EPIDEMIOLOGIC TRAITS OF AND FACTORS POTENTIALLY CONTRIBUTING TO THE HIGHLY PATHOGENIC AVIAN INFLUENZA OUTBREAK AMONG COMMERCIAL POULTRY FARMS IN MINNESOTA AND IOWA BETWEEN APRIL AND JUNE OF 2015 By Aisling Nolan A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Epidemiology - Master of Science 2017 ABSTRACT EPIDEMIOLOGIC TRAITS OF AND FACTORS POTENTIALLY CONTRIBUTING TO THE HIGHLY PATHOGENIC AVIAN INFLUENZA OUTBREAK AMONG COMMERCIAL POULTRY FARMS IN MINNESOTA AND IOWA BETWEEN APRIL AND JUNE OF 2015 By Aisling Nolan The largest outbreak of highly pathogenic avian influenza (HPAI) virus among commercial poultry farms in the United States, most notably in Minnesota (MN) and Iowa (IA), took place between December 2014 and June 2015. This study addressed potentially contributing environmental factors through a case-control logistic regression analysis in which environmental traits of MN and IA counties were compared to county infection status. Outcomes were: 0=no infected farms and 1=at least one infected farm in terms of turkey (HPAIT); laying hen and pullet (HPAIC); and turkey, laying hen, and pullet (HPAICOMB) farms. The primary exposure was the percent of total county area which is corn grown for grain. To assess if results were influenced by counties with the greatest number of infected farms, one MN county and two IA counties were removed from the datasets. Prior to removal, significant associations were found between the corn variable and HPAIT for MN, corn variable and HPAIT for IA, corn variable and HPAICOMB for MN, and corn variable and HPAICOMB for IA. After removal, regression coefficient estimates remained approximately the same; these counties did not unduly impact the findings. More consideration might be given to the environmental component of the HPAI virus transmission pathway when attempting to predict spread during an outbreak. ACKNOWLEDGEMENTS I appreciate the support of my thesis committee, Dr. A. Mahdi Saeed, Dr. Dorothy R. Pathak, and Dr. Lixin Zhang; my defense auditor, Dr. David Barondess; my graduate secretary, Nancy Bieber; and my parents, relatives, and friends. Thank you. iii TABLE OF CONTENTS LIST OF TABLES .......................................................................................................................... v LIST OF FIGURES ...................................................................................................................... vii KEY TO ABBREVIATIONS ...................................................................................................... viii CHAPTER 1: INTRODUCTION ................................................................................................... 1 1.1 Objectives ............................................................................................................................. 1 1.2 Hypotheses ............................................................................................................................ 2 1.3 Hypothesized Transmission Mechanism .............................................................................. 2 1.4 Biology.................................................................................................................................. 3 1.5 Disease Presentation ............................................................................................................. 5 1.6 Early History ......................................................................................................................... 5 1.7 History in the U.S. ................................................................................................................ 6 1.8 History in the World ............................................................................................................. 8 1.9 Events Pertinent to the 2014/2015 HPAI Outbreak in Minnesota and Iowa ...................... 12 1.9.1 Asia .............................................................................................................................. 12 1.9.2 Europe .......................................................................................................................... 12 1.9.3 North America ............................................................................................................. 12 1.10 Study Justification ............................................................................................................. 13 CHAPTER 2: METHODS ............................................................................................................ 14 2.1 Study Population and Design .............................................................................................. 14 2.2 Measures ............................................................................................................................. 14 2.3 Analysis............................................................................................................................... 15 CHAPTER 3: RESULTS .............................................................................................................. 17 CHAPTER 4: DISCUSSION ........................................................................................................ 30 4.1 Temporal Distribution ......................................................................................................... 32 4.2 Strengths ............................................................................................................................. 34 4.3 Future Research .................................................................................................................. 34 APPENDICES .............................................................................................................................. 35 APPENDIX A: Tables .............................................................................................................. 36 APPENDIX B: Figures ............................................................................................................. 57 REFERENCES ............................................................................................................................. 73 iv LIST OF TABLES Table 1a: Distribution of infected farms among states and individual MN and IA counties in the USDA APHIS “HPAI 2014-2015 Infected Premises” (state, county, number of birds, flock type, species, HPAI strain confirmed) dataset (N=232 infected farms) and the analytic sample (N=180 infected farms) ......................................................................... 37 Table 1b: Characteristics of infected farms in MN and IA counties in the USDA APHIS “HPAI 2014-2015 Infected Premises” (state, county, number of birds, flock type, species, HPAI strain confirmed) dataset (N=232 infected farms) and the analytic sample (N=180 infected farms) ................................................................................................................. 40 Table 2: Characteristics of Minnesota and Iowa counties in the 2010 U.S. Census and the 2012 USDA Census of Agriculture datasets ........................................................................... 41 Table 3: Explanation of acronyms ............................................................................................... 43 Table 4: Description of the highly pathogenic avian influenza outbreak in Minnesota and Iowa between April and June of 2015 ................................................................................... 44 Table 5a: Results of logistic regression with HPAIC outcome based on data from all counties. 45 Table 5aa: Results of logistic regression with HPAIC outcome following removal of Kandiyohi (MN), Buena Vista (IA), and Sioux (IA) Counties ...................................................... 47 Table 5b: Results of logistic regression with HPAIT outcome based on data from all counties. 48 Table 5bb: Results of logistic regression with HPAIT outcome following removal of Kandiyohi (MN), Buena Vista (IA), and Sioux (IA) Counties ...................................................... 50 Table 5c: Results of logistic regression with HPAICOMB outcome based on data from all counties ......................................................................................................................... 52 Table 5cc: Results of logistic regression with HPAICOMB outcome following removal of Kandiyohi (MN), Buena Vista (IA), and Sioux (IA) Counties .................................... 54 Table 6: Characteristics of the counties in Minnesota and Iowa with the greatest number of infected farms ............................................................................................................... 56 v LIST OF FIGURES Figure 1: Derivation of the analytic sample for infected farms ................................................... 58 Figure 2: Epidemic curve: Laying hen, turkey, and pullet farms in Minnesota reported as positive for highly pathogenic avian influenza (HPAI) virus during April 2015-June 2015................................................................................................................................ 59 Figure 3: Epidemic curve: Laying hen, turkey, and pullet farms in Iowa reported as positive for highly pathogenic avian influenza (HPAI) virus during April 2015-June 2015 ............ 60 Figure 4: Derivation of the analytic sample for all MN and IA counties .................................... 61 Figure 5: Minnesota counties with at least one infected turkey, laying hen, or pullet farm ........ 62 Figure 6: Iowa counties with at least one infected turkey, laying hen, or pullet farm ................. 63 Figure 7: Distribution of highly pathogenic avian influenza (HPAI)-infected laying hen, turkey, and pullet farms in Minnesota counties during April 2015-June 2015 .......................... 64 Figure 8: Distribution of highly pathogenic avian influenza (HPAI)-infected laying hen, turkey, and pullet farms in Iowa counties during April 2015-June 2015................................... 65 Figure 9: Days at risk of HPAI virus infection contributed by laying hen, turkey, and pullet farms during the observation period .............................................................................. 66 Figure 10: Distribution of the cases of HPAI virus infection in Minnesota and Iowa counties over the observation period ............................................................................................ 67 Figure 11: Yield of corn grown for grain in Minnesota counties (USDA NASS Agriculture Counts 2015) .................................................................................................................. 68 Figure 12: Yield of corn grown for grain in Iowa counties (USDA NASS Agriculture Counts 2015) .............................................................................................................................. 69 Figure 13: Farms with turkey inventory on December 31, 2012 (USDA NASS 2012 Census of Agriculture) .................................................................................................................... 70 Figure 14: Farms with layer inventory on December 31, 2012 (USDA NASS 2012 Census of Agriculture) .................................................................................................................... 71 Figure 15: Hogs and pigs: Inventory 2012 (USDA NASS 2012 Census of Agriculture) ........... 72 vi KEY TO ABBREVIATIONS adjOR adjusted odds ratio OR odds ratio CI confidence interval USDA United States Department of Agriculture APHIS Animal and Plant Health Inspection Service HPAI highly pathogenic avian influenza LPAI low pathogenic avian influenza highly pathogenic avian influenza infection status, where 0=no turkey HPAIT farms with virus infection and 1=at least one infected turkey farm highly pathogenic avian influenza infection status, where 0=no laying HPAIC hen or pullet farms with virus infection and 1=at least one infected laying hen or pullet farm highly pathogenic avian influenza infection status, where 0=no laying HPAICOMB hen or pullet or turkey farms with virus infection and 1=at least one infected laying hen or pullet or turkey farm ALLPOUFA number of all types of poultry farms per county TURKFA number of turkey farms per county LAYFA number of laying hen farms per county PULFA number of pullet farms per county LAYPULFA combined number of laying hen and pullet farms per county TOTAREAC acres of total area per county vii CROPAC acres of total cropland per county CROPPRC percent of total county area which is cropland CORNGRAINAC acres of corn grown for grain per county CORNGRPRC percent of total county area which is corn grown for grain FARMAC acres of farm land per county FARMPRC percent of total county area which is farm land WATAC acres of water per county WATSQPRC percent of total county area which is water viii CHAPTER 1: INTRODUCTION Between December 2014 and June 2015, roughly 45 million domestic poultry died or were culled due to the highly pathogenic avian influenza (HPAI) virus infection. Culling is a mandatory practice in domestic poultry farms when one or more HPAI-infected birds are detected57. The two most affected states, (MN) and Iowa (IA), are ranked highly in agriculture product production and are located within the Mississippi migratory bird route (Figures 13, 14). Asymptomatic wild migratory birds that use farm, particularly corn, fields for shelter and food, have tested positive for HPAI virus76. The primary exposure of interest in this study was the percent of total county area which is corn grown for grain (Figures 11, 12). The U.S. Department of Agriculture (USDA) has conducted surveys of domestic poultry operation staff and investigated factors such as wind which may have contributed to the outbreak; however, it has not looked deeply into agricultural and geographic factors. Data on the number of birds on uninfected farms were unavailable; therefore, this study compared the species-specific HPAI virus infection status of MN and IA counties, in terms of infected and uninfected farms per county, to those agricultural and geographic characteristics of the MN and IA counties possibly involved in the mechanism underlying the transmission of HPAI virus from wild migratory birds to domestic birds. 