NEW UNDERSTANDING OF THE EPIDEMIOLOGY OF RIFT VALLEY FEVER

VIRUS IN KENYA

By

ANGELLE DESIRÉE LABEAUD, MD

Submitted in partial fulfillment of the requirements

For the degree of Master of Science

In Clinical Research

Thesis Committee

Dr. Neal V. Dawson (Thesis Advisor)

Dr. Charles H. King

Dr. Anna M. Mandalakas

Department of Graduate Studies

Clinical Research Scholars Program

CASE WESTERN RESERVE UNIVERSITY

May 2009

i CASE WESTERN RESERVE UNIVERSITY

SCHOOL OF GRADUATE STUDIES

We hereby approve the thesis/dissertation of

______Angelle Desiree LaBeaud______candidate for the ______M.S.______degree *.

(signed)______Neal Dawson______

(chair of the committee)

______Anna Mandalakas______

______Charles King______

______

______

______

(date) _____March 6, 2009_____

*We also certify that written approval has been obtained for any proprietary material

contained therein.

i

Dedicated to all those who suffer from the neglected tropical diseases

of the developing world.

ii TABLE OF CONTENTS

List of Tables…………………………………………………………………………...viii

List of Figures……………………………………………………………………………ix

Acknowledgements……………………………………………………………………....x

List of Abbreviations……………………………………………………………………xi

ABSTRACT…………………………………………………………………………...….1

CHAPTER 1

1. INTRODUCTION AND GOALS

1.1 Introduction………………………………………………………………………2

1.2 Rationale………………………………………………………………………….3

1.3 Justification………………………………………………………………………5

1.4 Goals

1.4.1 General Objectives………………………………………………………5

1.4.2 Specific Objectives……………………………………………………….6

CHAPTER 2

2. LITERATURE REVIEW

2.1 Rift Valley Fever Virus Virology……………………………………………..…9

2.2 Biodefense Significance………………………………………………………….9

2.3 Epidemiology of Rift Valley Fever Virus……………………………….……..10

2.4 Prior Kenyan Outbreaks……………………………………………………….12

iii 2.5 Climate-Disease Connections…………………………………………………..12

2.6 Implications for Animal Health………………………………………………..15

2.7 Implications for Human Health………………………………………………..16

2.8 Ophthalmologic Complications………………………………………………..17

2.9 Diagnosis of Rift Valley Fever…………………………………………………18

2.10 Unanswered Questions…………………………………………………………19

CHAPTER 3

3. STUDY DESIGN

3.1 Inclusion Criteria……………………………………………………………….21

3.2 Exclusion Criteria………………………………………………………………21

3.3 Special Populations……………………………………………………………..22

3.4 Outcome Measurements………………………………………………………..22

3.4.1 Study Outcome Measures………………………………………………22

3.4.2 Primary Outcome Measures…………………………………………...23

3.4.3 Secondary Outcome Measures…………………………………………23

CHAPTER 4

4. METHODS

4.1 Retrospective Study…..………………………………………………………...24

4.1.1 Study Areas………………………………………………………...…...24

4.1.2 Populations Surveyed…………………………………………………..25

4.1.3 Laboratory Testing……………………………………………………..25

iv 4.1.4 Statistical Analysis……………………………………………………...26

4.2 Prospective Study……..………………………………………………………...26

4.2.1 Study Areas…………………………………………………………...... 26

4.2.2 Sampling Strategy……………………………………………………....28

4.2.3 Populations Surveyed…………………………………………………..29

4.2.4 Subject Participation…………………………………………………...29

4.2.5 Examination Procedures…………………………………………….....30

4.2.6 Specimen Collection……………………………………………...... 31

4.2.7 Laboratory Testing……………………………………………...... 32

4.2.8 Statistical Analysis……………………………………………...... 33

4.2.9 Data processing……………………………………………...... 34

CHAPTER 5

5. ETHICAL CONSIDERATIONS

5.1 Institutional Review Board…………………………………………...... 35

5.2 Informed Consent Process………………………………….………...... 35

5.3 Assent Process…………………………………………...... 36

5.4 Subject Confidentiality…………………………………………...... 37

5.5 Future Use…………………………………………...... 38

CHAPTER 6

6. RESULTS

6.1 Retrospective Study…………………………………………...... 39

v 6.1.1 Serological outcomes………………………………...... 39

6.1.2 Differences by location………………………………...... 40

6.1.3 Differences by age group……………………………...... 40

6.1.4 Differences by gender……………………………...... 41

6.1.5 Differences over time……………………………...... 41

6.1.6 Hypotheses generated……………………………...... 42

6.2 Prospective Study……………………………...... 43

6.2.1 Survey Results……………………………...... 43

6.2.2 Links between past exposure and seropositivity...... 45

6.2.3 Symptom history and current physical examination...... 49

6.2.4 Ophthalmological findings...... 50

CHAPTER 7

7. DISCUSSION...... 52

CHAPTER 8

8. CONCLUSIONS...... 63

8.1 General Conclusions...... 63

8.2 Future Studies...... 64

CHAPTER 9

9. APPENDIX...... 67

vi 9.1 Detailed Study Summary...... 93

CHAPTER 10

10. BIBLIOGRAPHY...... 98

vii LIST OF TABLES

TABLES

Table 1: Retrospective Study Sample Details……………………………………………39

Table 2: Demographics of Masalani Town Survey Population………………………….43

APPENDIX TABLES

Table 1: Associations Between Variables and RVFV Seropositivity……………………72

Table 2: Association Between Location and Exposure Variables……………………….74

Table 3: Binary Logistic Regression Analysis…………………………………………...76

Table 4: Interaction Testing Between Variables…………………………………………77

Table 5: Logistic Models By Location…………………………………………………..78

Table 6: Association of Signs and Symptoms with RVFV Seropositivity………………80

Table 7: Eye Disease Association with RVFV Seropositivity…………………………...82

Table 8: Theory-Driven Variable Clusters………………………………………………83

Table 9: Associations Between Clusters and RVFV Seropositivity……………………..85

Table 10: Logistic Models Using Clusters……………………………………………….87

viii LIST OF FIGURES

Figure 1: Transmission Electron Micrograph of an RVFV-infected Tissue……………....9

Figure 2 : World Map of RVFV Distribution……………………………………………11

Figure 3 : NDVI Map of Kenya in 1996…………………………………………………13

Figure 4: NDVI Maps of Kenya in 1997 and 1998……………………………………...14

Figure 5 : African NDVI Anomaly Maps………………………………………………..14

Figure 6 : Map of Retrospective Study Sites…………………………………………….24

Figure 7 : Pictures of Prospective Study Sites…………………………………………...27

Figure 8 : RVFV Seropositivity by Location…………………………………………….40

Figure 9 : Lokichoggio RVFV Seroprevalence by Age Group………………………….41

Figure 10 : Flow Chart of Study Participants……………………………………………44

Figure 11: Village Exposure Chart…………………………………………………...... 46

Figure 12: RVFV Seroprevalence by Age and Village………………………………….48

APPENDIX FIGURES

Figure 1: Prospective Study Questionnaire…………………………..…………………..67

Figure 2: Data Entry Form for Physical and Eye Exam Findings……………………….70

ix ACKNOWLEDGEMENTS

I would like to thank Dr. Charles H. King, Dr. C.J. Peters and Dr. Eric M. Muchiri

for their mentoring and support of these studies. I would also like to give thanks to Dr.

Ronald E. Blanton, Dr. Chandy C. John, and Dr. Christopher L. King for their donation

of archived human sera for the retrospective study. For the prospective study, I thank the

survey participants, the Kenyan field team, the Masalani District Hospital laboratory

staff, the Kenya Medical Research Institute, and the DVBD laboratory team (Mariam

Mwanje, Adel Odek, and Ole Kesuka) without whom the study could not have been

performed. I would also like to thank Allison Glinka for her help with analysis. Finally,

I thank my husband and son for supporting and nurturing me while I spent numerous

hours at home and away performing studies, analyzing data, and writing my thesis.

x LIST OF ABBREVIATIONS

CDC Centers For Disease Control and Prevention

CWRU Case Western Reserve University

DVBD Division of Vector Borne Diseases

ELISA Enzyme-Linked ImmunoSorbent Assay

ENSO El Nino-Southern Oscillation

KEMRI Kenya Medical Research Institute

NDVI Normalized Differentiated Vegetation Index

NTD Neglected Tropical Disease

OD Optical Density

OIE World Organisation for Animal Health

PCR Polymerase Chain Reaction

PRNT Plaque Reduction Neutralization Testing

RVF Rift Valley Fever

RVFV Rift Valley fever virus

SST Sea Surface Temperature

UTMB University of Texas Medical Branch, Galveston

VBD Vector-Borne Disease

VERO “VErda REno” cells (African green monkey kidney epithelial cells)

xi New Understanding of the Epidemiology

of Rift Valley Fever Virus in Kenya

Abstract

By

ANGELLE DESIRÉE LABEAUD, MD

Rift Valley fever virus (RVFV) is an emerging high priority pathogen that threatens livestock and human health. In order to define the range of human transmission during interepidemic and epidemic periods in Kenya, archived sera (N = 1263) were tested for anti-

RVFV IgG by ELISA and plaque reduction neutralization testing. RVFV seroprevalence was 10.8% overall and varied significantly by location, gender and age. Because most RVF outbreaks occur in remote locations following floods, environmental risk factors and human sequelae are not fully known. During an interepidemic period in 2006 a randomized household survey (N=248) was performed to examine age- and location- specific RVFV seroprevalence rates, and their association with health status and animal/non-animal exposures. Overall RVFV seroprevalence was 13% and interepidemic

RVFV transmission was documented. Seropositivity increased with older age, rural village, male sex, and abortus disposal. The extent of epidemic and interepidemic RVFV transmission is greater than previously documented.

1 CHAPTER 1: INTRODUCTION AND GOALS

1.1 Introduction

Rift Valley fever virus (RVFV) is a high priority pathogen because of its potential

for severe economic harm to livestock 1, its ability to cause life-threatening hemorrhagic

fever in humans 2,3, and its potential for non-vector aerosol spread during epizootics and

epidemics 4,5. RVFV is adaptable to weaponization, as shown by past studies of the US

biological warfare program 6,7. It is also an emerging pathogen, whose range has recently

expanded from East Africa, across sub-Saharan Africa to North Africa, and to the

Arabian Peninsula 5,8,9. Relatively little is known about the natural history of RVFV

transmission and infection because natural outbreaks are sporadic and are typically

associated with severe flooding events in very remote locations 4,5.

Although RVFV transmission is known to be both focal and episodic in nature

and closely tied with flooding rainfall events, there may be sporadic transmission in

enzootic areas that helps to perpetuate and/or extend the range of RVFV transmission

areas. While most RVFV-infected humans have symptoms of limited duration and

appear to make a full recovery, we hypothesize that the late-onset non-hemorrhagic complication of Rift Valley Fever (RVF) retinitis is associated with permanent, measurable injury to the affected organ 10,11. The aim of the studies outlined in this protocol is to examine the Kenya-wide human RVFV seroprevalence rates and determine the late outcomes and possible continuing enzootic/endemic presence of RVFV in Ijara

District, Kenya. The studies involve a retrospective serosurvey of human sera samples

collected outside of a known RVFV transmission zone and a clinical and serological

resurvey of the human population in Masalani town, Kenya, a known epidemic focus

2 during the 1997-1998 and 2006-2007 RVFV outbreaks. The results are able to provide a picture of previous and current RVFV human transmission activity in Kenya. If no new

RVFV transmission is occurring, the results also provide a picture of the late (> 6 yr) effects of RVF in affected human populations. This information is valuable to assess the continuing human transmission, the risk factors associated with human transmission, and the long-term disabilities and health-care needs resulting from RVFV infection. This study also has the capacity to enrich Kenya’s capacity for RVFV serology testing in order to improve its capabilities for disease surveillance.

1.2 Rationale

Recent experience of inadvertent West Nile virus introduction into North America indicates that 'exotic' arboviral pathogens can quickly become persistent in local ecosystems,

provided the necessary vectors and animals reservoirs are present. RVFV is perhaps the

most threatening of all tropical viral infections because it can infect so many different

animals and be carried by so many vectors. All of the necessary animal hosts and vectors required for RVFV introduction and persistence are present in North America. Lessons learned studying RVFV in its current endemic setting are useful for public health protection in both the developing and developed world. Because of the threat of natural or bioterrorist introduction of RVFV into new areas of the world, such as the United States, and the likelihood of its regional persistence once introduced, it is essential to learn more about how

RVFV is spread (and contained) under natural circumstances.

In developing strategies for control of this important pathogen, is critical to determine its transmission patterns during the intervals between the recognized large-

3 scale enzootics/epidemics. Indirect evidence suggests that RVFV is transmitted at low

levels in interepidemic years, a process that may act to provide a continuing reservoir of

disease in high-risk, semi-arid regions, and which may contribute significantly to overall burden of disease in marginal communities. During outbreaks, intensive efforts have been made to determine the immediate nature and extent of RVFV transmission and the associated case/infection and case-fatality rates. However, long-term outcomes of RVFV

infection are not well known, and the late post-infection physical and serodiagnostic

findings are not as well defined as the acute manifestations of infection. In particular, the

diagnostic characteristics of screening serological testing are not well understood for

communities that have experienced multiple past outbreaks, and the long-term outcomes

of RVFV-related eye disease are not well known.

The data derived from serologic testing may help to improve the interpretation of

laboratory findings in populations which have both remote and acute exposure to RVFV

transmission. When diagnostic tests are used outside of an epidemic period, they may be

less sensitive or specific, therefore testing the Ijara population now yields useful insight

into the overall predictive value of current RVFV serologic testing. Analysis of

sensitivity and specificity can determine the best use of these assays during Kenya’s next

RVF outbreak. An additional, important feature of this project is the implementation of

new testing capability at the Nairobi headquarters of the foreign partner, the Division of

Vector Borne Diseases (DVBD) of the Ministry of Health, Kenya.

4 1.3 Justification

This project accomplishes several aims, including expanding the knowledge of the epidemiology and epizoology of RVFV infection, and expanding Kenya’s capabilities for prediction and detection of RVF outbreaks. The goal of the study is to better define the extent of RVFV transmission across Kenya, the most important human risk factors for infection, how Kenyans were affected by the RVF outbreak in 1997-98, and what long- term disabilities and health-care needs may persist. Ultimately, the information gathered in this study helps to improve health care in RVFV-infected individuals and may be used to prevent further spread of this disease.

1.4 Goals

The goals of the study are: 1) to examine the RVFV seroprevalence across

Kenyan regions; 2) to identify which animal and non-animal exposures are associated with RVFV seropositivity, 3) to evaluate whether seropositivity, exposures, and risks differ between town and village settings in a high-risk region of northeastern Kenya, and

4) to assess if interepidemic human RVFV transmission occurs.

1.4.1 General Objectives

The general objective of this research project is to identify novel information about the disease course, transmission, ecologic risk factors, and epidemiology of Rift

Valley Fever. The objective of the studies outlined in this proposal is to better define the extent and timing of RVFV transmission and its related chronic disease/infection attack rate, during an interepidemic period in a high-risk region that has seen repeated RVF

5 outbreaks. The results can be used to develop and refine predictive algorithms for RVFV

transmission, based on epidemiological, environmental, and remote sensing data, with the

ultimate goal of providing improved early detection of significant RVF outbreaks. It is

expected that the analysis of test-performance characteristics may determine the best use of serologic assays in Kenya’s next RVF epizootic/epidemic.

1.4.2 Specific Objectives

1. Define the regional extent of RVFV infection across different Kenyan regions- in

areas not known to have ongoing RVFV human transmission.

a. By measuring evidence of prior RVFV exposure in stored sera specimens

collected in human surveys across Kenya, interepidemic RVFV human

transmission can be estimated.

2. Contrast the earlier disease/infection attack rate during an epidemic period with

present findings in an interepidemic period, in a region that has had prior RVF

outbreaks (primary outcome measure).

a. By measuring the prevalence of specific anti-RVFV antibodies in the Ijara

population now, while no epidemic is occurring, seroprevalence levels

found during the 1997-98 epidemic can be compared. By asking detailed

questions about signs and symptoms of RVFV infection, the determination

of whether acute or remote RVFV infection has occurred in conjunction

with present serology results can occur. Specific methods involve

Enzyme-Linked ImmunoSorbent Assay (ELISA) testing of blood samples

6 for serum anti-RVFV IgG antibodies, with confirmatory testing via

standard plaque-reduction neutralization testing (PRNT).

