 stealing poultry 8 5 5 dead animals 3 3 5 9 fish from the river 5 vermin 6 faeces from livestock and children 5 Antiga Kelusa Kaler (mixed) Ketug (mixed) Kelod (mixed) Kelikikawan (mixed) Roban (male) Roban (female) Yehtengah (female) Ayah (male) Ayah (female) Peliatan (male) Peliatan (female) Triwangsa (male) Yehtengah (male) 162 Appendix 3.7 Rank of food sources of unhealthy dogs in order of importance (1 = most important). a opinion split 50: 50 as to whether this is a source; b equal with livestock faeces; c equal with fish from river leftovers from the owner 6 6 leftovers from the neighbour 4 a stealing food from the neighbour 5 offerings 6 4 4 4 5 rubbish 3 2 4 1 1 3 3 3 3 2 3 spoiled food in the ditch 1 2 2 school rubbish 2 5 4 waste from the local stalls 4 2 7 waste food from someone other than the owner or neighbour 7 pig food from the sty 3 3 3 3 5 2 6 4 stealing poultry 5 dead animals 2 5 1 2 4 1 1 2 1 3 1 fish from the river 1 vermin 1 3 faeces 1 c 3 2 livestock faeces 1 2 1 2 1 4 children's faeces 1 b 1 b 4 b grass 4 Antiga Kelusa Kaler (mixed) Ketug (mixed) Kelod (mixed) Kelikikawan (mixed) Roban (male) Roban (female) Yehtengah (female) Ayah (male) Ayah (female) Peliatan (male) Peliatan (female) Triwangsa (male) Yehtengah (male) 163 Appendix 3.8 Rank of food sources of unowned dogs in order of importance (1 = most important). a equal with livestock faeces stealing leftovers from householders 6 1 offerings 3 6 3 5 7 3 5 2 4 4 7 2 rubbish 2 1 3 2 1 5 6 3 5 2 spoiled food in the ditch 4 5 4 1 school rubbish 3 4 2 7 7 5 waste from the local stalls 5 2 5 1 6 1 3 waste food from someone other than the owner or neighbour 6 7 7 5 pig food from the sty 1 2 8 9 3 4 3 2 3 5 3 6 stealing food from the local stalls 7 6 stealing poultry 4 6 3 dead animals 5 4 6 2 1 1 2 4 1 4 fish from the river 8 vermin 3 faeces 1 1 4 livestock faeces 2 4 1 2 4 5 children's faeces 2 a 5 a grass 6 cassava in dry fields 9 Antiga Kelusa Kaler (mixed) Ketug (mixed) Kelod (mixed) Kelikikawan (mixed) Roban (male) Roban (female) Yehtengah (female) Ayah (male) Ayah (female) Peliatan (male) Peliatan (female) Triwangsa (male) Yehtengah (male) 164 Appendix 4 Appendix 4.1 Summary of the study methodology. a upper outliers were those dogs in Zenzele (n=7) with histories and post-vaccinal titres suggestive of vaccination with Rabisin by the DoA through a vaccination point in Zenzele in May 2006 or undertaken independently by the owner b upper outliers were those dogs in Kelusa and Antiga (n=4 and n=15 respectively) with histories and post-vaccination titres suggestive of vaccination undertaken independently by the owner or as part of vaccination campaigns outside of Kelusa and Antiga longitudinal data analysis all data summarised models were fitted with and without upper outliers a all data summarised models were fitted to titres from vaccinated dogs models were fitted with and without upper outliers b all data summarised titres excluded from models all data summarised titres excluded from models Rabisin door-to-door for 16 days every available dog in January 2010 not vaccinated by the DoL in December 2009 study area location cohort vaccine used vaccination strategy blood sampling strategy research Rabisin door-to-door for 10 days every available dog in February 2010 not vaccinated by the DoA in October 2009 every available dog vaccinated in February 2010; samples collected day 0 (immediately prior to vaccination) and then 30, 90, 180 and 360 days after vaccination Zenzele informal settlement in Gauteng Province, South Africa Department of Agriculture (DoA) every available dog vaccinated in October 2009; samples collected 8-10, 30, 90, 180 and 360 days after vaccination Rabisin Galaxy DA2PPv ivermectin one day vaccination point (VP) October 2009 coastal village in Bali Province, IndonesiaAntiga all data summarised titres excluded from models every available dog vaccinated and unvaccinated; samples collected 180 and 360 days after vaccination one day vaccination point (VP) in two banjars in the study area December 2009 Rabivet Supra 92 Department of Livestock (DoL) inland village in Bali Province, Indonesia Kelusa research Rabisin door-to-door for 12 days every available dog in January 2010 every available dog vaccinated and unvaccinated; samples collected 180 and 360 days after vaccination one day vaccination point (VP) in one banjar outside the study area February 2010 Rabisin Department of Livestock (DoL) all data summarised models were fitted to titres from vaccinated dogs models were fitted with and without upper outliers b research 165 Appendix 4.2 Summary of vaccination coverage. a there were an additional four households with at least one dog; three of these households owned mostly greyhounds that were confined to the yard b in Kelusa 146 dogs and in Antiga 124 dogs were caught for blood sampling at both time points (including vaccinated and unvaccinated dogs) c there was one household where the details of the number of dogs owned was not available d see Appendix 4.24 Note: puppies born in the study area within ~12 weeks prior to vaccination/sampling but not observed during the vaccination/sampling period were assumed to have been lost prior to the commencement of vaccination/sampling owner awareness of and access to vaccination point consent declined: 3 dogs consent declined: 7+ dogs c nobody home to obtain consent: 10 dogs unable to properly restrain: 11 dogs missed: 25 dogs % sampled day 180 b _ _ 75.2% (252/335) 76.8% (205/267) % sampled & vaccinated day 360 b _ _ 81.4% (250/307) 86.0% (202/235) vaccinated by the DoL: 16 dogs d missed: 16 dogs could not be caught: 45 dogs missed: 19 dogs could not be caught: 29 dogs consent declined for sampling: 13 dogs consent declined for sampling: 14 dogs consent declined for sampling: 13 dogs nobody home for consent: 21 dogs nobody home for consent: 5 dogs declined consent for vaccination day 360: 1 dog declined consent for vaccination day 360: 1 dog unable to properly restrain: 12 dogs unable to properly restrain: 7 dogs number consent declined or unable to restrain during the study period consent declined for sampling: 3 dogs reasons for not vaccinating day 0 consent declined: 10 dogs method vaccine delivery door-to-door 82.2% (259/315) 85.6% (332/388) incl. 73 dogs vaccinated by the DoA still present Feb-10 a vaccination coverage day 0 Antiga research cohort vaccination point (VP) door-to-door door-to-door 26.9% (97/361) 80.9% (284/351) 79.2% (259/327) Zenzele research cohort Zenzele DoA cohort Kelusa research cohort 166 Appendix 4.3 The number of dogs in the research cohorts and the number of unvaccinated controls in Bali that were blood sampled at each time point. day 0 a day 30 day 90 day 180 day 360 Zenzele vaccinated dogs 190 183 148 134 103 Kelusa vaccinated dogs _ _ _ 168 124 Kelusa unvaccinated dogs _ _ _ 70 79 Antiga vaccinated dogs _ _ _ 163 126 Antiga unvaccinated dogs _ _ _ 35 49 a day 0 immediately prior to vaccination for the research cohort Note 1: of the 97 dogs in Zenzele that were vaccinated by the DoA in October 2009, 60 were blood sampled 8- 10 days after vaccination, and 77 were sampled 30 days, 65 were sampled 90 days, 64 were sampled 180 days, and 47 dogs were sampled 360 days after vaccination Note 2: in addition to the dogs recorded in Appendix 4.3 - Kelusa day 180: 11 dogs vaccinated by the DoL Dec-09 sampled, 1 dog taken by the owner to an NGO for vaccination sampled, 1 dog sampled but its vaccination history was uncertain, 1 missing sample; there were an additional 32 puppies <6-8 weeks of age present in the study area during the sampling period - Kelusa day 360: 12 dogs vaccinated by the DoL Dec-09 sampled, 1 dog taken by the owner to an NGO for vaccination sampled, 1 dog sampled but its vaccination history was uncertain, 33 dogs vaccinated only (mostly puppies <6-8 weeks of age) - Antiga day 180: 3 dogs vaccinated by the DoL Apr-10 sampled, 1 dog taken by the owner to an NGO for vaccination sampled, 1 dog sampled but its vaccination history was unknown, 2 missing samples; there were an additional 19 puppies <6-8 weeks of age present in the study area during the sampling period - Antiga day 360: 3 dogs vaccinated by the DoL Apr-10 sampled, 9 dogs in Ketug inadvertently vaccinated by the NGO involved in the island wide vaccination campaign sampled ~30 days after vaccinated (titres ranged from 2-181 IU/ml), 2 dogs taken by the owner to an NGO for vaccination sampled, 13 dogs vaccinated only (mostly puppies <6-8 weeks of age) 167 Appendix 4.4(i) Sample selection and sampling technique for deep skin scrapes. Deep skin scrapes (DSS) were taken on day 180 in Bali to investigate generalised demodicosis. Having to catch affected dogs by net precluded true random selection, therefore a convenience sample of 15 dogs in Kelusa and 17 in Antiga were selected from the pool of 168 dogs diagnosed with generalised dermatitis during the preceding survey. To estimate the true prevalence of demodicosis in free-roaming dogs with dermatitis a sample size of approximately 90 dogs was required (AusVet EpiTools) [assuming an expected prevalence of 12% (Nayak et al. 1997; Rodriguez-Vivas et al. 2003), precision of 5%, and sensitivity