SYSTEMS FOR THE PREVENTION AND CONTROL OF INFECTIOUS DISEASES IN PIGS
Katharina D.C. Stärk
A thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy at Massey University, Palmerston North, New Zealand
1998
Printed by the Massey University Printery, Palmerston North, 1998
To obtain a copy of this thesis write to:
EpiCentre Wool Building Massey University Palmerston North New Zealand
Ph +64 6 350 52 70 Fax +64 6 350 57 16
Or send an email to [email protected]
The results of science remain hypotheses that may have been well tested, but not established: not shown to be true. Of course, they may be true. But even if they fail to be true, they are splendid hypotheses, opening the way to still better ones.
Karl R. Popper, A World of Propensities, 1990
Abstract
An expert system (RestiMATE) was designed that assists veterinary practitioners in assessing the respiratory health status of a pig farm. RestiMATE uses classification rules to identify patterns of environmental risk factors for respiratory diseases and to select optimal manage- ment interventions to control and prevent respiratory diseases. The classification rules are based on expert interviews and on empirical data collected in New Zealand. Recursive parti- tioning and neural network techniques have been applied for rule induction. These methods were compared with logistic regression and appeared to be similarly efficient in terms of clas- sification while providing additional insight into the structure of a data set. Non-parametric analytical methods appear to be particularly suitable when analysing complex data sets and for exploratory data analysis.
EpiMAN-SF is an advanced decision-support system designed to manage and analyse data accumulated during an African swine fever or classical swine fever emergency. EpiMAN-SF offers state-of-the-art technology for managing data related to a swine fever epidemic, in- cluding laboratory results. An expert system was developed to support rapid classification of contacts between pig farms in terms of the risk of virus transmission. These classifications are used to set priorities in visiting farms for laboratory investigations. The validation of the ex- pert system showed that its evaluation was more consistent and generally more risk-averse than that of human experts. A stochastic simulation model was developed to investigate the spread of swine fever infection within a farm and a second model (INTERSPREAD-SF) was designed to forecast the dynamics of the epidemic within a region and to evaluate control strategies. INTERSPREAD-SF has been validated using real outbreak data from Germany and was shown to be capable of realistically replicating the behaviour of classical swine fever.
However, more research is needed to complete our knowledge about the detailed epidemiol- ogical processes during a swine fever epidemic.
A prerequisite for efficient disease control in pig populations is reliable animal identification.
A series of trials was conducted in order to compare electronic ear tags and implantable iden- tification chips with visual ear tags. It was shown that the difficulties with respect to implants are loss rates of up to 18.1% within 4 weeks after implantation while electronic ear tags were lost or damaged by processing at the abattoir in up to 23.4% of pigs.
Infectious aerosols were reviewed as an additional aspect of the causative network of infec- tious diseases in pigs. An air sampling system based on air filtration was developed and ap- plied in combination with polymerase chain reaction assays. Using this technique, Myco- plasma hyopneumoniae, the major causative agent of enzootic pneumonia was isolated from air samples for the first time. However, the attempt to isolate classical swine fever virus from the air was unsuccessful, probably due to technical difficulties.
iii
Acknowledgements
Doing a PhD is all about learning. I certainly learnt a lot during these last three years in the Epidemiology Group (now EpiCentre) at the Department of Veterinary Clinical Sciences (now Institute of Veterinary, Animal and Biomedical Sciences). Probably the experience of the most lasting value to me was to realise that no matter how good or bad a situation, I can always learn something: either how to do something or how not to do it. Therefore, first of all, I would like to thank everyone who has helped me learn.
In this context, I am particularly grateful to my chief supervisor Prof. Roger Morris, who pro- vided me with the ideas and vision, which are the strong foundations of this thesis. The help of my second supervisor Dr. Dirk Pfeiffer was indispensable in successfully realising many aspects of my work. His ‘you can do it’ attitude helped me tackle many problems and allowed me to put analytical issues into perspective. I also worked with Mr. Mark Stern on software design issues and the INTERSPREAD model. Many other people have contributed signifi- cantly to the content of this thesis, and their help is acknowledged at the end of each chapter.
I would probably never have come to New Zealand if it had not been for the motivation and enthusiasm of my mentor and friend Prof. Ueli Kihm. He taught me the importance of integ- rity and modesty. I have also not forgotten the repeated advice given by my first boss and teacher in animal health, Prof. Hermann Keller. He showed me that research needs to serve the purpose of solving problems, a principle that I believe is very much present in this thesis.
As I am definitely a social person whose performance depends on personal discussions and a motivating atmosphere to a large extent, I thank all my friends within the Epidemiology Group, particularly Dirk, Barb, Deb, Jo and Ron for letting me share with them my thoughts, ideas and frustrations every day.
This PhD project was funded by the Swiss National Science Foundation (Grant No. 823B- 040072) and supported by the Swiss Federal Veterinary Office. For the trials described in CHAPTER 2.4 approval from the Massey University Animal Ethics Committee was obtained.
The preparation of a PhD inevitably impacts on family life. I would therefore like to acknowl- edge the support of my husband Marcus and of my family in Europe.