1.1 Objectives This study aims to summarize the epidemiologic features of the outbreaks in MN and IA and address potential factors which may have contributed to the outbreak by conducting a casecontrol analysis. contributed to the outbreak by conducting a case-control analysis. contributed to the outbreak by conducting a case-control analysis. contributed to the outbreak by conducting a case-control analysis. contributed to the outbreak by conducting a case-control analysis. 1 1.2 Hypotheses 1. The greater the percent of total county area which is cropland, the more likely a county is to have at least one infected 1) laying hen or pullet or 2) turkey farm. 2. The greater the percent of total county area which is corn grown for grain, the more likely a county is to have at least one infected 1) laying hen or pullet or 2) turkey farm. 3. The greater the percent of total county area which is farm land, the more likely a county is to have at least one infected 1) laying hen or pullet or 2) turkey farm. 4. The greater the percent of total county area which is water, the more likely a county is to have at least one infected 1) laying hen or pullet or 2) turkey farm. 5. The greater the total number of turkey farms per county, the more likely a county is to have at least one infected turkey farm. 6. The greater the total number of laying hen farms per county, the more likely a county is to have at least one infected laying hen farm. 7. The greater the total number of pullet farms per county, the more likely a county is to have at least one infected pullet farm. 1.3 Hypothesized Transmission Mechanism MN and IA lie within the Mississippi migratory bird route. Asymptomatic wild migratory birds, particularly water birds, have tested positive for HPAI virus and are suspected to have transmitted the virus to domestic birds during the 2014 to 2015 outbreak70. Authors of a 2009 study found that there were more migratory wild birds and more bird species in harvested 2 sunflower and corn fields than in harvested soybean and small grain fields. Corn field residue “provides considerable vertical dimension” and contains kernels which wild migratory birds find appetizing and relatively easy to eat65. In 2015, IA and MN were the first and forth greatest corn grain harvesting states53. Additionally, northern Iowa contains a wetland region that sees more than half of the North American migratory wild water birds, while around 19 percent of Minnesota is wetland62. A 2012 study described several elements possibly involved in the relation of avian influenza virus to wetlands67. Wild birds rest in wetland during migration and also utilize it for feeding, breeding, and nesting68. MN and IA are also ranked highly in domestic bird production (Figures 13, 14). In the 2015 ranking of states by turkey production, MN and IA are ranked first and eighth, respectively80. In the 2016 ranking of states by laying hen inventory, IA is ranked first and MN eighth79. IA is also ranked first in the 2014 ranking of states by pullet inventory78. 1.4 Biology HPAI is an A virus that is one of four genera of the Orthomyxoviridae family, which have enveloped virions and a genome with eight single-stranded, negative-sense, ribonucleic acid (RNA) segments9. Characteristics of the viral lipoprotein envelope glycoproteins, hemagglutinin and neuraminidase, determine the subtype of the influenza A viruses. Hemagglutinin and neuraminidase antigens give infectious characteristics. Eighteen hemagglutinin subtypes and eleven neuraminidase subtypes are known to exist. All except H17N10 and H18N11, which have been found in bats alone, can infect birds. In common circulation among humans are H1N1 and H3N2. Low-pathogenic avian influenza A (LPAI) can cause mild or no disease in domestic birds. Highly-pathogenic avian influenza A (HPAI) can cause severe disease and mortality of up to ninety to one hundred percent in forty-eight hours in 3 domestic birds. H5, H7, and H9 viruses can infect both birds and humans. Nine H5 subtypes are known. The majority that are found in wild and domestic birds are LPAI, but HPAI viruses have been found on occasion, and sporadic infection in humans can cause sixty percent mortality3,11. Nine H7 subtypes are known, and the majority that are found in wild and domestic birds are LPAI. Human infection is rare and has occurred in people directly contacting infected birds. LPAI H7 infection in humans can cause mild to moderate illness, while HPAI H7 infection can cause mild to fatal illness. Nine H9 subtypes are known. All which are found in wild and domestic birds are LPAI. Human infection is rare and can cause mild illness3. The main reservoirs of avian influenza are the Anseriformes- and Charadriiformes-order waterfowl, which are ducks, geese, and swans and terns, gulls, and waders, respectively12. At the start of virus infection, several hemagglutinin glycoproteins bind to sialic acids on carbohydrate chains of glycolipids and glycoproteins on the surface of cells. After virus replication, neuraminidase removes the sialic acid from the surfaces of infected cells, and new viruses become free to infect other cells. The glycoproteins’ characteristics can change to evade the immune response during a pandemic10. In antigenic drift, mutations gradually alter the hemagglutinin and neuraminidase proteins, in some cases causing previous immune responses to be less effective. In antigenic shift, changes occur quickly when two viruses replicate in one cell and genetic material from both viruses gives rise to a new virion with a new neuraminidase protein, new hemagglutinin protein, or both a new neuraminidase protein and a new hemagglutinin protein. The two viruses can have similar or different animal hosts. If human influenza reassorts with avian influenza, avian influenza can become more transmissible to and among humans and other animals7,10. Antigenic shift can cause the previous immune responses to be completely ineffective7. 4 1.5 Disease Presentation In domestic birds, signs of HPAI can vary but include sudden death, discharge from the nares, coughing, sneezing, diarrhea, reduced appetite and coordination, swelling or discoloration of the body, and a decline in egg production. HPAI can cause up to 100 percent mortality, commonly in 48 hours, in domestic birds. Transmission among birds usually occurs through contact with airborne secretions and the feces of infected domestic and wild bird31. In humans, signs and symptoms of HPAI may include upper respiratory tract illness, myalgia, fever, and conjunctivitis, followed by lower respiratory tract illness, diarrhea, pneumonia, organ failure, encephalitis, and septic shock32. HPAI can cause up to 60 percent mortality in people33. H5N1 has caused more human cases of disease and death than other avian influenza viruses affecting humans. According to the World Health Organization, there were 258 infections and 154 deaths from H5N1 between 2004 and 2006. Most cases outside the U.S. have occurred in people improperly protected who have directly contacted infected birds or contaminated areas, been within six feet of infected birds, or travelled to a live domestic bird market. The virus is shed in the feces and oral, nasal, and optic fluids of birds31. The virus is shed in the feces and oral, nasal, and optic fluids of birds31. The virus is shed in the feces and oral, nasal, and optic fluids of birds31. 1.6 Early History Avian influenza was first recorded as “fowl plague” in northern Italy in 1878. It was referred to as Typhus exudatious gallinarum in 1880 and shown to be caused by a filterable virus in 1901. It was classified as a type A influenza virus after the discovery of the “type A influenza virus type-specific ribonucleoprotein” in 1955. In 1981, the description, “HPAI,” was used at the First International Symposium on Avian Influenza in the U.S.12 5 1.7 History in the U.S. Between 1924 and 1925, wet markets in New York City experienced the first outbreak of HPAI in the U.S. This may have resulted from containers of HPAI virus brought to the U.S. from France for a study in 1923. HPAI spread through infected poultry to New Jersey, Pennsylvania, and Connecticut. HPAI also spread to Indiana, Michigan, West Virginia, Missouri, and Illinois, due in part to contaminated trains. Live poultry travel regulations, quarantining poultry, depopulating farms, cleaning, and disinfecting helped control outbreaks which occurred again in New Jersey in 1929. Wild birds infected with avian influenza virus were discovered using serologic surveys in the U.S., Australia, and Russia in 196812. Pennsylvania chicken farms experienced a notable outbreak of low-pathogenic H5N2 virus in 1983. The same year, Pennsylvania chicken farms suffered an outbreak of a mutated, HPAI variant virus. Over 17 million chickens were culled, indirectly costing about 250 million dollars12. There was an outbreak of LPAI H7N2 virus on commercial turkey and chicken farms in Virginia in 2002. One of the culling staff developed influenza but survived and did not transmit the virus to other humans. LPAI H7N2 of unknown origin was detected in a person from New York in 2003. The person survived and did not transmit the virus to other humans. An outbreak of HPAI H5N2 virus occurred on a Texas farm with 7,000 chickens in 2004. No humans were affected. Outbreaks of HPAI H5 viruses took place on commercial poultry farms in 21 states between January and June of 201513. Roughly 45 million domestic poultry died from infection or were culled. Economic consequences included export bans on certain U.S. poultry products enforced by about 75 nations4,6. Wild birds were also found to be infected with HPAI viruses between 2014 and 2015. No humans were affected. Commercial turkey farms in Indiana experienced an outbreak of HPAI H7N8 in 2016. No humans were affected. An outbreak of 6 LPAI H7N2 virus took place among cats in New York City animal shelters in 2016. One person who had contacted the cats became infected but survived and did not transmit the virus to other humans13. The Morbidity and Mortality Weekly Reports (MMWR) on influenza activity from the Centers for Disease Control and Prevention (CDC) over the past 16 years have described emerging influenza strains and outbreaks of avian and swine influenza viruses among domestic animal farms and people. During the 2002 to 2003 influenza season, there was an outbreak of H7N2 virus in Virginia, and 4.7 million turkeys and chickens were culled. One of the culling staff became infected. During the 2003 to 2004 season, there were outbreaks of mostly H7N2 or H7N3 viruses in Delaware, Maryland, New Jersey, Pennsylvania, and Texas. There was an outbreak of H5N2 virus on one Texas poultry farm. During the 2008 to 2009 season, a person in Iowa became infected with swine influenza H1N1 virus after contacting pigs. One person in Texas and one person in South Dakota also became infected with swine influenza virus during the season. During the 2009 to 2010 season, a pandemic swine influenza H1N1 virus infection affected the U.S. The pandemic began in April 2009. There was evidence of transmission of the pandemic strains. Approximately 43 to 89 million people became infected. Additionally, there were three human cases of swine influenza H3N2 infection in Kansas, Iowa, and Minnesota. The person in Minnesota had visited a wet market, and the person in Kansas had contacted pigs. The three people survived. During the 2010 to 2011 season, there were five human cases of swine influenza H3N2 infection, one case in Wisconsin, two cases in Pennsylvania, and two cases in Minnesota. During the 2011 to 2012 season, there were 13 human cases of swine influenza H3N2 infection, two cases in Indiana, three cases in Iowa, two cases in Maine, three cases in Pennsylvania, one case in Utah, and two cases in West Virginia. Additionally, there was one 7 human case of swine influenza H1N2 infection in Minnesota and one human case of swine influenza H1N1 infection in Wisconsin. During the 2012 to 2013 season, there was one human case of swine influenza H3N2 infection each in Minnesota and Iowa. During the 2013 to 2014 season, there was one human case of swine influenza H3N2 infection in Iowa. During the 2014 to 2015 season, there were three human cases of swine influenza infection, one H3N2 infection in Wisconsin, one H1N1 infection in Minnesota, and one fatal H1N1 infection in Ohio. During the 2015 to 2016 season, there were three human cases of swine influenza infection, one H1N1 infection in Minnesota, one H3N2 infection in New Jersey, and one H1N2 infection in Minnesota12. 