of 90% and specificity of 99% based on operator experience and type of test (Fondati et al. 2009)]. Practical constraints limited testing to approximately 30 dogs with generalised dermatitis as a pilot study, considered sufficient to gauge the involvement of Demodex spp. Lesions were arbitrarily classified as generalised according to the general literature on demodicosis (Mueller 2004; Miller, Griffin & Campbell 2013). Dogs with generalised dermatitis had 12 or more localised lesions, and/or contiguous lesions that may have included complete involvement of an entire body region (e.g. face, flank), and/or complete involvement of two or more feet. Multiple deep skin scrapes, approximately 2x2cm, were taken from each affected dog according to standard protocols (Miller, Griffin & Campbell 2013). The cellular debris was placed into plain, coded 3ml blood tube and examined within 12 hours of sampling using a standard light microscope (from 4-40x magnification). Prior to microscopy, the samples were soaked in 10% potassium hydroxide for approximately 20 minutes to dissolve the cellular debris. Appendix 4.4(ii) Sample selection for faecal analysis (see Appendix 4.25). A prevalence of ~90% was expected for intestinal parasites in Zenzele, particularly Anycylostoma spp. (Minnaar, Krecke & Rajput 1999; Minnaar & Krecke 2001). In order to estimate true prevalence with a precision of 5% and test sensitivity and specificity of at least 95%, a sample of approximately 120 dogs was required. This estimate happened to coincide with the number of dogs (n=107) randomly selected to assess intestinal parasitism as a pilot study. One dog was positive for Hymenolepis nana, a tapeworm found in humans and rodents, which was probably acquired from eating a rodent (Palmer et al. 2011). 168 Appendix 4.5 Age-specific life expectancies for the entire (ecological study) research population derived from the observed ages at the last time point of the study period (see Appendix 2.5). current age (month of life) age-specific life expectancy (years) male female total Zenzele 2-12 2.8 3.0 2.9 13-24 3.1 2.3 2.7 25-36 3.5 4.1 3.7 37-48 4.8 3.6 4.3 49+ Kelusa 2-12 4.7 2.7 4.0 13-24 3.7 1.5 2.9 25-36 3.8 5.0 3.9 37+ Antiga 2-12 7.8 6.7 7.5 13-24 4.5 4.8 4.6 25-36 5.5 3.0 4.8 37+ 169 Appendix 4.6 Characteristics of the dogs in Bali (in January 2010) that were not vaccinated on day 0; the number that could not be caught for vaccination is shown in brackets. age class a (month of life) Kelusa b Antiga c male female male female 1-6 4 (0) 4 (0) 17 (6) 0 (0) 25 7-12 5 (2) 1 (0) 1 (1) 1 (1) 8 13-36 5 (4) 6 (5) 4 (4) 5 (4) 20 mature adult 18 (15) 5 (3) 26 (24) 7 (5) 56 32 16 48 13 109 a dogs in age class 1-6 months are between birth [start of their 1st month of life] and ~26 weeks of age [end of their 6th month of life], and so on; mature adults are mostly dogs first observed as adults at the start of the (larger ecological) study period (i.e. from March 2008), but also dogs first observed as adults after the start of this study period, therefore the exact age of these dogs could not be determined by direct observation [Note: dogs in age classes 1-6, 7-12 and 13-36 months were observed as pups or juveniles from March 2008, so their true age was known] b in addition, consent for vaccination was declined for 3 dogs including 1 male in its 20th month of life, 1 mature adult female, and 1 dog of unknown gender and age c in addition, consent for vaccination was declined for at least 7 dogs including 5 mature adult males, 1 male in its 13th month of life, and 1 dog of unknown gender and age 170 Appendix 4.7 Description of the dogs in Zenzele with baseline (day 0) titres ≥0.5 IU/ml. dog day 0 IU/ml day 30 IU/ml present May 2006 gender age (month of life) 1 1.00 45.25 yes m 36+ a 2 0.71 16.00 no m 21 3 0.50 8.00 yes m 36+ 4 0.71 1.41 no f 6 5 0.50 22.63 yes m 36+ a ≥36th month of life in February 2010 (at vaccination) 171 Appendix 4.8 Summary of titres from the dogs in Zenzele necessary to assess the inadvertent inclusion of dogs vaccinated by the Department of Agriculture in October 2009 in the research cohort. research cohort 0.06 79 0.09 40 0.13 15 0.18 25 0.25 16 0.35 10 0.50 2 0.71 2 1.00 1 day 0 IU/ml frequency research cohort upper outliers 1 0.09 181.02 2 0.25 128.00 3 0.09 128.00 4 0.09 256.00 5 0.09 128.00 6 0.09 362.04 7 0.06 256.00 dog day 0 IU/ml day 30 IU/ml day 90 titres ≤ 1 IU/ml of dogs vaccinated by the DoA 1 16.00 1.00 2 8.00 1.00 3 8.00 1.00 4 64.00 0.71 5 0.09 0.18 6 0.09 0.09 7 0.09 0.09 8 45.25 0.50 9 0.06 0.06 10 16.00 0.71 11 0.50 0.18 12 64.00 0.35 13 16.00 1.00 dog 30 day IU/ml 90 day IU/ml dogs vaccinated by the DoA 0.06 2 0.09 3 0.18 1 0.50 1 2.83 1 4.00 1 5.66 1 8.00 5 11.31 3 16.00 7 22.63 7 32.00 8 45.25 8 64.00 12 90.51 3 128.00 7 181.02 3 256.00 2 362.04 2 day 30 IU/ml frequency dogs vaccinated by the DoA with day 30 titres ≥128 IU/ml 1 128.00 8.00 2 362.04 1.41 3 181.02 5.66 4 128.00 ns 5 362.04 11.31 6 128.00 5.66 7 128.00 8.00 8 181.02 22.63 9 256.00 128.00 10 128.00 ns 11 128.00 16.00 12 128.00 ns 13 256.00 2.83 14 181.02 32.00 ns = no sample dog day 30 IU/ml day 90 IU/ml 172 Appendix 4.9 Summary of the titres of the dogs in the Zenzele research cohort present May 2006 and those that arrived into the population after May 2006 (excluding 8 dogs whose presence in May 2006 was unknown; including upper outliers). a 2 dogs with titres of 0.5 IU/ml and 1 dog with a titre of 1 IU/ml; b 2 dogs with titres of 0.71 IU/ml. The Mann-Whitney test was used to compare the mean titres between dogs present in Zenzele in May 2006 and those that arrived into the population after May 2006 (see Methods and materials section 4.2.4.2) mode IU/ml range IU/ml no. with titres ≥ 0.5 IU/ml Mann-Whitney test result mode IU/ml range IU/ml no. with titres < 0.5 IU/ml no. with titres ≤ 0.1 IU/ml day 0 observed GMT IU/ml present May 2006 0.06 0.06 - 1.00 3 (5.1%) a p=0.08 16.00 & 22.63 0.09 - 362.04 2 (3.7%) 1 (1.9%) arrived after May 2006 0.06 0.06 - 0.71 2 (1.6%) b 22.63 0.09 - 256.00 3 (2.5%) 3 (2.5%) 0.12 0.10 16.38 12.22 day 30 day 360 observed GMT IU/ml observed GMT IU/ml no. with titres < 0.5IU no. upper outliers Mann-Whitney test result mode IU/ml range IU/ml 2 (5.56%) p=0.15 6 (9.4%) no. with titres ≤ 0.1IU Mann-Whitney test result 12 (33.3%) 1.00 0.06-5.66 26 (40.6%) 4 p=0.19 (excluding upper outliers p=0.21) 1.00 0.09 - 16.000.53 0.771 173 Appendix 4.10 Characteristics of the unvaccinated controls in Bali with titres ≥0.5IU/ml (n=15). site and time point blood sampled male female Kelusa day 180 4 3 Kelusa day 360 4 1 Antiga day 180 2 a 0 Antiga day 360 2 a 0 a one dogs was sampled at both time points age class (month of life) a site and time point blood sampled 1-6 months 7-12 months 13-36 months mature adult Kelusa day 180 4 1 0 2 Kelusa day 360 1 0 1 3 Antiga day 180 0 1 0 1 b Antiga day 360 0 0 1 1 b a see Appendix 4.6 for the definition; b one dog was sampled at both time points titres IU/ml Kelusa Antiga 0.50 6 1 a 0.71 3 1 1.00 1 2 a 1.41 1 0 2.83 1 0 a one dog was sampled at both time points [day 180 titre = 1 IU/ml; day 360 titre = 0.5 IU/ml] 174 Appendix 4.11 Summary of the titres of the unvaccinated controls in Bali. day 180 day 360 observed GMT IU/ml mode IU/ml range IU/ml Mann-Whitney test result observed GMT IU/ml mode IU/ml range IU/ml Mann-Whitney test result Kelusa 0.10 0.06 0.04 - 1.41 p=0.13 0.09 0.06 0.04 - 2.83 a p<0.001 excluding Kelusa 2.83IU p<0.001 Antiga 0.09 0.04 0.04 - 1.00 0.06 0.04 0.04 - 0.71 a 1 dog with a titre 2.83 IU/ml was acquired as a pup from outside the survey area (Ubud) and may have been vaccinated in Ubud; the Mann-Whitney test p-value remains <0.001 after removing this dog from the analysis. The Mann-Whitney test was used to compare the mean titres between unvaccinated dogs in Kelusa and Antiga for the same time points (see Methods and materials section 4.2.4.2) 175 Appendix 4.12 Characteristics (at vaccination) of the dogs in the Zenzele research and DoA cohort with peak (day 30) titres <0.5 IU/ml. id day 30 IU/ml gender age (months) present May-06 minimum body condition score maximum body condition score clinical signs pregnancy or lactation research cohort 1 0.09 m 8 no 4 4 yes NA 2 0.13 m 36+ a yes me me no NA 3 0.09 f 15 no 2 3 me lac 4 0.09 f 21 no 2 5 yes lac 5 0.09 m 36+ yes 4 4 no NA 6 me c m 36+ yes me me me NA DoA cohort 1 0.09 m 32+ b yes 3 3 no NA 2 0.09 m 11 no 5 5 no NA 3 0.09 f 32+ yes 3 4 no preg 4 0.06 m 10 no 3 3 no NA 5 0.06 f 3 no 4 4 me NA 6 0.18 m 6 no me me me NA a ≥36th month of life in February 2010 (at vaccination); b ≥32nd month of life in October 2009 (at vaccination) c peak titre missing but titres <0.5IU/ml at all other time points; key: me = missing entry, NA = not applicable 176 Appendix 4.13 Summary of the day 180 and 360 titres in the research cohorts [vaccinated dogs] (excluding upper outliers). a 16 dogs had titres of 1 IU/ml, 16 dogs had titres of 1.41 IU/ml, and 16 dogs had titres of 2 IU/ml. The Mann-Whitney test was used to compare the mean titres between vaccinated dogs in Zenzele, Kelusa and Antiga for the same time points (see Methods and materials section 4.2.4.2) 0.04 - 11.31 25 (21.6%) 3 (2.6%)Antiga 1.03 