Katharina D.C. Stärk Palmerston North, March 1998
Table of contents
Abstract ... i
Acknowledgements... iii
Table of contents ... v
List of Figures ... x
List of Tables... xii
INTRODUCTION... 1
PART I Endemic infectious diseases Example: Respiratory diseases ... 7
CHAPTER 1.1 EPIDEMIOLOGICAL INVESTIGATION OF THE INFLUENCE OF ENVIRONMENTAL RISK FACTORS ON RESPIRATORY DISEASES IN SWINE – A LITERATURE REVIEW... 9
1. Summary ... 11
2. Introduction... 11
3. Methods... 12
3.1 Study design and sample size ...12
3.2 Case definition ...14
3.3 Exposure definition and measurement ...18
3.4 Data analysis ...18
4. Results and discussion ... 19
4.1 Infection pressure ...21
4.2 Susceptibility...24
4.3 Path model hypothesis ...27
5. Conclusion ... 27
CHAPTER 1.2 RISK FACTORS FOR RESPIRATORY DISEASES IN NEW ZEALAND PIG HERDS.. ... 39
1. Abstract ... 41
2. Introduction... 41
3. Material and methods ... 41
3.1 Farm recruitment...41
3.2 Abattoir data recording ...42
3.3 Farm data collection...44
3.4 Data management and analysis...44
4. Results ... 45
5. Discussion ... 57
CHAPTER 1.3 THE ROLE OF INFECTIOUS AEROSOLS IN DISEASE TRANSMISSION IN PIGS (A LITERATURE REVIEW) ... 63
1. Introduction... 65
2. Definitions ... 65
3. The airborne pathway ... 66
3.1 Factors influencing aerosol production...66
3.2 Factors influencing aerosol decay...67
3.3 Factors influencing aerosol inhalation and infection ...69
4. Aerosol sampling... 71
5. Aerosol sample analysis ... 73
vi
6. Airborne diseases in pigs ... 73
6.1 Foot and mouth disease... 74
6.2 Swine vesicular disease... 75
6.3 Aujeszky’s disease... 75
6.4 Influenza ... 76
6.5 Porcine respiratory and reproductive syndrome (PRRS)... 76
6.6 Classical and African swine fever ... 77
6.7 Porcine respiratory corona virus (PRCV) ... 77
6.8 Enzootic pneumonia (EP)... 78
6.9 Pleuropneumonia ... 78
6.10 Atrophic rhinitis ... 79
6.11 Other diseases... 79
7. Prevention of airborne disease in pig production... 79
CHAPTER 1.4 DETECTION OF MYCOPLASMA HYOPNEUMONIAE BY AIR SAMPLING WITH A NESTED PCR ASSAY ... 91
1. Abstract... 93
2. Introduction ... 93
3. Material and Methods ... 94
3.1 Strain growth conditions and DNA extraction. ... 94
3.2 Air sampling system... 94
3.3 Air-sample processing for PCR assay... 96
3.4 Design of specific oligonucleotide primers and PCR reactions ... 96
3.5 Field sampling ... 97
3.6 Statistical analysis ... 99
4. Results ... 99
4.1 Specificity and sensitivity of the nested PCR assay... 99
4.2 Field sampling ... 100
5. Discussion ...104
CHAPTER 1.5 ALTERNATIVE METHODS TO SOLVE CLASSIFICATION PROBLEMS IN COMPLEX DATA SETS ...109
1. Introduction ...111
2. Recursive partitioning (classification trees) and machine learning ...111
3. Neural networks...113
4. Example: Classification of pig farms with respect to the prevalence of enzootic pneumonia...115
4.1 Data set... 115
4.2 Methods and software... 115
5. Results ...118
6. Discussion ...122
CHAPTER 1.6 RestiMATE – THE DESIGN OF AN EXPERT SYSTEM FOR DIAGNOSING AND MANAGING RESPIRATORY DISEASES ON PIG FARMS...129
1. Introduction ...131
2. Problem definition ...133
3. Users ...133
4. System structure ...133
5. Input ...135
5.1 Farm variables... 135
5.2 Target values... 136
6. Processing ...137
6.1 Comparison of farm variables with target values...137
6.2 Farm classifications...137
6.3 Identification of problem areas...141
6.4 Recommendation of interventions...142
7. Output... 143
7.1 Output of empirical classification method ...144
7.2 Output of expert-based classification method ...144
8. Discussion ... 145
PART II Exotic infectious diseases Example: Classical and African swine fever ... 149
CHAPTER 2.1 REVIEW OF THE EPIDEMIOLOGY OF CLASSICAL AND AFRICAN SWINE FEVER ... 151
1. Epidemiology of Classical swine fever ... 153
1.1 Introduction...153
1.2 Aetiology ...153
1.3 Affected species ...153
1.4 Disease course ...154
1.5 Diagnosis ...154
1.6 Virus transmission...155
1.7 Prevention and control ...158
2. Epidemiology of African swine fever ... 160
2.1 Introduction...160
2.2 Aetiology ...160
2.3 Affected species ...160
2.4 Sylvatic cycle ...161
2.5 Disease in domestic pigs...161
2.6 Virus transmission...161
2.7 Diagnosis ...163
2.8 Prevention and control ...164
CHAPTER 2.2 FAILURE TO ISOLATE CLASSICAL SWINE FEVER VIRUS FROM THE AIR OF ROOMS HOUSING EXPERIMENTALLY INFECTED PIGS... 169
1. Introduction... 171
2. Animals and Methods ... 171
2.1 Animals ...171
2.2 Virus...172
2.3 Air sampling...172
2.4 Sample analysis...172
3. Results ... 172
4. Discussion and conclusions ... 173
CHAPTER 2.3 ELICITATION OF EXPERT KNOWLEDGE ON RISK FACTORS FOR CLASSICAL SWINE FEVER TRANSMISSION IN SWITZERLAND ... 177
1. Abstract ... 179
2. Introduction... 179
3. Material and methods ... 180
4. Results ... 181
5. Discussion ... 182
CHAPTER 2.4 COMPARISON OF ELECTRONIC AND VISUAL IDENTIFICATION SYSTEMS IN PIGS ... 185
1. Abstract ... 187
2. Introduction... 187
viii
3. Methods ...188
3.1 Sow trial ... 188
3.2 Piglet trials... 189
3.3 Data handling and analysis ... 190
4. Results ...190
4.1 Sow trial ... 190
4.2 Piglet trials... 192
5. Discussion ...195
5.1 Application of internal EID ... 196
5.2 Loss rate and performance on farm ... 196
5.3 Problems at slaughter ... 197
6. Conclusions...198
CHAPTER 2.5 QUANTIFICATION OF CONTACTS BETWEEN PIG FARMS TO ASSESS THE POTENTIAL RISK OF CLASSICAL SWINE FEVER TRANSMISSION ...201
1. Introduction ...203
2. Material and Methods ...203
2.1 Dutch data set ... 203
2.2 Swiss data set ... 204
2.3 Data analysis... 204
3. Results ...206
3.1 Description of participating farms... 206
3.2 Number of contacts... 208
3.3 Distance of contacts ... 208
3.4 Risk of contacts ... 210
4. Discussion ...211
4.1 Data quality ... 211
4.2 Classification of contacts... 212
4.3 Analysis of contacts ... 212
CHAPTER 2.6 EpiMAN-SF – DESIGNING A DECISION-SUPPORT SYSTEM FOR THE MANAGEMENT OF SWINE FEVER EPIDEMICS ...217
1. Introduction ...219
2. Problem domain definition ...219
3. Users ...220
4. Tasks...220
4.1 Virus strain identification ... 221
4.2 Farm data management... 221
4.3 Spatial data management ... 222
4.4 Management of traces ... 223
4.5 IP and PE management ... 223
4.6 Movement control management ... 223
4.7 Surveillance and laboratory management ... 224
4.8 Epidemic analysis ... 228
4.9 Documentation and reporting ... 229
5. Technical principles and requirements of EpiMAN-SF ...229
6. Discussion ...230
CHAPTER 2.7 EXPERT SYSTEM COMPONENTS OF EpiMAN-SF ...233
1. Introduction ...235
2. Rule base of EpiMAN-SF ...235
2.1 Rules for calculating time periods ... 235