1.8 History in the World The World Organization for Animal Health Office International des Epizooties (OIE) has been summarizing reports of HPAI by participating countries since 2004. The program began as a response to the outbreak of H5N1 in Southeast Asia at the end of 2003. In 2004, outbreaks of H5N1 were reported by Cambodia, the People’s Republic of China, Hong Kong, Indonesia, Japan, the Republic of Korea, Laos, peninsular Malaysia, Thailand, and Vietnam. H5N2 was reported by Chinese Taipei21. In 2005, the OIE received reports of H5N1 from Cambodia, the People’s Republic of China, Croatia, Hong Kong, Indonesia, Kazakhstan, peninsular Malaysia, Mongolia, Romania, Russia, Thailand, Turkey, Ukraine, and Vietnam. H5N2 was reported by Zimbabwe. H5 was reported by the Philippines. H7N7 was reported by the Democratic People’s Republic of Korea21. In 2006, H5N1 was reported by Afghanistan, Albania, Austria, Azerbaijan, Bosnia and Herzegovina, Bulgaria, Burkina Faso, Cambodia, Cameroon, the People’s Republic of China, 8 Croatia, the Czech Republic, Cote d’Ivoire, Denmark, Djibouti, Egypt, France, Georgia, Germany, Greece, Hong Kong, Hungary, India, Indonesia, Iran, Iraq, Israel, Italy, Jordan, Kazakhstan, the Democratic People’s Republic of Korea, the Republic of Korea, Laos, peninsular Malaysia, Mongolia, Myanmar, Niger, Nigeria, Pakistan, Palestinian Autonomous Territories, Poland, Romania, Russia, Serbia and Montenegro, Slovakia, Slovenia, Spain, Sudan, Sweden, Switzerland, Thailand, Turkey, Ukraine, the United Kingdom, and Vietnam. H5N2 was reported by Zimbabwe and South Africa21. In 2007, the OIE saw reports of H5N1 by Afghanistan, Bangladesh, Benin, Cambodia, the People’s Republic of China, the Czech Republic, Cote d’Ivoire, Djibouti, Egypt, France, Germany, Ghana, Hong Kong, Hungary, India, Japan, the Republic of Korea, Kuwait, Laos, peninsular Malaysia, Myanmar, Pakistan, Poland, Romania, Russia, Saudi Arabia, Slovenia, Sudan, Thailand, Togo, Turkey, Ukraine, the United Kingdom, and Vietnam. Canada reported H7N321. In 2008, reports of H5N1 were made by Bangladesh, Benin, Cambodia, the People’s Republic of China, Egypt, Germany, Hong Kong, India, Iran, Israel, Japan, the Republic of Korea, Laos, Myanmar, Nigeria, Pakistan, Poland, Romania, Russia, Saudi Arabia, Switzerland, Thailand, Togo, Turkey, Ukraine, the United Kingdom, and Vietnam. The United Kingdom also reported H7N7. Canada reported H7N321. In 2009, H5N1 was reported by Afghanistan, Bangladesh, Cambodia, the People’s Republic of China, Germany, Hong Kong, India, Japan, Laos, Mongolia, Nepal, Nigeria, Russia, Thailand, Togo, and Vietnam. Spain reported H7N721. 9 In 2010, reports of H5N1 were made by Bangladesh, Bhutan, Bulgaria, Cambodia, the People’s Republic of China, Hong Kong, India, Israel, Japan, the Republic of Korea, Laos, Myanmar, Nepal, Romania, Russia, and Vietnam. Spain reported H7N7. Mongolia reported H517. In 2011, H5N1 was reported by Bangladesh, Cambodia, the People’s Republic of China, Hong Kong, India, Indonesia, Iran, Israel, Japan, the Republic of Korea, Mongolia, Myanmar, Nepal, Palestinian Autonomous Territories, and Vietnam. H5N2 was reported by South Africa21. In 2012, H5N1 was reported by Bangladesh, Bhutan, Cambodia, the People’s Republic of China, Chinese Taipei, Hong Kong, India, Iran, Israel, Myanmar, Nepal, and Vietnam. Australia reported H7N7. Chinese Taipei and South Africa reported H5N2. Israel reported H5. Mexico reported H7N321. In 2013, H5N1 was reported by Bangladesh, Bhutan, Cambodia, the People’s Republic of China, Hong Kong, India, the Democratic People’s Republic of Korea, Nepal, and Vietnam. H7N7 was reported by Australia and Italy. H7N2 was reported by Australia. H5N2 was reported by the People’s Republic of China, Chinese Taipei, and South Africa. H7N3 was reported by Mexico21. In 2014, H5N1 was reported by Cambodia, the People’s Republic of China, India, the Democratic People’s Republic of Korea, Libya, Nepal, Russia, and Vietnam. H7N2 was reported by Australia. H5N2 was reported by Canada, the People’s Republic of China, the U.S. of America, Chinese Taipei, and the U.S. of America. H5N8 was reported by Germany, Italy, Japan, the Republic of Korea, the Netherlands, Russia, the United Kingdom, the U.S. of America, and the People’s Republic of China. H5N6 was reported by Laos, Vietnam, and the 10 People’s Republic of China. H7N3 was reported by Mexico. H5N3 was reported by the People’s Republic of China21. In 2015, H5N1 was reported by Bhutan, Bulgaria, Burkina Faso, Cambodia, Canada, the People’s Republic of China, Cote d’Ivoire, France, Ghana, India, Iran, Israel, Kazakhstan, Libya, Myanmar, Niger, Nigeria, the Palestinian Autonomous Territories, Romania, Russia, Turkey, the U.S. of America, and Vietnam. H5N2 was reported by Canada, the People’s Republic of China, Chinese Taipei, France, and the U.S. of America. H5N8 was reported by Canada, Chinese Taipei, Germany, Hungary, Italy, Japan, the Republic of Korea, the Netherlands, Russia, Sweden, the United Kingdom, and the U.S. of America. H5N6 was reported by the People’s Republic of China, Hong Kong, Laos, and Vietnam. H5N3 was reported by Chinese Taipei. H5N9 was reported by France. H7N7 was reported by Germany and the United Kingdom. H7N3 was reported by Mexico. H5 was reported by the Palestinian Autonomous Territories21. In 2016, H5N1 was reported by Bangladesh, Bhutan, Burkina Faso, Cambodia, Cameroon, the People’s Republic of China, Cote d’Ivoire, France, Ghana, India, Iraq, Laos, Lebanon, Myanmar, Niger, Nigeria, Togo, and Vietnam. H5N6 was reported by Vietnam, the Republic of Korea, Japan, Hong Kong, and the People’s Republic of China. H5N2 was reported by the U.S. of America, France, Chinese Taipei, and the People’s Republic of China. H7N8 was reported by the U.S. of America. H5N8 was reported by the United Kingdom, Tunisia, Switzerland, Sweden, Serbia, Russia, Romania, Poland, Nigeria, Netherlands, the Republic of Korea, Israel, Iran, India, Hungary, Greece, Germany, France, Finland, Egypt, Denmark, Croatia, Chinese Taipei, Canada, and Austria. H5 was reported by Ukraine, Tunisia, Russia, Palestinian Autonomous Territories, and Bulgaria. H5N5 was reported by the Netherlands and Montenegro. 11 H7N3 was reported by Mexico. H7N7 was reported by Italy. H5N9 was reported by France. H5N3 was reported by Chinese Taipei. H7N1 was reported by Algeria21. 1.9 Events Pertinent to the 2014/2015 HPAI Outbreak in Minnesota and Iowa 1.9.1 Asia An outbreak of HPAI H5N8 virus infection occurred among poultry in South Korea and Japan between January and April of 2014. In China, outbreaks of HPAI H5N1, H5N2, H5N6, and H5N8 virus infection followed in September. South Korea had additional cases of H5N8 virus infection among its commercial poultry in late September. In November, cases of H5N8 virus infection were seen in Japan. Taiwan had outbreaks of H5N2, H5N3, and H5N8 virus infection in early 201556. 1.9.2 Europe In Germany, H5N8 virus was detected in commercial poultry and in a wild duck in early November 2014. The United Kingdom, The Netherlands, and Italy detected H5N8 virus in commercial poultry in mid-December56. 1.9.3 North America Wild waterfowl using the East Asia / Australia migration route probably brought the HPAI H5N8 virus to North America. A bird in North America became infected with both the HPAI H5N8 virus and an endemic LPAI virus, and genetic reassortment occurred to bring about the H5N1 and H5N2 viruses. British Columbia, Canada detected H5N2 virus in its commercial poultry in late November 2014. H5N8 virus was detected in captive wild birds in Washington state in early December. Oregon found H5N8 virus in backyard poultry in December. In late December, Washington state detected H5N1 virus in a wild duck. California had commercial 12 turkeys infected with H5N8 virus in mid-January 2015. In March 2015, H5N2 virus was detected in poultry in Arkansas, Kansas, Minnesota, and Missouri56. There were over 200 confirmed detections of HPAI A in wild birds and domestic poultry between December 2014 and May 2015 in the U.S., with over 40 million birds in 20 states exposed or infected27. Domestic and wild birds in the U.S. were affected by the subtypes H5N1, H5N2, and H5N8, the majority of domestic birds being affected by H5N228. By June 2015, Minnesota had the greatest number of cases followed by Iowa. Birds on all infected farms were culled. Throughout the U.S., turkeys were affected most followed by laying hens, mixed poultry, pullets, chickens in general, and, with an equal number of cases, ducks, mixed game fowl, and pheasants. After January 2016, one report of LPAI H7N8 virus infection came from a turkey farm in Indiana, where 43,000 turkeys were culled30. 1.10 Study Justification Consequences of the outbreak included lost jobs and export bans on certain U.S. poultry products. The prices of poultry products in the U.S. increased between December 2014 and the spring of 2015, and about 75 nations restricted or banned the import of these products4,6. By August 2015, 8,444 jobs and therefore 427 million dollars in wages and 145 million dollars in taxes were lost in Iowa. Veterinarians, suppliers of poultry feed, truck transporters, poultry processors, and others suffered negative economic effects5. The government granted indemnities to owners of affected farms82. Understanding what may have contributed to the outbreak is necessary to protect the U.S. agricultural industry, domestic bird welfare, and human health. As MN and IA have large poultry and swine inventories, and influenza tends to undergo antigenic changes when passing between hosts, there is no assurance that HPAI will not become more transmissible to and among humans in the future. 13 CHAPTER 2: METHODS This study was deemed as exempt from approval by the Michigan State University Institutional Review Board (IRB). 2.1 Study Population and Design The USDA Animal and Plant Health Inspection Service (APHIS) outbreak dataset provided data on the infected commercial laying hen, turkey, and pullet farms US states and counties74. Since information was inconsistently available for the number of farms that were not infected for IA and MN, the analysis had to be performed at the county level. Therefore, 186 counties were included in the analytic sample at the county level. This study compared the species-specific HPAI virus infection status of MN and IA counties to those agricultural and geographic characteristics of the MN and IA counties possibly involved in the mechanism underlying the transmission of HPAI virus from wild migratory birds to domestic birds. The 2012 USDA Census of Agriculture provided information on the agricultural traits of the counties, while the 2010 U.S. Census provided information on the geographic traits72,73. 