0.71 0.04 - 8.00 28 (18.9%) observed GMT IU/ml mode IU/ml range IU/ml no. with titres < 0.5 IU/ml no. with titres ≤ 0.1 IU/ml Mann-Whitney test result Mann-Whitney test result __ 1.03 1.41 a observed GMT IU/ml mode IU/ml range IU/ml 1.08 0.71 0.04 - 8.00 37 (22.6%) 8 (4.9%) 8 (8.1%) no. with titres < 0.5 IU/ml no. with titres ≤ 0.1 IU/ml day 180 4 (2.7%) Kel: Ant p=0.11Kel: Ant p=0.48 0.77 1.00 0.04 - 8.00 34 (27.4%) 8 (6.6%) 0.81 1.00 0.06 - 11.31 32 (24.8%) 4 (3.1%) Zen: Kel p=0.015 Zen: Ant p=0.052 Zen: Kel p=0.025 Zen: Ant p=<0.001 0.59 1.00 0.06 - 16.00 39 (39.4%) day 360 Kelusa Zenzele 177 Appendix 4.14 Characteristics (at vaccination) of the dogs in the research cohorts with day 360 titres ≤0.1 IU/ml. dog day 30 IU/ml day 180 IU/ml present May-06 gender age (month of life) minimum body condition score maximum body condition score clinical signs generalised dermatitis pregnancy or lactation Zenzele research cohort (8.1%) 1 0.09 0.18 No M 8 4 4 yes NA NA 2 2.83 0.18 No f 5 4 5 no NA NA 3 5.66 0.06 No m 2 me me me NA NA 4 2.83 0.13 yes m 36+ a 4 5 yes NA NA 5 4.00 me No f 7 me me me NA NA 6 1.41 0.25 No m 23 4 4 no NA NA 7 0.09 0.13 no f 21 2 5 yes NA lac 8 0.09 0.09 yes m 36+ 4 4 no NA NA Zenzele DoA cohort (6.4%) 1 2.83 0.18 no f 32+ a me me me NA no 2 me me no m 32+ me me me NA NA 3 0.06 0.25 no f 3 4 4 me NA NA Kelusa research cohort (6.6%) 1 NA 0.35 NA m 1 me me me me NA 2 NA 0.25 NA m 34+ a 4 4 yes yes NA 3 NA 0.09 NA m 3 me me me me NA 4 NA 0.06 NA f 10 me me me me no 5 NA 0.18 NA f 34+ me me me me no 6 NA 0.06 NA m 34+ 5 5 yes yes NA 7 NA 0.25 NA m 9 4 4 me me NA 8 NA me NA m 25-33 b 2 4 no no NA Antiga research cohort (2.6%) 1 NA me NA m 13-21 b me me me me NA 2 NA me NA m 33+ a me me me me NA 3 NA 0.25 NA m 33+ me me me me NA a ≥36th month of life in February 2010 in Zenzele for the research cohort, 32nd month of life in Zenzele for the DoA cohort, 34th month in January 2010 in Kelusa, and 33rd month in January 2010 in Antiga (at vaccination); b month of life is in this range; key: me = missing entry, NA = not applicable 178 Appendix 4.15 Correlation coefficients for the dogs in the Zenzele research cohort that were blood sampled at every time point (i.e. day 30, 90, 180 and 360) (see Figure 4.1). correlation coefficient (r) time points compared including dogs with day 30 titres <0.5 IU/ml and excluding upper outliers including dogs with day 30 titres <0.5 IU/ml and including upper outliers IU log IU IU log IU day 30 / 90 0.50 0.71 0.74 0.74 day 30 / 360 0.40 0.55 0.29 0.57 day 30 / mean 90-360 0.52 0.72 0.71 0.74 day 30 / mean 180-360 0.44 0.61 0.53 0.64 day 90 / 180 0.56 0.72 0.79 0.74 day 90 / 360 0.52 0.62 0.40 0.64 day 180 / 360 0.68 0.78 0.59 0.79 179 Appendix 4.16 Models restricted to natural log of the titre as the response variable and time as the covariate for the research cohorts. Natural logs are shown in the tables. Model 1: Zenzele quadratic model fitted to four time points (day 30, 90, 180 and 360) with the intercept adjusted to day 30. Model 2: Zenzele linear model fitted to three time points (day 90, 180 and 360) with the intercept adjusted to day 90. Model 3: Zenzele linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 180. <0.001 std dev of dog-level random effect intercept 0.9661 0.8400 1.1112 – <0.001 time (Δ per day) -0.0297 -0.0319 -0.0274 <0.001 std dev of dog-level random effect intercept 1.0029 0.8785 1.1449 – excluded 547 1666.7 intercept 2.4957 time (Δ per day) -0.0303 -0.0326 -0.0281 2.3117 2.6796 time^2 (Δ per day) 0.000063 0.000057 0.000070 <0.001 time^2 (Δ per day) 0.000064 0.000058 0.000071 <0.001 97.5% CI p-value included 568 1749.9 intercept 2.5917 2.4064 2.7769 <0.001 outliers observations (n=) AIC parameter estimate 2.5% CI time (Δ per day) -0.0033 -0.0039 -0.0026 <0.001excluded 371 1013.3 intercept 0.2542 0.0745 std dev of dog-level random effect intercept 0.9588 0.8351 1.1009 – -0.0040 -0.0027 <0.001 std dev of dog-level random effect intercept 0.9747 0.8507 1.1167 – 0.4339 0.006 97.5% CI p-value included 385 1058.2 intercept 0.3016 0.1224 0.4807 0.001 outliers observations (n=) AIC parameter estimate 2.5% CI time (Δ per day) -0.0034 excluded 228 610.2 intercept -0.2120 -0.3979 -0.0261 0.026 time (Δ per day) -0.0018 -0.0026 -0.0010 <0.001 std dev of dog-level random effect intercept 0.9493 0.8234 1.0945 – time (Δ per day) -0.0018 -0.0026 -0.0010 <0.001included 237 635.7 intercept -0.1819 -0.3658 std dev of dog-level random effect intercept 0.9611 0.8359 1.1050 – outliers observations (n=) AIC parameter estimate 2.5% CI 97.5% CI p-value 0.0021 0.054 180 Appendix 4.16 continued Model 4: Kelusa linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 180. Model 5: Antiga linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 180. Note: in addition to excluding the 15 dogs with day 180 titres ≥11.3 IU/ml, excluding the 3 dogs with day 360 titres of 11.3 IU/ml results in observations = 258, intercept = 0.0907 p = 0.518, time = -0.0006 p = 0.213 Model 6: Kelusa and Antiga linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 180. a confidence intervals could not be obtained on the variance-covariance components using the intervals function with lme{nlme}; therefore, estimates of the confidence intervals were derived using the lme4 package with lmer, profile and confint functions 1.3050 – 97.5% CI p-value included 292 870.6 intercept 0.1464 -0.0443 0.3372 0.133 outliers observations (n=) AIC parameter estimate 2.5% CI time (Δ per day) -0.0030 -0.0038 -0.0022 <0.001 std dev of dog-level random effect intercept 1.1544 1.0212 1.2655 – excluded 285 838.7 intercept 0.0918 -0.0960 0.2795 0.338 time (Δ per day) -0.0029 -0.0038 -0.0021 <0.001 std dev of dog-level random effect intercept 1.1158 0.9839 1.0608 0.9176 1.2263 – excluded 264 765.2 intercept 0.0208 -0.1568 std dev of dog-level random effect intercept 0.8917 0.7579 1.0490 – 0.1983 0.818 time (Δ per day) -0.0005 -0.0015 0.0005 0.356 97.5% CI p-value included 289 908.1 intercept 0.2687 0.0741 0.4634 0.007 outliers observations (n=) AIC parameter estimate 2.5% CI time (Δ per day) -0.0012 -0.0022 -0.0001 0.030 std dev of dog-level random effect intercept outliers observations (n=) AIC parameter estimate 2.5% CI 97.5% CI p-value included 581 1788.8 intercept 0.2071 0.0295 0.3847 0.023 time (Δ per day) -0.0021 -0.0027 -0.0014 <0.001 std dev of site/dog-level random effect intercept 1.1060 1.0063 excluded a 549 1616.6 intercept 0.0582 -0.0973 0.2168 0.378 time (Δ per day) -0.0017 -0.0024 -0.0010 <0.001 std dev of site/dog-level random effect intercept 1.0054 0.9074 1.1106 – std dev of site-level random effect intercept 0.0820 0.0091 0.7410 – std dev of site-level random effect intercept 0.00005 0.0000 0.2440 – 1.2157 – 181 Appendix 4.17 Observed and predicted geometric mean titres for each time point. Predictions are derived from the models restricted to natural log of the titre as the response variable and time as the covariate (see Appendix 4.16). median (range) including upper outliers 22.63 (0.09 - 362.04) 1.41 (0.06 - 45.25) 1.00 (0.06 - 16.00) 0.71 (0.06 - 16.00) observed GMT including upper outliers 16.32 1.53 0.84 0.61 median (range) excluding upper outliers 22.63 (0.09 - 90.51) 1.41 (0.06 - 16.00) 1.00 (0.06 - 11.31) 0.71 (0.06 - 16.00) observed GMT excluding upper outliers 14.80 1.45 0.81 0.59 Appendix 4.16, model 1 predicted GMT (95CIs) including upper outliers 13.36 (11.09 - 16.07) 2.73 (2.31 - 3.24) 0.60 (0.49 - 0.74) 0.67 (0.54 - 0.85) predicted GMT (95CIs) excluding upper outliers 12.13 (10.09 - 14.58) 2.57 (2.17 - 3.03) 0.59 (0.48 - 0.71) 0.65 (0.52 - 0.82) Appendix 4.16, model 2 predicted GMT (95CIs) including upper outliers – 1.35 (1.13 - 1.62) 1.00 (0.85 - 1.18) 0.54 (0.44 -0.66) predicted GMT (95CIs) excluding upper outliers – 1.29 (1.08 - 1.54) 0.96 (0.81 - 1.13) 0.53 (0.43 - 0.65) Appendix 4.16, model 3 predicted GMT (95CIs) including upper outliers – – 0.83 (0.69 - 1.00) 0.60 (0.50 - 0.73) predicted GMT (95CIs) excluding upper outliers – – 0.81 (0.67 - 0.97) 0.59 (0.48 - 0.71) median (range) including upper outliers – – 1.41 (0.04 - 16.00) 1.00 (0.04 - 11.31) observed GMT including upper outliers – – 1.16 0.80 median (range) excluding upper outliers – – 1.41 (0.04 - 8.00) 1.00 (0.04 - 8.00) observed GMT exlcuding upper outliers – – 1.08 0.77 Appendix 4.16, model 4 predicted GMT (95CIs) including upper outliers – – 1.16 (0.96 - 1.40) 0.67 (0.55 - 0.83) predicted GMT (95CIs) excluding upper outliers – – 1.10 (0.91 - 1.32) 0.64 (0.53 - 0.79) median (range) incl upper outliers – – 1.41 (0.04 - 45.25) 1.00 (0.04 - 16.00) observed GMT incl upper outliers – – 1.33 1.17 median (range) excl upper outliers – – 1.00 (0.04 - 8.00) 1.00 (0.04 - 11.31) observed GMT excl upper outliers – – 1.03 1.03 Appendix 4.16, model 5 predicted GMT (95CIs) incl upper outliers – – 1.31 (1.08 - 1.59) 1.06 (0.86 - 1.31) predicted GMT (95CIs) excl upper outliers – – 1.02 (0.85 - 1.22) 0.93 (0.77 - 1.14) day 360 IU/ml Zenzele Kelusa Antiga day 30 IU/ml day 90 IU/ml day 180 IU/ml 182 Appendix 4.18 Description of the covariates in the models described in Methods and materials section 4.2.4.2 (see Appendices 4.19-4.22). The covariates were evaluated by direct observation, by