2.2 Rules for classification of conveyors ... 238
2.3 Rules for re-classification of conveyors after the virus strain becomes known...247
2.4 Validation of rules for the risk classification of conveyors ...249
2.5 Rules for the classification of farms...252
2.6 Validation of rules for the risk classification of farms ...260
3. Discussion ... 263
CHAPTER 2.8 WITHIN-FARM SPREAD OF CLASSICAL SWINE FEVER VIRUS – A BLUEPRINT FOR A STOCHASTIC SIMULATION MODEL... 267
1. Summary ... 269
2. Introduction... 269
3. Modelling concepts ... 270
4. Model characteristics ... 271
4.1 Start of infection...271
4.2 Management groups ...272
4.3 Transition probabilities ...273
4.4 Virus strain ...274
4.5 Carrier sow syndrome ...277
5. Model output ... 277
6. Validation ... 278
7. Discussion ... 279
CHAPTER 2.9 ANALYSIS OF A CLASSICAL SWINE FEVER OUTBREAK IN LOWER SAXONY, GERMANY... 283
1. Introduction... 285
2. Material and methods ... 285
2.1 Outbreak data...285
2.2 Spatial data and mapping...288
2.3 Analytical methods and simulation ...288
3. Results ... 293
3.1 Maps of outbreak area ...293
3.2 Epidemic curve ...293
3.3 Network of spread...295
3.4 Survival curve and dissemination rate...296
3.5 Proportion of IP caused by episode types ...297
3.6 Simulation modelling ...298
4. Discussion ... 303
4.1 Suitability of field data ...303
4.2 Outbreak dynamics ...304
4.3 Simulation and selection of control strategy...305
GENERAL DISCUSSION... 311
1. What is a system?... 313
2. Biological systems: investigating the web of causation... 313
3. Information systems ... 315
3.1 Decision-support systems ...315
3.2 Simulation models as elements of decision-support systems ...317
3.3 How to deal with uncertainty...317
3.4 Validation of knowledge-based systems and simulation models...318
3.5 Use and success of information systems ...319
4. Systems thinking ... 321
List of related publications ... 325
x
List of Figures
FIGURE 1. Causal web based on relationships between several risk factors ... 13
FIGURE 2. Distribution of number of study units in reviewed articles on respiratory diseases in pigs... 14
FIGURE 3. Alternative pathways for respiratory disease to be influenced by environmental risk factors... 21
FIGURE 4. Path model hypothesis for the associations between risk factors for respiratory diseases in pigs. 28 FIGURE 5. Typical lesion classified as ‘enzootic-pneumonia-like lesion’... 43
FIGURE 6. Typical lesion classified as ‘Actinobacillus pleuropneumoniae-like lesion ... 43
FIGURE 7. Location of abattoirs ... 46
FIGURE 8. Violin plots for farm prevalence of enzootic pneumonia and pleurisy/pleuropneumonia in New Zealand pig herds ... 48
FIGURE 9. Association of enzootic pneumonia prevalence and mean enzootic pneumonia score per inspected pig in New Zealand pig herds ... 49
FIGURE 10. Relationship between prevalence of enzootic pneumonia and pleuropneumonia in New Zealand pig herds ... 50
FIGURE 11. Air sampling in the field... 99
FIGURE 12. PCR analysis of air sampling polyethersulfone membranes... 101
FIGURE 13. Charts of two pig rooms illustrating the distribution of positive PCR results from 6 air samples . 103 FIGURE 14. Diagram of a multi-layer perceptron... 114
FIGURE 15. Classification trees grown using a) ID3 and CART (gini rule), b) C4.5, c) CHAID, and d) CART (twoing rule) ... 120
FIGURE 16. Graphical comparison of classification methods using multidimensional scaling... 122
FIGURE 17. Structure of an expert system... 132
FIGURE 18. System architecture of RestiMATE ... 134
FIGURE 19. Example of data entry screen for general farm management area ... 136
FIGURE 20. Example of an output window... 144
FIGURE 21. Example of a report produced by the expert-based classification method ... 145
FIGURE 22. Piglets with electronic ear tag and visual ear tag ... 191
FIGURE 23. Implantation site at right ear base... 191
FIGURE 24. Lesions observed with visual ear tags ... 194
FIGURE 25. Comparison of mean migration distance of two different injectable electronic identification transponders ... 195
FIGURE 26. Boxplots of animal numbers on 21 Swiss and 94 Dutch pig farms ... 207
FIGURE 27. Distance to point of origin/destination for contacts on and off 21 Swiss and 96 Dutch pig farms during a 2-week period ... 209
FIGURE 28. Components of EpiMAN-SF ... 220
FIGURE 29. State-transition flowchart for farm status during a swine fever epidemic... 222
FIGURE 30. System architecture of EpiMAN-SF ... 230
FIGURE 31. Time frames for forward and backward tracing and relationship to risk classification of traces ... 239
FIGURE 32. Agreement among experts when classifying contacts in an imaginary classical swine fever outbreak expressed as difference from the median ... 250
FIGURE 33. Differences between median risk ratings and individual ratings for 10 tracing officers ... 251
FIGURE 34. Differences between median risk ratings and individual ratings for 31 imaginary traces classified by human experts... 251
FIGURE 35. Agreement between human experts and an expert system when classifying contacts in an imaginary classical swine fever outbreak expressed as difference from the human rating... 252
FIGURE 36. Influence of number of conveyor contacts with different risk categories on the survival probability of
a farm to remain free of swine fever...255
FIGURE 37. Mean ranked survival probability and 5% and 95% percentiles of 13 farms with different episode and conveyor scenarios ...256
FIGURE 38. Cumulative probability of survival for a neighbouring farm with one medium risk conveyor...257
FIGURE 39. Tornado diagrams for farms with A) 1, B) 2, C) 3, or D) 4 very-low risk contacts, respectively....258
FIGURE 40. Agreement among experts when ranking farms in an imaginary classical swine fever outbreak expressed as difference from the group median ...262
FIGURE 41. Differences between farm rankings performed during an imaginary classical swine fever outbreak by human experts and an expert system...262
FIGURE 42. Elements of simulation process on a farm consisting of several management groups...270
FIGURE 43. Starting point of classical swine fever infection depending on virus source...272
FIGURE 44. Distribution of incubation time for a low-moderately virulent classical swine fever strain...276
FIGURE 45. Percentage of pigs incubating or shedding virus in an infected pig unit during a 30-day simulation period of classical swine fever transmission ...277
FIGURE 46. Spread of classical swine fever in a pen with 100 pigs...278
FIGURE 47. Geographic location of German classical swine fever epidemic analysed in this chapter...286