2.2 Measures The 2012 USDA Census of Agriculture provided information on the percent of total county area which is cropland, the percent of total county area which is corn grown for grain, the percent of total county area which is farm land, the total number of turkey farms per county, the total number of laying hen farms per county, and the total number of pullet farms per county72. The 2010 U.S. Census provided information on the percent of total county area which is water73. The USDA APHIS outbreak dataset provided information on the HPAI virus infection status of the counties74. Measurement methods for the 2012 USDA Census of Agriculture are described in the Census Appendix B52. It defines a farm as an operation which produces or sells greater than 14 or equal to $1000 of agricultural products. Measurement methods for the 2010 U.S. Census are described in the technical assessment supplement75. The USDA APHIS outbreak data were collected in part via survey of farm personnel and on-farm sampling71. Data from the 2012 USDA Census of Agriculture were arranged in a PDF file. For the analysis, the data were manually imported into a Microsoft Excel spreadsheet in comma delimited values format. Data from the 2010 U.S. Census were arranged in a Microsoft Excel file. They were copied and pasted into the comma delimited value file. Data from the USDA APHIS outbreak dataset were arranged in a PDF file and manually imported into the comma delimited value file. 2.3 Analysis Logistic regression was performed with three dichotomous outcomes of interest and continuous exposure variables using SAS version 9.4 (SAS Institute, Inc., Cary, NC). Significance was considered at p<0.05 level. Original exposures were the acres of total cropland per county, the acres of corn grown for grain per county, the acres of farm land per county, the square miles of water per county, the total number of turkey farms per county, the total number of laying hen farms per county, and the total number of pullet farms per county. With the exception of the farm numbers and the square miles of water per county, exposures were converted to square miles. Next, all exposures but the farm numbers were calculated as proportions of the total county area. Finally, percents of the total county area were created for all exposures but the farm numbers, and the number of laying hen and pullet farms per county was combined. Pullets are simply young laying hens, and the individual number of laying hen and pullet farms per county was small. Final exposures were the percent of total county area which is cropland (CROPPRC), the percent of total county area which is corn grown for grain (CORNGRPRC), the percent of total county area which is farm 15 land (FARMPRC), the percent of total county area which is water (WATSQPRC), the total number of turkey farms per county (TURKFA), and the combined number of laying hen and pullet farms per county (LAYPULFA). Two species-specific dichotomous outcomes and one unspecific dichotomous outcome were used in the analysis. They were HPAIC, where 0=no laying hen or pullet farms with virus infection and 1=at least one infected laying hen or pullet farm; HPAIT, where 0=no turkey farms with virus infection and 1=at least one infected turkey farm; and HPAICOMB, where 0=no laying hen or pullet or turkey farms with virus infection and 1=at least one infected laying hen or pullet or turkey farm. To assess if the counties with the greatest number of infected farms influenced the regression coefficient estimates, one county from the MN dataset (Kandiyohi) and two counties from the IA dataset (Buena Vista, Sioux) were removed and logistic regression analyses with these counties were compared to those without these counties. These counties were infected early and frequently during the observation period. 16 CHAPTER 3: RESULTS Table 1a shows the number of infected farms for each state and for IA and MN on a county level. Table 1b shows the characteristics of the cases on the farm level. Data on the characteristics were provided by the USDA APHIS outbreak dataset. Commercial domestic poultry operations in MN and IA were selected for this study. Within the original outbreak dataset (N=232), which consisted of infected farms, 77 infected IA farms made up 33.2 percent of the sample, while 110 infected MN farms made up 47.4 percent of the sample. After applying inclusion and exclusion criteria, no MN or IA counties were excluded based on the commercial operation criterion; counties with excluded “backyard” operations still contained infected commercial operations. 71 infected IA farms made up 39.4 percent of the analytic sample, while 109 infected MN farms made up 60.6 percent of the analytic sample. Within IA, Buena Vista County, with 15 infected farms, made up the largest percentage, 21.1 percent, of the analytic sample (Figure 8). Sioux County (14 infected farms, 19.7 percent) and Sac County (8 infected farms, 11.3 percent) followed. Within MN, Kandiyohi County, with 40 infected farms, made up the largest percentage, 36.7 percent, of the analytic sample (Figure 7). Stearns County (14 infected farms, 12.8 percent) and Renville County (8 infected farms, 7.3 percent) followed. Considering MN and IA together, infected farms with between zero and 100,000 birds, 135 infected farms, made up the largest percentage, 75.8 percent, of the analytic sample. Infected farms with greater than 500,000 birds (16 infected farms, 9.0 percent) and infected farms with between greater than 100,000 birds and 200,000 birds (11 infected farms, 6.2 percent) followed. Considering MN and IA together, turkey farms, 139 infected farms, made up the largest percentage, 77.2 percent, of the analytic sample. Laying hen farms (27 infected farms, 15 percent) and pullet farms (14 infected farms, 7.8 percent) followed. Considering MN and IA 17 together, the majority of infected farms (173 infected farms, 96.1 percent) were infected with HPAI H5N2 virus in the analytic sample. The unspecific “Dangerous Contact Highly Pathogenic” (6 infected farms, 3.3 percent) and HPAI H5 (one infected farm, 0.6 percent) categories followed. Figure 1 shows the derivation of the analytic sample on the farm level through the application of inclusion and exclusion criteria. Farms in states other than MN and IA (N=45) were first excluded from the analytic sample. Farms that are “backyard,” rather than commercial, operations (N=7) were next excluded for MN and IA to give a count of 180 farms for the infected farms analytic sample. Figures 2 and 3 show the full outbreak for descriptive purposes (the outbreak took place between 4/13/2015 and 6/16/2015 in IA and between 3/5/2015 and 6/4/2015 in MN); however, in order to compare the infection rates in the two states, outbreak dates were restricted to April 2015 through June of 2015 in the analysis. Figure 4 shows that the derivation of counties to be considered in the county analytic sample. MN has 87 counties, while IA has 99 counties. Two MN counties were missing information on the acres of corn grown for grain per county, which was used to create the percent of total county area which is corn grown for grain variable. These counties were not excluded since the percent of total county area which is corn grown for grain was only one of the exposures of interest. After applying inclusion and exclusion criteria, no MN or IA counties were excluded based on the commercial operation criterion; counties with excluded “backyard” operations still contained infected commercial operations. 18 Table 2 illustrates the geographical characteristics of the MN and IA counties in the analytic sample (N=186). The 2012 USDA Census of Agriculture and the 2010 U.S. Census were used to determine the characteristics. The percentage of infected counties for the two states differed with 18.2 percent for IA and 26.4 percent for MN. Compared to infected IA counties, infected MN counties had a greater mean value for the acres of total area per county, greater mean value for the acres of water per county, smaller mean value for the acres of farm land per county, smaller mean value for the acres of total cropland per county, smaller mean value for the acres of corn grown for grain per county, greater mean value for the number of all types of poultry farms per county, greater mean value for the number of laying hen farms per county, greater mean value for the number of turkey farms per county, and greater mean value for the number of pullet farms per county. Considering laying hens, pullets, and turkeys, there were 64 HPAI-uninfected MN counties and 23 HPAI-infected MN counties (Figure 5). Table 2 shows that the infected MN counties relative to the uninfected had a smaller mean value for the acres of total area per county, smaller mean value for the acres of water per county, greater mean value for the acres of farm land per county, greater mean value for the acres of total cropland per county, greater mean value for the acres of corn grown for grain per county, roughly equal mean value for the number of all types of poultry farms per county, smaller mean value for the number of laying hen farms per county, greater mean value for the number of turkey farms per county, and smaller mean value for the number of pullet farms per county. There were 81 HPAI-uninfected IA counties and 18 HPAI-infected IA counties (Figure 6). Table 2 shows that the infected IA counties relative to the uninfected had a greater mean value for the acres of total area per county, smaller mean value for the acres of water per county, 19 greater mean value for the acres of farm land per county, greater mean value for the acres of total cropland per county, greater mean value for the acres of corn grown for grain per county, smaller mean value for the number of all types of poultry farms per county, smaller mean value for the number of laying hen farms per county, greater mean value for the number of turkey farms per county, and smaller mean value for the number of pullet farms per county. Table 4 summarizes the HPAI outbreak in MN and IA between April and June of 2015. Within IA, the greatest incidence rate per 100,000 farm-days of observation was seen among turkey farms, followed by pullet farms and laying hen farms. Within MN, the greatest incidence rate was also seen among turkey farms, followed by pullet and laying hen farms. Compared to IA, MN had a greater incidence rate in terms of turkey farms, a smaller incidence rate in terms of pullet farms, and a smaller incidence rate in terms of laying hen farms. The MN/IA incidence rate ratio was greatest for turkey farms, followed by laying hen and pullet farms. Table 4 and Figure 9 shows that, overall, the greatest cumulative incidence (CI) of virus infection was seen in turkey farms, followed by pullet and laying hen farms. The same was true for incidence density (ID). State specific: 1) MN turkey farms had greater CI and ID values than IA turkey farms; 2) IA laying hen farms had greater CI and ID values than MN laying hen farms; and 3) IA pullet farms had greater CI and ID values than MN pullet farms. The MN/IA incidence rate ratio for turkey farms was 2.1 (95 percent confidence interval: 1.4-3.1), for laying hen farms was 0.2 (0.05-0.5), and for pullet farms was 0.07 (0.009-0.5). Table 5a shows the results of the logistic regression analysis with the HPAIC outcome, where 0=no laying hen or pullet farms with HPAI virus infection and 1=at least one infected laying hen or pullet farm, based on data from all counties. In the univariate analysis for MN, the percent of total county area which is farm land was significant at p<0.05 and the percent of total 20 county area which is cropland was significant at p<0.1. The odds ratio estimates were 7.8 and 2.6, respectively. In the univariate analysis for IA, the percent of total county area which is cropland and the percent of total county area which is farm land were significant at p<0.05. The odds ratio estimates were 1.9 and 2.5, respectively. In the multivariate analysis for MN, no exposures remained significant after adjustment for the other exposures in the model due to high correlation between exposures. The same was true for IA. Table 5aa shows the results of the logistic regression analysis following the removal of the counties with the greatest number of infected farms: Kandiyohi (MN), Buena Vista (IA), and Sioux (IA). The HPAIC outcome, where 0=no laying hen or pullet farms with virus infection and 1=at least one infected laying hen or pullet farm, is used. In the univariate analysis for MN, the percent of total county area which is cropland was significant at p<0.1, and the percent of total county area which is farm land was significant at p<0.05. In the univariate analysis for IA, the percent of total county area which is cropland and the percent of total county area which is farm land were significant at p<0.1. In the multivariate analysis for MN, no exposures remained significant after adjustment for the other exposures in the model due to high correlation between exposures. The same was true for IA. Table 5b shows the results of the logistic regression analysis with the HPAIT outcome, where 0=no turkey farms with HPAI virus infection and 1=at least one infected turkey farm, based on data from all counties. In the univariate analysis for MN, the number of turkey farms per county was significant at p<0.05; the percent of total county area which is cropland, the percent of total county area which is corn, and the percent of total county area which is farm land were significant at p<0.01; and the percent of total county area which is water was significant at p<0.1. In the univariate analysis for IA, the number of turkey farms per county was significant at 21 p<0.01, and the combined number of laying hen and pullet farms per county, the percent of total county area which is cropland, the corn variable, and the percent of total county area which is farm land were significant at p<0.05. In the multivariate analysis for MN, for the outcome HPAIT (Table 5b), during the model building process, only two exposures retained their statistical significance when adjusted for each other: the number of turkey farms per county at p<0.05 and the corn variable at p<0.01. The odds ratio estimates were 1.1 and 2.0, respectively. In the multivariate analysis for IA, the model building process yielded three models