the primary researcher and the enumerators, and by owner questionnaire. The questionnaire was developed and tested in February 2008 through community focus groups, including modified participatory rural appraisal (PRA) techniques (Chambers 1994a; Chambers 2007; Kumar 2007), and subsequent pilot studies. In Bali the questionnaires were bilingual (English and Bahasa with the Bahasa back-translated). Several languages were spoken in Johannesburg, including English, therefore the questions were written in English and the accuracy of the various translations checked regularly with the multi-lingual enumerators throughout the study period. The respondent was the person/s in the household that the householders collectively identified as most knowledge about the dog, which was not necessarily the owner. Respondents under 16 years of age were always interviewed with an adult present. Data collection was standardised through detailed enumerator training at the start of the study period for each enumerator and repeated on the first day of each survey. Surveys were undertaken every 6-12 weeks from March 2008 until April 2011. Each survey included the direct observations, by the primary investigator and enumerators, and the owner questionnaires. Apart from the primary researcher, all enumerators were local residents. All clinical examinations were undertaken by a qualified veterinarian. The age of the dogs was reported by owners and/or visually assessed by the enumerators, including from the dentition in dogs in puppies and juveniles (i.e. dogs ≤12th month of life [≤ ~52 weeks of age] (Dyce, Sack & Wensing 1987) in Zenzele. The exact age (month of life) at vaccination was known for most dogs ≤ their 35th month in Zenzele, ≤33rd month in Kelusa, and ≤32nd month in Antiga (i.e. for dogs acquired from March 2008 as puppies or juveniles). Body condition includes the minimum and maximum score recorded during the surveys immediately prior to and following vaccination (i.e. within 6 weeks of vaccination). Body condition was evaluated using a standard 9-point (emaciated score 1 to obsess score 9) scoring system validated in adults (Laflamme 1997) and modified to assess body condition score (BCS) without palpation. The modified system was validated using dual energy x-ray absorptiometry in 71 dogs, including a small number of growing dogs (German & Holden 2006; German et al. 2006). Each dog was body condition scored independently by two enumerators during each survey. The minimum and maximum scores were generally the same or one score point different (e.g. 3 and 3, or 3 and 4 respectively). Clinical signs were 183 Appendix 4.18 continued observed by the primary researcher and enumerators at the time of vaccination or during the survey immediately prior to vaccination. As part of the owner questionnaire, owners reported clinical signs they had observed during the previous 7 days and 3 months based on a set of pictures, each of a dog with a different clinical sign. With the exception of generalised dermatitis, clinical signs were those associated with serious illness and likely to cause weight loss, e.g. vomiting, lethargy. Almost all of the dogs with generalised dermatitis were in Bali, and were the majority of those with observed clinical signs. There was insufficient variation in reported protein intake in Zenzele for analysis. In Bali, protein intake (either never/rarely=0 or more frequent than never/rarely=1) reported during the survey immediately prior to vaccination was generally consistent with that fed throughout the study period. 184 Appendix 4.18 continued covariate description, factor levels included in the analysis, and period of evaluation methods of evaluation house and dog identification gender (male or female) What is the dog’s name? direct observation direct observation at day 0 (vaccination): age class in months (month of life: 1-6, 7-12, 13-36, mature adult) a sterilization status (yes=1 / no=0) pregnancy or lactation (yes=1 / no=0) at day 0 (vaccination): How old is this dog or When did you get this dog? How old was the dog when you got it? direct observation Has your (female) dog been sterilized or Has she had an operation to stop her having puppies? direct observation of male dogs direct observation; whelping and vaccination dates Zenzele Jan-10 - Mar-10; Bali Dec-09 - Feb-10: body condition minimum and maximum (thin [1-3] or fat [4+]; or 2, 3, 2-3, 4, 5, 5+ [also 6+ in Zenzele]) direct observation Zenzele Jan-10/Feb-10; Bali Dec-09/Jan-10: clinical signs associated with serious illness b and generalised dermatitis (present=1 / not present=0) generalised dermatitis as a separate category (present=1 / not present=0) Zenzele Jan-10/Feb-10; Bali Dec-09/Jan-10: direct observation and owner reporting based on a set of pictures In reference to the pictures: Has this dog had any of these problems the past 7 days? Has this dog had any other problems the past 7 days? Has this dog had any of these problems since our last visit? Has this dog had any other problems since our last visit? Bali Dec-09/Jan-10: protein intake (never or rarely=0 / more frequent than never or rarely=1) Bali Dec-09/Jan-10: owner reporting based on detailed discussions a see Appendix 4.6 for the definition b including, but not limited to, vomiting; diarrhoea; dysuria; dyschezia; constipation; swollen stomach; drooling/salivation; dehydration; increase or decrease in eating or drinking; recent weight loss; jaundiced, pale, hyperaemic or cyanotic mucous membranes; coughing; dyspnoea or tachypnoea; ataxia; lethargy/depression; recumbency; severe injury 185 Appendix 4.19 Zenzele models. The full range of models were tested with natural log of the titre as the response variable and the covariates described in Methods and materials section 4.2.3. All models with the lowest AIC retained time as the only covariate (i.e. for the quadratic models, see Appendix 4.16) or were null (i.e. for models fitted to a single time point) with the exception of the models shown below. Natural logs are shown in the tables. Model 1: Zenzele quadratic model fitted to four time points (day 30, 90, 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, pregnancy and lactation as the covariates. Only lactation at the time of vaccination was retained in the model with the lowest AIC. Model 2: Zenzele analysis of variance using stepAIC fitted to a single time point (day 30). The model included natural log of the titre as the response variable and age, gender, pregnancy and lactation as the covariates. Only lactation at the time of vaccination was retained in the model with the lowest AIC. excluded 547 1663.3 intercept 2.5428 2.3563 2.7294 <0.001 lactating (relative to not lactating) -0.7886 -1.4541 -0.1232 0.021 time (Δ per day) -0.0297 -0.0319 -0.0274 <0.001 time^2 (Δ per day) 0.000063 0.000056 0.000070 <0.001 std dev of dog-level random effect intercept 0.9520 0.8293 1.0929 – outliers observations (n=) AIC parameter estimate 2.5% CI 97.5% CI p-value included 568 1745.9 intercept 2.6410 2.4534 2.8286 <0.001 lactating (relative to not lactating) -0.8562 -1.5414 -0.1710 0.015 time (Δ per day) -0.0303 -0.0325 -0.0281 <0.001 time^2 (Δ per day) 0.000064 0.000058 0.000071 <0.001 std dev of dog-level random effect intercept 0.9869 0.8618 1.1302 – outliers observations (n=) AIC parameter estimate 2.5% CI 97.5% CI p-value included 183 135.8 intercept 2.8577 2.6408 3.0746 <0.001 lactating (relative to not lactating) -1.0902 -1.9748 -0.2056 0.016 excluded 176 116.4 intercept 2.7563 2.5436 2.9690 <0.001 lactating (relative to not lactating) -0.9888 -1.8395 -0.1381 0.023 186 Appendix 4.19 Zenzele models continued Model 3: Zenzele analysis of variance using stepAIC fitted to a single time point (day 30). The model included natural log of the titre as the response variable and age, gender, pregnancy, lactation, clinical signs and body condition as the covariates. Only clinical signs at the time of vaccination was retained in the model with the lowest AIC. Clinical signs included generalised dermatitis because, in contrast to Bali, very few dogs in Zenzele had generalised dermatitis at the time of vaccination therefore this factor was not treated separately. Model 4: Zenzele analysis of variance using stepAIC fitted to a single time point (day 0). The model included natural log of the lymphocyte counts as the response variable and age, gender, body condition and the natural log of day 30 titres as the covariates. Only body condition (in this case the lower of two independent scores) was retained in the model with the lowest AIC. 0.039 97.5% CI p-value included 115 89.6 intercept 3.0339 2.7034 3.3644 <0.001 outliers observations (n=) AIC parameter estimate 2.5% CI 3.1810 with clinical signs (relative to without clinical signs) -0.6068 -1.1817 -0.0319 excluded 109 75.8 intercept 2.8509 2.5208 <0.001 with clinical signs (relative to without clinical signs) -0.4239 -0.9830 0.1352 0.136 133 -0.189 intercept body condition score 1-3 (relative to body condition 4+) -186.64 1.276 NA outliers observations (n=) AIC 0.032 1.1669 -0.3607 <0.0011.3860 p-value2.5% CI 97.5% CI -0.0170 parameter estimate 187 Appendix 4.19 Zenzele models continued Note 