FIGURE 48. Pig density in Germany based on State figures ...287
FIGURE 49. Map of pig farm locations in District 2 ...294
FIGURE 50. Example of a thematic map ...294
FIGURE 51. Epidemic curve of classical swine fever outbreaks two districts in Lower Saxony between October 1993 and October 1995...295
FIGURE 52. Network of spread of classical swine fever in two districts in Lower Saxony, Germany, between October 1993 and October 1995 ...296
FIGURE 53. Survival function for farms in District 2 between 01/Oct/93 and 18/May/95 ...297
FIGURE 54. Typical epidemic curves of mean weekly numbers of classical swine fever outbreaks simulated with INTERSPREAD and applying different control strategies...301
FIGURE 55. Survival curves for duration of classical swine fever epidemics ...302
FIGURE 56. Relationship between artificial intelligence, expert systems and knowledge-based systems ...317
xii
List of Tables
TABLE 1. Factors influencing respiratory disease occurrence or the incidence of re-infection of respiratory
disease-free herds ... 19
TABLE 2. Influence of herd size on the frequency of respiratory lesions at slaughter expressed as odds ratios ... 22
TABLE 3. Recommended maximal values for air contaminants in swine buildings ... 25
TABLE 4. Definition of macroscopic lung lesions in slaughter pigs ... 42
TABLE 5. The effect of season on the prevalence of respiratory lesions in slaughter-weight pigs in New Zealand... 46
TABLE 6. Occurrence of respiratory lesions in slaughter-weight pigs in New Zealand during winter 1995 (and summer 1996 ... 47
TABLE 7. Descriptive statistics of continuous management variables of New Zealand pig farms... 51
TABLE 8. Farm management for North and South Island pig farms and odds ratios for risk factors for enzootic pneumonia and pleurisy/pleuropneumonia in New Zealand pig herds: Binary variables . 52 TABLE 9. Farm management for North and South Island pig farms and odds ratios for risk factors for enzootic pneumonia and pleurisy/pleuropneumonia in New Zealand pig herds: Continuous variables... 53
TABLE 10. Random-effects logistic regression model for enzootic pneumonia in New Zealand pig herds... 56
TABLE 11. Random-effects logistic regression model for pleurisy / pleuropneumonia in New Zealand pig herds ... 57
TABLE 12. Relationship between body weight of pigs and heat producing units ... 70
TABLE 13. Air sampling methods and their characteristics ... 72
TABLE 14. Airborne infectious diseases in pigs: accumulated evidence ... 74
TABLE 15. Porcine Mycoplasma and Acholeplasma strains used in this study and their reaction in the two steps of the nested PCR assay. ... 95
TABLE 16. Oligonucleotide primers used in this study ... 96
TABLE 17. Description of Swiss pig farms and rooms sampled... 98
TABLE 18. Results for air samples analysed with a nested PCR to detect DNA of Mycoplasma hyopneumoniae ... 102
TABLE 19. Factors associated with the outcome of a nested PCR assay to detect DNA from Mycoplasma hyopneumoniae in air samples ... 103
TABLE 20. Variables used in the analysis of risk factors affecting the prevalence of enzootic pneumonia lesions in New Zealand pig herds ... 116
TABLE 21. Comparative performance of classification schemes using data on enzootic pneumonia prevalence from 86 farms ... 118
TABLE 22. Multinomial logistic regression models for the classification of 3 levels of enzootic pneumonia prevalence in 86 pig farms ... 119
TABLE 23. Ranking of variable selection in different classification schemes... 121
TABLE 24. Results of 15-fold cross-validation ... 121
TABLE 25. Structure of target value table and some examples of variables... 137
TABLE 26. Farm classification groups for classification tree... 138
TABLE 27. Rule base to classify farms with respect to the prevalence of EP and PLPN using a classification tree ... 139
TABLE 28. Farm classification groups for human expert ... 140
TABLE 29. Rule base for classifying farms with respect to EP ... 140
TABLE 30. Recommendation examples as they will be stored in the recommendation table of RestiMATE. 143
TABLE 31. Rules for selecting advice components for reporting...143
TABLE 32. Comparison of tree-based and expert-based methods in RestiMATE...146
TABLE 33. Sources of classical swine fever outbreaks in European epidemics...156
TABLE 34. Persistence of classical swine fever virus in various pork products ...157
TABLE 35. Incubation period and onset of viraemia of African swine fever ...162
TABLE 36. Earliest onset of African swine fever virus excretion in relation to onset of clinical signs...162
TABLE 37. Persistence of African swine fever virus in various pork products...163
TABLE 38. Serological and virological results of blood samples from pigs infected with classical swine fever virus ...173
TABLE 39. Relative importance of risk factors for the transmission of classical swine fever within Switzerland as estimated by experts ...181
TABLE 40. Source of introduction of classical swine fever virus to 121 farms during the 1993-1995 outbreak in Germany ...182
TABLE 41. Evaluation of adaptive conjoint analysis workshop by participants ...182
TABLE 42. Performance of pig identification tags as percentage tags in place and working at different trial stages ...193
TABLE 43. Summary of results from different electronic identification trials in pigs...197
TABLE 44. Contact-related attributes recorded by farmers...204
TABLE 45. Risk classification rules for visitor contacts on and off farms ...205
TABLE 46. Risk classification rules for family contacts on and off farms ...205
TABLE 47. Pig and dairy cow inventory for 21 Swiss and 94 Dutch farms ...207
TABLE 48. Number of contacts per farm for 21 Swiss and 96 Dutch farms in a 2-week period ...208
TABLE 49. Distance (km) between origin/destination of contacts and farm for 21 Swiss farms ...209
TABLE 50. Number of contacts associated with different risk levels for spread of classical swine fever in 21 Swiss and 96 Dutch pig farms during a 2-week period...210
TABLE 51. Descriptive statistics for the number of contacts per farm for 21 Swiss and 96 Dutch farms during a 2-week period ...211
TABLE 52. Visit schedules and task protocol according to European Union legislation...224
TABLE 53. Rules for scheduling farm visits based on laboratory test results...227
TABLE 54. Chronology of events during a classical swine fever outbreak on different levels with respect to procedures performed in EpiMAN-SF ...227
TABLE 55. Differences between low-moderate virulence strains and high-virulence strains and effect of assumption on EpiMAN-SF decisions ...236