3. Determine the late outcomes and diagnostic characteristics of serology 6+ years

after an epidemic/epizootic period.

a. By testing the present serology of those exposed to RVFV during the last

epidemic, the background operating characteristics of RVFV serology

tests can be determined, and the limitations of serologic testing when

future outbreaks occur can be anticipated.

4. Provide a picture of possible continuing enzootic/endemic presence of RVFV in

Ijara District, Kenya.

a. By testing children less than age 7 years, the presence of any

interepidemic transmission of RVFV can be determined, as these subjects

would not have been able to be infected during the 1997-98 outbreak.

5. To determine behavioral risk factors for recent or remote RVFV infection

(secondary outcome measure).

a. By asking detailed questions about non-animal exposures and animal

exposures (via questionnaire at the time of study examination), the

significant behavioral risk factors for RVFV seropositivity in this area can

be identified.

7 6. Conduct a population-based sample survey to define the late (6+ year) medical

consequences of Rift Valley fever virus infection (secondary outcome measure).

a. By performing physical exam and ophthalmologic exam of the study

subjects by clinical officers blinded to the RVFV serostatus of the

subjects, the late medical consequences associated with prior RVFV

infection can be investigated.

8 CHAPTER 2: LITERATURE REVIEW

2.1 RVFV Virology

Rift Valley fever virus is a member of the Phlebovirus genus, one of the five genera in the family Bunyaviridae 1. The virus was first identified in 1931 during an outbreak investigation of a sheep epizootic on a farm in the Great Rift Valley of Kenya

1,12. RVFV, like other Bunyaviruses, has a single-stranded, tripartite negative or ambisense coded RNA genome composed of the L, M, and S segments encoding four structural proteins, viral polymerase (L segment), glycoproteins (M segment) and nucleocapsid protein N (S segment) 13. RVFV is an enveloped virus with a diameter of

90 to 110 nm and a core element of 80 to 85 nm 14. Viral replication occurs in the cytoplasm and viral particles bud into the Golgi 15. As in other single-stranded RNA viruses, double-stranded RNA is likely to be generated as intermediates during viral replication, which likely activate innate receptors and lead to effective host defense.

Figure 1. Transmission electron micrograph depicts a highly magnified view of a tissue that has been infected with Rift Valley fever virus. Note the spherical viral particles.

Image provided by CDC.

2.2 Biodefense Significance

RVFV has been studied as a potential agent of biological warfare both by the US and the former USSR, and is adaptable to weaponization 6,7. High titers of RVFV can be

9 achieved by growth in several artificial systems. Furthermore, in formal aerosol studies

in the laboratory, RVFV demonstrates aerosol stability and infectivity, making it a

dangerous candidate for a biological weapon or terrorist use. The Soviet Union is said to

have produced such weapons and the US offensive program was developing such a

weapon in 1968 when it was disestablished 6,7.

Equally important as the threat to humans is the effect of RVFV on livestock.

The transmission of the virus among US domestic sheep and cattle would result in death

of about 1/3 of infected animals and abortion of all pregnant animals. In the event of

RVFV introduction to the United States, the USDA plans to call for the freezing of livestock movement, in order to prevent viral spread. RVFV infection of domestic livestock would be economically devastating to a complex $100 billion US industry.

Experimental infection of North American mosquitoes with RVFV suggests that endemicity could be established in this country 16. Given the recent US experience with

West Nile virus, it should be expected that after either accidental or intentional introduction, RVFV would become a widespread persistent multi-state or multinational

problem in North America.

2.3 RVFV Epidemiology

Relatively little is known about the natural history of RVFV transmission and infection because natural outbreaks are sporadic and explosive 4,5. However, epizootic

outbreaks do not occur at random, but instead are strongly linked to excessive rainfall and

local flooding events. RVFV is maintained in nature at least in part by transovarial

transmission in flood-water Aedes mosquitoes 17,18. Certain floodwater Aedes mosquito

10 species allow RVFV to become embedded in endemic ecosystems by means of vertical

transmission to their offspring 1. The transovarially-infected mosquito eggs can then

remain viable for many years during dry spells 17,18. When rainfall occurs, shallow

depressions in the landscape, called dambos, fill with water and allow the eggs to hatch.

The newly hatched transovarially RVFV-infected mosquitoes then feed on livestock that

come to the dambos to drink, and an RVFV epizootic is initiated. These sites also later

breed Culex spp. mosquitoes, which are competent to transmit RVFV from animal-to- animal and from animal-to-human 8,19, thus amplifying the outbreak as ‘epidemic’

vectors.

RVFV is endemic to countries of East Africa, South Africa, and the Senegal River

valley 2,20-22. RVFV has been introduced repeatedly into Egypt since the 1970s, and most

recently to the Arabian peninsula (Yemen and Saudi Arabia) in 2000 5,8,9(Figure2). The

latest Kenyan RVF outbreak occurred in association with El Niño rains between

November 2006- April 2007 23,24. The previous and largest RVF outbreak in Kenya took

place in a similar El Niño-related flooding period in 1997-1998 4.

Endemic Sporadic

Figure 2. Map of the world showing the current distribution of Rift Valley fever virus. Provided by CDC.

11 Prior Kenya Outbreaks

In 1997-98, the El Niño/Southern Oscillation (ENSO) event resulted in extensive

heavy rains and flooding in East Africa with epidemic RVF disease activity in Ethiopia,

Sudan, Somalia, Tanzania and Kenya 4. The epicenter of the Kenyan epidemic was Garissa

District (see map, Figure 6, black circle), in Northeastern Province, where in December

1997, 170 hemorrhagic fever-associated deaths were reported 4. Systematic multistage cluster sampling across Garissa District in 1997-1998 indicated a 14% prevalence of acute

(IgM-positive) cases, with an estimated 20-26% of the population having either recent or past infection with RVFV. Some populations had RVFV IgG seropositivity as high as 32% in high-risk areas during the epidemic. An estimated 27,500 infections occurred in Garissa

District, making it the largest recorded outbreak of RVFV in East Africa. However, the nationwide extent of RVFV transmission during the 1997-1998 outbreak was not studied.

Even within different climate zones, RVFV transmission may vary significantly as a function of fine-scale differences in local environment. In order for surveillance, prediction, and containment programs to be most effective, it is important that knowledge of

RVFV transmission be determined both on national as well as regional and district levels

during interepidemic and epidemic periods 25.

2.4 Climate-Disease Connections

RVFV epizootics typically occur when ENSO-related periods of heavy rainfall

flood natural landscape depressions to breed transovarially-infected “floodwater” Aedes

spp. mosquitoes 26,27. Because persistence of RVFV is linked to floodwater Aedes

species, epizootic outbreaks are closely linked to excess rainfall 26, and particularly to El

12 Nino-Southern Oscillation and sea surface temperature (SST) anomalies in the Indian and

Pacific Oceans. The increased precipitation associated with these events causes an

increase in the availability of breeding grounds for the disease-carrying mosquitoes 27.

Due to the associated expansion of plant cover, an increase in the satellite- detected normalized differentiated vegetation index (NDVI), a measure of greenness or plant chlorophyll, can be observed 4. NVDI anomalies in Kenya are significantly

correlated with RVFV activity one to two months later 27. NDVI derived from NOAA

polar orbiting satellite data can identify areas of abnormally high green vegetation

development resulting from above normal rainfall (Figure 3). Elevated NDVI (Figure 4)

and NDVI anomalies (the difference between expected NDVI and observed

NDVI)(Figure 5) were observed for East Africa starting in October and extending to

March and April 1998, which coincided with the large RVFV outbreak in this area. A

better prediction of RVFV activity is obtained when NDVI is combined with Indian

Ocean SST in a moving average predictive model.

Figure 3. An NDVI map of Kenya in 1996, depicting a normal year of rainfall. Each row of 6 boxes represents two months of satellite data. The upper left three boxes are January 1996 data and the lower right three boxes are December 1996 data.

13 Figure 4. NDVI maps of Kenya in 1997 (left) and 1998 (right), depicting abnormally heavy rainfall (orange color). Each row of 6 boxes represents two months of satellite data.

Figure 5. African NDVI anomaly maps depicting excessive rainfall in East Africa.

14 The ability to forecast regional RVF virus activity in Kenya, based upon Pacific and Indian Ocean SST anomalies, and NDVI two months before RVF activity could

prevent outbreaks by permitting vaccination of domestic animals, and pretreatment of

mosquito habitats adjacent to domestic animal herds and human habitations.

Implications for Animal Health

Rift Valley Fever is an emerging mosquito-borne zoonosis that is expanding in its range in Africa and the Middle East. RVF can have catastrophic economic impact on meat and dairy producers, causing high morbidity and mortality among affected livestock herds 1,12, and invoking 2-3 year World Organisation for Animal Health (OIE)-mandated

international embargoes of livestock exports during epizootics.

RVFV is able to infect many species of animals causing severe disease in

domesticated animals including cattle, sheep, camels and goats. Sheep appear to be the

most susceptible. Age of the animal is an important risk factor for severe disease: young

sheep are particularly susceptible to RVF-related illness—mortality among lambs is 90%

as compared to 10% among adult sheep 1. The virus is also an abortifacient causing

almost 100% of pregnant ewes to abort 1,2. Epizootics of RVF in animals frequently

manifest as a wave of unexplained abortions among livestock and may signal the start of a human epidemic.

RVF epizootics are very devastating for pastoral nomads and local herding

economies through the loss of many adult animals, the devastation of the next crop of

newborns, and the danger to locals who are dependent on milk and meat for survival

during the epizootic. Because outbreaks of RVF in animals precede human cases, the

15 establishment of an active animal health surveillance system to detect new cases is

essential in providing early warning for veterinary and human public health authorities.

Prevention of RVF epizootics can be prevented by sustained animal vaccination programs with either the modified live attenuated virus and inactivated virus vaccines1.

Only one dose of the live vaccine is required to provide long-term immunity but the vaccine may result in spontaneous abortion if given to pregnant animals. The inactivated virus vaccine does not cause abortion, but requires multiple doses to provide protection.

In order to prevent outbreaks, animal vaccination must occur prior to the epizootic, which can be difficult to implement in remote settings.

Implications for Human Health

Epizootics and epidemics can result in massive loss of livestock, consequent export embargoes, and significant human morbidity and mortality, all of which can be economically devastating to affected areas 1,2,28. During large RVF animal outbreaks,

significant numbers of human infections occur as well, leading to substantial healthcare

challenges in resource-limited settings. Because of RVFV's ability to cause retinitis,

encephalitis, and hemorrhagic fever, episodic epidemics of RVF present a significant

natural threat to human health in endemic countries 2,3.

Human RVF disease is not well characterized, but the best estimates are that most infections are minimally symptomatic and that a minority, perhaps 10% result in eye

disease and ~1%, result in severe hemorrhagic fever 2. Among humans, RVF typically

manifests as a symptomatic febrile disease with fever, myalgia, and malaise; but in a

significant minority of cases it can lead to retinitis, encephalitis, hemorrhagic fever, and

16 death, with ~ 1% overall mortality rate 2,28. In non-hemorrhagic cases, humans suffer 4 to

7 days of influenza-like symptoms and a small proportion lead to meningoencephalitis

(1%). Anti-RVFV IgM and IgG antibodies emerge in the bloodstream, and the virus

disappears from circulation. When hemorrhagic disease occurs, it begins 2 to 4 days after

the onset of illness and is manifested by jaundice, hematemesis, hematochezia, purpura,

and gingival bleeding. Patients with hemorrhagic disease remain viremic for prolonged

periods (up to 10 days) and suffer a high mortality risk of 20-50%.

Ophthalmologic Complications

The most common complication following RVF recovery is eye injury 10,29,30.

RVFV affects the uvea and posterior chorioretinal area and is associated with permanent

visual loss resulting from macular and paramacular scarring, vascular occlusion, and optic atrophy. In a recent Saudi outbreak, 15% of severe RVF cases had retinal disease and 31% of mild cases had anterior uveitis 10. The mean interval between the onset of

RVF and visual symptoms ranged from 4 to 15 days. Macular or paramacular retinitis

was identified in all the affected eyes at the time of initial assessment. Lesions included

retinal hemorrhages (40%), vitreous reactions (26%), optic disc edema (15%), and retinal

vasculitis (7%). Anterior uveitis was present in 31% of outpatients. Initial visual acuity was less than 20/200 in 80% of eyes in the outpatient group; their vision improved, deteriorated, or remained the same in 13%, 15%, or 72%, respectively. Evaluation at the

last follow-up showed macular (60%) or paramacular (9%) scarring, vascular occlusion

(23%), and optic atrophy (20%) in the outpatient group. The Saudi study demonstrated

that RVF was associated with major ocular morbidity. The ocular manifestations of RVF

17 occurred with a relatively higher frequency than reported previously and were not limited

to severe infections. This study also demonstrated for the first time that transient

nongranulomatous anterior uveitis is associated with RVF 10.

While most RVFV-infected humans have symptoms of limited duration and

appear to make a full recovery, this study’s hypothesis is that the late-onset non-

hemorrhagic complication of RVF retinitis is associated with permanent, measurable

injury to the affected eye. This study discovers new information not only about the

disease course, transmission, behavioral risk factors, and epidemiology of Rift Valley

Fever, but also provides valuable information about the long-term consequences of

infection. This information can be used by the DVBD and the Kenyan Ministry of Health

to improve RVF care standards and manage long-term disabilities resulting from RVFV

infection.

Diagnosis of RVFV Exposure

RVFV can be detected in many ways. Usually the environmental signals of heavy

rain, livestock abortions, and mortality in young animals signals a RVF outbreak. Viral

isolation via VERO cell (African green monkey kidney epithelial cell) culture remains

the gold standard for diagnosis 31. Because propagation of the virus requires strict

laboratory guidelines (Biosafety Level 3+ or 4), other methods are often utilized in the

field for diagnosis. Detection of RVFV-specific IgM or IgG in animal or human sera by enzyme-linked immunosorbent assays (ELISA) are widely used and are more sensitive than previous methods of complement fixation and hemagglutination inhibition 31. The

presence of anti-RVFV IgG lends evidence of prior RVFV infection, whereas the

18 presence of anti-RVFV IgM suggests a more recent RVFV exposure 32,33. Because

ELISAs are not highly specific and can give false positive results with closely related

viruses, highly specific methods, such as plaque-reduction neutralization, can be used to confirm RVFV diagnosis. More recently, detection of the viral nucleic acid by

polymerase chain reaction (PCR) amplification has been developed and can be used for

rapid presumptive diagnosis and later detailed characterization of viral strains 31.

Unanswered Questions

Because RVF outbreaks typically occur in remote locations under extreme weather conditions, relatively little is known about interepidemic transmission of RVFV to humans or the health consequences of RVF disease. Likewise, debate continues regarding the likely dominant mode of animal-to-human transmission during combined epizootics and epidemics. RVFV re-emergence via floodwater mosquitoes is followed by significant amplification in high-risk animal populations, with progressively greater animal prevalence. When epizootic conditions are right, additional mosquito species will feed on viremic animals and subsequently transmit RVFV to humans, potentially creating an epidemic. Humans can also become infected through exposure to infectious animal tissues or bodily fluids such as abortus, birthing fluids, milk, or blood. Among pastoral nomads and other herders in the semi-arid regions of Africa, family members could be differentially exposed depending on cultural assignment of tasks, and thereby alter the risk modifying effects of age or sex. The specifics of what types of animal exposure are most risky, and what non-animal exposures are important, have not yet been elucidated.

Knowing which forms of exposure provide the greatest RVFV transmission risk may be

19 useful for endemic or epidemic public health education, and for targeting interventions

(such as animal vaccination) that can decrease infection or morbidity during an epidemic.

20 CHAPTER 3: STUDY DESIGN

This is a two-part study. The first part is a retrospective cross-sectional,

observational study of Rift Valley fever virus exposure and extent in three regions of

Kenya using stored serum samples. The second is a randomized prospective cross-

sectional household cluster serosurvey of a systematic stratified sample.

RETROSPECTIVE STUDY:

Stored sera samples were used from serum banks obtained as part of well defined

cross-sectional village or family surveys 34,35(Blanton, unpublished). No further inclusion or exclusion criteria were imparted.

PROSPECTIVE STUDY:

3.1 Inclusion criteria: Those adults and children residing in selected survey

households in the area of Masalani Town and its 5 km environs who are over one year of

age were eligible, provided: 1) Informed consent was obtained and signed, and assent was obtained and signed from children age 7 and older and 2) The subject was able to comply with all study procedures.

