than children from middle (37 percent) and upper classes (45 percent) (Deressa et al 2007). Malaria incidence is ...... Adobe Illustrator v10.0.3 (27). Results.
i
UNDERSTANDING REGIONAL PATTERNS OF VECTOR-BORNE INFECTIOUS DISEASE IN A CHANGING ENVIRONMENT
by
Sarah H Olson
A dissertation submitted in partial fulfillment of the requirements for the degree of
Doctor of Philosophy (Population Health Sciences & Environment and Resources)
at the University of Wisconsin-Madison 2009
ii
Abstract
UNDERSTANDING REGIONAL PATTERNS OF VECTOR-BORNE INFECTIOUS DISEASE IN A CHANGING ENVIRONMENT Sarah H. Olson Under the supervision of Professor Jonathan A. Patz At the University of Wisconsin-Madison
The aim of this dissertation is to develop an integrated, new regional perspective of how climate and landscape conditions affect critical vector-borne diseases in the tropics and temperate latitudes. In particular, my research focuses on the effects of climate and land use and cover change on malaria in the Amazon Basin, and the effects of landscape fragmentation on Lyme disease in the United States. This work lays a spatial ecological foundation for new and better predictive models of human vector-borne infectious diseases.
Disease vectors and agents are dependent on their environment, and I first reviewed the known links of malaria and Lyme disease to climate, land use and cover, and ecological risk factors. My research builds on in situ and in vitro observations of the agents, vectors and human hosts, which identify what environmental factors combine to create high entomological and human disease risk. These studies show that some abiotic risk factors, such as temperature and pH, are consistent and can apply across continents, and other
iii ecological risk factors, such as biodiversity and deforestation, do not yet extended beyond isolated research locations, a limitation which my dissertation addresses.
Following the overview of current literature, I examined how climatic factors, specifically precipitation, affect the ecology of malaria across the Brazilian Amazon. Regional patterns of malaria and climate in the Amazon have not been studied extensively until now. This is an important concern, as an estimated 1.5 million cases occur in the basin annually, and climatic patterns are anticipated to change under scenarios of global warming and deforestation. With different mosquito vectors and habitat, it is not surprising that simplified, global-scale models of malaria suitability – based on the African malaria experience – perform poorly when they are applied to Amazon. I employed a unique, spatially-extensive health data set at the county (município) resolution in the Amazon region, along with detailed monthly, spatial temperature and precipitation data to derive new representations of precipitation and wetland drivers of malaria in the Brazilian Amazon. This study finds that areas of the Amazon Basin with few wetlands show a variable relationship between precipitation and malaria, while areas with extensive wetlands show a negative relationship with malaria incidence.
Next, I built on literature reporting malaria and deforestation connections at isolated research sites, and identified the same eco-epidemiological association operates in Mâncio Lima, a county in Acre State, Brazil. These health districts are often points of care, and equipped to administer public health interventions. I used a slide confirmed surveillance
iv dataset collected by the Brazilian Ministry of Health to assess the correlation between deforestation and malaria incidence within health districts (localidade). Such health districts are the smallest spatial unit in a standardized surveillance network that covers 5.1 million km2. With this dataset I find significant associations between malaria and deforestation. After controlling for access to care, health district size, inter-health district variability, and spatial trends a six percent change in percent deforestation over the prior decade associates with a 40 percent higher malaria incidence. Extensive and rapid land use and cover change continues to occur across the Basin, and I demonstrate that this type of changing ecology associates with higher malaria incidence.
The first studies to pioneer relationships of ecology, biodiversity, and vector-borne disease risk were those of Lyme disease. An important theory that resulted, called the dilution effect, states that increased biodiversity can limit the population of the predominant host, the white-footed mouse (Peromyscus leucopus), via trophic cascades and thereby reduce the likelihood of host to vector transmission. Despite evidence that fragmentation of deciduous forests decreases biodiversity and increases abundance of the white-footed mouse, there has not been a regional assessment of fragmentation on Lyme disease risk. I build on existing research to investigate the role of fragmentation on the Lyme disease vector over a large regional scale. The roll of fragmentation is theoretically based in landscape ecology, the study of relationships between landscape patterning and ecological function, which is an emerging field with promising relevance to vector-borne diseases. I developed and piloted a methodology that uses landscape metrics – such as fragmentation,
v edge, and landscape mixtures – and their context in the regional landscape to display the linkages of landscape ecology and the adult tick Lyme vector that persist in the Mid-Atlantic region of the United States.
The main achievement of this thesis has been to show that scientific findings of ecological disease risk do translate across spatial scales and associate with human health risk. High-resolution spatial health and environmental data connect the dots between vector ecology, vector abundance, and human disease risk.
vi Acknowledgements
Behind this dissertation is an amazing team of supporters and teachers who deeply enriched my learning.
My thanks go out to dedicated public school teachers in Poplar, MT, Bozeman, MT, Whitewater, WI, and Gardiner, MT, and college professors and students at Montana State University Bozeman. Special thanks go to Jean Buckingham, Jerry Reisig, Jeff Hostteter, Michael Franklin, Stephanie Thomas, David Ward, Joyce Hannula, Jack Horner, Clifford Montagne, Al Scharen, and Dale and Anne Olson.