TABLE 56. Rules for estimating the period when a farm became infected...238
TABLE 57. Rules for classifying conveyors when tracing back to identify the source of infection and the source of the conveyor is an IP...240
TABLE 58. Rules for classifying conveyors when tracing back to identify source of infection and source of conveyor is not an IP...244
TABLE 59. Rules for classifying conveyors when tracing forward to identify secondary outbreaks and source of conveyor is an IP ...244
TABLE 60. Rules for classifying conveyors when tracing forward to identify secondary outbreaks and the source of conveyor is not an IP...247
TABLE 61. Risk classification table for conveyors (source = IP) ...248
TABLE 62. Frequency (%) of tracing classifications by human experts and expert system ...252
TABLE 63. Transmission probabilities for conveyors and episodes ...254
xiv
TABLE 64. Characteristics of scenarios used to explore farm risk classifications ... 256
TABLE 65. Rules for scheduling farms for farm visits based on directive 80/217 EEC... 259
TABLE 66. Example of a table containing information on management groups... 273
TABLE 67. State transition possibilities for the simulation of spread of classical swine fever virus ... 273
TABLE 68. Epidemiological differences between classical swine fever virus strains of low, moderate and high virulence... 275
TABLE 69. Characteristics of classical swine fever virus isolates ... 275
TABLE 70. Descriptive statistics of farms in two classical swine fever-infected districts in Germany ... 286
TABLE 71. Input parameters related to between-farm spread as used by INTERSPREAD to simulate the dynamics of classical swine fever ... 290
TABLE 72. Control measures applied in all scenarios... 291
TABLE 73. Combination of control strategies ... 292
TABLE 74. Further assumptions for classical swine fever simulation with INTERSPREAD... 292
TABLE 75. Distribution of sources of classical swine fever infection for 37 outbreaks in Germany... 298
TABLE 76. Descriptive statistics of INTERSPREAD simulations ... 299
TABLE 77. Number of farms affected by different infection and control mechanisms ... 300
INTRODUCTION
3
Under intensive production systems pigs are kept at high density and slaughtered at a young age. Consequently, infectious diseases are more important than degenerative or proliferative problems. In a morbidity and mortality survey of swine conducted in the United States in 1990/1991 scours alone accounted for 58% of all observed conditions and illnesses (Anony- mous, 1992). The course of infectious diseases is particularly severe in young piglets where most of the mortality is due to infectious pathogens (Wegmann, 1990; Zimmer et al., 1997).
Infectious diseases cause great losses to the producer and are important also from the animal welfare perspective. Additionally, drugs used to treat infectious diseases in pigs, if not prop- erly applied, have the potential to remain as residues in meat destined for human consump- tion. For these reasons the prevention of infectious diseases is of great interest not only to the pig producer but also to the consumer. In the United States according to a major study, 57.2 % of farmers vaccinate all piglets against at least one infectious pathogen and 77.5 % vaccinated all sows (Anonymous, 1992).
If prevention is not possible or is not successfully achieved, infectious diseases have to be treated using medication. In the above cited US survey, 18.8% of farmers used oral antibiotics and 32.7% injected antibiotics to prevent and treat infectious diseases in piglets (Anonymous, 1992). Additional interventions addressing changes in housing or management may also be effective. As such interventions are expensive, they need to be based on sound epidemiologi- cal knowledge. A detailed understanding of the risk factors involved and their interaction is required. However, the number of relevant risk factors may be large and the most appropriate intervention difficult to select.
For each country, two categories of infectious diseases can be distinguished in relation to pig health: endemic and exotic diseases. An endemic disease is defined as a disease that is con- stantly present with only relatively minor fluctuations in its frequency within a defined geo- graphic area or population (Last, 1988; Smith, 1991). Typically, there is no official control programme for endemic diseases, and it is up to the individual farmer and his/her veterinary consultant to plan therapeutic or preventive interventions or both.
If an infectious pathogen is not normally present in a region or a country, it is called exotic and any outbreak of the disease will qualify as an epidemic, i.e. provoke an occurrence ex- ceeding the expected frequency (Smith, 1991). Most countries will adopt an eradication strat- egy for exotic diseases, particularly if the disease is harmful to humans (zoonosis) or will cause significant production losses. If a disease will prompt trade restrictions from trade part- ners who are free from the disease, eradication is also the strategy of choice. New occurrences of highly contagious diseases that are easily transmitted between animals and between farms are considered particularly serious. Because pigs are social animals, which are typically housed in groups, disease spread can be rapid and hard to prevent. Therefore, highly conta- gious exotic diseases require fast and powerful response coordinated by the veterinary serv- ices at national or regional level. Particularly in areas with high pig densities, a delayed re- sponse can have devastating consequences and disrupt production for several months (Davies, 1995). Obviously, an epidemic of such size will have severe direct and indirect economic consequences, out of which the actual costs of measures for the eradication of the disease will probably be a relatively small part (Vantemsche, 1995). Drastic measures such as the compul- sory stamping-out strategy, where the entire stock of an infected farm is destroyed, may there- fore be justified. Losing all stock, particularly breeding stock, is a tragedy for the individual farmer. The media and the public in general will keep a close watch over how the epidemic is
addressed and whether the ‘right’ decisions are being made. However, making the right deci- sion under pressure is difficult, particularly if a new disease has been introduced or if a dis- ease has not occurred for a long time, and there is a lack of knowledge and relevant expertise in the veterinary services.