1: Models 1 and 2 treat the following dogs as not lactating at the time of vaccination: (i) three dogs that whelped in November 2009 and still had at least one of their pups present at vaccination (February 2010) but these pups were probably fully weaned (at vaccination), (ii) one dog that whelped in February 2010 but the survival/number of pups present at vaccination was uncertain, and (iii) two dogs that whelped in December 2009 and still had at least one of their pups present at vaccination but it was uncertain if the pups were fully weaned [treating these two dogs as lactating at vaccination in the quadratic model with upper outliers intercept = 2.6424, lactating = -0.7435 p = 0.022, and without upper outliers intercept = 2.5504, lactating = -0.8385 p = 0.010; and in the single time point model with upper outliers intercept = 2.8485, lactating = -0.7931 p = 0.059, and without upper outliers intercept = 2.7583, lactating = -0.9359 p = 0.025] Note 2: One additional linear model, fitted to four time points (day 30, 90 180 and 360) with the intercept adjusted to day 30, with the lowest AIC retained covariates other than time. The model included natural log of the titre as the response variable and time, age, gender, pregnancy, lactation and body condition as covariates. The results were similar to Model 1 (i.e. the larger data set). Apart from time, the model with the lowest AIC retained lactation at the time of vaccination only. Treating the two dogs described under Note 1(iii) as not lactating at vaccination, with upper outliers observations = 449, intercept = 2.6664, lactating = -1.0848 p = 0.009, and without upper outliers observations = 432, intercept = 2.5599, lactating = -1.0097 p = 0.012. Treating the two dogs described under Note 1(iii) as lactating at vaccination, with upper outliers observations = 449, intercept = 2.6686, lactating = -0.8981 p = 0.016, and without upper outliers observations = 432, intercept = 2.5697, lactating = -1.0511 p = 0.006. Note 3: One additional linear model, fitted to a single time point (day 30) with the lowest AIC, retained covariates. The model included natural log of the titre as the response variable and age, gender, pregnancy, lactation and body condition as covariates. The results were similar to Model 2 (i.e. the larger data set). The model with the lowest AIC retained lactation at the time of vaccination only. Treating the two dogs described under Note 1(iii) as not lactating at vaccination, with upper outliers observations = 144, intercept = 2.8825, lactating = -1.4486 p = 0.008, and without upper outliers observations = 138, intercept = 2.7756, lactating = -1.3417 p = 0.011. Treating the two dogs described under Note 1(iii) as lactating at vaccination, with upper outliers observations = 144, intercept = 2.8712, lactating = -0.9963 p = 0.043, and without upper outliers observations = 138, intercept = 2.7783, lactating = -1.2342 p = 0.013. Note 4: One additional linear model, fitted to a single time point (day 30) with the lowest AIC, retained covariates. The model included natural log of the titre as the response variable and age, pregnancy, lactation and clinical signs as covariates. The results were similar to Model 3 (i.e. the smaller data set). The model with the lowest AIC retained clinical signs at the time of vaccination only. With upper outliers observations = 122, intercept = 2.9553, with clinical signs = -0.5365 p = 0.063. Including gender in the model, with upper outliers, treating the two dogs described under Note 1(iii) as lactating intercept = 2.9553, with clinical signs = -0.5365 p = 0.063; and, treating the two dogs described under Note 1(iii) as not lactating intercept = 3.2086, with clinical signs = -0.5583 p = 0.054, lactating = -0.8403 p = 0.14, male = -0.4230 p = 0.13. Without upper outliers all the models were null. 188 Appendix 4.20 Kelusa models. The full range of models were tested with natural log of the titre as the response variable and the covariates described in Methods and materials section 4.2.3. All models with the lowest AIC retained time as the only covariate (see Appendix 4.16) with the exception of the models shown below. Natural logs are shown in the tables. Model 1: Kelusa linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, protein intake, sterilization, clinical signs and generalised dermatitis as the covariates. Only generalised dermatitis at the time of vaccination was retained in the model with the lowest AIC. Clinical signs included dermatitis and in Kelusa the majority of dogs with clinical signs had generalised dermatitis Model 2: Kelusa linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, protein intake, sterilization and body condition as the covariates. Only body condition at the time of vaccination was retained in the model with the lowest AIC. In this case body condition was the lower of the two independent scores. Note 1: Two dogs with incomplete observational data for generalised dermatitis during December 2009 and January 2010 but with chronic, generalised dermatitis diagnosed by direct observation prior to December 2009 and after January 2010 almost certainly had generalised dermatitis when vaccinated. Model 1 treats these dogs as having generalised dermatitis at vaccination. Note 2: One additional linear model, fitted to two points (day 180 and 360) with the intercept adjusted to day 30, with the lowest AIC retained covariates other than time. The model included natural log of the titre as the response variable and time, age, gender, protein intake, sterilization, body condition, clinical signs and generalised dermatitis as covariates. The results were similar to Model 1 (i.e. the larger data set). Apart from time, the model with the lowest AIC excluded upper outliers and retained generalised dermatitis at the time of vaccination only [observations = 183, intercept = 0.8961, with generalised dermatitis = -0.3358 p = 0.151]. outliers excluded observations (n=) AIC 205 590.8 parameter estimate p-value intercept 0.8625 0.5362 1.1888 <0.001 with generalised dermatitis (relative to without generalised dermatitis) std dev of dog-level random effect intercept 1.0696 0.9200 1.2435 – -0.3164 -0.0024 <0.001 2.5% CI 97.5% CI -0.7458 0.1130 0.150 time (Δ per day) -0.0034 -0.0043 0.9160 excluded 204 601.7 intercept 0.8966 std dev of dog-level random effect intercept 1.0693 body condition score 1-3 (relative to body condition 4+) -0.3204 time (Δ per day) -0.0035 1.2483 – 0.5572 1.2359 <0.001 -0.7414 0.1007 0.138 -0.0045 -0.0025 <0.001 outliers observations (n=) AIC 2.5% CI 97.5% CIparameter estimate p-value 189 Antiga 4.21 Antiga models. The full range of models were tested with natural log of the titre as the response variable and the covariates described in Methods and materials section 4.2.3. All models with the lowest AIC retained time as the only covariate (see Appendix 4.16) when upper outliers were included or were null when upper outliers were excluded with the exception of the models shown below. Natural logs are shown in the tables. Model 1: Antiga linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, pregnancy and lactation as the covariates. Only age class at the time of vaccination was retained in the model with the lowest AIC. Model 2: Antiga linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, pregnancy, lactation, clinical signs and generalised dermatitis as the covariates. Age class and generalised dermatitis at the time of vaccination were retained in the model with the lowest AIC. Clinical signs included generalised dermatitis and in Antiga the majority of dogs with clinical signs had generalised dermatitis. 1.3667 std dev of dog-level random effect intercept 1.0247 0.8831 1.1891 age class mature adult (relative to age class 1-6) -0.5920 outliers observations (n=) AIC parameter estimate age class 7-12 months (relative to age class 1-6) -0.7492 -1.4647 -0.0338 2.5% CI 97.5% CI included 289 905.6 intercept 0.8839 0.4012 age class 13-36 months (relative to age class 1-6) -0.1143 -0.7661 0.5376 -1.0751 -0.1089 time (Δ per day) p-value 0.042 0.732 – 0.018 0.037 <0.001 -0.0011 -0.0022 -0.00008 97.5% CI p-value included 230 719.6 intercept 1.3406 0.7037 1.9774 <0.001 outliers observations (n=) AIC parameter estimate 2.5% CI with generalised dermatitis (relative to without generalised dermatitis) -0.5383 -0.9292 -0.1475 0.008 age class 7-12 months (relative to age class 1-6) -0.9527 -1.7801 0.070 std dev of dog-level random effect intercept 0.9872 0.8301 1.1741 – -0.1253 0.026 age class 13-36 months (relative to age class 1-6) -0.3015 -1.0850 0.4821 0.454 age class mature adult (relative to age class 1-6) -0.7956 -1.4182 -0.1729 0.014 – excluded 208 596.0 0.3925 0.227 with generalised dermatitis (relative to without generalised dermatitis) -0.3619 -0.7196 -0.0042 0.049 intercept 0.1497 -0.0931 std dev of dog-level random effect intercept 0.8648 0.7206 1.0380 time (Δ per day) -0.0011 -0.0023 0.00008 190 Appendix 4.21 Antiga models continued Model 3: Antiga linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, pregnancy, lactation, body condition, clinical signs and generalised dermatitis as the covariates. Body condition and generalised