The University of Wisconsin-Madison has provided an ideal academic setting. I am indebted to students and professors across campus in Public Health, Zoology, Veterinary School, Medical School, and the Nelson Institute for Environmental Studies. At the trunk of these various disciplinary branches, was the Center for Sustainability and the Global Environment (SAGE), where my co-advisors, Jonathan A. Patz and Jonathan A. Foley, provided a welcoming environment and inspired me to dream bigger. In addition to Jonathan2, my dissertation committee members included, Lisa Naughton (also a co-advisor), Ronald Gangnon, and Karen Cruickshanks, all of who fertilized new ideas and always had their doors open when something was wrong. Jonathan Patz cannot be thanked enough for his utmost kindness, infectious energy, unwavering support, and friendship.
vii Funding from a NASA-LBA grant, the Weston Fellowship, the Robert Wood Johnson Working Group on Interdisciplinary Perspectives on Health and Society Award, and the NSF-FACE exchange program was essential. Thanks to all of the support staff at SAGE, the Nelson Institute, and the Department of Population Health that helped me through the administrative hoops, consoled and repaired when a computer crashed, and took out my trash, doing it all with a smile or a laugh.
Abroad, research opportunities and adventures opened up because of the kindness and generosity of numerous friends and researchers. In Brazil, I am indebted to Jonas Brant and many other students connected to the CDC sponsored Field Epidemiologist Training Program. The malaria data was made accessible thanks to Jóse Ladislau (Director SIVEP), Guilherme Silveria, Carolina Santelli, and Eduardo M Mácario. During my travels abroad, Carlos Corvalan (PAHO), Jean-François Guégan (IRD), Jean Issaly (Pasteur Institute, Cayenne, French Guiana), and Laurent Durieux were important mentors. In France, Frederick Teyssier, Jean Quefelec, Christophe Gernigon, Hélèn Hecht, and the rest of club Vélo Saint-Mathieu 34, showed me the utmost kindness biking together in the Central Massive.
Outside the official classroom I had peers and colleagues who offered considerable guidance, including Holly Gibbs, David Zaks, Scott Spak, Bill Sacks, Carol Barford, Chris Kucharik, Margot Parkes, Maya Golden-Krasner, and Mary Sternitzky. My co-authors, Ronald Gangnon, Eric Elguero, Laurent Durieux, Jean-François Guégan, Jonathan A. Foley,
viii Jonathan A. Patz, Guilherme Silveria, Jeffrey A. Cardille, Murray K. Clayton, Joseph E. Bunnell, and Scott Heckle, have taught me so much with enduring patience and support. Micah Hahn, Vijay Limaye, and Jack Teng, are my honorary committee members as they graciously and meticulously helped me throughout the dissertation editing process.
Family and friends from both near and far provided intangible support. Very near, I was lucky to live and study with the extended 1814ner family, including Brian Harahan, James Davis, Chelsea Anderson, Matt Jarosz, Sally Gallagher, Christopher Thomas, and Aistis Tumas. Mary and Tom Karau, Barb and Denny Luster, Laura and William Gentry, Helen Pinter, and Ewan and Meredith Wolff made sure I never went hungry. Heidi Ploeg, Greg Ferguson, John Curtin, and Jonathan Patz made sure I kept biking and skiing. And lastly, my parents in Alaska, where the distance could never diminished their enthusiastic support and love.