With both endemic and exotic diseases, effective decision making is a crucial component of disease control. At the same time, decision making is potentially difficult, because the conse- quences of a particular decision are not known with certainty. The objective is to choose the best option amongst a number of possible decisions given the current circumstances, either on a farm or during an epidemic. Decisions have to be justifiable in terms of their efficacy and cost-effectiveness. In order to be able to make an informed choice, all information that is available should be considered and used as the basis. Because the amount of information may be large and complex, computer-aided decision support is likely to accelerate and improve the decision process by making information accessible. Today, computers are used to assist hu- man experts in a vast number of areas. They are used to collect, record, store, retrieve and display data (Teichroew, 1993), but also to perform more demanding tasks such as data proc- essing and data interpretation for reasoning. Such advanced information systems have also been developed in the veterinary field (Morris, 1991). A good overview is provided in an edi- tion of the Revue scientifique et technique de l’Office international des Epizooties dedicated to epidemiological information systems (1991, issue no. 1).
In this thesis, the term ‘information system’ is used in the sense of being a structured ap- proach to the definition and solution of a problem as defined by Morris (1991). Information systems are designed with the objective to support decision-makers. The user will work with a computer in order to provide data or models to “recognise, understand and formulate a prob- lem and make use of analytical aids to evaluate alternatives” (Klein and Methlie, 1995).
In this thesis, the case of endemic and exotic infectious diseases in pigs is being considered.
In each area, one disease was selected as an example. Respiratory diseases are used as a typi- cal endemic disease problem complex (PART 1 of the thesis) and classical and African swine fever are used to illustrate the case of exotic diseases (PART 2 of the thesis). Respiratory dis- eases, more specifically enzootic pneumonia and pleuropneumonia were selected because they are equally important in all intensive pig-producing countries worldwide (Christensen and Mousing, 1992). Also, the causal web of factors related to respiratory diseases is com- plex, so that choosing the right strategy for an intervention is not straightforward. African and particularly classical swine fever (CSF) were chosen because swine fever is currently the sin- gle most devastating exotic pig disease in Europe. Huge outbreaks of CSF have destroyed significant parts of the national pig populations in Belgium, Germany and the Netherlands.
The control of CSF is difficult because the disease can remain unnoticed for considerable time periods resulting in large numbers of secondary outbreaks due to unrestricted movements. The amount of data related to these outbreaks is likely to quickly become unmanageable without the help of computerised information technology.
The literature related to both diseases is reviewed in CHAPTERS 1.1 and 2.1. Because respi- ratory diseases are also airborne diseases, the literature on this additional aspect is also cov- ered in CHAPTER 1.3. After having reviewed the available knowledge on the epidemiology of the example diseases, additional studies were conducted to complete the information. Two levels were considered: within-farm spread and between-farm spread.
5
With respect to between-farm spread, a key aspect of disease transmission is the movement of people, animals and goods between farms. These can be numerous (see CHAPTER 2.5) and when a disease outbreak is investigated, the farmer will probably not be able to remember all contacts. The movement of susceptible animals is of greatest risk. In order to be able to trace these movements, animals need to be individually identifiable. Without accurate animal iden- tification, it is impossible to verify movements of animals between farms. For this reason compulsory identification of pigs has been introduced in many countries. Part of the current discussion on this topic involves issues related to the practicality and reliability of the current identification system for pigs. This question has been specifically addressed in a series of field trials described in CHAPTER 2.4. Although the chapters on contacts between farms and ani- mal identification are in PART 2 of the thesis dedicated to exotic diseases, they are equally relevant and applicable to endemic diseases. There is some overlap between the areas related to endemic and exotic diseases.
In addition to observational field studies and questionnaire surveys, the relatively new tech- nique of expert knowledge elicitation was used to obtain data on the epidemiology of infec- tious diseases (CHAPTER 2.3). Experiments to investigate the possibility of aerosol transmis- sion were conducted for both diseases (CHAPTERS 1.4 and 2.2). Where the conduct of stud- ies was not possible, simulation models were developed to identify critical points where our understanding of disease transmission is still incomplete (CHAPTER 2.8). State-of-the art techniques for data analysis were used to make the results available for use in an information system. A series of non-parametric techniques were also applied to investigate their potential in the analysis of complex data sets (CHAPTER 1.5). Finally, two decision-support systems were developed.
RestiMATE is a diagnostic guide to assist decision-makers in the control of respiratory dis- eases on individual farms. It uses data from a field survey (CHAPTER 1.2) and expert knowl- edge to assess the respiratory health status of a farm and to provide advice on effective inter- ventions (CHAPTER 1.6). EpiMAN-SF (CHAPTER 2.6) is a more complex system designed to support veterinary services in containing and eradicating swine fever epidemics. Its expert system components are described in detail in CHAPTER 2.7. The use of EpiMAN-SF as an analytical tool is illustrated using data from a German classical swine fever outbreak in CHAPTER 2.9.
References
Anonymous (1992). National swine survey: Morbidity/mortality and health management of swine in the United States. National Animal Health Monitoring System, USDA:APHIS:VS, Fort Collins, CO.
Christensen, G. and Mousing, J. (1992). Respiratory System. In: A.D. Leman, B.E. Straw, W.L. Mengeling, S.
D'Allaire, and D.J. Taylor (eds.) Diseases of Swine, 7th Edition. Iowa State University Press, Iowa, 128-162.
Davies, G. (1995). An overview of recent epidemics of animal diseases. In: Diijkuizen, A.A., and Davies, G.
(eds.) Animal health and related problems in densely populated livestock areas of the Community. Office for Official Publications of the European Communities, Luxembourg, EUR 16609, 35-40.
Klein, M.R., and Methlie, L.B. (1995). Knowledge-based decision support systems with applications in business.
John Wiley & Sons, Chichester.
Last, J.M. (1988). A dictionary of epidemiology. Oxford University Press, Oxford.
Morris, R.S. (1991). Information systems for animal health: objectives and components. Revue scientifique et
technique de l’Office International des Epizooties 10:13-23.
Smith, R.D. (1991). Veterinary clinical epidemiology. Butterworth-Heinemann, Boston.
Teichroew, D. (1993). Information system. In: Ralston, A., and Reilly, E.D. The Encyclopedia of Computer Science 3rd Edition, Van Nostrand Reinhold Co., New York, 666-669.
Vantemsche, P. (1995). Classical swine fever 1993-1994 Belgium. In: Diijkuizen, A.A., and Davies, G. (eds.) Animal health and related problems in densely populated livestock areas of the Community. Office for Offi- cial Publications of the European Communities, Luxembourg, EUR 16609, 69-79.