where mutually adjusted exposures remained significant: 1) the number of turkey farms per county, the corn variable; 2) the combined number of laying hen and pullet farms per county, the corn variable; and 3) the number of turkey farms per county, the combined number of laying hen and pullet farms per county, the corn variable. Estimates were significant 1) at p<0.01 for the number of turkey farms per county and at p<0.05 for the corn variable; 2) at p<0.05 for the combined number of laying hen and pullet farms per county and the corn variable; and 3) at p<0.05 for the number of turkey farms per county and the combined number of laying hen and pullet farms per county and at p<0.1 for the corn variable. The odds ratio estimates were 1) 1.3 and 3.3; 2) 0.9 and 5.1; and 3) 1.4, 0.9, and 4.1, respectively. Table 5bb shows the results of the logistic regression analysis following the removal of the counties with the greatest number of infected farms: Kandiyohi (MN), Buena Vista (IA), and Sioux (IA). The HPAIT outcome, where 0=no laying hen or pullet farms with HPAI virus infection and 1=at least one infected laying hen or pullet farm, is used. In the univariate analysis for MN, the percent of total county area which is water was significant at p<0.1; the number of turkey farms per county was significant at p<0.05; and the percent of total county area which is 22 cropland, the percent of total county area which is corn grown for grain, and the percent of total county area which is farm land were significant at p<0.01. In the univariate analysis for IA, the combined number of laying hen and pullet farms per county and the percent of total county area which is farm land were significant at p<0.1. The number of turkey farms per county, the percent of total county area which is cropland, and the percent of total county area which is corn grown for grain were significant at p<0.05. In the multivariate analysis for MN, a model with the number of turkey farms per county and the percent of total county area which is corn grown for grain was possible. Estimates were significant at p<0.05 for the number of turkey farms per county and at p<0.01 for the percent of total county area which is corn grown for grain. Odds ratio estimates were 1.1 and 2.0, respectively. Table 5c shows the results of the logistic regression analysis with the HPAICOMB outcome, where 0=no laying hen or pullet or turkey farms with HPAI virus infection and 1=at least one infected laying hen or pullet or turkey farm, based on data from all counties. In the univariate analysis for MN, the number of turkey farms per county and the percent of total county area which is water were significant at p<0.1; the percent of total county area which is cropland and the percent of total county area which is farm land were significant at p<0.01; and the corn variable was significant at p<0.001. In the univariate analysis for IA, the combined number of laying hen and pullet farms per county and the percent of total county area which is water were significant at p<0.1; the percent of total county area which is cropland and the percent of total county area which is farm land were significant at p<0.01; and the corn variable was significant at p<0.05. In the multivariate analysis for MN, for the outcome HPAICOMB (Table 5c), during the model building process, only two exposures retained their statistical significance when adjusted 23 for each other: the number of turkey farms per county at p<0.05 and the corn variable at p<0.001. The odds ratio estimates were 1.1 and 2.1, respectively. In the multivariate analysis for IA, two exposures retained their statistical significance when adjusted for each other: the combined number of laying hen and pullet farms per county at p<0.1 and the corn variable at p<0.05. The odds ratio estimates were 1.0 and 1.8, respectively. Table 5cc shows the results of the logistic regression analysis following the removal of the counties with the greatest number of infected farms: Kandiyohi (MN), Buena Vista (IA), and Sioux (IA). The HPAICOMB outcome, where 0=no laying hen or pullet or turkey farms with HPAI virus infection and 1=at least one infected laying hen or pullet or turkey farm, is used. In the univariate analysis for MN, the number of turkey farms per county and the percent of total county area which is water were significant at p<0.1; the percent of total county area which is cropland and the percent of total county area which is farm land were significant at p<0.01; and the percent of total county area which is corn grown for grain was significant at p<0.001. In the univariate analysis for IA, the percent of total county area which is water was significant at p<0.1; the combined number of laying hen and pullet farms per county and the percent of total county area which is corn grown for grain were significant at p<0.05; and the percent of total county area which is cropland and the percent of total county area which is farm land were significant at p<0.01. In the multivariate analysis for MN, for the outcome HPAICOMB (Table 5cc), during the model building process, only two exposures retained their statistical significance when adjusted for each other: the number of turkey farms per county at p<0.1 and the percent of total county area which is corn grown for grain at p<0.001. Odds ratio estimates were 1.1 and 2.1, respectively. In the multivariate analysis for IA, only two exposures retained their statistical 24 significance when adjusted for each other: the combined number of laying hen and pullet farms per county at p<0.05 and the percent of total county area which is corn grown for grain at p<0.05. Odds ratio estimates were 1.0 and 1.8, respectively. A review of the univariate and multivariate analyses showed that the corn variable is the main exposure which was most consistently statistically significant for all three outcome variables. Following is a summary of the results for the corn variable as observed in Tables 5a through 5cc. Table 5a shows that, in the unadjusted model for MN and the HPAIC outcome, for a 10 percent increase in the percent of total county area which is corn, the odds that a county has at least one infected laying hen or pullet farm versus having no laying hen or pullet farms with virus infection increase by a factor of two. This is not significant but shows a trend toward significance. After the counties with the greatest number of infected farms were removed in Table 5aa, unadjusted, every 10 percent increase in the corn variable is associated with a 101 percent increase in the odds that a county has at least one infected laying hen or pullet farm. Again, this shows a trend toward significance but is not significant. In the unadjusted model for IA and HPAIC as shown in Table 5a, for every 10 percent increase in the corn variable, the odds that a county has at least one infected laying hen or pullet farm versus having no laying hen or pullet farms with virus infection increase by a factor of 1.4. This is not significant. After the aforementioned counties were removed in Table 5aa, unadjusted, every 10 percent increase in the corn variable is associated with a 34 percent increase in the odds that a county has at least one infected laying hen or pullet farm. This is not significant. In the unadjusted model for MN and HPAIT in Table 5b, for every 10 percent increase corn, the odds that a county has at least one infected turkey farm versus having no turkey farms with virus infection increase by a factor of 1.8. This is significant. Adjusting for the number of 25 turkey farms per county, for every 10 percent increase in corn, holding the number of turkey farms per county fixed, the odds that a county has at least one infected turkey farm versus having no turkey farms with virus infection increase by a factor of two. Significance has increased with adjustment. After the aforementioned counties were removed in Table 5bb, unadjusted, every 10 percent increase in corn is associated with an 82 percent increase in the odds that a county has at least one infected turkey farm. This is significant. Adjusting for the number of turkey farms per county, every 10 percent increase in corn is associated with a 99 percent increase in the odds that a county has at least one infected turkey farm. The significance remains approximately the same with adjustment. In the unadjusted model for IA and HPAIT as shown in Table 5b, for every 10 percent increase in corn, the odds that a county has at least one infected turkey farm versus having no turkey farms with virus infection increase by a factor of 3.2. This is significant. Adjusting for the number of turkey farms per county, for every 10 percent increase in corn, holding the number of turkey farms per county fixed, the odds that a county has at least one infected turkey farm versus having no turkey farms with virus infection increase by a factor of 3.3. This coefficient for corn becomes less significant with this adjustment. Adjusting for the combined number of laying hen and pullet farms per county, for every 10 percent increase in corn, holding the combined number of laying hen and pullet farms per county fixed, the odds that a county has at least one infected turkey farm versus having no turkey farms with virus infection increase by a factor of 5.1. The significance remains approximately the same. Adjusting for the number of turkey farms per county and the combined number of laying hen and pullet farms per county, for every 10 percent increase in corn, holding the number of turkey farms per county and the combined number of laying hen and pullet farms per county fixed, the odds that a county has at least one infected turkey farm versus having no turkey farms with virus infection increase by a 26 factor of 4.1. Significance of the coefficient for corn decreases with this adjustment. After the counties were removed in Table 5bb, unadjusted, every 10 percent increase in corn is associated with a 223 percent increase in the odds that a county has at least one infected turkey farm. This is significant. Table 5c shows that, in the unadjusted model for MN and HPAICOMB, for every 10 percent increase in corn, the odds that a county has at least one infected laying hen or pullet or turkey farm versus having no laying hen or pullet or turkey farms with virus infection increase by a factor of 1.9. This is highly significant. Adjusting for the number of turkey farms per county, for every 10 percent increase in corn, holding the number of turkey farms per county fixed, the odds that a county has at least one infected laying hen or pullet or turkey farm versus having no laying hen or pullet or turkey farms with virus infection increase by a factor of 2.1. Significance decreases slightly with this adjustment. After the aforementioned counties were removed in Table 5cc, unadjusted, every 10 percent increase in corn is associated with a 92 percent increase in the odds that a county has at least one infected laying hen or pullet or turkey farm. This is highly significant. Adjusting for the number of turkey farms per county, every 10 percent increase in corn is associated with a 105 percent increase in the odds that a county has at least one infected laying hen or pullet or turkey farm. Significance decreases slightly with this adjustment. As shown in Table 5c, in the unadjusted model for IA and HPAICOMB, for every 10 percent increase in corn, the odds that a county has at least one infected laying hen or pullet or turkey farm versus having no laying hen or pullet or turkey farms with virus infection increase by a factor of 1.9. This is significant. Adjusting for the combined number of laying hen and pullet farms per county, for every 10 percent increase in corn, holding the combined number of laying hen and pullet farms per county fixed, the odds that a county has at least one infected 27 laying hen or pullet or turkey farm versus having no laying hen or pullet or turkey farms with virus infection increase by a factor of 1.8. Significance remains unchanged. After the counties were removed in Table 5cc, unadjusted, every 10 percent increase in corn is associated with an 81 percent increase in the odds that a county has at least one infected laying hen or pullet or turkey farm. This is significant. Adjusting for the combined number of laying hen and pullet farms per county, every 10 percent increase in corn is associated with an 80 percent increase in the odds that a county has at least one infected laying hen or pullet or turkey farm. Significance remains unchanged. The first three counties to be infected in MN were Pope, Lac Qui Parle, and Stearns (Figure 10). Pope is ranked 34th out of 87 counties for the acres of corn grown for grain per county, while Lac Qui Parle is ranked 10th, and Stearns is ranked fifth. For the square miles of water per county, Pope is ranked 19th, Lac Qui Parle 53rd, and Stearns 20th. For the number of turkey farms per county, Pope is ranked 44th, Lac Qui Parle 71st, and Stearns first. For the number of laying hen