ix Table of Contents
Abstract
ii
Acknowledgements
vi
Table of Contents
ix
Table and Figure Directory
xii
Introduction and Literature Review Historic and current relationships between climate factors and malaria epidemiology
1 1
Global malaria endemicity and minimum thresholds of malaria transmission
1
The spatial realization of climate thresholds
5
Specific environmental conditions most suited to Anopheles mosquitoes
13
Malaria and climate in the Amazon Basin
16
How deforestation can impact malaria in the Amazon Basin
19
Climate, landscape, and ecological risk factors of Lyme disease
21
How biodiversity impacts the enzootic cycle
22
The role of fragmentation
26
Summary
27
References
29
Figures
44
Paper #1: Links between Climate, Malaria, and Wetlands in the Amazon Basin References
49 56
x Figures Paper #2: Deforestation Links to Malaria in the Amazon
58 62
Abstract
62
Introduction
63
Materials and Methods
65
Health Data
66
Remote Sensing
66
Analysis and Modeling
67
Results
68
Discussion
71
References
77
Table
82
Figures
83
Paper #3: Landscape Ecology influences Lyme Disease Vector Abundance in the MidAtlantic Region, USA
89
Abstract
89
Methods
94
Results
96
Discussion
98
References
102
Table
105
Figure legends
106
xi Figures
107
Conclusion
113
Overview and relevance of results
113
Study limitations, strengths and weaknesses
118
Reflections and recommendations for future research
121
Summary and policy synthesis
124
References
126
Figures
130
xii Table and Figure Directory
Tables 3.1. Summary statistics of risk factors 4.1. Log-linear model of tick abundance
82 105
Figures 1.1. Levels of global malaria endemicity based on parasite ratios
44
1.2. Global distributions of important malaria vectors
45
1.3. Mean precipitation and temperature in the Amazon 1996–2000
46
1.4. Temperature and malaria incidence in the Amazon
47
1.5. Land use change in rural America 1950–2000
48
2.1. Malaria incidence per 1,000 and mean monthly precipitation
58
2.2. Malaria risk in the Amazon Basin
59
3.1. Malaria incidence and health districts in Mâncio Lima
83
3.2. Deforestation in Mâncio Lima
84
3.3. Plots of confirmed malaria cases in Mâncio Lima from 2003–2008
85
3.4. Cloropleths of selected malaria risk factors in Mâncio Lima
86
3.5. Joint relative risk of access to care and health district spatial area
87
xiii
4.1. Explanatory and response variable histograms
107-108
4.2. Example calculation of landscape metrics
109
4.3. Plot of the observed and predicted tick counts
110
4.4. Mid-Atlantic map of adult ticks
111
4.5. Tick abundance estimates for ten landscapes
112
5.1. Map of annual malaria incidence 1966 in Brazil
130
5.2. Map of malaria risk areas circa 1980s in French Guiana
131
5.3. Conceptual model of spatial epidemiology in practice
132
1 Chapter 1 Introduction and Literature Review
Historic and current relationships between climate factors and malaria epidemiology
Global malaria endemicity and minimum thresholds of malaria transmission Malaria is a fascinating disease that has global reach and dynamic relationships with climate. This review focuses on malaria parasites responsible for 247 million of malaria cases in 2006, Plasmodium vivax and P. falciparum (WHO 2008). Climate is fundamentally linked to malaria incidence through the biological requirements of both the vector and parasitic agent. Mathematical models and laboratory experiments have identified the basic thresholds of temperature and humidity on mosquito and parasite development, and these climate underpinnings are essential to explaining the global geography of malaria (Craig et al 1999; Guerra et al 2008; Martens et al 1999; Rogers et al 2000).
The climate malaria story begins with an examination of the isolated effects of temperature on the agent and vector. The rate at which parasitic replication occurs inside a mosquito vector has an L-shaped relationship with temperature. At 18˚C P. falciparum requires 56 days to complete sporogony, a period of time also called extrinsic incubation, which is an unlikely life span for an adult mosquito, but as temperature increases from 22˚C to 28˚C the development time drops from about 20 days to eight days (Craig 1999). The temperature range for P. vivax transmission is slightly larger (18˚–30˚C) than P. falciparum (20˚–30˚C) and
2 sporongony requires a day or two less at matched temperatures (Hamoudi and Sachs 1999). Raising the temperature from 20˚C and 30˚C considerably shortens the extrinsic incubation period from 18 to eight days in the malaria transmission cycle (Gilles and Warrell 1993). When temperatures exceed 30˚C the limiting factor in transmission is no longer the parasite – the weakened mosquito vectors begin to suffer high mortality and population turnover (Le Seuer 1991; Maharaj 1995). At temperatures above 40˚C mosquitoes desiccate and daily survival is zero in the laboratory (Martens 1997).
Even within the ideal Anopheles temperature window, small temperature changes can result in large biological changes. A study in the Kenyan highlands shows that just a half-degree centigrade rise in temperature from 1950 to 2002 can increase Anopheles gambiae mosquito population by 30–100 percent (Pascual et al 2006). Afrane et al. find a 1.8˚ C rise throughout the dry season drops the duration of the first and second gonotrophic stages by 1.7 days (59 percent) and 0.9 days (27 percent) respectively (2005). Temperature increases in the ‘sweet spot’ between 27˚C and 30˚ C, would lead to shortened latent periods and theoretically not compromise vector survivability. However, maximum and minimum daily temperatures may be more important than changes in mean temperature for the crepuscular active mosquito.
The temperature theme often correlates with the role of precipitation as the Anopheles mosquitoes require some amount of humidity to avoid desiccation as adults and precipitation to create pools of water near or on which females lay their eggs. When seasonal rainfall triggers an abundance of vectors malaria incidence tends to rise. Rainfall peaks two to three months prior to
3 malaria incidence in Africa and has been used with an early warning system that forecasts the timing and intensity malaria epidemics (Bomblies et al 2008; Eisele et al 2005; Thomas et al 2006; Ceccato et al 2007; WHO 2001; WHO 2004). In the Amazon, it appears that seasonal rainfall may suppress incidence for a longer period in areas with more wetlands (Olson et al 2009). While rainfall can affect the date of peak malaria incidence, total annual rainfall does not necessarily reflect annual incidence. In Madhya Pradesh, India, a longitudinal study over four decades showed no clear relationship between malaria incidence and annual rainfall (Singh and Sharma 2002). Hence, there is no consensus in the literature on general precipitation and humidity limits. Gilles and Warrell suggest the best environments maintain 60 percent mean relative humidity (1993). Craig et al. examine eight sites spread throughout Africa and determine that sites with stable transmission have at least 80 mm of rainfall per month (1999). Others use an annual index that counts the number of months with 60 mm of rainfall or more each month or an arbitrary minimum of 1.5 mm of precipitation a day (Kleinschmidt et al 2001; Martens et al 1995). The normalized difference vegetation index (NDVI), a satellite measure of greenness and a proxy measure for precipitation, is also used as a significant predictor of malaria in southern Africa and Kenya (Craig et al 1999; Hay et al 1998; Patz et al 1998; Ceccato et al 2007).