3.2 Exclusion criteria: The following residents were not eligible to participate: 1)

Age less than one year, 2) Acutely ill, or 3) Adults and children who could not provide informed consent or who could not participate fully in the study procedures.

21 3.3 Special Populations--Children: Children aged 1 year and older accompanied by

an adult were recruited to participate in the study. Appropriate precautions were taken to

obtain informed parental consent and child assent when age-appropriate. Children may

represent a special high- or low-risk subgroup for RVFV infection and disease and

therefore the survey design included a representative sample of children under 15 years

of age that matched the local at-risk childhood population of Ijara District, in order to

determine the RVFV prevalence and infection-associated disease rates in children and

adolescents as compared with adults.

3.4 Measurements

The primary outcome measure is RVFV seropositivity as measured by anti-

RVFV IgG detection via ELISA of serum (retrospective and prospective surveys).

ELISA results are confirmed by plaque reduction neutralization testing (PRNT). The secondary outcome measures are age- and location-specific rates of previous RVFV exposure (retrospective and prospective surveys), behavioral risk factors for recent or remote RVFV infection (prospective survey), and RVFV association with current health

status and its link to animal and non-animal exposures (prospective survey).

3.4.1 Study Outcome Measures

The primary outcome measure is the RVFV ELISA result, a test that has been

shown to be a reliable and accurate marker of RVFV exposure in prior studies4,33. This

variable is considered binary (positive or negative) for the purpose of evaluation. The

22 secondary outcome measures, RVFV retinitis or physical abnormalities associated with

RVFV infection, are considered as categorical variables.

3.4.2 Primary Outcome Measures

The primary outcome measure is the current RVFV-specific seropositivity profile for each region studied. It is determined via ELISA detection of RVFV-specific

IgG antibodies in serum obtained from study participants. Determination of ELISA positivity is according to previous RVFV ELISA test cut-offs. ELISA test results are confirmed by plaque reduction neutralization testing (PRNT). Analysis of ELISA results and linkage with questionnaire information is performed using standard statistical techniques.

3.4.3 Secondary Outcome Measures

The secondary outcome measures to be studied are: 1) an assessment of the behavioral factors associated with risk of seropositivity (previous infection) in the study population as determined by subject questionnaire; and 2) the association of selected long-term ophthalmological and clinical outcomes with seropositivity as identified on standard physical examination. These are determined via examination and are analyzed accordingly. Another secondary outcome is the determination of possible non-animal and animal RVF risk factors from questionnaire data.

23 CHAPTER 4: METHODS

4.1 RETROSPECTIVE SURVEY

4.1.1 Study areas. This retrospective study tested archived anonymous human serum

samples taken from well-defined cross sectional

studies performed during 1994, 1996, 1997 and

1998 in three distinct areas of Kenya outside the

recognized RVFV transmission zone. The

primary focus of the official RVF outbreak

investigation and the epicenter of the RVF

outbreak in 1997-98 was Garissa District of

Northeastern Province4 —centered at 0° 27.5’ S,

39° 39’E (Figure 6, black circle). For the present

study banked serum was tested derived from

populations in: 1) Kabobo Division, Uasin Gishu Figure 6. Map depicting retrospective study sites. District (located in the western highlands at 0°

40’N, 35° 30’E and 260 miles west of Garissa) (Figure 6, brown circle), 2) Msambweni and

Daragube villages, Kwale District (located on the Indian Ocean at 4° 18’S, 39° 30’E and 260 miles south of Garissa) (Figure 6, blue circle) and 3) Lokichoggio area in Turkana District

(located in NW Kenya, 4° 13’N, 34° 17’E, and 230 miles north of Kabobo) (Figure 6, red circle). These rural village sites, distributed across Kenya and separated by distances of at least 250 miles, represent a spectrum of different habitats, environmental conditions, and

24 population lifestyles. The study sites span a range of climates from highlands (Kabobo), to monsoonal coastal plain (Msambweni and Daragube), to semi-arid (Lokichoggio).

4.1.2 Populations surveyed. The serum samples tested were from serum banks obtained as part of well defined cross-sectional village or family surveys for infection and immune response to Echinococcus granulosus in 1994 (cross-sectional village survey of Lokichoggio

(R Blanton, unpublished)), malaria in 1996 and 1997 (cross-sectional village survey of

Kabobo 34), and schistosomiasis and filariasis in 1996–1998 (matched case-control study of

Msambweni and Daragube 35). Serum samples were stored at -80˚C prior to RVFV testing.

Informed consent was obtained from all human adult participants and from parents or legal guardians of minors. Institutional review board approval was obtained from University

Hospitals of Cleveland and the Kenya Medical Research Institute.

4.1.3 Laboratory Testing. In order to test for evidence of past RVFV infection, the serum specimens were screened for the presence of anti-RVFV IgG via ELISA as previously described 4,32, using lysates of VERO cells infected with the MP-12 strain (vaccine strain) of

RVFV as the test antigen and mock infected cells as the internal control antigen. This ELISA assay has been established and validated in previous survey studies 33. Serum samples diluted at 1:100 were read at 405 nm and those scoring an optical density (OD) value greater than mean + 2 SD for control sera, and absolute value greater than 0.2 were deemed positive. Each sample was run in duplicate and OD values are averaged. Any significant discrepancy in the

ODs (measuring at least 0.05) between duplicate tests was resolved by repeat testing. Pooled

RVF-positive Lokichoggio sera were used as the positive control and pooled RVF-negative

25 North American sera were used as the negative control (for cross-contamination) to assure accurate ELISA assay performance. All positive samples (N=143) and an age- and location- matched set of negative samples (N=142) had confirmatory testing via plaque reduction neutralization testing (PRNT) as previously described 36. RVFV PRNT titers > 1:40 were considered positive. The PRNT results followed a Gaussian distribution with the majority of positive ELISAs clustered around the 1:320 PRNT titer range.

4.1.4 Statistical Analysis. Initial univariate analysis was done to describe demographic variables. Bivariate analysis was calculated based on χ2 testing (or Yates’ correction to the

χ2 where appropriate) of several potential predictors (location, age group, gender) of RVFV seropositivity. All bivariate analysis was initially performed using R software, version 2.3.1 and confirmed using SPSS, version 15.

4.2 PROSPECTIVE SURVEY

4.2.1 Study Areas. This prospective study was based on a location-stratified household- based cluster sampling of the human populations residing in two areas near Masalani

Town, Ijara District, situated in a semi-arid region of Northeastern Province, Kenya. The estimated total area population is 7610. This location was selected for study as a representative, previously-active focus of RVFV transmission, recording 2 proven cases and 8 suspected cases of RVFV-associated hemorrhagic fever in the epidemic outbreak of

Nov. 1997- Feb.1998. Ijara is situated along the Tana River plain, which is prone to flooding from heavy seasonal rainfall and from heavy upland rains (Figure 6, black circle).

26 Flatness of the local terrain, combined with poor drainage makes the area a prime environment for RVFV transmission during flooding events, as evidenced by ongoing RVF outbreaks. The study was performed in March and April of 2006, approximately 8.5 years after the last RVF outbreak of 1997-98, and well before the flooding events of fall 2006 that were associated with the most recent RVF epizootic/epidemic. Based on the objectives of this study, the balanced sampling frame for selection of the planned 250 participants was divided between a rural village area, Gumarey (centered at 1°, 40’, 12” S, 40°, 10’, 48”E), and a town area, Sogan-Godud (centered at 1°, 41’, 24” S, 40°, 10’, 12”E) (Figure 6, black circle). Both are sub-locations defined within the Kenya Census and are located within

500m of each other. Gumarey has a largely semi-nomadic pastoralist population and local homes consist of traditional grass huts. Sogan-Godud is a larger town with more permanent tin-roofed dwellings and stores (see Figure 7).

Figure 7. Prospective study sites. Gumarey (left) and Sogan-Godud (right).

27 4.2.2 Sampling Strategy. Participants were randomly selected for inclusion in the prospective survey on a household basis using stratified cluster sampling design. The estimated total area population is 7610. In consultation with the TB-health coordinators and Medical Officer of Ijara District Hospital, who know the area well, a preliminary demographic census with GPS mapping determined the current local population and its distribution within-town (Sogan-Godud) and rural (Gumarey) strata. These strata were selected because of the expected significant differences in livestock exposure between town-dwelling and more rural and nomadic families, and the need to adequately sample both sub-populations. From this population frame, a stratified cluster sample based on households, with selection probability proportional to size, was used to select potential survey households from Masalani village area (rural and town). All censused individuals were serially enumerated. A random number table was then used to select a random number corresponding to an individual. Once selected that chosen individual and their household members were enrolled in the study. It was intended that each individual within a stratum would initially have an equal probability of selection, and the sample was expected to provide a representative sample of the population for study. Approximately equal numbers of subjects were enrolled from each stratum, in order to provide more precise measurement for the prevalence of outcomes among the smaller group (within- town). In this setting, the household-based cluster sampling strategy was considered to be the optimal means of randomization due to its cultural acceptability, its enhanced yield of subject participation, and its provision of a direct assessment of the immediate environment of the participants.

28 Using a stratified sampling of village and nomadic households in the Masalani area, approximately 250 healthy individuals were surveyed to determine the planned primary (serological) and secondary (epidemiological and clinical) outcomes of the study.

4.2.3 Population surveyed. Study recruitment was begun after consultation and approval by local administrators and religious leaders. After an initial demographic census to determine the current local population and its distribution, > 250 survey participants were selected by randomized cluster sampling of households in the two designated subsections of Masalani town. Children aged <1 year and those residing in the area < 2 years were excluded. All adult participants provided informed consent. Parents provided informed consent for participating children and children >7 years provided individual assent. The study sample consisted of a locally representative ethnic mix of >99% Somali or Bantu and <1% Indian or other Asian. Sample distribution was planned to be representative of the local population at risk, and (based on the latest Kenya census (1999)) of the population of Ijara District in general. Participating households were sampled using a probability proportionate to size (PPS) approach. To substitute for non-participating households, additional, randomly-selected households were chosen according to sampling rules established at the outset of the survey.

4.2.4 Subject Participation. The purpose of the study, the procedures, the eligibility criteria, the risks/benefits of participation, and the voluntary nature of the participation were described on the consent form and explained orally by study personnel in Kiswahili and Somali. The initial consent process for possible study participants took about 30-60

29 min/household, so that all questions could be answered satisfactorily and both adults and children fully understood the study’s purpose and what study participation entailed. Study participation lasted about 3 hours, in order for the questionnaire, physical examination, ophthalmologic examination, and venipuncture to be completed. Following completion of the eye exam, participants were further monitored closely for any side effects of pupillary dilation, and were not allowed to depart until their vision had recovered. Subjects participated only once in this study, in a single session of questionnaire survey, examination, and phlebotomy.

4.2.5 Examination procedures. Study participants received: 1) a structured interview regarding housing, animal exposure, motor function, visual function and recent or remote

RVF-related symptoms (Appendix Figure 1). (For the questionnaire process, accompanying parents served as proxies for children when necessary); 2) a complete physical examination, 3) a vision test and indirect ophthalmoscopic exam to check for signs of current or previous retinal inflammation, and 4) phlebotomy (i.e. venous blood samples-

5 mL in those older than 5 years and 1 mL in children less than 5 years old). All study personnel who administered questionnaires and performed exams were blinded to the RVF serostatus of participants. Blood samples were drawn and serum was prepared according to established protocols by standard techniques. Serum samples were aliquoted into sealed coded tubes and transported under refrigeration to Nairobi for testing. Testing for anti-

RVFV antibodies was then performed by screening ELISA and PRNT (see below).

30 4.2.6 Specimen Collection, Preparation, Handling and Shipping. Serum specimens were collected utilizing standard precautions and aseptic technique. Blood was aliquoted into sterile glass collection microtainers that were labeled with unique participant identifiers. Samples were initially stored on ice, and then serum was separated by centrifugation at the DVBD facility, and aliquoted into labeled storage tubes at Ijara

District Hospital. After transfer to DVBD headquarters in Nairobi, they were stored at -20 degrees F in a locked freezer at the DVBD national laboratory. Only protocol-designated

KEMRI or DVBD scientists and US collaborators had access to the samples. Aliquots of each specimen were then transported to CWRU laboratories for RVFV ELISA testing, and were tested concurrently at DVBD for confirmation and quality assurance. [As part of this project I traveled to Kenya and trained DVBD laboratory staff to perform RVFV ELISA testing. Results from DVBD and CWRU testing of samples were identical.] Because

RVFV ELISAs were not an established assay in DVBD labs, transportation of sera to the

US testing at CWRU and University of Texas Medical Branch, Galveston (UTMB) was considered necessary for optimization of serologic assay performance.

Blood specimens were shipped in coolers on wet ice from Ijara to processing, aliquotting, and temporary storage in the DVBD laboratory at Msambweni District

Hospital, thence to DVBD headquarters laboratories at the completion of the study

(Division of Vector Borne Diseases; Kenyatta Hospital, Ministry of Health; Nairobi,

Kenya). After freezing, serum aliquots for each subject were then shipped to CWRU for concurrent testing via expedited air freight shipment. After initial ELISA testing, all positive samples and an age-, sex-, and location-matched set of negative samples were sent to UTMB for confirmatory testing (by PRNT assays) via overnight air freight.

31

4.2.7 Laboratory testing. The primary measure of RVFV exposure was seropositivity measured by serum anti-RVFV IgG detection via enzyme-linked immunosorbent assay

(ELISA). Specimens were screened for the presence of anti-RVFV IgG via ELISA using lysates of VERO cells infected with the MP-12 strain (vaccine strain) of RVFV as the test antigen, and lysates of mock-infected cells as the internal control antigen. This ELISA assay has been established and validated in previous survey studies 33,37. Serum samples diluted at

1:100 were read at 405 nm and those scoring an OD value (corrected for reactivity on normal cell antigen) greater than mean + 2 SD for control sera, and absolute value greater than 0.2 were deemed positive. Each sample was run in duplicate and OD values were averaged. Any significant discrepancy in the ODs (measuring at least 0.05) between duplicate tests was resolved by repeat testing. Pooled RVFV-positive sera were used as the positive plate control and pooled RVFV-negative North American sera were used as the negative plate control.

Serologic screening was performed at the Division of Vector Borne Diseases in Nairobi, and confirmed at Case Western Reserve University with identical results. Confirmatory plaque reduction neutralization testing (PRNT) was performed at the UTMB to assess the risk of false positive results secondary to ELISA cross reaction with related viruses. All positive samples (N=33) and an age- and location-matched set of negative samples (N=33) had confirmatory testing via PRNT as previously described 36. All ELISA positive samples had PRNT titers of 1:80 or greater, the majority having titers of 1:320. All but one ELISA negative sample had titers of <1:10. This apparently ‘false negative’ sample had a PRNT titer of 1:80 on repeated testing.

32 4.2.8 Statistical Analysis. Initial univariate analysis was done to describe demographic variables. Bivariate analysis was calculated based on χ2 testing (or Yates’ correction to the

χ2 where appropriate) of several potential predictors of RVFV seropositivity (see Appendix

Table 1) as well as bivariate comparisons between villages (see Appendix Table 2).

Following initial bivariate analysis of RVFV seropositivity outcomes, predictor variables were further tested for association with RVFV seropositivity with the use of multivariable logistic regression. All 248 participant data were modeled using predictor variables that had been determined to be associated with RVFV seropositivity via bivariate analysis (see

Appendix Table 3). Individual predictors were tested for multicolinearity using a χ2 test

(see Appendix Table 4). Logistic models were also constructed by village to determine local predictors of RVFV seropositivity (see Appendix Table 5). Hosmer-Lemeshow

Goodness of Fit Chi-squared tests were calculated for all logistic models and indicated that model predictors sufficiently described the observed data (see Appendix Tables 3 and 5).

Because each variable assessed was unlikely to be independent from all others, repeat analysis using theory driven clusters were tested in repeat logistic modeling.

Clusters were devised using all known scientific evidence about these variables in relation to RVFV and were checked by an RVFV expert (Appendix Table 8). Initial bivariate analysis of clusters and RVFV seropositivity was performed (Appendix Table 9). Repeat logistic analysis was performed using clusters to proxy a measurement of exposure intensity (Appendix Table 10). Using Appendix Table 9, variables which were at least mildly associated (p<0.1) with RVFV seropositivity were included in the model. In addition, other relevant variables were included: age, sex, village. All variables were placed in the model and "backward (stepwise) conditional" was used to remove variables with the

33 removal threshold set at 0.1. All bivariate analysis and logistic modeling was initially performed using R software, version 2.3.1 and confirmed using SPSS, version 15.