Wegmann, P. (1990). Pathology of swine – a portrait of economic loss in pig production in Switzerland. Pro- ceedings of the 11th IPVS Congress, Lausanne, 295.
Zimmer, K., Zimmermann, Th., and Heß, R.G. (1997). Todesursachen bei Schweinen. Der praktische Tierarzt 78:772-780.
PART I
Endemic infectious diseases
Example: Respiratory diseases
CHAPTER 1.1
EPIDEMIOLOGICAL INVESTIGATION OF THE INFLUENCE OF ENVIRON- MENTAL RISK FACTORS ON RESPIRATORY DISEASES IN SWINE –
A LITERATURE REVIEW
11
1. Summary
In this chapter, the influence of environmental factors on respiratory diseases in pigs is re- viewed from an epidemiological point of view. The suitability of methods for the investiga- tion of risk factors is discussed including aspects of study design, case definition, exposure assessment and data analysis. The results of published studies suggest a causal web of factor interaction, the analysis of which provides considerable challenge for current epidemiological techniques. New approaches to the problem, such as knowledge-based data interpretation systems, should be further explored in the future in order to provide reliable advice to deci- sion makers.
2. Introduction
Respiratory diseases are a common problem in swine populations world-wide (Blaha, 1992;
Christensen and Mousing, 1992). Most prevalent are pneumonia, pleuropneumonia and pleu- risy commonly with multiple infectious agents contributing (Goiš et al., 1980; Morrison et al., 1985; Ciprian et al., 1988; Falk et al., 1991; Amass et al., 1994). Generally, the morbidity of these diseases is high, while mortality is variable depending on the causative agents involved.
There are two major subsets of lung disease within the respiratory disease complex: a high proportion of farms with respiratory problems are infected with Mycoplasma (M.) hyopneu- moniae plus secondary invaders, while a somewhat lower proportion have disease due to Ac- tinobacillus (A.) pleuropneumoniae. For this reason, these two agents deserve special atten- tion and will be treated accordingly in this review.
The economic impact of respiratory diseases is considerable, due mainly to reduced growth and feed efficiency (Huhn, 1970; Goodwin, 1971; Braude and Plonka, 1975; Hoy et al., 1987;
Pointon et al., 1985; Straw et al., 1989; Straw et al., 1990) and possibly reduced fertility (Hoy, 1994). In fact, respiratory diseases are still among the most devastating diseases in in- tensive swine production (Guerrero, 1990). The economic impact has been found to be more severe if respiratory disease occurs early in life (Wallgren et al., 1990, 1993a; Morris et al., 1995a) or is aggravated by other diseases (Bernardo et al., 1990) or an adverse environment (Done, 1990a; Straw, 1991).
Besides the economic aspect, the animal welfare side of the problem has been addressed (Blaha, 1993) and public health concerns with regard to the syndrome have been raised. It has been shown that the probability of pigs being treated is higher among animals with respiratory disease (Willeberg et al., 1978; Madsen, 1980; Elbers et al., 1992a; Singer, 1993, Blaha et al., 1994), particularly in the finishing period. Blaha et al. (1994) consequently postulated that the detection rate of antimicrobial residues is also expected to be higher in carcasses with lung lesions, as compared with others without pathological abnormalities.
Finally, the environmental factors causally related to respiratory problems in swine may also have a negative effect on the health of people working on pig farms (Larsson et al., 1992).
Increased incidences of respiratory conditions in swine confinement building workers (Cormier et al., 1991) and veterinarians (Tielen et al., 1996) have been reported.
Numerous attempts have been made to control respiratory problems by different methods (MacInnes and Rosendal, 1988; Zimmermann, 1990; Straw, 1992; Wallgren et al., 1993b;
Plonait and Gindele, 1995). The development of specific pathogen free (SPF) herds has been the most effective (Christensen and Mousing, 1992; Kuiper et al., 1994; Bækbo et al., 1996), although many of these herds have become re-infected at a later stage due to airborne disease transmission (Goodwin, 1985; Jorsal and Thomsen, 1988; Stärk et al., 1992a). Besides eradi- cation, various control measures designed to reduce infection pressure within the herd, e.g.
management changes, medication and vaccination, have been applied (Christensen and Mousing, 1992).
Since the early seventies, it has been suggested that respiratory diseases in swine may be in- fluenced not just by the presence of specific organisms but by a rather complex interaction between a number of factors related to the agent, the host and the environment (Kalich, 1970a, 1970b). Knowledge of these factors is essential for disease control and prevention (Hoy et al., 1987). Epidemiological methods have helped to identify the most important risk factors and to investigate their interaction. The methods applied and the results of these stud- ies are reviewed in this article.
3. Methods
The causal relationship between environmental risk factors and respiratory diseases is com- plex. One of the reasons for this is the possibility of direct and indirect effects of such factors on the respiratory system of the pig (FIGURE 1). A factor could have one or both effects.
Using the concept of causal inference described by Rothman (1986) there are a number of so- called component causes involved in any disease process. A set of such components consti- tutes the minimal conditions required to start the disease and represents a ‘sufficient cause’.
For a given outcome, there may exist several sets of sufficient causes and consequently a large set of possible factor constellations having a similar effect, in this case influencing respi- ratory disease in swine. Within a sufficient cause interaction of factors may occur, thus changing the impact of a given factor depending on the level of another factor.
The objectives of epidemiological studies in this context can be: 1) identification of risk fac- tors, 2) quantification of risk factor influence and 3) quantification of risk factor interaction and assessment of possible confounding.
In order to be able to achieve these goals, epidemiological studies have to be planned care- fully. A number of requirements have to be met for the results to be valid.
3.1 Study design and sample size
Done (1991) lists alternative study designs for the investigation of the relationships between risk factors and respiratory diseases in swine. Basically, these are:
1) Assessment of the health status of animals and of the environmental status of the farm of origin, either subjectively or based on physical measurements at one point in time to investi- gate associations (cross-sectional approach, case-control studies).
13 F A C T O R 1
F A C T O R 2
F A C T O R 3
E F F E C T
FIGURE 1. Causal web based on relationships between several risk factors 2) Monitoring of the situation on a farm(s) over a longer time period, collecting data on a number of putative risk factors and assessing their relationship with health measures (longitu- dinal approach).