In addition to basic parameters of precipitation and temperature, synergistic effects and the variability of climate can also drive malaria outbreaks. Daily records of seven sites in the East African highlands between 1978 and 1998 indicate that climate variability, defined as, “short-term fluctuations around the mean climate state,” explained the variance for 65–81
4 percent of monthly malaria, and case numbers were also effected by, “non-linear and synergistic effects of temperature and rainfall” (Zhou et al 2004). Intraseasonal oscillations of climate are also associated with malaria risk. El Niño, caused by warming sea surface temperatures in the central tropical Pacific, increased malaria mortality 17–37 percent in Venezuela and Columbia, and in sub-Saharan Africa, higher malaria incidence was observed during warm La Niña phases of the southern oscillation index (Bouma et al 1997a, 1997b; Mabaso et al 2007). Similarly, 1956–1998 El Niño events set off malaria epidemics in Columbia, Venezuela, Peru, and Guyana (Gagnon et al 2002). El Niño events in Columbia are connected with warmer temperatures, higher dew points, and less precipitation and river discharge. These climatic changes are connected to increases in malaria in the second half of El Niño years and in the following year (Poveda et al 2001).
These complex climate patterns have greatly challenged those who would model the worldwide malaria burden. Another challenge is the limited resolution of global climate data. Presently, crude monthly mean and minimum temperature windows are used along side minimal monthly precipitation averages, so crude that some argue they are worthless for projecting malaria transmission under future scenarios of climate change (Rogers et al 2002). At the global scale, very little is known about the links between patterns of malaria and precipitation intensity and frequency. The widely used Mapping Malaria Risk in Africa/Atlas du Risque de Malaria (MARA/ARMA) model uses an 80 mm monthly threshold based on eight independent studies that are spread throughout Africa, but it is difficult to judge whether this approach is appropriate for other regions (Craig et al 1999). Furthermore, these models overlook the influence of rainfall
5 on the aquatic developmental stages of the vector. Mosquito eggs and larvae rely on persistent pools of water, of which the hydroperiod will depend on the daily frequency and intensity of rainfall events as well as soil types, landscape, such as the presence of wetlands of forest cover, and topography.
The spatial realization of climate thresholds Mapping human immunological adaptations and the phylogentic structure of malaria parasite species over time indicates that the dispersal of malaria beyond Africa followed a course of events set in motion by the thawing of the last glacial period 10,000 years ago. The warming temperatures and advent of agriculture practices in Africa were pivotal events in the emergence of malaria as a global infectious disease (Coluzzi 1999; Livingstone 1958). Agriculture allowed humans to settle in larger populations, but also provided mosquitoes with copious blood sources and plentiful larval habitat (Carter 2002). The reach of malaria expanded as humans migrated around the globe. Because of freezing temperatures, P.vivax probably did not cross the Siberian land bridge, more likely reaching South America after new and old world contact. The transmission of malaria exists along a continuum of levels, from stable to unstable and finally, to areas of no transmission. The restrictions of temperature and water availability, in both amount and seasonality, play a large role in determining the level of transmission or endemicity for any given region (see Figure 1.1 from Hay et al 2004). For similar climate regimes, regional variations of transmission develop from interactions amongst climate, landscape, socioeconomics, biology, and historical land use.
6 There is no universal theory for geographic variations in incidence of malaria in the literature. Some evidence points to general characterizations of human activities and life-styles. Historically, malaria first appeared in regions at the same time as land clearing and as humans were pioneering into new frontiers. The rise and fall of malaria on the Campagna, the marshy countryside surrounding Rome, “is noteworthy that in the one-and-one half millennia until the present time, periods of several centuries of high malaria incidence on the Campagna seem to have alternated with similar periods when malaria was apparently absent…these long episodes of absence and then presence of malaria have been associated with corresponding rising and falling agricultural and economic prosperity” (Carter 2002). Poverty and malaria persistently feed into one another. Today, per capita income is 43 percent lower in countries with unchecked malaria in comparison to healthy states (Gollin 2007). Other evidence claims that, “location and severity of malaria are mostly determined by climate and ecology, not poverty per se,” suggesting that climatic and ecologic drivers, rich in complex relationships between the parasites, vectors, host and environment, are the main perpetrators (Gallup and Sachs 2001).