farms per county, Pope is ranked 55th, Lac Qui Parle 77th, and Stearns first. Finally, for the number of pullet farms per county, Pope is ranked 43rd, Lac Qui Parle 60th, and Stearns 15th. Respectively, Pope, Lac Qui Parle, and Stearns accounted for about two, one, and 13 percent of all HPAI virus-infected commercial poultry farms in MN. The first three counties to be infected in IA were Buena Vista, Osceola, and Sac. Buena Vista is ranked 28th out of 99 counties for the acres of corn grown for grain per county, while Osceola is ranked 67th, and Sac is ranked 25th. For the square miles of water per county, Buena Vista is ranked 25th, Osceola 80th, and Sac 34th. For the number of turkey farms per county, Buena Vista is ranked first, Osceola 38th, and Sac second. For the number of laying hen farms per county, Buena Vista is ranked 97th, Osceola 90th, and Sac 88th. Finally, for the number of 28 pullet farms per county, Buena Vista is ranked 53rd, Osceola 76th, and Sac 40th. Respectively, Buena Vista, Osceola, and Sac accounted for about 21, four, and 11 percent of HPAI virusinfected commercial poultry farms in IA. Counties with the greatest number of HPAI virus-infected commercial poultry farms in MN were Kandiyohi, Stearns, and Meeker (Table 6). Kandiyohi is ranked 14th out of 87 counties for the acres of corn grown for grain per county, while Stearns is ranked fifth, and Meeker is ranked 28th. For the square miles of water per county, Kandiyohi is ranked 15th, Stearns 20th, and Meeker 23rd. For the number of turkey farms per county, Kandiyohi is ranked fourth, Stearns first, and Meeker 15th. For the number of laying hen farms per county, Kandiyohi is ranked 35th, Stearns first, and Meeker 14th. Finally, for the number of pullet farms per county, Kandiyohi is ranked 30th, Stearns 15th, and Meeker 27th. Respectively, Kandiyohi, Stearns, and Meeker accounted for about 37, 13, and nine percent of all HPAI virus-infected commercial poultry farms in MN. Counties with the greatest number of HPAI virus-infected commercial poultry farms in IA were Buena Vista, Sioux, and Sac. Buena Vista is ranked 28th out of 99 counties for the acres of corn grown for grain per county, while Sioux is ranked 6th, and Sac is ranked 25th. For the square miles of water per county, Buena Vista is ranked 11th, Sioux 84th, and Sac 81st. For the number of turkey farms per county, Buena Vista is ranked first, Sioux 69th, and Sac second. For the number of laying hen farms per county, Buena Vista is ranked 97th, Sioux eighth, and Sac 88th. Finally, for the number of pullet farms per county, Buena Vista is ranked 53rd, Sioux third, and Sac 40th. Respectively, Buena Vista, Sioux, and Sac accounted for about 21, 20, and 11 percent of all HPAI virus-infected commercial poultry farms in IA. 29 CHAPTER 4: DISCUSSION This study summarized the epidemiologic features of the outbreaks in MN and IA and, using a case-control logistic regression analysis, addressed factors which may have potentially contributed to the outbreak. In MN, it was observed that one county had about a three-fold difference in the number of infected farms relative to other counties. In IA, two counties had about a two-fold difference in the number of infected farms compared to other counties. These three counties not only had the greatest number of infected farms but were also infected early and frequently during the observation period. Therefore, analyses with all counties and removing these three counties were performed to determine if these potentially influential observations distorted the regression estimation. The three dichotomous infection status outcomes used involved turkey farms; laying hen and pullet farms combined; and laying hen, pullet, and turkey farms combined. Prior to removing the counties with the greatest number of infected farms, in both the unadjusted and adjusted analyses, associations were found between the percent of the total county area which is corn and the HPAIT outcome for MN (OR: 1.8, p=0.002, 95% CI: 1.32.7; adjOR: 2.0; p=0.001, 95% CI: 1.3-3.1), between the corn variable and HPAIT for IA (OR: 3.2, p=0.03, 95% CI: 1.1-9.2; adjOR: 3.3, p=0.04, 95% CI: 1.1-10.0; adjOR: 5.1, p=0.04, 95% CI: 1.1-23.3; adjOR: 4.1, p=0.07, 95% CI: 0.9-18.3), between the corn variable and HPAICOMB for MN (OR: 1.9, p=0.0007, 95% CI: 1.3-2.8; adjOR: 2.1, p=0.0006, 95% CI: 1.4-3.2), and between the corn variable and HPAICOMB for IA (OR: 1.9, p=0.02, 95% CI: 1.1-3.1; adjOR: 1.8, p=0.02, 95% CI: 1.1-3.1; adjOR: 1.6, p=0.08, 95% CI: 0.9-2.7). For the HPAIC outcome, only univariate analyses were possible. For MN, the percent of total county area which is farm land was significant at p<0.05 (OR: 7.8, p=0.04, 95% CI: 1.1-56.2), and the percent of total county area which is cropland was significant at p<0.1 (OR: 2.6, p=0.09, 95% CI: 0.9-7.5). For 30 IA, the percent of total county area which is cropland (OR: 1.9, p=0.03, 95% CI: 1.1-3.3) and the percent of total county area which is farm land (OR: 2.5, p=0.03, 95% CI: 1.1-5.5) were both significant at p<0.05. It was predicted that the removal of the counties with the highest number of infected farms would attenuate the magnitude and significance of the regression coefficient estimates. After removal, associations were found between the corn variable and HPAIT for MN (OR: 1.8, p=0.002, 95% CI: 1.2-2.7; adjOR: 2.0, p=0.002, 95% CI: 1.3-3.1), between the corn variable and HPAIT for IA (OR: 3.2, p=0.04, 95% CI: 1.1-9.9), between the corn variable and HPAICOMB for MN (OR: 1.9, p=0.0009, 95% CI: 1.3-2.8; adjOR: 2.1, p=0.0007, 95% CI: 1.4-3.1), and between the corn variable and HPAICOMB for IA (OR: 1.8, p=0.03, 95% CI: 1.1-3.1; adjOR: 1.8, p=0.03, 95% CI: 1.1-3.1). Again, only univariate analyses were possible for the HPAIC outcome. In the univariate analysis for MN, the percent of total county area which is cropland was significant at p<0.1(OR: 2.5, p=0.09, 95% CI: 0.9-7.4), and the percent of total county area which is farm land was significant at p<0.05 (OR: 7.7, p=0.04, 95% CI: 1.1-56.0). In the univariate analysis for IA, the percent of total county area which is cropland (OR: 1.6, p=0.08, 95% CI: 0.9-2.9) and the percent of total county area which is farm land (OR: 2.1, p=0.07, 95% CI: 0.9-4.6) were both significant at p<0.1. The magnitude and significance of the regression coefficient estimates were generally slightly attenuated by the removal of Kandiyohi County from the MN dataset and of Buena Vista and Sioux Counties from the IA dataset; however, these observations did not notably affect estimates. The percent of the total county area which is corn grown for grain was the most consistently significant exposure across all three outcome variables. The percent of the total county area which is corn grown for grain was the most consistently significant exposure across all three outcome variables. 31 4.1 Temporal Distribution Generally, farms in central MN became infected first, followed by farms in southern MN, northern IA, and central IA (Figure 10). Few farms in southern IA became infected. Most corn appears to be produced in central and southern MN and in northern and central IA (Figures 11, 12). Additionally, domestic poultry production is concentrated in central MN (Figures 13, 14). Spatially, the counties with the greatest number of HPAI virus-infected commercial poultry farms in MN; Kandiyohi, Stearns, and Meeker; all share borders with one another and accounted for about 59 percent of all HPAI virus-infected commercial poultry farms in MN. For the counties with the greatest number of HPAI virus-infected commercial poultry farms in IA, Buena Vista and Sac share a border, and Buena Vista is one county away from Sioux. These counties accounted for about 52 percent of all HPAI virus-infected commercial poultry farms in IA. Being highly-ranked producers of domestic birds, perhaps the Kandiyohi, Stearns, and Meeker region distributed the virus via supply trucks, equipment or workers moving on and off the farms and through the transport of live birds on major interstate highways to the poultry processing plants. Buena Vista, Sac, and Sioux Counties are also high-ranking domestic bird producers and may have distributed the virus in this manner. Domestic poultry production and the attraction of migrating wild bird carriers of HPAI virus to areas with high corn production may have impacted the temporal distribution of infected farms. The outbreak appears to be correlated with the major interstate highways leading to the top poultry processing plants and with popular wetland areas for viewing water birds. Several factors related to both poultry production (concentrated poultry production, shipping of live birds to poultry processing plants) and to migratory water birds (corn production, proximity to wet areas) may have played a role in the outbreak. Migratory water birds, which carry the HPAI virus 32 and are attracted to corn-growing land and water, may have infected domestic poultry in both Canada and the U.S. Movement of equipment to and from regions with concentrated poultry production and the shipping of live birds to poultry processing plants may have contributed to the spread of HPAI. Results may be generalizable to U.S. counties similar in terms of geography, crop and livestock production, and HPAI susceptibility (e.g. proximity to wild bird migration routes). In terms of HPAI management, if the associations are true, considering these geographic and agricultural factors as potential predictors of HPAI virus spread during an outbreak may be the only practical use of these results. Any modification of agricultural land or wetland to control HPAI is not a practical option. In particular, modification of wetland may cause infected birds to scatter64. Wetland also lessens flood damage, filters water, and serves as a habitat for several species69. The Ramsar Handbook for the Wise Use of Wetlands recommends consulting zoonoses epidemiology, wildlife, ornithological, and wetland specialists when combating HPAI virus infection64. The environment is one aspect considered when determining the extent of an infection control region83. Perhaps HPAI virus infection responders should give more consideration to an environmental factor outside of establishing an infection control area. Accounting for this factor outside of establishing an infection control area may contribute to controlling the outbreak quickly and mitigating its impact. The transmission ability of current HPAI strains among humans is still not a concern in the U.S., but outbreaks among domestic poultry should be well documented and analyzed in order to minimize the opportunity for strain mixing and mutation, especially in agricultural areas producing large numbers of different domestic animal species. IA and MN were ranked first and third, respectively, in the 2015 ranking of states by swine inventory (Figure 15)77. Pigs are 33 susceptible to multiple influenza strains, including avian and human influenza31,81. Different strains can reassort and give rise to strains which are more transmissible to and among humans and other animals4,31,60. The CDC MMWR has documented outbreaks of both swine and avian influenza viruses in Minnesota and Iowa over the years and described the transmission ability of the 2009 pandemic swine influenza virus among humans. The potential for the mutation and reassortment of influenza and the noted resistance of influenza to antiviral drugs continues to be alarming66. 4.2 Strengths The USDA has examined the U.S. as a whole with regard to the outbreak. It has conducted surveys of domestic poultry operation staff and investigated factors such as wind which may have contributed to the outbreak; however, it has not specifically focused on agricultural and geographic factors, which have a role in HPAI virus transmission. This study evaluated the possible association between geographic and agricultural traits of counties in the two most affected states and county infection status. All exposure variables were continuous in the analysis. 4.3 Future Research Future studies should examine the environmental portion of the HPAI virus transmission pathway more closely and clarify the size of the role of geographic and agricultural factors in transmission. Studies should be conducted on the farm level so that more specific inferences can be made. With more complete data, an outcome involving the proportion of farms in a county that are infected should be used in linear regression. 