4.2.9 Data processing and analysis. The results of questionnaires and of clinical and laboratory testing was entered into individual case report forms for archiving. Data were then double-entered from the report forms into password-protected databases using predetermined metafile coding values (and internal error-checking protocols) in order to minimize coding errors prior to analysis. A modified version of Microsoft Access was used to generate auditable linked data spreadsheets and the database was stored in password protected desktop computers at DVBD and CWRU, according to established protocols.

34 CHAPTER 5: ETHICS/PROTECTION OF HUMAN SUBJECTS

This study was conducted at an international site (Kenya). Thus, the investigators have ensured that this study was conducted in full conformity with the current revision of the

Declaration of Helsinki, or with the International Conference for Harmonisation Good

Clinical Practice (ICH-GCP) regulations and guidelines, whichever afforded the greater protection to the subject.

5.1 Institutional Review Board

The Case Western Reserve University Institutional Review Board (IRB) approved the protocol for this study (approval number 10-04-09). In addition, the Kenya Medical

Research Institute (KEMRI) also approved this protocol. KEMRI is an institution holding a current U.S. Federal-Wide Assurance issued by OHRP. Both IRBs approved the English and translated versions of the consent forms, assent forms, questionnaire, and data collection forms. Adult Consent involved a full length written and an Oral Script.

Consent for minor participants was made by a parent or guardian and included written assent, when appropriate. Much of this research was conducted among a non-English speaking population, and consent documents were translated into Kiswahili or Somali by the research staff from DVBD. Any amendments to the protocol or consent materials were approved before they were placed into use.

5.2 Informed Consent Process

Informed consent is a process that is initiated prior to the individual’s agreeing to participate in the study and continuing throughout the individual’s study participation.

35 Extensive discussion of risks and possible benefits of participation in this study was

provided to the subjects and their families in the language of their choice. Consent forms

describing in detail the study procedures and risks were given to the subject and written

documentation of informed consent was required prior to enrolling in the study. All

consents were obtained using the full-length text. For people who were literate, they

were asked to read the full text and if they consented to participate to sign the document.

For those who were not literate, field staff read the entire text to them in their language of

choice, and asked them to sign. This was done for both adults and for parents/guardians who were consenting on behalf of minors for whom they are responsible. Consent forms were IRB approved. Upon reviewing the document, the investigator explained the research study to the subject and answered any questions that arose. The subjects signed the informed consent document prior to being enrolled in the study. The subjects had the opportunity to discuss the study with their surrogates or think about it prior to agreeing to participate. The subjects were able to withdraw consent at any time throughout the course of the study without any ill effects. A copy of the informed consent document was given to the subjects for their records. The rights and welfare of the subjects was protected by emphasizing to them that the quality of their medical care would not be adversely affected if they declined to participate in this study.

5.3 Assent Process

For children whose parents agreed to their participation in the study, an assent process was performed. Children were allowed to ask questions, and then the assent form was read to them, in their language of choice. They were then asked to sign the assent

36 form and the child consent form, if they agreed to comply with the study. The parents or

legal guardian of the child were also asked to sign the child consent form. Children could

decline to participate in the study at any time for any reason.

5.4 Subject Confidentiality

The study protocol, documentation, data and all other information generated were

held in strict confidence. No information concerning the study or the data was released to

any unauthorized third party. All data were collected by trained Kenyan staff and were

stored in locked cabinets at Ijara and Msambweni District Hospitals to maintain

confidentiality and limit accessibility to pertinent personnel. Maintenance of

confidentiality by study workers and protection of research records minimized the

potential for social harm due to inadvertent disclosure of research data. Subject identity

was masked by use of coded ID numbers for the samples. During handling of outcomes

data from this study for statistical analysis, identifying features were made unavailable in

the databases by using random tracking numbers to enter each subject's data. By doubly

masking subject identity, and by maintaining data in password protected data files, the

participants' confidentiality was maintained at the highest possible level. In order to afford more identity protection, the identity of participants, the data were aggregated in any tables and maps were presented in publications and reports. Unique identifiers for individuals and households were removed when compiling and analyzing the data, and were stored separately so that confidentiality was maintained. Kenyan staff hired by

CWRU and the Division of Vector-Borne Diseases entered results, and data was maintained in an Access database on password-protected computers. The computers

37 have limited access, and the Access database was also password protected. Other original documents, such as memoranda, checklists, and notes were saved and stored in locked cabinets. Serologic samples bearing coded ID numbers were tested concurrently at DVBD and CWRU, and stored in locked freezers with limited access.

5.5 Future Use of Stored Specimens

All serum specimens that were collected as part of this protocol were stored bearing a unique identifier only and no personal information. Specimens were transported to CWRU as previously detailed. Transportation of sera to the US was necessary for optimization of serologic testing at CWRU and UTMB. If necessary in the future, a written request and approved protocol will be submitted to KEMRI for use of these stored specimens in any other study or research.

38 CHAPTER 6: RESULTS

6.1 RETROSPECTIVE SURVEY

6.1.1 Serological outcomes for study subjects. Initially, 1263 samples were screened for

anti-RVFV IgG via ELISA (see Table 1). Of the 143 ELISA-positive samples,

confirmatory PRNT indicated that the overall anti-RVF seropositivity of the study sample was 10.8% (136 samples/1263 total). This confirmatory testing indicated a relatively low

ELISA false-positive rate of 4.9% (95% CI 2.02%-10.08%, 7 PRNT negative samples/ 143

ELISA-positive samples). All of the 142 sex-, age-, and location-matched ELISA-negative

samples had negative PRNT results (titers <1:10 for 140 samples, and 1:10 for 2 samples);

and thus no ELISA false-negatives were found in this relatively small sample (0%, 95% CI

0-2.6%). It is possible that a few positives may have been missed by ELISA pre-screening,

because all study samples did not undergo PRNT (only the ELISA positives and a sex-, age-

and location-matched set of ELISA negatives had PRNT). All data presented below pertain

to PRNT results.

Table 1. Collection Date, Number, Gender, and Seropositivity of RVF Serum Samples by Study Site

Location Date Total Positive Males Females Positive Positive Prevalence of 95% N N Males Females Anti-RVFV Confidence IgG (%) Intervals (%) Lokichoggi 1994 675 129 286 384 42 86 19.1 15.76–21.66 o Kabobo 1996 119 0 47 72 0 0 0 0–3.03 Kabobo 1997 101 1 31 69 0 1 1.0 0.03–5.45 Daragube 1996 169 5 81 75 4 1 3.0 0.94–6.78 Msambweni 1997 106 1 0 106 N/A 1 0.9 0.02–5.19 Msambweni 1998 93 0 0 93 N/A 0 0 0–3.89 Total 1263 136 445 799 46 89 10.8 9.13–13.11

39 6.1.2 Differences by location. Seropositivity varied

significantly according to study area, with the

Lokichoggio area of Turkana District in

Northeastern Province having the highest RVFV

seroprevalence of 19.1% (129 samples/675 total)(see

Table 1) which was significantly higher than all

other locations (Fisher’s exact test, P<0.001).

Daragube, located on the monsoonal coastal plain,

had the second highest RVFV seroprevalence of Figure 8. RVFV Seropositivity by location 3.0% (5 samples/169 total) with lower

seroprevalence in the nearby coastal Msambweni area (1%), and in the highland Kabobo

area (1%) (See Figure 8).

6.1.3 Differences by age group. Seroprevalence also varied by age group, with those

persons 15 years old and younger from Lokichoggio having a significantly lower RVFV

seroprevalence, 10.0% (26 samples/260 total) vs. 25.0% (102 samples/408 total)(Χ²=

23.067, p ≤ 0.001) among those 16 years and older (see Figure 9). The Daragube sample had similar age group seroprevalence trends with those 15 years and younger having a

seroprevalence of 0% (0 samples/71 total) vs. 6.0% (5 samples/83 total)(Χ²= 4.421, p ≤

0.05) among those 16 years and older.

40 Lokichoggio 1994 RVF Seroprevelance by Age Group

0.4 0.39

0.34 0.34 0.34 0.33

0.3 0.30

0.25 0.25 0.22 0.2 0.21

0.17 0.17

0.14 RVF Seroprevelance 0.11 0.1 0.10 0.11

0.07

0.01 0 5-15 16-25 26-35 36-45 46-55 56-65 Age Group (years)

N=260 N=106 N=105 N=124 N=53 N=19

Figure 9. Lokichoggio RVFV seroprevalence by age group.

6.1.4 Differences by gender. Seropositivity varied significantly according to gender in the

Lokichoggio sample. Women were at greater risk of infection than men, 22.4% (86 samples/384 total) vs. 14.7% (42 samples/286 total)(Χ²= 6.306, p ≤ 0.025). Gender differences from other regions could not be analyzed secondary to small number of positive samples.

6.1.5 Differences over time. Msambweni, a coastal region, had one seropositive sample in

March of 1997 (1 sample/105 total), but had no seropositive samples in March of 1998.

This sample consisted only of women so is gender-biased and may not reflect community seroprevelance accurately. This pattern is quite distinct from the large RVF outbreak in nearby Garissa that occurred in the interim period. Kabobo, in the central highlands, had no

41 seropositive samples in August of 1996 (0 samples/119 total), but had one seroconversion detected in April of 1997, yielding a RVF seroprevalence rate of 1% (1 sample/100 total) prior to the observed period of epizootic/epidemic transmission in late 1997. This finding is remarkable because interepidemic transmission of RVFV in a non-arid area has not been previously documented.

6.1.6 Hypotheses generated from retrospective study data. Many hypotheses were generated from the retrospective study that led to the design and performance of the prospective study. First, the highest RVFV seroprevalence was documented in semi-arid northwestern Kenya. Was the greater exposure due to the specific location and cultural influence or rather to the semi-arid environmental effects? This question was addressed in the prospective study by studying another semi-arid region of Kenya (northeastern) that differed in location. Second, the highest RVFV seroprevalence was documented in those greater than 15 years of age. Could this finding be repeated and are children at less risk of exposure due to determinable factors? In the prospective study, a greater number of children were enrolled and tested to confirm the finding and detailed information about risk factors was collected. Third, women were documented to have higher RVFV seroprevalence rates in the Lokichoggio, Turkana sample, which contradicted previous studies that reported male gender as a risk factor. Was this finding accurate and repeatable?

Did Kenyan women put themselves at risk for RVFV exposure or was this a sampling bias?

In the prospective study, a balanced sample of men and women were included to evaluate these possibilities.

42 6.2 PROSPECTIVE SURVEY

6.2.1 Survey Results. 270 potential participants were initially invited to participate, based on selection by randomized cluster sampling of 66 households in the two designated administrative sublocations of Masalani Town in Ijara district. Two hundred forty-eight (91.9%) of this selected sample group completed all study procedures, including serum testing (see Table 2).

Gumarey Sogan-Godud

Participants 143 127

With Complete Data 122 126

Households 30 36

With Complete Data 28 35

Participants per Household 1-18 (mean 4.4) 1-12 (mean 3.6)

Setting Village Town

Age Range (years) 2-81 1-86

Median 16 16

Mean 25.8 23.1

Female 73 (60%) 82 (65%)

Seropositive 25 (20.5%) 8 (6.4%)

Seropositive Age Range (years) 12 - 81 4 - 71

Seropositive Females 11 (15.2%) 3 (3.7%)

Seropositive Males 14 (28.6%) 5 (11.4%)

Table 2. Demographics of Masalani Town survey population

43 The final study cohort consisted of 248 participants, of whom 33 (13%, C.I.95% 9.3-18.1) were RVFV seropositive. Forty-nine percent of subjects (N= 122) were from Gumarey, and of these, 25 (20%, C.I.95% 14.0-29.2) were seropositive. Fifty-one percent of subjects

(N = 126) were from Sogan-Godud, and of these, 8 (6%, C.I.95% 2.7-11.8) were seropositive (see Figure 10). 118 (47.6%) of all samples were from children 15 years and under, and 3.4% (4/118) of these were seropositive. The youngest seropositive subjects were aged 4, 12, 13, and 14 years old and all were long-term permanent residents of the study area. Of the adults in the sampled cohort, 29/130 (22.3%) were positive for anti-

RVFV IgG, the oldest of whom was 81 years old.

270 Survey participants 248 with complete data

13% Seropositive; N=33

47% 53% Age ≤15y Age ≥16y N=117 N=131

3% 22% 49% 51% Seropositive Seropositive Gumarey Sogan-Godud N=4 N=29 N=122 N=126

20% 6% Seropositive Seropositive N=25 N=8

Figure 10. Flow chart of study participants: Youngest seropositive aged 4 y; oldest, 81 y.

44

6.2.2 Links between past exposure and seropositivity. Many exposures, both non-

animal and animal, were associated with RVFV seropositivity (see Appendix Table 1).

In bivariate statistical analyses, RVFV seropositivity varied significantly according to the following factors: age (subjects ≥15 years old had more risk, p=.0001), sex (males had more risk, p=0.011), location (those from Gumarey had more risk, p=0.001), as well as the following exposure histories: home flooding (p=0.024), contact with a dead human body (p=0.0001), cattle contact (p=0.012), and involvement in sheltering (p=0.003), , butchering (p=0.0001), skinning (p=0.0001), cooking (p=0.005), milking (p=0.0001), or birthing livestock (p=0.0001), or disposing of an aborted animal fetus (p=0.0001).

Other reported exposures varied significantly between the two sub-location groups, with those from Sogan-Godud more likely to use mosquito nets (OR 5.2, P =

0.0001) and mosquito coils (OR=8.2, P = 0.0001) to reduce insect exposure; whereas those from Gumarey were more likely to have had goat contact (OR=2.6, P =0.046), cattle contact (OR= 4.7, P = 0.0001), drink raw milk (OR= 4.1, P = 0.0001), shelter livestock (OR= 2.6, P ≤ 0.002), butcher livestock (OR=1.5, p=0.0001), birth livestock in the home (OR= 2.1, P = 0.005), dispose of a livestock fetus (OR=1.7, p=0.005), or to have had direct contact with dead human remains (OR=2.1, p=0.026)(see Figure 11 and

Appendix Table 2).

45 Exposure Differences Between Villages

Fetus Disposal*

Birthed Livestock**

Drank Raw Milk***

Butchered Livestock***

Sheltered Livestock**

Cattle Contact***

Goat Contact*

Dead Human Contact*

Mosquito Coil***

Fire Use***

Mosquito Net Use***

0 102030405060708090100 Number of Participants Gumarey Sogan-Godud

Figure 11. Exposures between villages differed: Gumarey had more animal exposure and Sogan-Godud had more mosquito control. *P<0.05; **P<0.01; ***P≤0.001 by Chi-square testing.

Because many of the exposure variables were related, an attempt to cluster exposure variables into groups was made, using previous knowledge of RVFV risk factors (Appendix Tables 8 and 9). The mosquito protection factor (combined use of mosquito nets, use of fire, and use of mosquito coils) was protective, with every unit increase in frequency of behaviors associated with using mosquito protection at 0.68

(C.I.95% 0.48, 0.96) the odds of being RVFV seropositive. Those who had partaken in

animal processing (slaughtering, butchering, skinning, or cooking livestock) had over two times (OR 2.22, C.I.95% 1.57, 3.13) the odds of being RVFV seropositive for every unit

46 increase in frequency of behaviors associated with animal processing, compared to those

who did not. Of great interest is the risk conferred by the animal living factor (milking,

birthing, or sheltering livestock). Those with this factor had over three times the odds of

being RVFV seropositive (OR 3.42, C.I.95% 2.07, 5.65) for every unit increase in

frequency of behaviors associated with animal living compared to those who did not.

Because data were not collected on the intensity of animal exposure, these theory-driven

factor clusters yield insight into the important role that animal contact has in RVFV

exposure in this region.

The final logistic model to predict RVFV seropositivity included age, location,

sex, and disposal of an aborted animal fetus (see Appendix Tables 3 and 4). In multivariable logistic regression models used to predict adjusted odds of RVFV seropositivity, location was significant controlling for age and sex, with those residing in

Gumarey being at four times the odds of those in Sogan-Godud (Adjusted OR = 4.15;

C.I.95% 1.59 – 10.87). Seropositivity also varied by sex controlling for age and location,

with males having nearly 3 times greater odds ratio than females (20% vs. 9%, Adjusted

OR = 2.78; C.I.95% 1.18 - 6.58 for men vs. women), but this difference did not remain

significant within sub-location analysis. Those who had disposed of an aborted animal fetus, controlling for age, sex, and location, had 3 times the odds of being seropositive

(72.7% vs. 35.7%, Adjusted OR 2.78, C.I.95% 1.03-7.52). Age and location, but not sex,

were associated with disposal of an aborted animal fetus, such that those who were older

or from Gumarey were more likely to do so.