The regional to local scale ecology of malaria transmitting mosquitoes
Globally, there are over 60 different species of anopheline mosquitoes that transmit malaria, but in any particular region just a few are dominant (see Figure 1.2 from Kiszewski et al 2004). Higher mosquito abundance and life expectancy usually increase malaria transmission risk and are associated to the suitability of the local climate and landscape. An exception is the unresolved so-called, ‘paddies paradox,’ observed in Africa, where irrigation projects for rice cultivation and agricultural purposes in areas of stable transmission appear to increase the
7 abundance of mosquitoes but not the incidence of malaria, but similar projects in areas of unstable transmission increase both the number of vectors and malaria incidence (Ijumba and Lindsay 2001; Matthys et al 2006).
In general species abundance, specific habitat niche, and behavior can explain temporal and spatial differences of malaria incidence. Adult abundance is dependent on the availability and quality of local aquatic larval habitat, as adults are unlikely to fly more than a few kilometers (Gilles and Warrell 1993). Behavior of a single species can also vary over time and space. In the 1940’s two out of every three A. darlingi mosquitoes were captured indoors, but at present and following an unsuccessful indoor DDT spraying campaign, the outdoor capture rates have increased five to ten fold over the indoor rates (Deane 1948; Charlwood 1980; Tadei and Thatcher 2000). In Columbia, the gravid females rest near the house floor while in Acre, Brazil, they are more often found on the ceiling, and the biting cycle is observed to vary greatly from location to location (Quinnes and Suarez 1990; Roberts et al 1987; Vittor et al 2006; Tadei and Thatcher 2000; Charlwood 1996). Peak river flow also appears to determine A. darlingi abundance differently across regions (Charlwood et al 1980; Hudson et al 1984; Rozendaal et al 1987; Rozendaal et al 1992; Magris et al 2007). Climate or human behaviors can create or minimize suitable habitat. Limited habitat increases the energy and time expenditures of gravid females searching for habitat and resources, and decreases the abundance of hatched adult mosquitoes. Increasing larval density in the remaining available habitat leads to smaller adults and longer development time, lowering malaria risk.
8 Local anthropogenic development also has the potential to create larval habitat (Killeen et al 2004). The main malaria vector in the Amazon, A. darlingi, is detected in areas of human disturbances such as irrigation projects, dams, suburban expansion, and deforestation (Tadei 1998). Plantation of agricultural crops over natural swamp vegetation in Uganda is observed to increase the ambient local temperature, improve the habitat for A. gambiae, and increase malaria incidence (Lindblade et al 2000). Each species identifies with a specific ecological niche. As a result, while one mosquito species may thrive along riverways, that same mosquito may not be well adapted to forested conditions. Deforestation throughout the tropics is associated with changes in malaria, but the direction of trend depends on the ecological niche of the resident Anopheles species (Sawyer 1998; Vittor et al 2006; Guerra et al 2006). Expansive deforestation has caused declines in numbers of A. dirus and A. culicfacies and malaria cases in northeast India (Dev et al 2003). In the heart of the Amazon, the association of deforestation and malaria is just the opposite, malaria risk is increasing with increasing deforestation, and in another twist, the vector is absent behind the advancing frontier (de Castro et al 2006; Vittor et al 2006; Vittor et al 2009).
If mosquitoes are abundant and surviving, the proximity and arrangement of mosquito resources becomes a key factor for successful malaria transmission. Female anopheline mosquitoes usually require a blood meal for the development of their eggs. Species of mosquito that prefer a human source are called anthropophilic and those that prefer animals or cattle are called zoophilic. Animals do not harbor the parasites of human falciparum or vivax malaria, so this preference is an important characteristic of each mosquito species, but the preference is
9 relative, as mosquitoes in the absence of a favored host will feed on other hosts. The more anthropophilic the species, the more likely consecutive blood meals will be taken from humans and elevate the risk of transferring a malaria parasite. Abundance and proximity of anthropophilic mosquitoes to both humans and larval habitats increases transmission risk. For example, the distance between A. gambiae (highly anthropophilic) larval habitats is significantly correlated with the adult mosquito density in houses, such that 90 percent of the adults are found within 300 m of nearest habitats (Minakawa et al 2002). In India, annual parasitic incidence is highest in villages near A. minimus habitat (Dev et al 2004).
The regional to local scale susceptibility of human populations
Human communities have traits that may alter the likelihood of mosquito bites and the proportion of bites that are infective, often measured as the annual parasitic incidence (API). The community traits are linked to development status, livelihood, and levels of poverty within the community, which may also be associated with land cover and land use, and incidence rates of other infectious diseases.