34 APPENDICES 35 APPENDIX A: Tables 36 Table 1a: Distribution of infected farms among states and individual MN and IA counties in the USDA APHIS “HPAI 2014-2015 Infected Premises” (state, county, number of birds, flock type, species, HPAI strain confirmed) dataset (N=232 infected farms) and the analytic sample (N=180 infected farms) USDA APHIS Dataset (Max N=232 infected farms) N (%) State Arkansas California Iowa Idaho Indiana Kansas Minnesota Missouri Montana North Dakota Nebraska Oregon South Dakota Washington Wisconsin 1 2 77 1 1 1 110 3 1 2 6 2 10 5 10 0.43 0.86 33.19 0.43 0.43 0.43 47.41 1.29 0.43 0.86 2.59 0.86 4.31 2.16 4.31 Infected Farms Analytic Sample (Max N=180 infected farms) N (%) 0 0 71 0 0 0 109 0 0 0 0 0 0 0 0 37 0 0 39.44 0 0 0 60.56 0 0 0 0 0 0 0 0 Table 1a (cont’d) USDA APHIS Dataset Iowa and Minnesota (Max N=187 infected farms) N (%) County Iowa: Adair Buena Vista Calhoun Cherokee Clay Hamilton Kossuth Lyon Madison O’Brien Osceola Palo Alto Plymouth Pocahontas Sac Sioux Webster Wright 1 15 2 5 2 4 1 1 1 3 4 1 1 3 8 18 1 6 1.30 19.48 2.60 6.49 2.60 5.19 1.30 1.30 1.30 3.90 5.19 1.30 1.30 3.90 10.39 23.38 1.30 7.79 Infected Farms Analytic Sample Excluding Backyard Farms (Max N=180 infected farms) N (%) 1 15 2 5 2 4 1 1 1 2 3 1 1 3 8 14 1 6 38 1.41 21.13 2.82 7.04 2.82 5.63 1.41 1.41 1.41 2.82 4.22 1.41 1.41 4.22 11.27 19.72 1.41 8.45 Table 1a (cont’d) USDA APHIS Dataset Iowa and Minnesota (Max N=187 infected farms) N (%) County Minnesota: Blue Earth Brown Chippewa Clay Cottonwood Kandiyohi Lac Qui Parle Le Sueur Lyon Meeker Nicollet Nobles Otter Tail Pipestone Pope Redwood Renville Roseau Stearns Steele Swift Wadena Watonwan 1 5 1 1 2 40 1 1 1 10 1 1 4 2 2 4 8 1 14 1 7 1 1 0.91 4.55 0.91 0.91 1.82 36.36 0.91 0.91 0.91 9.09 0.91 0.91 3.64 1.82 1.82 3.64 7.27 0.91 12.73 0.91 6.36 0.91 0.91 Infected Farms Analytic Sample Excluding Backyard Farms (Max N=180 infected farms) N (%) 1 5 1 1 2 40 1 1 1 10 1 1 4 1 2 4 8 1 14 1 7 1 1 39 0.92 4.59 0.92 0.92 1.83 36.70 0.92 0.92 0.92 9.17 0.92 0.92 3.67 0.92 1.83 3.67 7.34 0.92 12.84 0.92 6.42 0.92 0.92 Table 1b: Characteristics of infected farms in MN and IA counties in the USDA APHIS “HPAI 2014-2015 Infected Premises” (state, county, number of birds, flock type, species, HPAI strain confirmed) dataset (N=232 infected farms) and the analytic sample (N=180 infected farms) USDA APHIS Dataset (Max N=232 infected farms) N (%) Flock Type commercial backyard Number of Birds 0-100,000 >100,000-200,000 >200,000-300,000 >300,000-400,000 >400,000-500,000 >500,000 Species laying hen turkey pullet other HPAI Strain H5N2 H5 DCHP Infected Farms Analytic Sample Excluding Backyard Farms (Max N=180 infected farms) N (%) 180 7 96.26 3.74 180 0 100 0 142 11 10 4 2 16 76.76 5.95 5.41 2.16 1.08 8.65 135 11 10 4 2 16 75.84 6.18 5.62 2.25 1.12 8.99 27 139 14 7 14.44 74.33 7.49 3.74 27 139 14 0 15 77.22 7.78 0 179 2 6 95.72 1.07 3.21 173 1 6 96.11 0.56 3.33 USDA=United States Department of Agriculture APHIS=Animal and Plant Health Inspection Service HPAI=highly pathogenic avian influenza DCHP=Dangerous Contact Highly Pathogenic 40 41 42 Table 3: Explanation of acronyms USDA APHIS HPAI LPAI HPAIT HPAIC HPAICOMB ALLPOUFA TURKFA LAYFA PULFA LAYPULFA TOTAREAC CROPAC CROPPRC CORNGRAINAC CORNGRPRC FARMAC FARMPRC WATAC WATSQPRC United States Department of Agriculture Animal and Plant Health Inspection Service highly pathogenic avian influenza low pathogenic avian influenza highly pathogenic avian influenza infection status, where 0=no turkey farms with virus infection and 1=at least one infected turkey farm highly pathogenic avian influenza infection status, where 0=no laying hen or pullet farms with virus infection and 1=at least one infected laying hen or pullet farm highly pathogenic avian influenza infection status, where 0=no laying hen or pullet or turkey farms with virus infection and 1=at least one infected laying hen or pullet or turkey farm number of all types of poultry farms per county number of turkey farms per county number of laying hen farms per county number of pullet farms per county combined number of laying hen and pullet farms per county acres of total area per county acres of total cropland per county percent of total county area which is cropland acres of corn grown for grain per county percent of total county area which is corn grown for grain acres of farm land per county percent of total county area which is farm land acres of water per county percent of total county area which is water 43 Table 4: Description of the highly pathogenic avian influenza outbreak in Minnesota and Iowa between April and June of 2015 44 Table 5a: Results of logistic regression with HPAIC outcome based on data from all counties Outcome (HPAIC): 0 = no laying hen or pullet farms with HPAI virus infection 1 = at least one infected laying hen or pullet farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 MN univariable TURKFA 0.065 0.050 1.67 0.20 1.07 0.97, 1.18 LAYPULFA 0.011 0.0077 2.02 0.16 1.01 0.99, 1.03 CROPPRC 0.094 0.055 2.88 0.090† 2.55a 0.87, 7.50 CORNGRPRC 0.070 0.044 2.52 0.11 2.01a 0.85, 4.78 * FARMPRC 0.21 0.10 4.15 0.042 7.79a 1.08, 56.17 0.0020, WATSQPRC -0.20 0.22 0.78 0.38 0.14a 11.07 multivariable N/A IA univariable TURKFA LAYPULFA CROPPRC CORNGRPRC FARMPRC -0.019 -0.011 0.062 0.035 0.091 0.086 0.014 0.029 0.027 0.041 0.050 0.53 4.53 1.68 5.04 0.82 0.47 0.033* 0.19 0.025* WATSQPRC -1.59 1.02 2.43 0.12 0.98 0.99 1.86a 1.42a 2.49a <0.0 010a 0.83, 1.16 0.96, 1.02 1.05, 3.28 0.84, 2.43 1.12, 5.51 <0.0010, 59.54 multivariable N/A *N/A indicates that no variables remained statistically significant after mutual adjustment in the multivariate model 45 Table 5a (cont’d) 46 Table 5aa: Results of logistic regression with HPAIC outcome following removal of Kandiyohi (MN), Buena Vista (IA), and Sioux (IA) Counties Outcome (HPAIC): 0 = no laying hen or pullet farms with HPAI virus infection 1 = at least one infected laying hen or pullet farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 MN univariable TURKFA 0.068 0.050 1.86 0.17 1.07 0.97, 1.18 LAYPULFA 0.011 0.0077 2.00 0.16 1.01 0.996, 1.03 CROPPRC 0.092 0.055 2.82 0.093† 2.52a 0.86, 7.39 CORNGRPRC 0.070 0.044 2.53 0.11 2.01a 0.85, 4.72 * a FARMPRC 0.20 0.10 4.07 0.044 7.70 1.06, 55.96 WATSQPRC -0.19 0.22 0.74 0.39 0.15a 0.002, 11.52 multivariable N/A IA univariable TURKFA LAYPULFA CROPPRC CORNGRPRC FARMPRC WATSQPRC -0.17 -0.018 0.050 0.029 0.073 -2.22 0.13 0.018 0.029 0.028 0.040 1.42 1.57 1.06 2.98 1.03 3.28 2.45 0.21 0.30 0.084† 0.31 0.070† 0.12 0.85 0.98 1.64a 1.34a 2.08a <0.0010a 0.65, 1.10 0.95, 1.02 0.94, 2.88 0.76, 2.33 0.94, 4.57 <0.0010, 268.50 multivariable N/A *N/A indicates that no variables remained statistically significant after mutual adjustment in the multivariate model 47 Table 5b: Results of logistic regression with HPAIT outcome based on data from all counties Outcome (HPAIT): 0 = no turkey farms with HPAI virus infection 1 = at least one infected turkey farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 MN univariable TURKFA 0.087 0.038 5.13 0.024* 1.09 1.01, 1.18 LAYPULFA -0.0021 0.0055 0.14 0.71 0.99 0.99, 1.01 CROPPRC 0.036 0.012 8.30 0.0040◊ 1.43a 1.12, 1.82 CORNGRPRC 0.060 0.019 9.67 0.0019◊ 1.83a 1.25, 2.67 FARMPRC 0.042 0.015 7.68 0.0056◊ 1.53a 1.13, 2.06 0.051, WATSQPRC -0.14 0.082 2.80 0.094† 0.25a 1.26 multivariable TURKFA 0.10 0.041 6.24 0.013* 1.11 1.02, 1.20 CORNGRPRC 0.070 0.022 10.20 0.0014◊ 2.02a 1.31, 3.10 48 Table 5b (cont’d) Outcome (HPAIT): 0 = no turkey farms with HPAI virus infection 1 = at least one infected turkey farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 IA univariable TURKFA 0.25 0.086 8.29 0.0040◊ 1.28 1.08, 1.52 * LAYPULFA -0.11 0.047 5.06 0.025 0.90 0.82, 0.99 1.19, * a CROPPRC 0.15 0.069 4.88 0.027 4.55 17.43 * a CORNGRPRC 0.12 0.054 4.74 0.029 3.21 1.12, 9.17 1.10, * a FARMPRC 0.14 0.068 4.42 0.035 4.16 15.71 <0.0010, WATSQPRC -0.51 0.65 0.61 0.43 0.0060a >999.99 multivariable TURKFA 0.36 0.16 5.22 0.022* 1.43 1.05, 1.94 LAYPULFA -0.15 0.070 4.85 0.028* 0.86 0.75, 0.98 0.90, CORNGRPRC 0.14 0.077 3.30 0.069† 4.05a 18.34 TURKFA CORNGRPRC 0.23 0.12 0.091 0.057 6.69 4.30 0.0097◊ 0.038* 1.26 3.27a 1.06, 1.51 1.07, 9.99 LAYPULFA -0.14 0.065 4.72 0.030* 0.87 CORNGRPRC 0.16 0.077 4.45 0.035* 5.11a 0.77, 0.99 1.12, 23.29 49 Table 5bb: Results of logistic regression with HPAIT outcome following removal of Kandiyohi (MN), Buena Vista (IA), and Sioux (IA) Counties Outcome (HPAIT): 0 = no turkey farms with HPAI virus infection 1 = at least one infected turkey farm ML Wa ML Model Variable Estimat ld Pr>X2 OR CL SE 2 e X MN univariable TURKFA 0.076 0.038 4.00 0.046* 1.08 1.00, 1.16 LAYPULFA -0.0021 0.0056 0.15 0.70 0.99 0.99, 1.01 CROPPRC 0.036 0.013 8.07 0.0045◊ 1.43a 1.12, 1.83 CORNGRPRC 0.060 0.020 9.31 0.0023◊ 1.82a 1.24, 2.67 FARMPRC 0.043 0.016 7.44 0.0064◊ 1.53a 1.13, 2.08 0.029, WATSQPRC -0.17 0.094 3.25 0.072† 0.18a 1.16 multivariable TURKFA 0.094 0.041 5.15 0.023* 1.10 1.01, 1.19 CORNGRPRC 0.069 0.022 9.90 0.0017◊ 1.99a 1.30, 3.06 50 Table 5bb (cont’d) Model Outcome (HPAIT): 0 = no turkey farms with HPAI virus infection 1 = at least one infected turkey farm ML ML Wald Variable Pr>X2 OR Estimate SE X2 CL IA univariable TURKFA LAYPULFA CROPPRC CORNGRPRC FARMPRC 0.21 -0.089 0.14 0.12 0.13 0.096 0.045 0.069 0.057 0.069 4.55 3.83 4.27 4.22 3.72 0.033* 0.0502† 0.039* 0.040* 0.054† 1.23 0.92 4.15a 3.23a 3.77a WATSQPRC -0.78 0.89 0.78 0.38 <0.0010a 1.02, 1.48 0.84, 1.00 1.08, 15.97 1.06, 9.91 0.98, 14.53 <0.0010, >999.99 multivariable N/A *N/A indicates that no variables remained statistically significant after mutual adjustment in the multivariate model 51 Table 5c: Results of logistic regression with HPAICOMB outcome based on data from all counties Outcome (HPAICOMB): 0 = no laying hen or pullet or turkey farms with HPAI virus infection 1 = at least one infected laying hen or pullet or turkey farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 MN univariable TURKFA 0.070 0.036 3.76 0.052† 1.07 0.99, 1.15 LAYPULFA -0.0032 0.0054 0.34 0.56 0.99 0.99, 1.01 CROPPRC 0.041 0.013 10.26 0.0014◊ 1.51a 1.17, 1.94 ∆ a CORNGRPRC 0.066 0.019 11.47 0.00070 1.93 1.32, 2.82 FARMPRC 0.050 0.016 9.14 0.0025◊ 1.64a 1.19, 2.27 WATSQPRC -0.15 0.081 3.24 0.072† 0.24a 0.048, 1.14 multivariable TURKFA 0.084 0.039 4.75 0.029* 1.09 1.01, 1.17 CORNGRPRC 0.073 0.021 11.82 0.00060∆ 2.08a 1.37, 3.15 52 Table 5c (cont’d) Outcome (HPAICOMB): 0 = no laying hen or pullet or turkey farms with HPAI virus infection 1 = at least one infected laying hen or pullet or turkey farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 IA univariable TURKFA 0.070 0.064 1.19 0.27 1.07 0.95, 1.22 LAYPULFA -0.027 0.015 3.22 0.073† 0.97 0.95, 1.00 CROPPRC 0.088 0.029 9.17 0.0025◊ 2.42a 1.37, 4.28 * a CORNGRPRC 0.062 0.026 5.66 0.017 1.87 1.12, 3.12 FARMPRC 0.11 0.037 9.14 0.0025◊ 3.09a 1.49, 6.43 WATSQPRC -1.32 0.73 3.30 0.069† <0.0010a <0.0010, 2.84 multivariable LAYPULFA -0.026 0.015 2.97 0.085† 0.97 0.95, 1.00 CORNGRPRC 0.061 0.027 5.31 0.021* 1.84a 1.10, 3.10 53 Table 5cc: Results of logistic regression with HPAICOMB outcome following removal of Kandiyohi (MN), Buena Vista (IA), and Sioux (IA) Counties Outcome (HPAICOMB): 0 = no laying hen or pullet or turkey farms with HPAI virus infection 1 = at least one infected laying hen or pullet or turkey farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 MN univariable TURKFA 0.060 0.036 2.74 0.098† 1.06 0.99, 1.14 LAYPULFA -0.0033 0.0055 0.35 0.55 0.997 0.99, 1.01 CROPPRC 0.041 0.013 10.02 0.0015◊ 1.51a 1.17, 1.95 CORNGRPRC 0.065 0.020 11.10 0.0009∆ 1.92a 1.31, 2.81 ◊ a FARMPRC 0.050 0.017 8.90 0.0029 1.65 1.19, 2.29 WATSQPRC -0.17 0.091 3.68 0.055† 0.17a 0.029, 1.04 multivariable TURKFA 0.076 0.039 3.77 0.052† 1.08 0.999, 1.17 CORNGRPRC 0.072 0.021 11.49 0.00070∆ 2.05a 1.35, 3.11 54 Table 5cc (cont’d) Outcome (HPAICOMB): 0 = no laying hen or pullet or turkey farms with HPAI virus infection 1 = at least one infected laying hen or pullet or turkey farm ML ML Wald Model Variable Pr>X2 OR CL Estimate SE X2 IA univariable TURKFA 0.035 0.075 0.22 0.64 1.04 0.90, 1.20 LAYPULFA -0.038 0.018 4.32 0.038* 0.96 0.93, 0.998 CROPPRC 0.080 0.029 7.69 0.0055◊ 2.22a 1.26, 3.90 * a CORNGRPRC 0.059 0.027 4.86 0.028 1.81 1.07, 3.06 FARMPRC 0.10 0.037 7.44 0.0064◊ 2.75a 1.33, 5.69 <0.0010, † a WATSQPRC -1.53 0.86 3.16 0.076 <0.0010 4.80 multivariable LAYPULFA -0.038 0.019 4.19 0.041* 0.96 0.93, 0.998 CORNGRPRC 0.059 0.028 4.54 0.033* 1.80a 1.05, 3.10 55 56 APPENDIX B: Figures 57 Figure 1: Derivation of the analytic sample for infected farms Inclusion Criteria Exclusion Criteria other states USDA APHIS outbreak data N = 45 farms N = 232 farms “backyard” operations Iowa N = 77 farms Iowa Minnesota N = 6 farms N = 110 farms Minnesota N = 1 farms commercial operations Iowa N = 71 farms Minnesota N = 109 farms USDA=United States Department of Agriculture APHIS=Animal and Plant Health Inspection Service 58 59 60 Figure 4: Derivation of the analytic sample for all MN and IA counties Inclusion Criteria total number of counties Iowa (IA) N=99 counties Exclusion Criteria Minnesota (MN) N=87 counties incomplete data N=0 counties 2 MN counties missing CORNGRAINAC will not be excluded complete data on agricultural and geographical traits and infection status IA N=99 counties MN N=87 counties IA=Iowa MN=Minnesota CORNGRAINAC=acres of corn grown for grain per county 61 Figure 5: Minnesota counties with at least one infected turkey, laying hen, or pullet farm 62 Figure 6: Iowa counties with at least one infected turkey, laying hen, or pullet farm 63 64 65 66 Figure 10: Distribution of the cases of HPAI virus infection in Minnesota and Iowa counties over the observation period 67 Figure 11: Yield of corn grown for grain in Minnesota counties (USDA NASS Agriculture Counts 2015) 68 Figure 12: Yield of corn grown for grain in Iowa counties (USDA NASS Agriculture Counts 2015) 69 Figure 13: Farms with turkey inventory on December 31, 2012 (USDA NASS 2012 Census of Agriculture) 70 Figure 14: Farms with layer inventory on December 31, 2012 (USDA NASS 2012 Census of Agriculture) 71 Figure 15: Hogs and pigs: Inventory 2012 (USDA NASS 2012 Census of Agriculture) 72 REFERENCES 73 REFERENCES 1. Boyce W, Sandrock C, Kreuder-Johnson C, et al. Avian influenza viruses in wild birds: A moving target. Comp Immun Microbiol Infect Dis. 2009;32:275-86. 