dermatitis at the time of vaccination were retained in the model with the lowest AIC. In this case body condition was the higher of two independent scores. Clinical signs included generalised dermatitis and in Antiga the majority of dogs with clinical signs had generalised dermatitis. Note 1: Three dogs with incomplete observational data for generalised dermatitis during December 2009 and January 2010 but with chronic, generalised dermatitis diagnosed by direct observation prior to December 2009 and after January 2010 almost certainly had generalised dermatitis when vaccinated. Models 2 and 3 treat these dogs as having generalised dermatitis at vaccination. Note 2: One additional linear models, fitted to two time points (day 180 and 360) with the intercept adjusted to day 30, with the lowest AIC retained covariates other than time. The model included natural log of the titre as the response variable and time, age, gender, pregnancy, lactation and body condition as covariates. The results were similar to Model 1 (i.e. the larger data set). Apart from time, the model with the lowest AIC included upper outliers and retained age class at the time of vaccination only [observations = 226, intercept = 1.0602, mature adults = - 0.7044 p = 0.019, the other factor levels were non-significant]. outliers observations (n=) AIC parameter estimate 2.5% CI 97.5% CI p-value intercept 1.1138 0.5280 1.6996 -1.1898 0.0099 0.057 -0.5792 0.5088 excluded 190 543.8 intercept 0.4577 body condition score 4 (relative to body condition score 2-3) -0.0352 0.900 <0.001 with generalised dermatitis (relative to without generalised dermatitis) -0.7244 -1.1520 -0.2968 0.001 included 212 664.5 time (Δ per day) -0.0014 std dev of dog-level random effect 1.0139 body condition score 5+ (relative to body condition score 2-3) -0.5899 0.6855 1.0194 – body condition score 4 (relative to body condition score 2-3) -0.0653 -0.5578 0.4273 0.796 -1.0535 0.0299 0.067 body condition score 5+ (relative to body condition score 2-3) -0.5118 std dev of dog-level random effect 0.8359 0.070 with generalised dermatitis (relative to without generalised dermatitis) -0.4893 -0.8742 -0.1044 0.014 -0.0027 -0.0002 0.022 0.8528 1.2055 – -0.0322 0.9475 191 Appendix 4.22 Bali models. The full range of models were tested with natural log of the titre as the response variable and the covariates described in Methods and materials section 4.2.3. All models with the lowest AIC retained time as the only covariate (see Appendix 4.16) with the exception of the models shown below. Natural logs are shown in the tables. Model 1: Bali villages (Kelusa and Antiga) combined linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, pregnancy and lactation as the covariates. Only lactation at the time of vaccination was retained in the model with the lowest AIC. Model 2: Bali villages (Kelusa and Antiga) combined linear model fitted to two time points (day 180 and 360) with the intercept adjusted to day 30. The model included natural log of the titre as the response variable and time, age, gender, pregnancy, lactation, clinical signs and generalised dermatitis as the covariates. Lactation and generalised dermatitis at the time of vaccination were retained in the model with the lowest AIC. Clinical signs included generalised dermatitis and in Bali the majority of dogs with clinical signs had generalised dermatitis. a confidence intervals could not be obtained on the variance-covariance components using the intervals function with lme{nlme}; therefore, estimates of the confidence intervals were derived using the lme4 package with lmer, profile and confint functions b possible identifiability issues precluded estimates of the confidence intervals using the intervals function with lme{nlme}; rather estimates of the confidence intervals were derived using the lme4 package with the lmer, profile and confint functions lactating (relative to not lactating) 0.7373 -0.0569 1.5315 0.069 1.1015 1.0017 1.2113 – time (Δ per day) -0.0020 -0.0027 97.5% CI p-value included 581 1787.5 intercept 0.4944 0.2790 0.7097 <0.001 outliers observations (n=) AIC parameter estimate 2.5% CI -0.00014 <0.001 std dev of site-level random effect intercept 0.0629 0.0029 1.3483 – std dev of site/dog-level random effect intercept lactating (relative to not lactating) 0.6723 -0.1960 1.5418 0.131 2.5% CI -0.0655 0.017-0.3590 -0.6527 -0.0842 0.011 outliers observations (n=) AIC parameter estimate 97.5% CI p-value included a 442 1353.7 intercept 0.7525 0.4987 1.0090 <0.001 time (Δ per day) -0.0022 -0.0030 -0.0014 <0.001 std dev of site-level random effect intercept with generalised dermatitis (relative to without generalised dermatitis) excluded 413 1199.4 intercept 0.5315 time (Δ per day) -0.0018 with generalised dermatitis (relative to without generalised dermatitis) -0.3578 std dev of site/dog-level random effect intercept 0.9552 0.8461 1.0783 – std dev of site-level random effect intercept 0.00004 0.0000 b 0.2334 b – – std dev of site/dog-level random effect intercept 1.0694 0.9537 1.1948 – 0.00005 0.0000 0.3021 -0.0025 -0.0010 <0.001 0.2870 0.7760 <0.001 -0.6313 192 Appendix 4.22 Bali models continued Note 1: Two dogs in Antiga had incomplete observational data for lactation at the time of vaccination but were almost certainly lactating when vaccinated (January 2010) based on the presence of puppies at vaccination and the bitches’ whelping and vaccination dates. Models 1 and 2 treat these dogs as lactating at vaccination. Note 2: Models 1 and 2 treat the following dogs as not lactating at the time of vaccination: (i) one dog in Antiga that did not whelp at the owner’s house and, consequently, the survival/number of pups present at vaccination was uncertain, (ii) one dog in Kelusa that whelped in October 2009 and still had at least one pup present January 2010 but the pups were probably fully weaned at vaccination, and (iii) one dog in Kelusa that whelped in January but the vaccination date of the bitch was different to the puppies [treating this dog as lactating at vaccination in Model 1 (with upper outliers), given the likely scenario that she was vaccinated immediately after whelping and vaccination of the puppies was delayed until they were bigger, intercept = 0.4919, lactating = 0.7545 p = 0.05; treating this dog as lactating at vaccination in Model 2 (with upper outliers) intercept = 0.7484, lactating = 0.6726 p = 0.11, with generalised dermatitis = -0.3547 p = 0.019]. Note 3: One additional linear model, fitted to two time points (day 180 and 360) with the intercept adjusted to day 30, with the lowest AIC retained covariates other than time. The model included natural log of the titre as the response variable and time, age, gender, pregnancy and lactation (with the same inclusion/exclusion criteria for lactation as Models 1 and 2, see Notes 1 and 2 above). Treating the dog in Kelusa vaccinated ~ 2 weeks before her puppies (as per Note 2(iii)) as lactating at the time of vaccination, apart from time the model with the lowest AIC excluded upper outliers and retained lactation at vaccination only [observations = 549, intercept = 0.3020, lactating = 0.5822 p = 0.13]. Note 4: Two dogs in Kelusa (as per Table S21, Note 1) and three dogs in Antiga (as per Table S22, Note 1) with incomplete observational data for generalised dermatitis during December 2009 and January 2010 but diagnosed with chronic, generalised dermatitis by direct observation prior to December 2009 and after January 2010 almost certainly had generalised dermatitis when vaccinated. Model 2 treats these dogs as having generalised dermatitis at the time of vaccination. Note 5: One additional linear model, fitted to two time points (day 180 and 360) with the intercept adjusted to day 30, with the lowest AIC retained covariates other than time. The model included natural log of the titre as the response variable and time, age, gender, pregnancy, lactation and body condition as covariates (with the same inclusion/exclusion criteria for lactation as Models 1 and 2, see Notes 1 and 2 above). The results were similar to Model 1 (i.e. the larger data set). Treating the dog in Kelusa vaccinated ~ 2 weeks before her puppies (as per Note 2(iii)) as not lactating at the time of vaccination, apart from time the model with the lowest AIC included upper outliers and retained lactation at vaccination only [observations = 437, intercept = 0.6646, lactating = 0.6837 p = 0.108]. Treating the dog in Kelusa vaccinated ~ 2 weeks before her puppies (as per Note 2(iii)) as lactating at vaccination, apart from time the model with the lowest AIC included upper outliers and retained lactation at vaccination only [observations = 437, intercept = 0.6617, lactating = 0.6959 p = 0.083]. Note 6: One additional linear model, fitted to two time points (day 180 and 360) with the intercept adjusted to day 30, with the lowest AIC retained covariates other than time. The model included natural log of the titre as the response variable and time, age, gender, pregnancy, lactation, body condition, clinical signs and generalised dermatitis as covariates (the same inclusion/exclusion criteria for lactation and generalised dermatitis as Models 1 and 2, see Notes 1, 2 and 4 above). The results were similar to Model 2 (i.e. the larger data set). With upper outliers and treating the dog in Kelusa vaccinated ~ 2 weeks before her puppies (as per Note 2(iii)) as not lactating at the time of vaccination, apart from time the model with the lowest AIC retained generalised dermatitis at the time of vaccination only [observations = 402, intercept = 0.8688, with generalised dermatitis = -0.3992 p = 0.012]; and, treating the dog in Kelusa vaccinated ~ 2 weeks before her puppies (as per Note 2(iii)) as lactating at vaccination, apart from time the model with the lowest AIC retained lactation and generalised dermatitis at vaccination [observations = 402, intercept = 0.8325, lactating = 0.6339 p = 0.137, with generalised dermatitis = -0.3705 p = 0.020]. Without upper outliers, apart from time the model with the lowest AIC retained generalised dermatitis at vaccination only [observations = 373, intercept = 0.5935, with generalised dermatitis = -0.3635 p= 0.014]. 