Subgroup analysis by village revealed the significant predictors of RVFV

seropositivity in Gumarey to be ill family member, disposal of an aborted fetus, and sex

47 (see Appendix Table 5a). Displacement by flood was also associated with RVFV

seropositivity in Gumarey, but could not be included in the model because every

seropositive individual was displaced by flood and this factor was overdetermined. Men

had over 3 times the odds of being seropositive compared to women: (Adjusted OR 3.45,

C.I.95% 1.17-10.19). Disposal of an aborted animal fetus (Adjusted OR 15.11, C.I.95%

4.445-51.35) and presence of an ill family member (Adjusted OR: 18, C.I.95%1.35-

246.97) were also associated with RVFV seropositivity. In Sogan-Godud, the logistic

model to predict seropositivity included age, such that the odds of seropositivity

increased 5% for every one year increase in age (see Appendix Table 5b). Children less

than 15 years had a much lower risk of RVFV seropositivity than those 15 years and

older. The adjusted OR for seropositivity (calculated from the overall logistic model)

was 1.05; C.I.95% 1.03 - 1.07 per year of age (see Figure 12). This difference persisted at

the sub-location level with those children in Sogan-Godud still with significantly lower

risk than adults (Adjusted OR for each year of age= 1.07; C.I.95% 1.028 - 1.114).

Total

y 1 Sogan-Godud 0.8 Gumarey 0.6 0.4 0.2

RVFV Seropositivit RVFV 0

5 5 1-5 -45 65 -75 6-15 6 76+ 16-25 26-3 3 46-5 56- 66 Age (years) Figure 12. Anti-RVFV IgG seropositivity by decade of age and village of residence: Gumarey had a higher rate in almost all age groups compared to Sogan-Godud.

48 A model constructed with the theory-driven clusters resulted in similar findings.

The final logistic model to predict RVFV seropositivity included age, location, sex, and

animal living factor (see Appendix Table 10). In multivariable logistic regression models

used to predict adjusted odds of RVFV seropositivity, location was significant controlling for age and sex, with those residing in Gumarey being at nearly four (Adjusted OR 3.91,

C.I.95% 1.49, 10.2) times the odds of those in Sogan-Godud. Seropositivity also varied by

sex controlling for age and location, with males having over 3 times (Adjusted OR 3.09,

C.I.95% 1.30, 7.35) greater odds of seropositivity than females. For every unit increase in

frequency of behaviors associated with animal living factor (milking, birthing, or

sheltering livestock) there was a 2 times (Adjusted OR 2.09, C.I.95% 1.16, 3.74) odds

increase in seropositivity. Finally each year in age resulted in a 3% (Adjusted OR=1.03,

C.I.95% 1.01, 1.06) increase in odds of being RVFV seropositive. Although this model

accounted for 36.9% of the variance of the outcome variable (RVFV seropositivity), the

goodness of fit testing was not favorable. It is likely that this model now has too many

predictor variables and is “over-fit” for the outcome variable. Overall correct

classification occurred 89.5% of the time when using this model.

6.2.3 Symptom history and current physical examination. In the survey questionnaire,

reported symptoms included fever, malaise, myalgia, chills, backache, eye pain,

headache, rash, red eyes, photophobia, poor appetite, flushing, nausea, vomiting, meningismus, poor vision, epistaxis, hematemesis, hematochezia, bruising, confusion,

vertigo, stupor, and coma (see Appendix Figure 1). Of these, a past history of myalgias

(OR=6.03, p=0.0001), backache (OR=3.86, p=0.003), eye pain (OR=2.28, p=0.034), red

49 eyes (OR=2.75, p=0.008), meningismus (OR=2.97, p=0.004), poor vision (OR=2.74,

p=0.008), and coma (OR=14.55, p=0.005) were statistically associated with RVFV

seropositivity in the study population (see Appendix Table 6). Upon physical

examination, no current, non-ocular examination finding was specifically associated with

RVFV seropositivity.

Symptoms of RVFV disease were also grouped into theory-driven clusters to

evaluate whether certain clinical syndromes were associated with RVFV seropositivity

(Appendix Tables 8 and 9). In this sample, meningoencephalitis and hemorrhage factor

were not associated with RVFV seropositivity.

6.2.4 Ophthalmological findings. Of the 18 identified cases of significant retinal

disease in the survey population, 7 were seropositive (7/33, 21.2%) compared with 11

who were seronegative (11/215, 5.1%) (p= 0.003, Χ² = 8.75). All of the subjects with eye disease were over 21 years old, and all of the seropositives with eye disease were at least

50 years old. The Odds Ratio of late eye disease associated with RVFV exposure

(seropositivity) was 4.99 (p = 0.003)(see Appendix Table 7). Measured visual acuity ranged from 6/5-6/60 (equivalent to 20/17-20/200) in the seronegative group and 6/5-

6/36 (20/17-20/120) in the seropositive group, although both groups had individuals with extremely poor vision who could decipher only large objects (measured by finger counting) or who could not perceive light. Current visual acuity differed statistically between groups and was more likely to be worse in the RVFV seropositive group (visual impairment defined as ≥ 20/80: 12% of seronegatives vs. 25% of seropositives; p =

0.047, Χ² = 3.94). Among the 18 people with retinal disease, 14 (78%) had visual

50 impairment, and among the 7 seropositives with retinal disease, 5 (71%) had visual

impairment. There was no one distinctive lesion associated with RVFV seropositivity,

though the eye diseases differed between the groups. Seropositive subjects with eye

disease had findings of optic atrophy (3), retinal hemorrhage (2), and retinal scarring (3).

One person had both retinal hemorrhage and scarring. By contrast, seronegative subjects

with eye disease had uveitis (1), vasculitis (1), maculopathy (3), peripapillitis (1), retinal

scar (1), optic scar (2), retinal atrophy (1), and retinal degeneration (1).

Retinitis factor (poor vision, eye pain, and red eyes) was associated with RVFV

seropositivity in this sample. Those with these symptom complaints had 1.5 (C.I.95%

1.15, 2.15) the odds of being RVFV seropositive for every unit increase in frequency of symptoms associated with retinitis factor (see Appendix Table 9).

51 CHAPTER 7: DISCUSSION

RVFV is native to sub-Saharan Africa and natural epizootic/epidemic outbreaks of

RVF have been detected 8 times during the second half of the 20th century in Kenya 27.

After sheep and cattle were introduced into Africa, the disease pattern became one of apparently low-level enzootic and endemic activity punctuated by massive transmission during periods of very high rainfall 26. Introduction of RVFV to new areas has been linked

to livestock movement 38, environmental modification 2,39,40, and weather changes 4. Thus far the knowledge of human RVF epidemiology has been limited to outbreak investigation and only sparse data exist about seroprevalence in endemic areas, such as Kenya. As witnessed in prior outbreaks, the detection of even low levels of human seroprevalence can have enormous significance for RVF outbreak potential in a given area 39.

By testing archived human serum samples collected as part of well-defined

community surveys across Kenya, this study demonstrated that RVF was present in

populations outside the Garissa District both before and during the 1997-1998 epidemic,

with a collective seropositivity rate of nearly 11%. Past serosurveys of humans and

primates from different parts of Kenya suggest that unrecognized transmission is occurring

on a regular basis across the country 20,22,41,42. The locations’ seroprevalence rates parallel

those recorded via a non-systematic sampling during an interepidemic period in the early

1980s, although the absolute rate values were dissimilar possibly secondary to temporal trends and testing procedures 42. This study confirmed the highest prevalence of anti-RVFV

antibodies in northwestern Kenya. Johnson, using the less sensitive indirect fluorescent antibody test, found a seroprevalence of 9% in Lodwar in 1983 42 which is significantly

52 lower than the 19% prevalence found in the same district in Lokichoggio in 1994 in this study. The Coast Province and highland samples were significantly lower in anti-RVF seroprevalence than the northwestern samples tested.

The variation in location may be secondary to multiple factors, including environmental differences. RVFV transmission is closely related to flooding events, and transmission can be widespread during a period of excessive rainfall and flooding 27,43.

Most regions of Kenya have two rainy seasons, the long rains falling between April and

June, and the short rains between October and December. Average annual rainfall varies from 5 inches a year in the most arid regions of the northern plains to 70 inches a year near Lake Victoria. The coast and highland areas have an average of 40 inches per year.

In times of very heavy rainfall, many Kenyan regions could be at extensive risk for epidemic RVF. In ordinary times, climatic and terrestrial factors define area risk. Areas that are arid in the dry season, such as Turkana, often have extensive standing water during the rainy season. These conditions enable the hatching of transovarially infected flood water Aedes and initiate local transmission. Thus, the semi-arid climate in northeast

(Masalani) and northwest (Lokichoggio) Kenya allows for annual RVFV transmission and results in high human seropositivity rates, whereas the highland climate of Kabobo may only permit RVFV transmission during periods of dramatic rainfall and manifests as low

RVF seropositivity.

Cultural differences can also modify RVF risk. The Turkana people are a nomadic pastoralist society that have substantial contact with their herds and depend on livestock for their survival and economics. Because the Turkana people live in such close proximity with their herds, they could easily be infected by mosquitoes or by blood/abortus from

53 the infected animals. While direct transmission from livestock is not thought to be

important, shed blood and abortuses contain large amounts of virus and the high viremias may also foster mechanical transmission 44. Tissue from RVFV-infected animals that abort

or die in the vicinity of mosquito larvae breeding habitats may contaminate the water and

infect mosquito larvae, a cycle that propagates transmission 45. Local factors, such as

proximity of populations to flood-water Aedes breeding sites, secondary vectors and their

biting habits, and the relationships of humans to their livestock, can foster RVFV

transmission independent of the overarching seasonal climactic factors. Therefore the

strikingly high RVF seroprevalence in the Turkana population is logical and important, for it

is this group who is most at risk for RVF infection because of the high local transmission

and who would be most impacted by a RVF outbreak because of their community’s reliance

on livestock. Local factors may also help to explain the slight non-significant variation seen

across Kenyan regions with similar climates (3% in Daragube vs. 1% in Msambweni).

Lokichoggio and Daragube seroprevalence rates were greatest among persons 16

years and older. The increased seroprevalence rates among older populations in these two

regions suggest a period of RVFV transmission in these areas around 1980. The age of the

youngest seropositive subject (5 years old in Lokichoggio and 22 years old in Daragube)

provides a maximum estimate of the time since the last period of RVFV circulation. It is

likely that interepidemic transmission is ongoing in Lokichoggio, given the high overall

seroprevalence and the detection of many seropositive young children (see Figure 9). The

seroprevalence spike in older Turkana children, both male and female, supports the idea that

there was an intense episode of transmission occurring >15 years before the samples were

taken. Out of 77 samples tested of Daragube children 18 years of age or less, no samples

54 were positive. This makes it less likely that significant interepidemic RVFV transmission

was occurring there in the two decades prior to sample collection. The community IgG age

profiles give evidence of past RVFV transmission that is out of phase with the past

outbreaks reported from these regions, and suggest the existence of previously undetected but potentially significant local transmission. Climate data reveals that an ENSO event occurred around 1977 and may have contributed to increased RVF transmission leading to increased seroprevalence among study subjects born previous to that time 27.

In the Lokichoggio sample population, women were more likely to be RVFV seropositive than men. Turkana women stay near the hut tending to cattle, caring for children and cooking, while men roam the pastures tending to their herds of sheep and goats.

Direct contact with infected animals is a known risk factor for RVF 46. Prior studies have

also shown that being male, having contact with sheep blood, amniotic fluid, or milk, and

sheltering livestock in the home were associated with RVFV infection 4. The initial

hypothesis was that the Turkana men have more direct contact with livestock and would,

therefore, be at greater risk for RVF, but women were more likely to be RVFV

seropositive in this subset. Perhaps competing mortality from other causes resulted in the

men dying at younger ages, thereby having less chance to be exposed to RVFV and a

lower RVFV seroprevalence. These data underscore the current lack of understanding of

the RVF risk factors. Household risks or certain animal exposures, such as cattle contact

or birthing of animals, may prove in future studies to be independent risk factors for

RVFV infection and explain the higher RVFV seropositivity rates detected among

Turkana women.

55 The seroconversion in Kabobo of a 22 year old female between 1996 and 1997 may reflect a very local RVF re-emergence or extension to the highlands region. The highlands people of Kabobo have limited travel or contact with other populations and therefore the

RVFV seroconversion detected likely represents local exposure and infection. Studies

conducted in this region have previously demonstrated significant variation in malaria transmission depending on total and seasonal rainfall effects 34,47. The 1996 samples were

collected in July and August of 1996 during a prolonged rainy season, whereas the samples

from 1997 were collected in March and April after the rainy season was complete. This

isolated seroconversion, combined with the detection of other foci of transmission during

the 1997/98 outbreak, indicate that RVF transmission may be occurring sporadically in

many non-arid agricultural areas of central Kenya, such as the highlands.

In addition, this prospective study highlights the variability in RVFV

seroprevalence in high risk settings. In semi-arid NE Kenya, older age, rural village

location, male sex, disposal of an aborted fetus, and eye disease were significantly

associated with RVFV seropositivity. RVFV seropositivity was relatively high in this

sample population in Masalani town, Kenya, particularly in the village area (Gumarey),

where seropositivity rates were nearly four times higher than the town area (Sogan-

Godud) separated by a distance of only 500m. Clues to the reasons for this discrepancy

in seroprevalence were identified in this study. Those from Gumarey were significantly

more likely to have mosquito and animal exposures than those from Sogan-Godud.

These risk factors coupled with the most important predictors of rural seropositivity, male

sex and disposal of an aborted animal fetus, yield evidence for disparate risks for RVFV

infection in different communities.

56 As identified in this retrospective study, RVFV seroprevalence can vary

significantly across Kenya 37. This study shows that large seroprevalence discrepancies

can also occur over very short distances. Spatial risk assessments of animal RVF in

Senegal have been developed using seasonal rainfall, land surface temperature, distance

to perennial water bodies, and time of year 48. Designing such risk maps with inclusion

of human risk factor data may allow for improved surveillance systems and better

prediction of the spatial distribution of RVFV. This information gathered with satellite

imagery 27, and large scale cluster analysis 49 can be used not only to predict large outbreaks, but also to identify local hot spots of RVFV transmission to optimize RVF

control in resource limited settings.

For each year of life, the odds of being RVFV seropositive increased by 5%.

Males had nearly three times the odds of being seropositive than females in this study, a

risk that was noted in the 1997 RVF outbreak investigation, but not in the retrospective

study 4. The difference in seropositivity between sexes is not explained on the basis of

reported animal or non-animal exposures which were comparable and not statistically

different between sexes. The increased seropositivity among males may have an

important biological basis, given that outcome of infection and resultant immune

response to other viruses have been linked to sex differences 50. Males may spend more

time tending to ill animals in herds and may increase their RVFV exposure in that way.

Disposal of an aborted animal fetus was associated with nearly three-times

increased odds of RVFV seropositivity. This finding may indicate the importance of

RVFV transmission by aerosolization of blood and amniotic fluid during animal birthing

in Kenya. It is unknown whether aerosol or vector-borne transmission is the dominant

57 form of transmission during interepidemic or epidemic periods. Study analysis indicates that disposal of an aborted animal fetus was an important associated risk factor at both the composite and sub-location level. Planned repeat sampling of this cohort since the last outbreak of 2006-7 will allow better evidence to address this important knowledge gap.

Evidence of interepidemic human transmission of RVFV was found, which has not been previously shown. The validation of seropositive young children, born after the last documented outbreak in 1997-1998, indicate that low-level interepidemic transmission to humans is continuing in the Masalani area, and likely in other areas in

Kenya 37. The natural reservoir for RVFV and the mechanism by which humans become infected during interepidemic periods are unknown. Wild animals have been shown to be infected with RVFV but further studies must determine whether these animals play a role in RVFV maintenance between outbreaks 51.

Important differences between the seropositivity rates of those with and without eye disease were demonstrated. Those with chronic retinal disease had 5 times the odds of being RVFV seropositive. A difference was observed in visual acuity between RVFV seropositives and seronegatives in the study sample tested 8 years following the last outbreak and perhaps greater changes may have been present during acute RVF disease.

Although there were no ocular findings that were pathognomonic for prior RVFV infection, this finding supports evidence from previous studies on the oculopathogenesis of RVFV 10.