2. Capua I, Alexander D. Avian influenza infection in birds: A challenge and opportunity for the poultry veterinarian. Poultry Science. 2009;88:842-6. 3. Centers for Disease Control and Prevention. Influenza type A viruses. Atlanta: Centers for Disease Control and Prevention; 9 Feb 2015. 4. Jhung M, Nelson D. Outbreaks of avian influenza A (H5N2), (H5N8), and (H5N1) among birds: United States, December 2014-January 2015. Atlanta: Centers for Disease Control and Prevention; 2015. 5. Center for Food Security and Public Health. Avian influenza [Internet]. Iowa State University; 2015. 6. Center for Infectious Disease Research and Policy. Report finds $1.2 billion in Iowa avian flu damage. Minneapolis: University of Minnesota; 18 Aug 2015. 7. Chen R, Holmes E. Avian influenza virus exhibits rapid evolutionary dynamics. Mol Biol Evol. 2006;23(12):2336-41. 8. Davlin S, Blanton L, Kniss K, et al. Influenza activity: U.S., 2015-16 season and composition of the 2016-17 influenza vaccine. Centers for Disease Control and Prevention Morbidity and Mortality Weekly Report. 2016;65(22):567-75. 9. Encyclopaedia Britannica. Orthomyxovirus. Chicago: Encyclopaedia Britannica, Inc. 10. Gamblin SJ, Skehel JJ. Influenza hemagglutinin and neuraminidase membrane glycoproteins. Biological Chemistry [Internet]. 2010 Sep;285:28403-09. 11. Guan Y, Smith G. The emergence and diversification of panzootic H5N1influenza viruses. Virus Research [Internet]. 2013 June;178:35– 43. 12. Lupiani B, Reddy S. The history of avian influenza. Comp Immun Microbiol Infect Dis. 2009;32:311-23. 13. National Center for Immunization and Respiratory Diseases. Outbreaks of avian influenza in North America. Centers for Disease Control and Prevention. 2017. 14. Neumann G, Macken C, Karasin A, et al. Egyptian H5N1 influenza viruses: Cause for concern? PLOS Pathogens. 2012;8(11):1-3. 15. Roche B, Drake J, Brown J, et al. Adaptive evolution and environmental durability jointly structure phylodynamic patterns in avian influenza viruses. PLOS Biology. 2014;12(8):1-12. 16. Sims L, Domenech J, Benigno C, et al. Origin and evolution of highly pathogenic H5N1 avian influenza in Asia. Veterinary Record. 2005;157:159-64. 74 17. Su S, Bi Y, Wong G, et al. Epidemiology, evolution, and recent outbreaks of avian influenza virus in China. Virology. 2015; 89(17):8671-6. 18. Understanding Evolution. Avian flu adapts to human hosts. University of California, Berkeley. 2015. 19. Understanding Evolution. Evolution and the avian flu. University of California, Berkeley. 2012. 20. Ward M, Lycett S, Avila D, et al. Evolutionary interactions between hemagglutinin and neuraminidase in avian influenza. BMC Evolutionary Biology. 2013;13:1-18. 21. ©World Organisation for Animal Health: Office International des Epizooties. Avian influenza portal. Paris: ©World Organisation for Animal Health: Office International des Epizooties; 2016. 22. Yang H, Carney P, Mishin V, et al. Molecular characterizations of surface proteins hemagglutinin and neuraminidase from recent H5Nx avian influenza viruses. Virology. 2016;90(12):5770-84. 23. Yee K, Carpenter T, Cardona C. Epidemiology of H5N1 avian influenza. Comp Immun Microbiol Infect Dis. 2009;32:325-40. 24. Yoon S, Webby R, Webster R. Evolution and ecology of influenza A viruses. Current Topics in Microbiology and Immunology. 2014;385:359-75. 25. Xu W, Berhane Y, Dube C, et al. Epidemiological and evolutionary inference of the transmission network of the 2014 HPAI H5N2 outbreak in British Columbia, Canada. Scientific Reports. 2016;6(30858):1-15. 26. Centers for Disease Control and Prevention. Highly pathogenic Asian avian influenza A (H5N1) virus. Atlanta: Centers for Disease Control and Prevention; 14 Oct 2015. 27. Census of Agriculture. Poultry and egg production. Washington, D.C.: USDA; Jan 2015. 28. Centers for Disease Control and Prevention. Highly pathogenic Asian avian influenza A (H5N1) in people. Atlanta: Centers for Disease Control and Prevention; 11 June 2015. 29. Centers for Disease Control and Prevention. Bird infections with highly-pathogenic avian influenza A (H5N2), (H5N8), and (H5N1) viruses: Recommendations for human health investigations and response. Atlanta: Centers for Disease Control and Prevention; 2 June 2015. 30. Animal and Plant Health Inspection Service. Confirmed avian influenza detections. Washington, D.C.: USDA; 2016. 31. Animal Production and Health Division. Avian influenza. New York: Food and Agriculture Organization of the United Nations; 2015. 32. World Health Organization. Manual for the laboratory diagnosis and virological surveillance of influenza. Geneva: World Health Organization; 2011. 33. Centers for Disease Control and Prevention. H5 viruses in the U.S.. Atlanta: Centers for Disease Control and Prevention; 4 August 2015. 75 34. FAO Regional Office for Asia and the Pacific. Bird flu rears its head again [Internet]. Food and Agriculture Organization of the United Nations; 2011. 35. Census of Agriculture. 2012 census highlights. Washington, D.C.: USDA; 19 Mar 2015. 36. Census of Agriculture. 2012 census volume 1, chapter 2: County level. Washington, D.C.: USDA; 2015. 37. Animal and Plant Health Inspection Service. USDA adds avian influenza vaccine doses to the national veterinary stockpile. Washington, D.C.: USDA; 13 Oct 2015. 38. Centers for Disease Control and Prevention. Update: Domestic H5 outbreak in birds. Atlanta: Centers for Disease Control and Prevention; 24 July 2015. 39. Centers for Disease Control and Prevention. HPAI A H5 virus background and clinical illness. Atlanta: Centers for Disease Control and Prevention; 2 June 2015. 40. Fry E. What the worst bird flu outbreak in U.S. history means for farms. Fortune [Internet]. 2015 [cited 2016 April 25]. 41. Gale P, Goddard A, Breed AC, et al. Entry of H5N1 HPAI virus into Europe through migratory wild birds: A qualitative release assessment at the species level. Applied Microbiology [Internet]. 2014 Feb;116:1405-17. 42. Harder T, Maurer-Stroh S, Pohlmann A, et al. Influenza A (H5N8) virus similar to strain in Korea causing HPAI in Germany. Emerging Infectious Diseases [Internet]. 2015 May;21(5):8603. 43. Harris MC, Miles AK, Pearce JM, et al. USGS HPAI research strategy [Internet]. U.S. Geological Survey; 2015. 44. Hovinga, Mary. Graduate Summer Session in Epidemiology. University of Michigan. Ann Arbor, MI. July 2015. 45. Ma W, Kahn RE, Richt JA. The pig as a mixing vessel for influenza viruses: Human and veterinary implications. Journal of Molecular and Genetic Medicine [Internet]. 2009 Jan;3(1):158-66. 46. Munster VJ, Baas C, Lexmond P, et al. Spatial, temporal, and species variation in prevalence of influenza A viruses in wild migratory birds. PLOS Pathogens [Internet]. 2007 May;3(5):6308. 47. National Turkey Federation. Statistics. Washington, D.C.: National Turkey Federation; 2014. 48. Pan American Health Organization [Internet]. Washington, D.C.: Regional Office for the Americas of the World Health Organization. Distribution of influenza viruses under surveillance by epidemiological week (EW), region/country, 2010-2015; 2015 [cited 2015 Oct 8]. 49. Reed C, Bruden D, Byrd KK, et al. Characterizing wild bird contact and seropositivity to HPAI A (H5N1) virus in Alaskan residents. Influenza and Other Respiratory Viruses [Internet]. 2014 May;8(5):516-23. 76 50. Shindo N. Making progress on the WHO public health research agenda for influenza. Influenza and Other Respiratory Viruses [Internet]. 2013 [cited 2015 Oct 3];7(2):1-3. 51. World Health Organization. Human infection with avian influenza A (H7N9) virus: China. Geneva: World Health Organization; 2015. 52. USDA NASS. Appendix B: General explanation and census of agriculture report form. Washington, D.C.: USDA NASS; 2012. 53. USDA NASS. Crop production 2015 summary. Washington, D.C.: USDA NASS; 2016. 54. Sonnberg S, Webby R, Webster R. Natural history of highly pathogenic avian influenza H5N1. Virus Res. 2013;178(1):63-77. 55. World Health Organization. H5N1 avian influenza: Timeline of major events. Geneva: World Health Organization; 2011. 56. Sandbulte M. H5 HPAI reports from East Asia: September 2014 through February 2015. Ames: Center for Food Security and Public Health. 57. USDA. FY2016 HPAI response: Contact premises. Washington, D.C.: USDA; 2015. 58. Torchetti M. OIE reference laboratory reports activities: Activities in 2015. Paris: OIE; 2015. 59. Tharakaraman K, Raman R, Stebbins N, et al. Antigenically intact hemagglutinin in circulating avian and swine influenza viruses and potential for H3N2 pandemic. Scientific Reports. 2013;1822(3):1-9. 60. Ito T, Couceiro J, Kelm S, et al. Molecular basis for the generation in pigs of influenza A viruses with pandemic potential. J Virol. 1998;72(9):7367-73. 61. Ludwig S, Zell R, Schwemmle M, et al. Influenza, a One Health paradigm: Novel therapeutic strategies to fight a zoonotic pathogen with pandemic potential. Int J Med Microbiol. 2014;304(7):894-901. 62. Blount S. Wetlands of the United States. Washington, D.C.: National Environmental Education Foundation; 2016. 63. SAS Software, Version 9.4. SAS Institute Inc., Cary, NC, 1989-2017. 64. Ramsar Convention Secretariat. Avian influenza and wetlands: Guidance on control of and responses to highly pathogenic avian influenza. Gland: Ramsar Handbooks for the Wise Use of Wetlands; 2010. 65. Galle A, Linz G, Homan H, et al. Avian use of harvested crop fields in North Dakota during spring migration. Western North American Naturalist. 2009;69(4):491-500. 66. Vandegrift K, Sokolow S, Daszak P, et al. Ecology of avian influenza viruses in a changing world. Ann N Y Acad Sci. 2010;1195:113-28. 67. Perez-Ramirez E, Acevedo P, Allepuz A, et al. Ecological factors driving avian influenza virus dynamics in Spanish wetland ecosystems. Plos One. 2012;7(11):1-10. 77 68. Office of Environment and Heritage. Birds in wetlands. Sydney: New South Wales Government; 2015. 69. U.S. Environmental Protection Agency. Why are wetlands important. Washington, D.C.: U.S. Environmental Protection Agency; 2017. 70. Kuiken T. Role for migratory wild birds in the global spread of avian influenza H5N8. Science. 2016;354(6309):213-6. 71. United States Department of Agriculture (USDA), Animal and Plant Health Inspection Service (APHIS), Veterinary Services (VS). Epidemiologic and other analyses of HPAI-affected poultry farms: September 9, 2015 report. Washington, D.C.: USDA, APHIS, VS; 2015. 72. 2012 Census of Agriculture. Introduction. Washington, D.C.: USDA; 2014. 73. 2010 Census. Population, housing units, area, and density: 2010 county census tract. Washington, D.C.: U.S. Census Bureau; 2010. 74. USDA APHIS. HPAI 2014-2015 infected premises. Washington, D.C.: USDA APHIS; 2016. 75. U.S. Census Bureau. Technical assessment of A.C.E. revision II. Washington, D.C.: U.S. Census Bureau; 2003. 76. Flint P, Pearce J, Franson J, et al. Wild bird surveillance for highly pathogenic avian influenza H5 in North America. Virology Journal. 2015;12:1-6. 77. Pork Checkoff. State rankings by hogs and pigs inventory. Des Moines: Pork Checkoff; 2015. 78. Economic Research Service. Leading 10 U.S. states by number of pullets in 2011 to 2014 (in 1,000s). Washington, D.C.: USDA; 2014. 79. American Egg Board. State/number of hens. Chicago: American Egg Board; 2016. 80. Economic Research Service. Turkey sector: Background and statistics. Washington, D.C.: USDA; 2015. 81. Lee J, Pascua P, Song M-S, et al. Isolation and genetic characterization of H5N2 influenza viruses from pigs in Korea. Journal of Virology. 2009;83(9):4205-15. 82. USDA. FY2016 HPAI response: Details for bird and egg appraisal and indemnity procedures. Washington, D.C.: USDA; 2016. 83. USDA. FY2016 HPAI response: Ready reference guide: Overview of zones. Washington, D.C.: USDA, APHIS, VS; 2016. 84. Gilbert M, Xiao X, Chaitaweesub P, et al. Avian influenza, domestic ducks, and rice agriculture in Thailand. Agriculture, Ecosystems, and Environment. 2007;119(3):409-15. 85. Gilbert M, Xiao X, Pfeiffer D, et al. Mapping H5N1 highly pathogenic avian influenza risk in Southeast Asia. PNAS. 2008;105(12):4769-74. 86. Brown J, Swayne D, Cooper R, et al. Persistence of H5 and H7 avian influenza viruses in water. Avian Diseases. 2007;51(1):285-9. 78