A historical case study of malaria in Guyana relates malaria increases in that country with changes in transportation powered first by horses, donkeys, and oxen, and then switching to tractors, cars, and buses. The culprit was, A. aquasalis, a vector that responded to the transition away from animal-powered transportation by shifting from zoophilic to anthropophilic biting behaviour (Spielman 2001). The World Health Organization (WHO) has recommended passive protection afforded by cattle, or zooprophylaxis, since 1982 as a useful intervention against
10 malaria and other mosquito-borne diseases. For instance, when 417 A. arabiensis mosquitoes were collected inside houses in northern Tanzania, the human blood meal rate was less for households with cattle (40 percent) than for households without cattle (70 percent) (Mahande et al 2007). However, other literature indicates the beneficial effect of zooprophylaxis is less consistent and suggests the importance of local dependencies. It was found Ethiopian children living in households shared with livestock actually have significantly greater odds of febrile illnesses, such as malaria, than children living in households that are not shared with livestock (Deressa et al 2007). A paired cohort study in Gambia attributes the effect there to unmeasured confounding of poverty – that is, the families with more cattle are associated with more wealth (Bøgh 2002).
If a community provides widespread access to care and rapid treatment, the pool of infected carriers will be reduced. But often public health clinics may not exist in remote frontier regions, and if they do, anti-malaria drugs may not be available. Communities with high levels of poverty will likely have less access to care. Ethiopian febrile children from households in the lowest levels of income are less likely to receive treatment (18 percent) within the first 24 hours than children from middle (37 percent) and upper classes (45 percent) (Deressa et al 2007). Malaria incidence is consistently lower in villages that are within 5 km of health care facilities in the Indian state of Assam (Dev 2004). Furthermore, communities with high birth rates may have higher malaria incidence. In Africa, children under the age of 5 carry nearly 10 fold higher risk of mortality (Snow et al 1999). Migrant populations may be less educated about protective measures and notably, native populations may be more asymptomatic (Duarte 2004; Laserson
11 1999; de Castro et al 2006). House construction materials, the number of openings, and smoke levels may affect the duration and location of African anophelines resting sites (Gillies and DeMeillon 1968). Poor housing construction and siting of communities may be related to socioeconomic status and or geographic features, and explain some variation in risk transmission.
In the Amazon, risk varies across socioeconomic groups. In some communities men carry double the risk of women, and in others, gold miners have three times the risk as urban residents (Duarte et al 2004). In another community setting there are no age-specific, occupational, or gender risks, but activities such as strolling outdoors after 6 pm, waking before 6 am, and attendance at church services in the evening, are significantly related to malaria risk (Roper et al 2000). A socio-demographic, epidemiologic, and environmental study of agricultural colonization in one frontier community in the Amazon summarizes the tripartite situation this way: “land use patterns, linked to social organization of the community and the structure of the physical environment, play a key role in promoting malaria transmission” (Singer and de Castro 2001).
Spatial patterns of malaria parasites and host immunity
Parasitic characteristics and innate human immunity will determine the outcome of a malaria infection. Globally there are differences in the capacity of anopheline vectors to harbor and transmit each parasite. Unfortunately for Africa, A. gambiae and P. falciparum together are a dynamic duo, which probably co-evolved in the Afrotropical rainforest and are responsible for
12 the majority of malaria mortality. The current temperate climate in North America is poorly suited for this highly efficient vector. In comparison the North American vectors, A. quadrimaculatusi and A. freeborni, while capable of transmitting malaria, are not nearly as efficient in transmitting the parasite (Gilles and Warrell 1993). Conversely, there is potential for colonization of South America by A. gambiae. A P. falciparum outbreak in South America at the beginning of the 20th century followed an accidental introduction of A. gambiae, and was only contained by militant larvaciding of 52,000 km2 (Killeen 2002).
Underlying the geographic variability of parasitic agents and vectors are geographic variations in levels of human susceptibility, resistance to anti-malarial drugs, and recrudescence rates. HIV weakens the immune system and models estimate it increases malaria incidence ten percent (Abu-Raddad et al 2006). Individuals are not equally genetically susceptible and can develop immunity over time. It is thought that falciparum malaria emerged in the last 5,000 to 10,000 years. Even in that relatively short evolutionary amount of time, several genetic conditions are believed to have persisted because they conferred some amount of protection against malaria. These include thalessemia, glucose-6-phosphate dehydrogenase deficiency, sickle cell trait, hemoglobin C, hemoglobin E, ovalyocytosis, and RBC duffy negativity (Carter 2002). Likewise, a genetic condition may explain why nearly 40 percent of Amerindians in French Guiana are asympotomatic carriers of P. vivax (Strobel et al 1985). Similar conjectures are made about native populations harboring malaria in regions of the Brazilian Amazon (Alves 2005). Genetic variations will influence malaria risk across populations and communities. In highly endemic regions the undeveloped immune systems of children are repeatedly challenged.
13 Children under ten in a longitudinal study in Venezuela had a 77 percent higher risk of malaria than persons over 50 (Magris et al 2007). Even fully mature immune systems are challenged by the parasites and the migration of carriers into a naïve population poses a great health risk, as was documented following an epidemic in Cacao, French Guiana, in the late 1980s (Mouchet 1989).