3) Experimental exposure of animals in a controlled environment (experimental approach).
The largest group of the studies reviewed in this chapter used the first approach because it al- lows rapid identification and quantification of risk factors under field conditions. The longitu- dinal approach is not often used as it is more time- and resource-consuming. Results from ex- perimental studies are hard to translate to field conditions and are therefore not well suited for this particular problem.
The second question when planning a study is: How many farms or animals are needed to detect a significant effect of a risk factor? The necessary sample size in observational studies basically depends on the following considerations:
- what difference between groups is relevant to be detected – what is the prevalence of exposure
– what are the test characteristics
– what power requirements (1- β error) are needed – what level of confidence (α error) is desired
If these figures can be estimated, the necessary sample size can be calculated using the for- mulas described in sampling text books (for example, Cannon and Roe, 1982).
FIGURE 2 shows the number of study units used in the publications reviewed in this article.
More than 50% of the studies used less than 50 epidemiological units (farms or animals). The consequence of this is a limited power to detect differences between groups. This has to be acknowledged when interpreting results. Particularly negative results are difficult to interpret with low-powered studies. This problem has been long recognised and widely discussed in the literature, yet, logistical and financial limitations sometimes prevent obtaining larger samples.
0 5 10 15 20 25 30 35 40
<10 10-50 51-100 101-200 201-400 >400 Number of study units
% publications
FIGURE 2. Distribution of number of study units in reviewed articles on respi- ratory diseases in pigs (n=32)
3.2 Case definition
When dealing with respiratory diseases, a case can either be an infected farm or an individual animal. In order to identify a case or a non-case, the disease needs to be diagnosed. It is rele- vant whether the outcome variable will be the occurrence of disease only (diseased vs. non- diseased, incidence, prevalence) or whether the severity of cases should also be measured.
Not all of the following diagnostic approaches are good indicators of the latter.
Basically, four diagnostic approaches can be used to define a case of respiratory disease: defi- nition by clinical signs, by serological analyses, by microbiological investigation or by lung scoring at slaughter or during post mortem investigations.
3.2.1 Diagnosis based on clinical signs
Respiratory diseases are commonly accompanied by the typical clinical signs of coughing, which can be used to estimate disease prevalence by defining a ‘cough index’ (Straw et al., 1986a; Bahnson et al., 1994). This system used on its own seems likely to miss cases because under good environmental conditions subclinical disease may develop (Keller, 1976). Applied at a herd level, clinical inspection failed to detect 30% of infected herds (Sørensen et al., 1993). Morris et al. (1995a) reported a sensitivity of clinical cough of 37.7 % in market pigs when compared with gross lesions at slaughter but a comparatively high specificity (76.3 %).
Coughing is also not considered to be a good indicator of severity (Straw et al., 1990), al- though the inclusion of clinical parameters in other measurement systems has proved to be helpful. Bahnson et al. (1994) for example, were able to predict optical density measurements of an M. hyopneumoniae ELISA system at slaughter by using among others a cough index and the time of onset of coughing as explanatory variables. The use of veterinary treatment has been used as an indirect measure of clinical disease (Elbers et al., 1992a), but is likely to be biased due to the influence of the farmer and the veterinarian.
15 3.2.2 Diagnosis based on antibody detection
Detection of antibodies is suitable both for clinical and subclinical cases. However, the dy- namics of the disease as well as the test parameters (sensitivity, specificity, predictive values) have to be considered when interpreting test results.
Blood antibodies against M. hyopneumoniae and A. pleuropneumoniae rise 2-4 weeks after infection, peak at around 11-14 weeks post inoculationem (p.i.) and disappear about 6 weeks later (Bachmann, 1972; Strasser et al., 1992; Yagihashi et al., 1993; Kobisch et al., 1993; Le Potier et al., 1994; Sitjar et al., 1994; Morris et al., 1995b; Sørensen et al., 1997). Under field conditions antibody levels against A. pleuropneumoniae rise from the age of 12 weeks with a peak at 23 weeks (Willson et al., 1987; Gardner et al., 1991; Sitjar et al., 1994). The situation is similar for M. hyopneumoniae with a peak at 12-14 weeks of age and about 4 weeks after peak of clinical signs (Sitjar et al., 1994). Seroconversion seems to coincide with the onset of coughing, but only moderate agreement between serology and occurrence of gross lesions at slaughter were reported (Falk and Lium, 1991; Morris et al., 1995a; 1995b).
The tests of choice for detecting antibodies against respiratory diseases are indirect or block- ing ELISA kits. With respect to M. hyopneumoniae the specificity of the ELISA can be re- duced due to cross-reactions with M. flocculare or M. hyorhinis (Strasser et al., 1992). Under field conditions, using a monoclonal blocking ELISA, the herd sensitivity and specificity were found to be 93-100% and 85.1-96%, respectively, if a farm was to be classified as in- fected with one out of 20 blood samples positive (Sørensen et al., 1992, 1993). If the criterion was changed to two or more positive samples for an infected herd, sensitivity and specificity were 69% and 98%, respectively. The test characteristics for an A. pleuropneumoniae ELISA on a herd level were 89% for both sensitivity and specificity and 97% and 96% respectively on an individual animal level (Willson et al., 1988).
A non-invasive method for antibody detection is the serology of colostrum samples (Zim- mermann et al., 1986; Volmer, 1994; Nielsen, 1995), but samples have to be collected imme- diately after farrowing, which may be unpractical because it requires constant monitoring of the sow. If colostrum samples are collected immediately after farrowing, positive ELISA test results are observed earlier and at a higher frequency than in serum among naturally infected pigs (Sørensen et al., 1993) indicating a higher herd sensitivity of this technique. However, at least 30 colostrum samples per herd are required to obtain an accurate picture (Rautiainen et al., 1996).
Another means for antibody detection is the analysis of saliva and fluid obtained by bron- choalveolar lavage for mucosal IgA which can be detected at an earlier stage of infection than humoral antibodies in blood serum (Loftager et al., 1993).
Recently new techniques for serological detection of past infections have been developed.
Frey et al. (1994) showed that antibodies against the species-specific L-lactate dehydrogenase of M. hyopneumoniae first occurred at 5 to 10 weeks p.i. when clinical signs and lung lesions were present. High titers persisted until 21 weeks p.i. which is much longer than antibodies against membrane proteins which are commonly used for diagnosis.