No specific non-ocular exam finding was associated with RVFV seropositivity, but there were several reported symptoms that were statistically more common in those

58 who were RVFV seropositive. The majority of these symptoms were severe neurologic

manifestations of disease, such as neck stiffness, confusion, and coma. It is known that

RVFV can cause encephalitis 1, and this type of inflammation may explain the higher

prevalence of these reported symptoms among seropositive subjects. Myalgia and

backache may be present in the majority of non-severe RVF cases, and are not specific to

RVFV infection. Poor vision, which was noted to be more common among RVFV

seropositives in this sample, may be an indicator for RVF retinitis, a common sequela of

RVFV infection 10,29.

RVFV IgG ELISA and PRNT antibodies are believed to last decades after

infection and so provide a reliable index of prior RVF exposure. In contrast, though less

well studied, it appears that IgM is lost in 50% of patients by day 45, and is absent in

100% by 4 months after infection 52,53. IgM testing was not performed in this study

although it might have yielded useful additional information about acute RVFV infection.

Although ELISA positivity may prove to be 'false-positive' due to cross-immunoreactivity

for viruses in the same family, it was low (4.9%) in this study, and the addition of

confirmatory PRNT testing of ELISA positives is able to greatly improve viral specificity 54.

New ELISA methods may prove to be even more sensitive and specific than PRNT tests in

the future 55.

Recent evidence from the Kenya RVF outbreak in 1997-98 indicates that case- finding based on clinical case-definition has poor specificity 4. At the same time, the low

incidence of the dramatic clinical manifestations of RVFV infections has made it difficult to

estimate the full extent of RVFV infection among humans during periods of lesser

transmission. Having high endemic transmission can significantly alter the positive

59 predictive value (PPV) of anti-RVF IgG detection by ELISA, reducing it to 50% in locations like Lokichoggio during an epidemic. Vaccine efficacy estimations could also be

flawed if endemic transmission and baseline seropositivity rates are not considered. It is

important to note that although the PPV of ELISA IgG is modest in highly endemic areas

during epidemic periods, the negative predictive value is excellent (100%, 142 negative

ELISA samples/142 negative PRNT samples). This suggests that RVF IgG testing may be

more useful for surveillance than outbreak investigations in these endemic regions when

outbreaks occur. In coastal or highland areas where baseline RVF seropositivity is low,

ELISA IgG is both useful for surveillance and outbreak screening, since the PPV improves

with increasing disease prevalence, and the likelihood of a false negative test result is low.

This retrospective study may have limited generalizability, since serum samples

were only tested in three distinct regions of Kenya. Although this study cannot comment on

the RVFV seroprevalence in other Kenyan regions, it highlights the valid point that one

cannot speak in general terms about RVF prevalence even on a provincial basis in Kenya.

This retrospective sample set was limited by its cross-sectional design and time- and gender-

confounding. Sampling was done cross-sectionally across different regions and in some

cases (Msambweni) only in certain genders. Different RVFV exposure data may have

resulted if longitudinal studies had been performed across all genders in the chosen areas.

Few samples were collected from those older than 50 years, because of the age distribution

of the community. Perhaps an enhanced sampling of those greater than age 50 years would

have yielded different RVFV seropositivity rates in this elder subset. The lower prevalence

documented in elders may be accurate or may be flawed due to many causes including, a

60 cohort effect, an age effect, a survivor effect, or limitations of ELISA testing in older

individuals.

This prospective study was limited by its cross-sectional design and therefore

cannot draw conclusions about whether the identified risk factors specifically caused

RVFV exposure. The validity of the associations in this study relies on accurate recall of

exposures by the study participants. Although questionnaires included information about

the timing of symptoms and exposures, during questioning language differences limited

accurate collection of these data. This study may have limited generalizability; risk

factors from a small population in Masalani was tested and risks may vary in other parts

of Kenya or in other countries. Data were collected on animal exposures in a binary

fashion, so no information about magnitude or duration of contact is known, which may

have significant impact on risk estimations. Unfortunately, no quantitative exposure data

for the RVFV vectors in this study area was included. In order to estimate intensity of

exposure, theory-driven clustering of exposure variables allowed for an ordinal scale and

proxy of environmental exposure.

The primary outcome measurement data for both the retrospective and prospective

studies, anti-RVFV IgG by ELISA, were collected as binary and not continuous titer

(conventional standard) important information may have been lost about the height and

duration of titer. These data may have yielded important clues to timing of exposure,

robustness of response, and duration of immunity that cannot be evaluated in the current

study. All laboratory methods, including ELISA, have limitations. In the retrospective

study, PRNT indicated relatively low rates of ELISA false-positives (4.9%) and false-

negatives (0%). PRNT is designed to confirm ELISA positivity and in this way can account

61 for the false-positive ELISA results, which result from cross-reactivity with other related viruses. Although no false negative ELISA samples resulted from the retrospective study, one prospective sample tested negative by ELISA, but was confirmed to be a low titer positive (1:80) by PRNT. ELISA positive samples in both studies had usually had titers of 1:320 or above (one prospective ELISA positive had a titer of 1:80). It may be that the one false negative ELISA was very early in infection, and had not yet seroconverted from

IgM to IgG positivity, which may take 4 or more weeks. Also, the PRNT cutoff of 1:40 for positivity may be too low during interepidemic periods when transmission rates are low.

62 CHAPTER 8: CONCLUSIONS

8.1 General Conclusions

In conclusion, the extent and the interepidemic spread of RVF in Kenya were greater than previously documented. This evidence expands the known range of RVFV circulation during the 1997-98 ENSO era, with the implication that broader surveillance and control measures are required. The finding that certain populations had baseline anti-RVFV IgG seroprevalence rates as high as 19% makes reliance on outbreak seroprevalence results problematic. More studies are needed to delineate these rates in order to reflect endemic transmission and improve the accuracy of serologic testing. Ongoing studies are assessing the continued enzootic/endemic presence of RVFV and the late outcomes of infection in the

Garissa area of Kenya. The results can be used to develop and refine predictive algorithms for RVFV transmission based on environmental and remote sensing data, with the ultimate goal of providing improved early prediction of RVFV outbreaks.

This prospective study highlights the large-scale variability in exposure and

RVFV seropositivity between Kenyan villages and emphasizes the impact of age, sex, location, and animal husbandry in RVFV transmission. Initial estimates of RVFV inter- epidemic transmission and disease parameters were provided, which are needed for evidence-based planning of RVFV control strategies in Kenya. This information is important for local public health agencies so that they can target protective interventions according to risk factors in different populations. Further studies are needed to examine the epidemiologic, biologic, and genetic basis for the increased risk among males, and to quantify the potential public health impact of modifying the rural environment. RVFV

63 transmission is known to be ongoing in livestock in RVFV endemic areas during interepidemic periods and this study shows that this extends to humans, confirming past

observations 56. Ongoing efforts to predict hot spots of infection on both small and large scales is useful only when at-risk communities are able to use the information to target mosquito or vaccine control efforts and prevent outbreaks.

Although environmental clues, such as excess rainfall, livestock abortions, and

mortality in young animals, can signal an RVFV epizootic, enhanced linkage between

animal and human surveillance must take place for effective control of this complex

pathogen. Too often human cases, and not animal infections, act as the sentinel

infections in RVF outbreaks and the opportunity for effective control via animal vaccination is missed. The failed attempts at animal vaccination during the recent RVF outbreak in Kenya (2006-7) demonstrated how crucial communication between surveillance teams can be. The focus should be not on animal health only or human health only, but “one health” for all in the case of RVFV control. As Rift Valley Fever expands its geographic range and becomes recognized as a disease of global importance for human and animal health, more research is needed to define the most accessible and effective modes of transmission control.

8.2 Future Studies

Although this study improves the understanding of the natural history, epidemiology, and human health impact of Rift Valley Fever, many fundamental research questions remain unanswered. Critical questions that remain include: 1) What amount and type of exposures lead to RVFV exposure during the last (2006-7) outbreak? 2) What

64 is the most common mode of RVFV human transmission during RVF outbreaks, aerosol exposure or vector-borne transmission? 3) Why do some humans suffer only a febrile illness while others succumb to hemorrhagic disease? 4) Does mode of RVFV transmission (aerosol or mosquito-borne) influence clinical manifestations of disease? 5)

What are the critical components of host immunity to RVFV and are they long-lasting?

In order to begin to answer these questions, future studies are ongoing. First, a repeat serosurvey of the Masalani area is planned to examine the risk for seroconversion during the last outbreak. An attempt will be made to repeat detailed questionnaires and serology testing on all 248 study participants. By comparing serology results drawn in

2003 (this study) and new serology results with exposure details, the most common modes of human transmission in this area of repeated outbreaks may be able to be identified.

In order to examine the determinants of human disease manifestations with RVF, a large prospective study in Northeast Province, Kenya is planned. Because hemorrhagic disease manifestations present early in the onset of illness (day 3), it is likely that innate and early adaptive immunity are critical determinants of disease severity. Both host immunity and host genetics are likely key factors in clinical outcome. Detailed immunologic and genetic testing will be performed on subsets of RVF patients (febrile illness, retinitis, and hemorrhagic disease) to determine their impact on clinical outcome.

By using animal models, detailed immunologic testing may reveal whether mode of transmission also contributes to disease severity.

Finally, detailed immunologic analysis will parcel out key immunologic factors for host immunity and B and T cell studies will reveal what types of immune function are

65 critical to long-lasting immunity. These analyses will be crucial to identify before an effective human RVFV vaccine can be implemented.

66 CHAPTER 9

9. APPENDIX

Appendix Figure 1. Questionnaire administered to prospective survey participants.

67

68

69 Appendix Figure 2. Data entry form for physical and eye exam findings.

70

71 Appendix Table 1. Associations of potential predictors with Rift Valley fever virus seropositivity*

Variable Test statistic p value‡ Odds ratio

Age (continuous) 6.184 0.0001 1.048 (1.029, 1.066)

Age (>15 y vs. 1–14 y) 18.772 0.0001 8.03 (2.73, 23.6)

Location (Gumarey vs. Sogan- 10.747 0.001 3.8 (1.64, 8.77) Godud)

Gender (Male vs. female) 6.546 0.011 2.586 (1.23, 5.45)

Home flooded 5.105 0.024 4.697 (1.09, 20.3)

Displacement by flood 2.696 0.101 2.432 (0.82, 7.23)

Contact with dead human body 36.97 0.0001 9.064 (4.087, 20.100)

Use of mosquito nets 2.655 0.103 0.545 (0.26, 1.139)

Use of fire 0.038 0.864 0.86 (0.187, 3.94)

Mosquito coils 3.6 0.058 0.392

Recent mosquito bite 0.038 0.538 0.744 (0.237, 2.330)

Recent illness 3.045 0.081 2.00 (0.908, 4.40)

Ill family member 0.154 0.430 1.692 (0.45, 6.35)

Sheep contact 2.22 0.136 3.464 (0.45, 26.66)

Goat contact 2.39 0.122 3.65 (0.474, 28.0)

Camel contact 4.75 0.029 0.224 (0.052, 0.972)

Cattle contact 6.288 0.012 6.436 (0.85, 48.63)

Sheltering 8.623 0.003 11.544 (1.542, 86.44)

Slaughtering 3.11 0.078 5.275 (1.125, 24.73)

72 Butchering 23.817 0.0001 5.97 (2.75, 12.93)

Skinning 15.237 0.0001 4.47 (2.02, 9.90)

Cooking 7.878 0.005 4.30 (1.46, 12.668)

Milking 16.64 0.0001 6.44 (2.396, 17.31)

Birthing livestock 32.447 0.0001 9.59 (3.96, 23.28)

Disposal of animal fetus 28.303 0.0001 7.57 (3.3, 17.27)

Drinking raw animal milk 3.077 0.079 6.44 (0.852, 48.63)

*All variables were dichotomous except age (continuous).

†Pearson χ2 test with Yates continuity correction was used for all variables except age

(continuous), which used independent samples 2-tailed t test.

‡p<0.005 was statistically significant.

73 Appendix Table 2. Association between (village) location and potential model predictors for RVFV seropositivity.

Odds ratio Test (Gumarey vs. Variable statistic† p value‡ Sogan-Godud) Age (continuous) 0.161 0.873 N/A

Age (>15 y vs. 1–14 y) 1.278 0.258 0.75

Location (Gumarey vs. Sogan-Godud) 10.747 0 = 0.001 3.8

Gender (Male vs. female) 0.727 0.394 1.25

Home flooded 0.57 0.45 0.79

Displacement by flood 0.194 0.66 0.876

Contact with dead human body 4.9 0.026 2.09

Use of mosquito nets 36.16 0.0001 0.194

Use of fire 17.67 0.0001 NA

Mosquito coils 41.75 0.0001 0.122

Recent mosquito bite 0.263 0.61 0.802

Recent illness 0.081 0.776 1.075

Ill family member 3.242 0.072 0.354

Sheep contact 3.27 0.07 2.43

Goat contact 3.98 0.046 2.64

Cattle contact 14.09 0.0001 4.74

Camel contact 4.498 0.034 0.503

Sheltering 9.41 0.002 2.62

Slaughtering 2.45 0.117 6.4

74 Butchering 1.776 0.0001 1.5

Skinning 0.875 0.35 1.28

Cooking 1.812 0.178 1.44

Milking 0.039 0.843 1.05

Birthing livestock 7.9 0.005 2.14

Disposal of aborted animal fetus 4.15 0.042 1.74

Drinking raw animal milk 12.45 0.0001 4.12

*All variables were dichotomous except age (continuous).

†Pearson χ2 test with Yates continuity correction was used for all variables except age

(continuous), which used independent samples 2-tailed t test.

‡p<0.005 was statistically significant.

75 Appendix Table 3: Binary Logistic Regression Analysis

Logistic Model 1:

Outcome variable: RVFV Seropositivity (coded as 0 vs. 1)

Predictor Variable Point Estimate (Odds Statistical Significance Variables Type Ratio)

Age Continuous 1.039 (1.017, 1.062) P=0.001

Location Dichotomous 0.241 (0.092, 0.628) P=0.004 (Sogan-Godud vs. Gumarey)

Sex (Male vs. Dichotomous 2.782 (1.176, 6.581) P=0.020 Female)

Disposal of Dichotomous 2.779 (1.026, 7.525) P=0.044 Aborted

Constant 0.119 P=0.005

Goodness-of-Fit: (Chi-squared analysis of observed vs. predicted values indicating model

is well fit to observed values).

Hosmer and Lemeshow Test

Chi-

Step square df Sig.

1 8.813 8 .358

76 Appendix Table 4. Testing of association between predictors of RVFV seropositivity*

Age Village Sex Disposal of Aborted Animal Fetus Age

Village t=0.161

(p=0.873)

Sex t=0.151 Χ2=0.727

(p=0.886) (p=0.394)

Disposal of t=10.25 Χ2=4.145 Χ2=0.278

Aborted (p=0.0001) (p=0.042) (p=0.598)

Animal Fetus

*Testing for interaction between associated predictors: (Disposal of aborted animal fetus)

× Age ; (Disposal of aborted animal fetus) × Village. Both interaction terms were not significant (p>0.05). The goodness-of-fit statistic indicated that the additional interaction terms were not beneficial to the predictive power of the model.

77 Appendix Table 5: Logistic Models by Location

Table 5a: Model 2a: Location= Gumarey

Outcome variable: RVFV Seropositivity

Predictor Variable Point Estimate Statistical Variables Type (Odds Ratio) Significance

Sex Dichotomous 3.454 P=0.025

(1.17,10.19)

Discarded aborted Dichotomous 15.12 P=0.0001 animal fetus (4.45,51.35)

Ill family member Dichotomous 18 (1.35,246.97) P=0.029

Constant 0.029 P=0.0001

Goodness of Fit: (Chi-squared analysis of observed vs. predicted values indicating model is well fit to observed values).

Hosmer and Lemeshow Test

Chi-

Step square df Sig.

1 5.493 2 .064

78 Table 5b: Model 2b: Location= Sogan-Godud

Outcome variable: RVFV Seropositivity

Predictor Variable TypePoint Estimate (Odds Statistical Significance

Variables Ratio)

Age Continuous 1.054 (1.019,1.091) P=0.0001

Constant 0.01 P=0.0001

Goodness of Fit: (Chi-squared analysis of observed vs. predicted values indicating model is well fit to observed values).

Hosmer and Lemeshow Test

Chi-

Step square df Sig.