193 Appendix 4.23 Contingency tables of the covariates for the models described in Methods and materials section 4.2.4.2 (see Appendices 4.19-4.22). The tables show the maximum number of dogs with that factor level in the models that included upper outliers. Zenzele fat 19 thin 20 fat 24 thin 15 fat 50 thin 28 fat 61 thin 17 clinical signs body condition yes 39 minimum maximum no 78 minimum maximum Kelusa fat 20 thin 24 fat 34 thin 10 fat 51 thin 25 fat 66 thin 10 generalised dermatitis body condition 76 44 no yes maximum minimum maximum minimum Antiga fat 28 thin 32 fat 43 thin 17 fat 49 thin 21 fat 60 thin 10 body conditiongeneralised dermatitis yes 60 minimum maximum no 70 minimum maximum 194 Appendix 4.23 continued Zenzele pregnancy lactation yes 5 yes 11 no 187 no 181 Kelusa pregnancy lactation yes 1 yes 2 no 182 no 181 Antiga lactationpregnancy 8yes 2 yes 173no 179 no Kelusa protein yes 132 no 51 Antiga protein yes 139 no 40 Kelusa sterilised (males only) yes 34 no 149 Antiga sterilised (males only) yes 75 no 101 195 Appendix 4.24 Dogs vaccinated by the Department of Livestock in Kelusa with Rabivet Supra 92. dog days since vaccination IU/ml gender age (month of life) 1 210 0.35 m 14 1 390 0.25 2 210 0.25 m 33+ a 2 390 0.50 3 210 0.35 m 4 3 390 0.18 4 210 0.06 m 33+ 4 390 0.06 5 210 0.35 f 5 5 390 0.71 6 210 0.09 f 3 7 210 0.18 m 15 7 390 0.13 8 390 0.25 f 8-16 b 9 390 0.25 m 13 10 210 0.25 m 23 10 390 0.35 11 210 0.50 m 4 11 390 0.35 12 210 0.06 m 8 12 390 0.04 13 390 0.25 m 14 14 210 0.35 f 2 a ≥33rd month of life in December 2009; b month of life in this range Note: this table excludes two mature adults (one male and one female) vaccinated by the Department of Livestock in December 2009 that died prior to the commencement of blood sampling in June 2010 196 Appendix 4.25 The number of dogs in Zenzele with intestinal parasites on day 0 (see Appendix 4.4.(ii)). parasite number of dogs diagnosed with the parasite Ancylostoma 103 Coccidia 2 Spirocerca 6 Toxocara 5 Hymenolepis nana 1 Note: only 3 (of the 107 dogs tested) were negative for parasites 197 Appendix 5 Appendix 5.1 Outbreak size for a range of R0 and transmission rates (𝜷) for the compartmental model (CM). Each vertical line represents the distribution in outbreak size generated from 1000 simulations. The dot on the vertical line represents the estimated median outbreak size. The top and bottom of the vertical line are equivalent to the estimated 95% and 5% quantile respectively. The top horizontal line indicates the upper limit in outbreak size (of approximately 35 cases) and the bottom horizontal line indicates the median outbreak size (of approximately 3 cases) observed in villages in the Serengeti District (Hampson et al. 2009). N = population size. 198 Appendix 5.2 Probability of an outbreak at the time of vaccination (t_0) and 12 and 24 months after vaccination for a range of vaccination coverage (from 0% to 80%) and transmission rates (beta). vax cov t_0 t_12 t_24 t_0 t_12 t_24 t_0 t_12 t_24 t_0 t_12 t_24 Zenzele Zenzele beta_0.0651 beta_0.0053 beta_0.0651 beta_0.0053 R0_0.7 0 0.414 – – 0.412 – – 0.372 – – 0.382 – – 20 0.373 0.364 0.364 0.368 0.359 0.358 0.325 0.348 0.369 0.320 0.360 0.370 40 0.305 0.313 0.340 0.315 0.333 0.352 0.273 0.319 0.354 0.268 0.326 0.369 60 0.217 0.274 0.295 0.227 0.275 0.331 0.208 0.309 0.323 0.192 0.306 0.337 80 0.154 0.242 0.311 0.152 0.236 0.313 0.128 0.265 0.325 0.122 0.261 0.333 R0_1.2 0 0.538 – – 0.519 – – 0.504 – – 0.466 – – 20 0.463 0.496 0.482 0.501 0.481 0.462 0.472 0.496 0.497 0.431 0.485 0.476 40 0.413 0.444 0.443 0.426 0.431 0.443 0.381 0.461 0.500 0.384 0.476 0.478 60 0.351 0.407 0.430 0.345 0.409 0.410 0.275 0.427 0.479 0.322 0.427 0.459 80 0.207 0.349 0.387 0.219 0.341 0.420 0.225 0.400 0.472 0.217 0.398 0.474 Braamfischerville Braamfischerville beta_0.0372 beta_0.003 beta_0.0372 beta_0.003 R0_0.7 0 0.393 – – 0.416 – – 0.392 – – 0.369 – – 20 0.341 0.338 0.361 0.336 0.345 0.344 0.330 0.348 0.369 0.307 0.356 0.376 40 0.295 0.301 0.341 0.274 0.295 0.359 0.284 0.336 0.365 0.271 0.335 0.360 60 0.221 0.272 0.337 0.213 0.286 0.340 0.218 0.296 0.354 0.229 0.314 0.347 80 0.146 0.227 0.302 0.127 0.222 0.300 0.125 0.286 0.305 0.124 0.261 0.327 R0_1.2 0 0.476 – – 0.528 – – 0.508 – – 0.493 – – 20 0.451 0.443 0.485 0.461 0.443 0.527 0.464 0.521 0.509 0.431 0.472 0.488 40 0.394 0.435 0.477 0.427 0.420 0.477 0.385 0.419 0.474 0.382 0.448 0.455 60 0.313 0.381 0.466 0.326 0.340 0.458 0.276 0.399 0.468 0.287 0.410 0.483 80 0.196 0.326 0.448 0.232 0.344 0.444 0.183 0.367 0.453 0.195 0.381 0.422 Kelusa Kelusa beta_0.0711 beta_0.0058 beta_0.0711 beta_0.0058 R0_0.7 0 0.390 – – 0.388 – – 0.405 – – 0.401 – – 20 0.344 0.367 0.400 0.336 0.353 0.399 0.343 0.373 0.394 0.335 0.367 0.366 40 0.259 0.319 0.388 0.254 0.335 0.362 0.258 0.325 0.337 0.309 0.332 0.356 60 0.208 0.268 0.321 0.214 0.295 0.347 0.232 0.296 0.341 0.209 0.299 0.321 80 0.138 0.260 0.295 0.114 0.237 0.295 0.136 0.229 0.319 0.135 0.241 0.322 R0_1.2 0 0.484 – – 0.528 – – 0.515 – – 0.511 – – 20 0.458 0.463 0.546 0.476 0.515 0.521 0.474 0.480 0.482 0.479 0.474 0.484 40 0.439 0.455 0.499 0.427 0.476 0.481 0.386 0.452 0.468 0.386 0.456 0.474 60 0.288 0.410 0.458 0.321 0.414 0.469 0.313 0.439 0.475 0.314 0.401 0.444 80 0.202 0.338 0.429 0.214 0.369 0.437 0.205 0.354 0.414 0.205 0.381 0.428 Kaler Kaler beta_0.1563 beta_0.0127 beta_0.1563 beta_0.0127 R0_0.7 0 0.423 – – 0.418 – – 0.419 – – 0.373 – – 20 0.330 0.358 0.371 0.308 0.325 0.373 0.340 0.328 0.389 0.297 0.373 0.384 40 0.289 0.323 0.357 0.281 0.287 0.335 0.299 0.338 0.366 0.281 0.327 0.359 60 0.197 0.280 0.303 0.220 0.286 0.319 0.216 0.293 0.355 0.207 0.285 0.354 80 0.135 0.225 0.298 0.142 0.250 0.301 0.132 0.253 0.308 0.126 0.261 0.315 R0_1.2 0 0.493 – – 0.553 – – 0.535 – – 0.494 – – 20 0.490 0.475 0.489 0.438 0.445 0.537 0.453 0.492 0.465 0.479 0.489 0.505 40 0.424 0.442 0.444 0.403 0.464 0.472 0.383 0.470 0.469 0.390 0.444 0.447 60 0.329 0.402 0.447 0.330 0.381 0.452 0.326 0.431 0.476 0.318 0.386 0.431 80 0.223 0.325 0.378 0.231 0.340 0.373 0.195 0.383 0.438 0.211 0.364 0.418 Ketug Ketug beta_0.0744 beta_0.006 beta_0.0744 beta_0.006 R0_0.7 0 0.391 – – 0.388 – – 0.378 – – 0.393 – – 20 0.366 0.353 0.381 0.360 0.345 0.394 0.327 0.369 0.371 0.337 0.352 0.355 40 0.297 0.319 0.344 0.281 0.319 0.374 0.245 0.345 0.399 0.274 0.327 0.365 60 0.214 0.281 0.333 0.221 0.282 0.340 0.197 0.295 0.321 0.225 0.309 0.315 80 0.146 0.247 0.324 0.124 0.239 0.301 0.141 0.263 0.291 0.131 0.278 0.295 R0_1.2 0 0.502 – – 0.524 – – 0.512 – – 0.517 – – 20 0.481 0.488 0.544 0.492 0.489 0.519 0.428 0.479 0.473 0.467 0.480 0.480 40 0.428 0.457 0.495 0.425 0.454 0.508 0.401 0.450 0.488 0.382 0.432 0.471 60 0.301 0.405 0.477 0.327 0.395 0.471 0.321 0.414 0.455 0.286 0.409 0.466 80 0.214 0.379 0.413 0.224 0.356 0.425 0.194 0.344 0.416 0.172 0.354 0.427 Individual-based model Compartmental model 199 Appendix 5.3 Mean estimates of vaccination coverage with time (95% confidence intervals). % of the population vaccinated at month 0 % vaccination coverage after month 12 % vaccination coverage after month 24 Zenzele 60 39.4 (39.3-39.5) 25.1 (25.0-25.2) 80 52.6 (52.5-52.6) 33.5 (33.5-33.6) Braamfischerville 60 38.3 (38.2-38.4) 24.0 (23.9-24.1) 80 50.9 (50.8-51.0) 29.3 (29.2-29.3) Kelusa 60 37.8 (37.7-37.8) 26.3 (26.3-26.4) 80 50.5 (50.4-50.6) 35.2 (35.1-35.3) Kaler 60 42.5 (42.4-42.5) 32.6 (32.4-32.7) 80 56.4 (56.4-56.5) 43.4 (43.2-43.4) Ketug 60 39.8 (39.7-40.0) 26.8 (26.6-26.9) 80 52.9 (52.7-53.0) 35.4 (35.3-35.5) 200 Appendix 5.4 Probability of outbreaks with human-mediated incursions for a range of vaccination coverage at the start of the time series and transmission rates (𝜷) for Kaler. The plots were generated using the individual-based model (IBM). The probability of an outbreak during the 37 month study period is shown on the y-axis. The average number of exposed dogs entering the study population during the study period are shown on the x-axis; and were estimated from empirical data described in section 5.2.3.3. The baseline probability that a dog gained into the study population during the study period was exposed was obtained by dividing 0.77 (number of exposed dogs) by the total number of dogs gained into the study population during the study period (n=266). This probability was included in the model. The number (and the probability) of exposed dogs gained during the study period was then increased incrementally by 10%, 20% and 50% from the baseline as shown. The horizontal line provides a reference to allow for comparisons. 