Specific environmental conditions most suited to Anopheles mosquitoes
Anopheles darlingi is the most widely distributed vector in the Americas and is highly suited to conditions associated with land disturbance. This exophagic and exophilic vector is mostly associated with rivers and forest edges. Prior to 1991 it was not found in appreciable abundances in locations beyond central South America. Today its range extends into the heart of the Amazon from southern Brazil north along the coast to Guyana, Venezuela, Belize, and the southern tip of Mexico, and west into Peru, Bolivia, and Paraguay (Need et al 1993; Kiszewski et al 2004). At Portuchuelo in Rondônia, Brazil, A. darlingi was collected at rates of 26 percent in 1940s to 77.7 percent in the 1980s and over 90 percent in 2003 (Gil et al 2003). Little is known about adult resting sites, but breeding sites are generally partially sunlit pools or along river and stream edges. The larvae can be found floating against dead leaves, flowers, and seed debris, but are notoriously difficult to sample because of low densities and a tendency of larvae to ‘dive’ during sampling (Charlwood et al 1996).
Two studies provide the most information about specific water conditions. In the first study in Belize, A. darlingi is found in 20 out of 30 larval habitats, with sites located along river
14 or streamways (15), lake margins (3), small lagoons (1), and ground pools (1). Shaded habitat with floating debris and submersed vegetation are significantly correlated with A. darlingi larvae. Depth of the pools varies from 0.1 to 2.0 meters. Water is generally fresh or slightly brackish and pH 6.49–6.93, but neither is significantly associated with absence or presence of larvae nor is presence or absence of algae (Manguin et al 1996). In a univariate analysis of breeding site characteristics suitable for A. darlingi in the Peruvian Amazon, the type of water body (i.e. fish farm), no/slow water current, algal mats, emergent grasses, aguaje palm, water depth greater than 0.5 meters, and less than 70 percent shaded, are significant environmental determinants. A. darlingi is most abundant between the rainy and dry seasons and following a month of higher rain. A small sampling of mean temperature and pH of sites is 25.2˚C /pH 5.64 for those with and 26.7˚C /pH 5.66 for those without A. darlingi (Vittor et al 2009).
A. gambiae and A. arabiensis prefer more sunlight than A. darlingi but these two very closely related species (which cannot be visually distinguished) maintain unique niches. Combined, these species are the most prolific malaria vectors in Africa. Like A. darlingi, these mosquitoes prefer warm climates (20–30˚C), humid conditions (~60 percent humidity), and respond biologically to forest cover. Controlling for host availability, the human blood index of A. arabiensis is much higher (0.923) than elsewhere in Africa and it is the main P. falciparum vector (Kent et al 2007). A study of A. arabiensis, which prefers drier environments, finds the percentage of larvae that matured into adults in forested habitat is four to nine percent, versus 60–82 percent in deforested habitat (notably the temperature indoors and outside is roughly one to two degrees Celsius warmer in deforested locations). The mosquitoes have 50 percent longer
15 median life expectancy in the deforested versus forested habitat and the reproductive rate is doubled. Where fecundity is greatest in deforested lowland, mean indoor relative humidity is 56 percent in the dry season and 64 percent in the rainy season (Afrane et al 2007). Drought virtually eliminated malaria transmission by A. arabiensis in Zambia during the 2004–2005 rainy season. The following year total rainfall doubled and the abundance of A. arabiensis resting indoors with humans was ten times higher. Three months following peak rainfall, the increase in human landing catches correlated with pediatric malaria admissions in a nearby hospital (Kent et al 2007). Landing catches are carried out by volunteers under prophylaxis to estimate numbers of biting adult mosquitoes and are the best measures of entomological malaria risk.
A. gambiae is an efficient reproductive species as it readily colonizes many types of water bodies and it is more likely to be found in farmlands and pastures than in forested areas (Minakawa et al 2005). For A. gambiae the story is similar to A. arabiensis larvae, where 1.5 percent of the larvae survive in the forest compared to 56 percent in open areas, but the adult median life is reduced five to seven days in deforested areas (Tuno et al 2005). However, the fecundity of females compensates for this reduction, such that the overall vectoral capacity of A. gambiae is 29–106 percent higher in deforested areas than forested areas (Afrane et al 2005; Afrane et al 2006). When A. gambiae is present the range of water temperature varies from 15.2–19.8˚C for forest pools, 15.6–24.2˚C for forest edge pools, and 14.2–32.2˚C for open pools. However, holding temperature constant across different locations still results in different rates of survivorship, suggesting other factors are responsible for larval mortality. Indeed, water temperature and the amount of debris are significantly associated with animal assemblages in the
16 pools (Tuno et al 2005). The presence of epizoic algae has a significant positive effect on water body productivity of A. gambiae (Tuno et al 2006). Cannibalism of younger larvae by older developmental stages, or instars, is also common (Koenraadt and Takken 2003). Of the physiochemical variables canopy cover, habitat size, water minerals (ammonium, nitrate, and phosphate), pH, and turbidity, only canopy cover significantly associates with A. gambiae abundance. The pH ranges from 6.1 to 6.3 and turbidity lies between 58.5 and 98.5 NTU (Minakawa et al 2005). The relative productivity of three types of habitat, drainage ditches, cow hoof prints, and abandoned goldmines is not statistically significant (Munga et al 2007). On the opposite side of Africa, the same vector had similar daily survival rates in the rainy season for rock pools (0.807), swamps (0.899), puddles (0.818), and artificial containers (0.863). In these habitats, mean temperature was 28˚–29˚C, pH was 7.02–7.25, and turbidity was 30.0 to 151.7 JTU (Edillo et al 2004).