1 9.318 7 .231

79 Appendix Table 6. Association of signs and symptoms with RVFV seropositivity*

Variable Test statistic p value Odds Ratio Personal illness 3.045 0.081 2.00

Family illness 0.154 0.43 1.692

Fever 0.109 0.741 1.97

Malaise 1.77 0.183 1.99

Myalgia 12.97 0.0001 6.03

Chills 2.962 0.085 2.22

Backache 9.059 0.003 3.864

Eye pain 4.511 0.034 2.275

Headache 0.035 0.852 1.11

Rash 0.134 0.714 0.822

Red eyes 7.036 0.008 2.75

Photophobia 3.63 0.057 2.09

Poor appetite 0.504 0.478 0.75

Flushing 0.752 0.386 2.03

Nausea 0.373 0.541 1.27

Vomiting 1.36 0.243 0.59

Meningismus 8.24 0.004 2.97

Poor vision 6.985 0.008 2.74

Epistaxis 0.181 0.67 1.28

Hematemesis 2.52 0.112 2.37

Hematochezia 0.372 0.542 1.35

80 Bruising 1.054 0.305 NA

Confusion 0.366 0.551 1.27

Vertigo 0.092 0.761 0.791

Stupor 1.173 0.279 1.516

Coma 7.93 0.005 14.55

*All variables were dichotomous.

†Pearson χ2 test with Yates continuity correction was used for all variables.

‡p<0.005 was statistically significant.

81 Appendix Table 7. Testing of association of eye disease with Rift Valley fever virus seropositivity

Variable Test statistic* p value Odds ratio Eye disease 11.011 0.003 4.99

*Pearson χ2 test with Yates continuity correction.

†p<0.005 was statistically significant.

82 Appendix Table 8: Theory-Driven Variable Clusters

Exposure Cluster Name Variables Included in References to Support Cluster Clustering Flood Factor Home flooding 4,5 Displacement by flood Mosquito Protection Factor Use mosquito net 8,17-19 Use fire Use mosquito coil Household Illness Factor Recent Illness 4 Ill family member Animal Contact Factor Sheep contact 1,4,12 Goat contact Cattle contact Animal Processing Factor Slaughtering 1,4,12 Butchering Skinning Cooking Animal Living Factor Sheltering 1,4,12 Milking Birthing

Symptom Cluster Name Variables Included in References to Support Cluster Clustering Meningoencephalitis Factor Headache 2,4 Photophobia Nausea Vomiting Meningismus Confusion Vertigo Stupor Coma Hemorrhage Factor Epistaxis 2-4 Hematemesis Hematochezia Bruising Retinitis Factor Poor vision 2,4,10 Eye pain Red eyes Exposure Clusters: The following exposure variables were left unclustered: age, village, sex, dead human body contact, recent mosquito bite, camel contact, disposal of aborted animal fetus, and drinking raw milk. These factors are likely to be important on their own (age, sex, location, dead human, fetus, milk) - or are not easily groupable (camel, mosquito bite).

83 Symptom Clusters: The following symptom variables were left unclustered: personal illness, family illness, fever, malaise, myalgia, chills, backache, rash, poor appetite, and flushing. These factors are nonspecific and likely to be witnessed in different clinical phenotypes of RVF disease.

84 Appendix Table 9. Associations of potential clustered predictors (grouped with non- clustered variables directly below) with Rift Valley fever virus seropositivity

Variable Test statistic p value‡ Odds ratio (95% CI) EXPOSURES Flood Factor 3.946 € 0.047 1.848 (0.979, 3.50) Home flooded 5.105 0.024 4.697 (1.09, 20.3) Displacement by flood 2.696 0.101 2.432 (0.82, 7.23) Mosquito Protection Factor 4.816 € 0.028 0.684 (0.484, 0.96) Use of mosquito nets 2.655 0.103 0.545 (0.26, 1.139) Use of fire 0.038 0.864 0.86 (0.187, 3.94) Mosquito coils 3.600 0.058 0.403 (0.15, 1.09) Household Illness Factor 3.495 0.062 1.856 (0.97, 3.57) Recent illness 3.045 0.081 2.00 (0.908, 4.40) Ill family member 0.154 0.43 1.692 (0.45, 6.35) Animal Contact Factor 5.305 € 0.021 1.275 (0.824, 1.972) Sheep contact 2.22 0.136 3.464 (0.45, 26.66) Goat contact 2.39 0.122 3.65 (0.474, 28.0) Cattle contact 6.288 0.012 6.436 (0.85, 48.63) Animal Processing Factor 23.889 € 0.001 2.22 (1.57, 3.13) Slaughtering 3.11 0.078 5.275 (1.125, 24.73) Butchering 23.817 0.0001 5.97 (2.75, 12.93) Skinning 15.237 0.0001 4.47 (2.02, 9.90) Cooking 7.878 0.005 4.30 (1.46, 12.668) Animal Living Factor 30.334 € 0.001 3.42 (2.07, 5.659) Milking 16.64 0.0001 6.44 (2.396, 17.31) Birthing livestock 32.447 0.0001 9.59 (3.96, 23.28) Sheltering 8.623 0.003 11.544 (1.542, 86.44)

SYMPTOMS Meningoencephalitis Factor 1.003 € 0.317 1.12 (0.899, 1.387) Headache 0.035 0.852 1.11 (0.363, 3.41) Vertigo 0.092 0.761 0.791 (0.174, 3.603) Stupor 1.173 0.279 1.516 (0.711, 3.23) Coma 7.93 0.005 14.55 (1.28, 165.56) Confusion 0.366 0.551 1.27 (0.576, 2.81) Vomiting 1.36 0.243 0.59 (0.243, 1.44) Nausea 0.373 0.541 1.27 (0.595, 2.69) Photophobia 3.63 0.057 2.09 (0.97, 4.50) Hemorrhage Factor 0.813 € 0.367 1.25 (0.77, 2.036) Epistaxis 0.181 0.67 1.28 (0.41, 3.99) Hematemesis 2.52 0.112 2.37 (0.797, 6.97)

85 Hematochezia 0.372 0.542 1.35 (0.513, 3.55) Bruising 1.054 0.305 NA* Retinitis Factor 8.651 € 0.003 1.574 (1.15, 2.15) Poor vision 6.985 0.008 2.74 (1.27, 5.92) Eye pain 4.511 0.034 2.28 (1.05, 4.923) Red eyes 7.036 0.008 2.75 (1.28, 5.93)

†Pearson χ2 test with Yates continuity correction was used for all variables except ordinal variables € Ordinal variables used a χ2 test statistic with linear-by-linear associations ‡p-value of p<0.05 was statistically significant. *Not applicable for OR because of zero cell count in 2x2 table

86 Appendix Table 10: Step Wise Logistic Modeling of RVFV Seropositivity Using Variable Clusters

B S.E. Wald df Sig. Exp(B)

Step 1a Sex 1.002 .499 4.036 1 .045 2.725

Age .046 .018 6.667 1 .010 1.047

Village -1.070 .582 3.384 1 .066 .343

Dis1ed_of_aborted_fet .580 .687 .714 1 .398 1.786 us

Contact_with_a_dead_ .359 .658 .299 1 .585 1.432 human_body

Flood_factor -.113 .415 .074 1 .786 .893

Mosq_protect_factor -.210 .253 .689 1 .406 .810

Illness_factor .807 .478 2.853 1 .091 2.241

Animal_processing_fac -.444 .352 1.590 1 .207 .642 tor

Animal_living_factor .795 .439 3.277 1 .070 2.215

Meningoencephalitis_f -.246 .174 2.001 1 .157 .782 actor

Hemorrhage_factor -.112 .342 .107 1 .743 .894

Retinitis_factor -.084 .265 .101 1 .750 .919

Constant -3.050 1.098 7.722 1 .005 .047 Step 2a Sex 1.025 .491 4.352 1 .037 2.787 Age .046 .018 6.616 1 .010 1.047 Village -1.075 .582 3.420 1 .064 .341 Dis1ed_of_aborted_fet .577 .683 .713 1 .399 1.780 us Contact_with_a_dead_ .356 .657 .294 1 .587 1.428 human_body Mosq_protect_factor -.206 .253 .667 1 .414 .814 Illness_factor .800 .477 2.814 1 .093 2.225 Animal_processing_fac -.442 .351 1.585 1 .208 .643 tor

87 Animal_living_factor .769 .426 3.263 1 .071 2.157 Meningoencephalitis_f -.243 .173 1.968 1 .161 .784 actor Hemorrhage_factor -.116 .341 .117 1 .733 .890 Retinitis_factor -.086 .265 .105 1 .746 .918 Constant -3.174 1.009 9.902 1 .002 .042 Step 3a Sex 1.022 .492 4.319 1 .038 2.779 Age .044 .017 6.661 1 .010 1.045 Village -1.074 .581 3.415 1 .065 .342 Dis1ed_of_aborted_fet .559 .678 .681 1 .409 1.749 us Contact_with_a_dead_ .339 .655 .269 1 .604 1.404 human_body Mosq_protect_factor -.212 .252 .708 1 .400 .809 Illness_factor .816 .474 2.965 1 .085 2.262 Animal_processing_fac -.450 .350 1.653 1 .199 .637 tor Animal_living_factor .772 .423 3.322 1 .068 2.163 Meningoencephalitis_f -.262 .163 2.569 1 .109 .770 actor Hemorrhage_factor -.133 .336 .157 1 .692 .875 Constant -3.164 1.006 9.890 1 .002 .042 Step 4a Sex .995 .486 4.191 1 .041 2.703 Age .044 .017 6.649 1 .010 1.045 Village -1.080 .581 3.457 1 .063 .340 Dis1ed_of_aborted_fet .561 .678 .685 1 .408 1.753 us Contact_with_a_dead_ .335 .649 .266 1 .606 1.398 human_body Mosq_protect_factor -.195 .247 .620 1 .431 .823 Illness_factor .796 .470 2.872 1 .090 2.216 Animal_processing_fac -.435 .349 1.556 1 .212 .647 tor

88 Animal_living_factor .762 .421 3.268 1 .071 2.142 Meningoencephalitis_f -.277 .160 3.019 1 .082 .758 actor Constant -3.145 1.000 9.884 1 .002 .043 Step 5a Sex 1.051 .475 4.908 1 .027 2.862 Age .048 .015 9.741 1 .002 1.049 Village -1.115 .579 3.715 1 .054 .328 Dis1ed_of_aborted_fet .633 .664 .907 1 .341 1.883 us Mosq_protect_factor -.204 .248 .677 1 .411 .816 Illness_factor .781 .467 2.798 1 .094 2.183 Animal_processing_fac -.404 .341 1.404 1 .236 .667 tor Animal_living_factor .749 .419 3.196 1 .074 2.115 Meningoencephalitis_f -.283 .159 3.171 1 .075 .753 actor Constant -3.165 1.002 9.979 1 .002 .042 Step 6a Sex 1.014 .470 4.651 1 .031 2.756 Age .044 .014 9.304 1 .002 1.045 Village -1.322 .529 6.252 1 .012 .267 Dis1ed_of_aborted_fet .569 .656 .753 1 .386 1.766 us Illness_factor .709 .460 2.367 1 .124 2.031 Animal_processing_fac -.361 .334 1.170 1 .279 .697 tor Animal_living_factor .788 .419 3.538 1 .060 2.199 Meningoencephalitis_f -.263 .158 2.785 1 .095 .769 actor Constant -3.088 1.000 9.535 1 .002 .046 Step 7a Sex 1.064 .467 5.202 1 .023 2.899 Age .044 .014 9.479 1 .002 1.045 Village -1.369 .524 6.824 1 .009 .254 Illness_factor .658 .454 2.099 1 .147 1.930

89 Animal_processing_fac -.292 .321 .827 1 .363 .747 tor Animal_living_factor .921 .387 5.674 1 .017 2.513 Meningoencephalitis_f -.238 .152 2.443 1 .118 .788 actor Constant -3.214 1.000 10.334 1 .001 .040 Step 8a Sex 1.032 .465 4.927 1 .026 2.806 Age .038 .013 9.176 1 .002 1.039 Village -1.305 .515 6.419 1 .011 .271 Illness_factor .545 .433 1.582 1 .208 1.725 Animal_living_factor .717 .313 5.247 1 .022 2.048 Meningoencephalitis_f -.211 .147 2.049 1 .152 .810 actor Constant -3.142 .990 10.069 1 .002 .043 Step 9a Sex 1.154 .454 6.451 1 .011 3.170 Age .038 .012 9.636 1 .002 1.039 Village -1.209 .498 5.884 1 .015 .298 Animal_living_factor .748 .306 5.969 1 .015 2.114 Meningoencephalitis_f -.196 .145 1.824 1 .177 .822 actor Constant -3.064 .993 9.512 1 .002 .047 Step 10a Sex 1.169 .451 6.713 1 .010 3.219 Age .033 .012 8.047 1 .005 1.034 Village -1.269 .497 6.527 1 .011 .281 Animal_living_factor .686 .301 5.200 1 .023 1.986 Constant -3.249 .971 11.206 1 .001 .039 a. Variable(s) entered on step 1: Sex, Age, Village, Dis1ed_of_aborted_fetus, Contact_with_a_dead_human_body, Flood_factor, Mosq_protect_factor, Illness_factor, Animal_processing_factor, Animal_living_factor, Meningoencephalitis_factor, Hemorrhage_factor, Retinitis_factor.

90 Primary Model Selection

Model Summary -2 Log Cox & Snell R Nagelkerke R Step likelihood Square Square 1 138.968a .201 .369 a. Estimation terminated at iteration number 6 because parameter estimates changed by less than .001.

Hosmer and Lemeshow Test Step Chi-square df Sig. 1 35.433 8 .000

Classification Tablea Predicted ELISA Percentage Observed RVF Negative RVF Positive Correct Step 1 ELISA RVF Negative 212 3 98.6 RVF Positive 23 10 30.3 Overall Percentage 89.5 a. The cut value is .500

Variables in the Equation

95.0% C.I.for EXP(B)

B S.E. Wald df Sig. Exp(B) Lower Upper

Step 1 Sex 1.128 .442 6.504 1 .011 3.089 1.298 7.348

Age .033 .012 7.832 1 .005 1.033 1.010 1.057

Village -1.364 .491 7.718 1 .005 .256 .098 .669

Animal_living_factor .736 .298 6.112 1 .013 2.088 1.165 3.743

91 Variables in the Equation

95.0% C.I.for EXP(B)

B S.E. Wald df Sig. Exp(B) Lower Upper

Sex 1.128 .442 6.504 1 .011 3.089 1.298 7.348

Age .033 .012 7.832 1 .005 1.033 1.010 1.057

Village -1.364 .491 7.718 1 .005 .256 .098 .669

Animal_living_factor .736 .298 6.112 1 .013 2.088 1.165 3.743

Constant -3.151 .953 10.943 1 .001 .043

92 DETAILED STUDY SUMMARY

Background: Rift Valley fever virus (RVFV) is a high priority pathogen because of its

potential for severe economic harm to livestock, its ability to cause life-threatening

hemorrhagic fever in humans, and its potential for non-vector aerosol spread during

epizootics and epidemics. The largest Kenyan Rift Valley Fever (RVF) epidemic

occurred in the Garissa District during 1997-1998; the most recent Kenyan outbreak

occurred in the same area in 2006-7. Limited surveillance leads to inaccurate estimates

of RVFV seropositivity in endemic areas. Because most RVF outbreaks occur in remote locations following flooding events, environmental risk factors and long-term sequelae of human RVF are not fully known. Exposure to an RVFV-infected animal is a clear risk factor for disease; however, the impacts of specific animal and non-animal exposures on

RVF risk are unknown.

Objective: Retrospective serosurvey: To examine RVFV seropositivity rates across

distinct regions of Kenya and the extent of RVFV transmission during epidemic and

interepidemic years.

Prospective serosurvey: To examine the association of certain animal and non-animal

exposures with RVFV seropositivity among Kenyans from two villages (Gumarey and

Sogan-Godud, located in Masalani Division, Ijara District, Northeast Province, Kenya)

with repeated RVF outbreaks. To compare seropositivity, exposures, and risk factors

between the two villages.

93 Classification: Retrospective serosurvey and a randomized prospective cross-sectional

household cluster serosurvey of a systematic stratified sample.

Setting/Participants: Retrospective serosurvey: participants enrolled in four prior Kenyan

studies.

Prospective serosurvey: all Masalani town residents more than 12 months old who

resided in Masalani for at least the last two years were eligible. Half of the samples were

from Gumarey (rural) and half from Sogan-Godud (developed).

Intervention: Retrospective serosurvey: no new intervention; testing of existing serum

samples for RVFV antibody seropositivity in different Kenyan populations both before

and during the 1997-98 Kenya epidemic.