201 Appendix 5.5 Probability of outbreaks with human-mediated incursions for a range of vaccination coverage at the start of the time series and transmission rates (𝜷) for Ketug. The plots were generated using the individual-based model (IBM). The probability of an outbreak during the 37 month study period is shown on the y-axis. The average number of exposed dogs entering the study population during the study period are shown on the x-axis; and were estimated from empirical data described in section 5.2.3.3. The baseline probability that a dog gained into the study population during the study period was exposed was obtained by dividing 0.36 (number of exposed dogs) by the total number of dogs gained into the study population during the study period (n=116). This probability was included in the model. The number (and the probability) of exposed dogs gained during the study period was then increased incrementally by 10%, 20% and 50% from the baseline as shown. The horizontal line provides a reference to allow for comparisons. 202 Appendix 5.6 Outbreak size with human-mediated incursions for a range of vaccination coverage at the start of the time series and transmission rates (𝜷) for all of the Bali study populations. The plots were generated using the individual-based model (IBM). The number of exposed dogs gained into the study population during the study period are shown on the x-axis; see the legends to Figure 5.3 and Appendices 5.4 and 5.5 for the derivation of the number (and probability) of exposed dogs gained into the study population during the study period. Each vertical line represents the distribution in outbreak size generated from 1000 simulations. The dot on the vertical line represents the estimated median outbreak size. The top and bottom of the vertical line are equivalent to the estimated 95% and 5% quantile respectively. 203 Appendix 6 Appendix 6.1 Rabies vaccine-induced virus neutralizing antibody (VNA) titres in pups in Zenzele. Excerpt from M.K Morters, S McNabb, D.L Horton, A.R Fooks, J.P Schoeman, H.R Whay, J.L.N Wood, S Cleaveland (in preparation) Effective vaccination against rabies in puppies in rabies endemic regions Every available puppy (≤12 weeks of age) in Zenzele was vaccinated in February 2010 as part of the serological study described in Chapter 4 [68 puppies were vaccinated from a total of 86 in the population]. The majority of the puppies were <6-8 weeks of age when vaccinated and deemed too small to blood sample immediately prior to vaccination without causing unnecessary distress to the puppy and/or owner. Therefore, pre-vaccinal virus neutralizing antibody titres were obtained from only four of the puppies. Titres for these puppies were ≤0.13 IU/ml, similar to pre-vaccinal titres in 32 dogs 1.5-4.5 months of age from dams vaccinated with high quality, inactivated vaccine against rabies in Thailand (Kasempimolporn et al. 1996). To measure post-vaccinal peak titres blood samples were collected approximately 30 days following vaccination. Thirty seven of the 68 vaccinated puppies remained in Zenzele 30 days after vaccination, and of these nineteen were big enough to blood sample; peak titres for these (19) puppies are shown in the table below. All of the puppies seroconverted to the vaccine (i.e. generated titres ≥0.5 IU/ml (Kennedy 1998)), with a geometric mean titre (GMT) of 20.7 IU/ml. 204 Appendix 6.1 continued Day 0 (pre-vaccination) and day 30 (peak) titres of the puppies vaccinated in Zenzele dog gender age at vaccination (weeks) puppy day 0 titres (IU/ml) puppy day 30 titres (IU/ml) dam day 0 titres (IU/ml) dam present May 2006 1 f 8-10 0.06 11.3 0.06 yes 2 m 8-10 0.06 2.0 0.06 yes 3 f 7-8 – 45.3 0.18 yes 4 m 6-7 – 22.6 0.06 no 5 m 6-7 – 45.3 0.06 no 6 f 7-8 – 16.0 0.06 no 7 m 5-6 – 64.0 0.06 no 8 f 4-6 – 45.3 0.06 yes 9 f 5-7 – 32.0 0.06 yes 10 f 5-7 – 64.0 0.06 yes 11 m 5-7 – 5.7 0.06 yes 12 m 4-6 – 45.3 0.18 yes 13 f 4-6 ‒ 90.5 0.18 yes 14 f 5 ‒ 22.6 0.06 yes 15 f 5 – 8.0 0.06 yes 16 f 5 – 5.7 0.06 yes 17 f 10 days – 5.7 0.09 no 18 f 6-8 – 32.0 ‒ ‒ 19 f 10-12 – 22.6 ‒ ‒ Seventeen of the nineteen puppies (blood sampled on day 30) were born in Zenzele to eight dams; all of the adult females were seronegative (<0.5 IU/ml) immediately prior to vaccination in February 2010 (titres ranged from 0.06-0.18 IU/ml). Five of the dams may have been vaccinated by the Department of Agriculture in May 2006, however none had an anamnestic response to vaccination (see Chapter 4, section 4.3.3) in February 2010 (day 30 titres ranged from 0.09-90.5 IU/ml). The other two puppies (blood sampled on day 30) were obtained from outside Zenzele, therefore the vaccination status of their dams was not known. Only five of the (68) vaccinated puppies were still in Zenzele 90 days after vaccination, and of these four remained 12 months after vaccination (with day 360 titres of 0.09, 0.35, 0.35 and 1 IU/ml). 205 Appendix 7 Appendix 7.1 Transcription errors. Paper questionnaires were used to collect dog and household data in the study areas (see Chapter 2, section 2.2.2 and Appendix 2.1). Each questionnaire included up to 100 questions, with the number of questioned asked during the interview dependent on the gender of the dog and the acquisition and loss of dogs from the household since the previous survey. In order to estimate the number of transcription errors incurred during the transfer of data from the questionnaires to the Excel spreadsheet (used to store and manipulate the data) approximately 400 questionnaires were randomly selected from all of the questionnaires completed during the study period (March 2008 – April 2011). A sample size of 400 gives 95% power to detect errors occurring at 1% frequency (Thrusfield 2005). The data in each sample questionnaire was checked against the corresponding data in the spreadsheet. An error rate of approximately 0.5% was detected. Summary of the transcription errors Zenzele Braamfischerville Kelusa Antiga total number of questionnaires selected 132 90 137 113 472 number of entries checked 8064 5202 8649 6889 28804 number of errors 41 18 42 49 150 % of entries with errors 0.51 0.35 0.49 0.71 0.52 number of questionnaires with errors 24 14 26 22 86 % of questionnaires with errors 18.18 15.56 18.98 19.47 18.22 206 Appendix 7.2 Repeatability of the questionnaires. Repeatability of the questionnaire was tested in Braamfischerville in September 2010 and in Antiga in April 2010 by re-interviewing the owners of 34 randomly selected dogs in each site. The first and second interviews were one day apart. The same questionnaire was used and the same respondent interviewed. The respondents had no prior knowledge of the repeat interview. The enumerator undertaking the repeat interview was randomly allocated. A comparison of the responses given during the first and second interview are shown in the table below. Comparisons exclude questions and observations specific to the previous 7 days. For example, questions such as Have you taken this dog out of the village the past 7 days? and When did you last feed this dog? and the clinical signs observed by the primary researcher and enumerator during the first interview were excluded from the analysis. This was on the basis that it would not be possible to differentiate any difference in the response due to an actual change in circumstance from variability in owner recall. Comparison of the responses given during the first and second interviews Braamfischerville Antiga current age same category 27 (79.4%) 21 (84.0%) (1-6, 7-12, 13-36, 37+ months) one category different a 7 (20.6%) 4 (16.0%) age at acquisition same category 34 (100%) 31 (91.2%) (≤3 b , 4-12, 13+ c months) different category 0 (0%) 3 (8.8%) source of a new dog d same location 6 (100%) 3 (100%) different location 0 (0%) 0 (0%) body condition score (1-9) same score 30 (88.2%) 12 (54.5%) (primary researcher) one score different 4 (11.8%) 10 (45.5%) owner reported clinical signs previous 3 months same 30 (85.3%) 31 (91.2%) different 4 (14.7%) 3 (8.8%) a there was ≤ 6 month difference in reported age for 6 dogs and >6 month difference in reported age for 5 dogs; b understood by the enumerators to be ≤12 weeks of age; c understood by the enumerators to be older than 1 year (52 weeks) of age d there were only 6 and 3 new dogs present in the selected households in Braamfischerville and Antiga respectively Note: In Antiga n=9 respondents reported ‘don’t know’ regarding the current age of the dog during one or both of the interviews. In Antiga n=12 dogs were not present during one or both of the interviews, therefore these dogs could not be given a body condition score; the variability in body condition scores in Antiga is probably the result of increased flightiness and movement of the dogs as a consequence most of the dogs having been recently caught by net for vaccination.