Prospective serosurvey: Study participants completed: 1) a questionnaire to determine

exposures; 2) a physical examination; 3) a vision test and ophthalmoscopic exam; and 4)

phlebotomy.

Outcome Measures: Retrospective and prospective serosurveys: The primary outcome

measure was RVFV seropositivity as measured by anti-RVFV IgG detection via ELISA of serum.

Prospective serosurvey: The secondary outcome measures were age- and location- specific rates of previous RVFV exposure, behavioral risk factors for recent or remote

RVF infection, and RVFV association with current health status and its link to animal and non-animal exposures.

94

Statistical Methods: Chi-squared test, Fisher’s exact test, Logistic Regression

Results: Retrospective serosurvey: 1263 archived serum specimens collected at different

periods from three 'unaffected' Kenyan areas outside of the known RVFV transmission

zone were tested for anti-RVFV IgG by ELISA and plaque reduction neutralization testing:

1) 675 samples from the Turkana District surrounding Lokichoggio in the northern Rift

Valley Province (from 1994), 2) 220 samples from the Division of Kabobo in the

highlands of central Rift Valley Province (from 1996 and 1997), 3) 199 samples from

Msambweni (from 1997 and 1998) and 169 samples from Daragube (from 1996) which

are both located in southern Coast Province. RVFV IgG seropositivity was detected using antibody capture ELISA against MP-12 antigen, a non-pathogenic vaccine strain of

RVFV. Overall 136 samples (11%) tested positive for anti-RVFV IgG. Seropositivity

varied significantly by location. The Turkana District had the highest seropositive rate

(19% in 1994) when compared to Kabobo (0% in 1996 and 1% in 1997), Msambweni

(1% in 1997), and Daragube (3% in 1996). Rates also varied by age. Turkana and

Daragube seroprevalence rates were greatest among persons 15 years and older (25% in

Turkana and 6% in Daragube). One seroconversion documented in 1997 in Kabobo

suggests previously unsuspected interepidemic transmission in this highland area.

Prospective serosurvey: The sample consisted of 248 participants (33 positive/248 total,

13% RVFV seropositive [95%CI 9.3-18.1]): 49% from Gumarey (25 positive/122 total,

20% RVFV seropositive [95%CI 14.0-29.2]), and 51% from Sogan-Godud (8

positive/126 total, 6% RVFV seropositivity [95%CI 2.7-11.8]). A best-fit logistic model

95 to predict RVFV seropositivity included age (OR 1.04, 95%CI 1.02-1.06), village

(Gumarey OR 4.14, 95%CI 1.66-11.5), sex (male OR 2.78, 95%CI 1.18-6.72), and

disposal of an aborted animal fetus (OR 2.77, 95%CI 1.04-7.76). Exposures varied

significantly between village subgroups: those from Sogan-Godud were much more

likely to use mosquito nets (P ≤ 0.001), fire (P ≤ 0.001), and mosquito coil (P ≤ 0.001);

whereas those from Gumarey were more likely to have goat (P ≤ 0.05) and cow contact

(P ≤ 0.001), birthed (P ≤ 0.01) and sheltered livestock (P ≤ 0.01), drunken raw animal

milk (P ≤ 0.001), and dead human body contact (P ≤ 0.05). In Gumarey, a best-fit model

included sex (males OR 3.7, 95%CI 1.24-12.07) and disposal of an aborted animal fetus

(OR 9.4, 95%CI 2.9-38.1). In Sogan-Godud, individual exposures were highly

correlated.

Limitations: This study may have limited generalizability. The retrospective serosurvey was performed on previously collected blood samples and may therefore introduce bias because of varied sample design, and may slightly alter seroprevalence estimates.

However, the large cross-sectional nature of the study helps to eliminate this bias.

Exposure data in the prospective study was self-reported which may introduce bias, and was collected in a binary fashion which did not include detailed information about the amount of animal contact and may affect risk estimations.

Discussion: This retrospective study demonstrated that RVF was present in populations

residing in Kenyan regions outside the Garissa District both before and during the 1997-

1998 epidemic. Certain regions, such as the Turkana District, had surprisingly high rates

96 of seropositivity. Increased seroprevalence rates among persons 15 years and older

suggests an undetected outbreak may have occurred in Kenya around 1980. The increase in Kabobo seropositivity between 1996 and 1997 may reflect RVFV extension to that region. The extent and the interepidemic spread of RVFV in Kenya are greater than previously reported.

This prospective study in Masalani demonstrated that older age, rural village, male sex, and disposal of an aborted animal fetus were significantly associated with

RVFV seropositivity. RVFV seropositivity was three times higher in rural Gumarey, where sex and disposal of an aborted animal fetus remained important predictors of seropositivity. Overall RVFV seropositivity by IgG ELISA (confirmed by neutralization assay) was 13%, and evidence of interepidemic RVFV transmission was detected.

Seropositives were more likely to have visual impairment and retinal lesions, but other physical findings did not differ. Differences in exposures between the two villages may help to explain the discrepant RVFV seropositivity rates: those from Sogan-Godud were

more likely to use mosquito control, and those from Gumarey were more likely to have

certain animal exposures, such as birthing and sheltering livestock. This study highlights

the variability in exposure and RVFV seropositivity between Kenyan villages and

emphasizes the impact of age, gender, and village in RVFV transmission. Further studies

are needed to examine the biologic basis for the increased risk among males, and to

quantify the potential public health impact of modifying the rural environment and high-

risk animal husbandry practices.

97 CHAPTER 10

10. BIBLIOGRAPHY

1. WHO. Rift Valley Fever. WHO Fact Sheet No. 207, 1-5 (2000). 2. CDC. Rift Valley Fever. Viral Hemorrhagic Fevers: Fact Sheets, 1-3 (2002). 3. Isaacson, M. Viral hemorrhagic fever hazards for travelers in Africa. Clin Infect Dis 33, 1707-1712 (2001). 4. Woods, C.W. et al. An outbreak of Rift Valley Fever in Northeastern Kenya, 1997-98. Emerging Infectious Diseases 8, 138-144 (2002). 5. CDC. Outbreak of Rift Valley fever--Yemen, August-October 2000. MMWR 49, 1065-6 (2000). 6. Peters, C.J. The Role of Antivirals in Responding to Biological Threats. in Biological Threats and Terrorism: Assessing the Science and Response Capabilities (eds. Knobler, S.L., Mahmoud, A.A.F. & Pray, L.A.) 119-130 (National Academy Press, Washington, 2002). 7. Peters, C.J., Spertzel, R. & Patrick, W. Aerosol Technology and Biological Weapons. in Biological Threats and Terrorism: Assessing the Science and Response Capabilities (eds. Knobler, S.L., Mahmoud, A.A.F. & Pray, L.A.) 66-77 (National Academy Press, Washington, 2002). 8. CDC. Update: Outbreak of Rift Valley Fever - Saudi Arabia, August-November 2000. MMWR 49, 982-5 (2000). 9. El-Akkad, A.M. Rift Valley Fever outbreak in Egypt. October-December 1977. J Egypt Publ Health Assoc 53, 137-46 (1978). 10. Al-Hazmi, A. et al. Ocular complications of Rift Valley fever outbreak in Saudi Arabia. Ophthalmology 112, 313-8 (2005). 11. Peters, C.J., Liu, C.T., Anderson, G.W., Jr., Morrill, J.C. & Jahrling, P.B. Pathogenesis of viral hemorrhagic fevers: Rift Valley fever and Lassa fever contrasted. Rev Infect Dis 11 Suppl 4, S743-9 (1989). 12. Daubney, R., Hudson, J.R. & Garnham, P.C. Enzootic hepatitis or Rift Valley fever: an undescribed virus disease of sheep, cattle and man from East Africa. Journal of Pathology and Bacteriology 34, 545-79 (1931). 13. Won, S., Ikegami, T., Peters, C.J. & Makino, S. NSm and 78-kilodalton proteins of Rift Valley fever virus are nonessential for viral replication in cell culture. J Virol 80, 8274-8 (2006). 14. Ellis, D.S., Simpson, D.I., Stamford, S. & Abdel Wahab, K.S. Rift Valley fever virus: some ultrastructural observations on material from the outbreak in Egypt 1977. J Gen Virol 42, 329-37 (1979). 15. Schmaljohn, C.S., Hasty, S.E., Harrison, S.A. & Dalrymple, J.M. Characterization of Hantaan virions, the prototype virus of hemorrhagic fever with renal syndrome. J Infect Dis 148, 1005-12 (1983). 16. Gargan, T.P., 2nd, Clark, G.G., Dohm, D.J., Turell, M.J. & Bailey, C.L. Vector potential of selected North American mosquito species for Rift Valley fever virus. Am J Trop Med Hyg 38, 440-6 (1988). 17. Linthicum, K.J., Davies, F.G., Kairo, A. & L., B.C. Rift Valley fever virus (family Bunyaviridae, genus Phlebovirus): isolations form Diptera collected

98 during an interepizootic period in Kenya. Journal of Hygiene (London) 95, 197- 209 (1985). 18. Davies, F.G. & Highton, R.B. Possible vectors for Rift Valley fever in Kenya. Trans R Soc Trop Med Hyg 74, 815-25 (1980). 19. Meegan, J.M., Khalil, G.M., Hoogstrall, H. & Adham, F.K. Experimental transmission and field isolation studies implicating Culex pipiens as a vector of Rift Valley fever virus in Egypt. Am J Trop Med Hyg 29, 1405-9 (1980). 20. Morrill, J.C. et al. Serological evidence of arboviral infections among humans of coastal Kenya. J Trop Med Hyg 94, 166-8 (1991). 21. Schrire, I. & Gear, J.H.S. The fevers of Africa. 3. Rift Valley fever. Central African Journal of Medicine 2, 237-240 (1956). 22. Johnson, B.K. et al. Viral haemorrhagic fever surveillance in Kenya, 1980-1981. Trop Geogr Med 35, 43-7 (1983). 23. Outbreaks of Rift Valley fever in Kenya, Somalia and United Republic of Tanzania, December 2006-April 2007. Wkly Epidemiol Rec 82, 169-78 (2007). 24. CDC. Rift Valley fever outbreak--Kenya, November 2006-January 2007. MMWR Morb Mortal Wkly Rep 56, 73-6 (2007). 25. Shears, P. Communicable disease surveillance with limited resources: the scope to link human and veterinary programmes. Acta Tropica 76, 3-7 (2000). 26. Davies, F.G., Linthicum, K.J. & James, A.D. Rainfall and epizootic Rift Valley Fever. Bull World Health Organ 63, 941-3 (1985). 27. Linthicum, K.J. et al. Climate and satellite indicators to forecast Rift Valley Fever epidemics in Kenya. Science 285, 397-400 (1999). 28. Laughlin, L.W., Meegan, J.M. & Strausbaugh, L.J. Epidemic Rift Valley fever in Egypt: observations of the spectrum of human illness. Trans R Soc Trop Med Hyg 73, 630-3 (1979). 29. Siam, A.L. & Meegan, J. Ocular disease resulting from infection with Rift Valley fever virus. Trans R Soc Trop Med Hyg 74, 539 (1980). 30. Siam, A.L., Meegan, J.M. & Gharbawi, K.F. Rift Valley fever ocular manifestation: observations during the 1977 epidemic in the Arab Republic of Egypt. British Journal of Ophthalmology 64, 366-374 (1980). 31. Flick, R. & Bouloy, M. Rift Valley fever virus. Curr Mol Med 5, 827-34 (2005). 32. Niklasson, B., Peters, C.J., Grandien, M. & Wood, O. Detection of human immunoglobulins G and M antibodies to Rift Valley fever virus by enzyme-linked immunosorbent assay. J Clin Microbiol 19, 225-229 (1984). 33. Niklasson, B., Grandien, M., Peters, C.J. & Gargan, T.P. Detection of Rift Valley fever virus antigen by enzyme-linked immunosorbent assay. J Clin Microbiol 17, 1026-1031 (1983). 34. John, C.C. et al. Cytokine responses to Plasmodium falciparum liver-stage antigen 1 vary in rainy and dry seasons in highland Kenya. Infect Immun 68, 5198-204 (2000). 35. King, C.L. et al. Schistosoma haematobium-induced urinary tract morbidity correlates with increased Tumor Necrosis Factor-alpha and diminished Interleukin-10 production. J Infect Dis 184, 1176-82 (2001). 36. Meadors, G.F., Gibbs, P.H. & Peters, C.J. Evaluation of a new Rift Valley fever vaccine: safety and immunogenicity trials. Vaccine 4, 179-184 (1986).

99 37. LaBeaud, A.D., Ochiai, Y., Peters, C.J., Muchiri, E.M. & King, C.H. Spectrum of Rift Valley fever virus transmission in Kenya: insights from three distinct regions. Am J Trop Med Hyg 76, 795-800 (2007). 38. Gad, A.M. et al. A possible route for the introduction of Rift Valley fever virus into Egypt during 1977. J Trop Med Hyg 89, 233-6 (1986). 39. Digoutte, J.P. & Peters, C.J. General aspects of the 1987 Rift Valley fever epidemic in Mauritania. Res Virol 140, 27-30 (1989). 40. Peters, C.J. Emergence of Rift Valley fever. in Factors in the Emergence of Arboviruses (eds. Saluzzo, J.F. & Dodet, B.) 253-64 (Elsevier, Paris, 1997). 41. Johnson, B.K. et al. Marburg, Ebola and Rift Valley Fever virus antibodies in East African primates. Trans R Soc Trop Med Hyg 76, 307-10 (1982). 42. Johnson, B.K. et al. Antibodies against haemorrhagic fever viruses in Kenya populations. Trans R Soc Trop Med Hyg 77, 731-3 (1983). 43. Anyamba, A., Linthicum, K.J. & Tucker, C.J. Climate-disease connections: Rift Valley Fever in Kenya. Cad. Saude Publica 17 (suppl), 133-40 (2001). 44. Hoch, A.L., Gargan, T.P., 2nd & Bailey, C.L. Mechanical transmission of Rift Valley fever virus by hematophagous Diptera. Am J Trop Med Hyg 34, 188-93 (1985). 45. Turell, M.J., Linthicum, K.J. & Beaman, J.R. Transmission of Rift Valley fever virus by adult mosquitoes after ingestion of virus as larvae. Am J Trop Med Hyg 43, 677-80 (1990). 46. Logan, T.M., Davies, F.G., Linthicum, K.J. & Ksiazek, T.G. Rift Valley fever antibody in human sera collected after an outbreak in domestic animals in Kenya. Trans R Soc Trop Med Hyg 86, 202-3 (1992). 47. John, C.C., Koech, D.K., Sumba, P.O. & Ouma, J.H. Risk of Plasmodium falciparum infection during a malaria epidemic in highland Kenya, 1997. Acta Trop 92, 55-61 (2004). 48. Clements, A.C. et al. Spatial risk assessment of Rift Valley fever in Senegal. Vector Borne Zoonotic Dis 7, 203-16 (2007). 49. Clements, A.C., Pfeiffer, D.U., Martin, V. & Otte, M.J. A Rift Valley fever atlas for Africa. Prev Vet Med 82, 72-82 (2007). 50. Hannah, M.F., Bajic, V.B. & Klein, S.L. Sex differences in the recognition of and innate antiviral responses to Seoul virus in Norway rats. Brain Behav Immun (2007). 51. Evans, A. et al. Prevalence of antibodies against Rift Valley fever virus in Kenyan wildlife. Epidemiol Infect, 1-9 (2007). 52. van Velden, D.J., Meyer, J.D., Olivier, J., Gear, J.H. & McIntosh, B. Rift Valley fever affecting humans in South Africa: a clinicopathological study. S Afr Med J 51, 867-71 (1977). 53. Madani, T.A. et al. Rift Valley fever epidemic in Saudi Arabia: epidemiological, clinical, and laboratory characteristics. Clin Infect Dis 37, 1084-92 (2003). 54. Tesh, R.B., Peters, C.J. & Meegan, J.M. Studies on the antigenic relationship among Phleboviruses. Am J Trop Med Hyg 31, 149-155 (1982). 55. Paweska, J.T., Mortimer, E., Leman, P.A. & Swanepoel, R. An inhibition enzyme-linked immunosorbent assay for the detection of antibody to Rift Valley

100 fever virus in humans, domestic and wild ruminants. J Virol Methods 127, 10-8 (2005). 56. Davies, F.G. Observations on the epidemiology of Rift Valley fever in Kenya. J Hyg (Lond) 75, 219-30 (1975).

101