Malaria and climate in the Amazon Basin
The Amazon receives limited attention from regional malaria researchers, who have published a suite of climate driven models for African nations. Moreover, as I explain below, the climatic fingerprint of malaria in the Amazon appears patently different than that of the African based MARA/ARMA model and hence the predictions of global malaria models based on conditions in Africa may not be applicable or relevant. Given the lack of a South American specific model of malaria and climate and the inadequate translation of models fashioned for Africa, the assessment of climate on malaria in the Amazon Basin relies heavily on recent trends and local studies.
17 Unlike most other regions of the world, there has been no increasing temperature trend in the Amazon over the past hundred years, although during that time annual precipitation increased by ~20–50 percent for many locations (Hulme and Sheard 1999). My preliminary research shows mean temperature patterns in the Basin very rarely cross the MARA/ARMA temperature suitability window, which lies between 22˚C and 34˚C (Craig et al 1999). The bottom graph of Figure 1.3 displays the mean temperature variability between 1996 and 1999 and periods of La Niña and El Niño events. Hence these temperature thresholds are not informative for the Basin. Figure 1.4a displays the relationship of temperature and pairwise complete observations of county reported malaria incidence from 1996 to 1999. The median temperature is 27˚C, very near the peak incidence rates. The mean is 26.9˚C and the 95 percent error bars around the mean fall at 24.3˚C and 29.6 ˚C. Above and below ~27˚C, peak malaria rates sharply decline. This suggests that malaria incidence is optimized at the median temperature and that increasing temperature may not lead to steeply increased risk of malaria. Below 27˚C, this observation is in agreement with prevailing theories of malaria risk and temperature. The distribution of temperature is positively skewed above the mean (Pearson’s measure of kurtosis is 7.67). A similar skew in the malaria incidence and temperature curve suggests increasing temperatures may create more suitable conditions for malaria in locations with similar temperature trends (Figure 1.4b). Currently the sites that are experiencing these temperature related increases in malaria fall along the main channel of the Amazon and extend up into Roraima. The majority (77 percent) of events (n=30) occur during El Niño months in 1997–1998. Seven events occur in July, 14 occur in August, and nine occur in October. Three counties experience two of these events. If future trends mimic the temperature conditions created by El Niño, the malaria rates in
18 this region may increase in a similar pattern as a result of shorted latent periods of the parasite. This back-of-the-envelope analysis requires further detailed review and has caveats. The locations are located in large population centers that are undergoing rapid development, thus the denominator in the calculation of incidence may be underestimated. As the Amazon River is a major route of transport, the probability that the reported cases did not originate in the local population is also a potential source of bias. Given these limitations, it is obvious that further research is needed to draw strong conclusions about the relationship between temperature and malaria risk in the Amazon.
The influence of precipitation on A. darlingi transmitted malaria is complicated. Drawing from the CRU TS 2.1 half degree resolution climate data set for 1996–1999, I find that the precipitation range for counties in the Amazon is zero to 840.6 mm per month (Mitchell and Jones 2005). The majority (62 percent) of all county months (n=35,240) report precipitation greater than the 80 mm. The seasonality of mean monthly precipitation is shown in Figure 1.3 along with the 80 mm threshold used in the MARA/ARMA model of malaria (Craig et al 1999). The average malaria incidence for the Basin follows the inverse of the precipitation curve, a considerable departure from the near synchronous (1–3 month lags) patterning of peak rainfall and malaria events in Africa (Olson et al 2009).
These broad trends suggest the MARA/ARMA model does not translate well across a large portion of the Amazon Basin, but several studies can provide more specific information about the sporadic seasonal and geographic nature of climate related conditions on adult A.
19 darlingi populations and malaria incidence. The mean number of human landing catches in the Yanomami Amerindian population (1,279 per 1,000 API) in the Upper Orinoco River is significantly and positively correlated with malaria incidence and peaked during the rainy season (September), and with a one month lag on the maximum river level. Village average annual rainfall is 2,487 mm and the average annual temperature is 24˚C, with 80 percent relative humidity (Magris et al 2007). In Roriama, there are more numbers of biting A. darlingi in the dry season for a tropical rainforest site and conversely more numbers in the wet season for a savannah near the alluvial forest of the Branco River (de Barros et al 2007). Overall in Roraima the rate ratios of malaria incidence for the middle of dry season (January–March) and early after the wet season (October–December) are 1.23 and 1.20 respectively and significant at the 95 percent confidence interval (Chaves and Rodrigues 2000). In Rondônia the river population (mostly natives) observes peak malaria incidence a month after the annual monthly peak rainfall and an inland lumber operation settlement malaria season peaks in the middle of the dry season (Gil et al 2003). In 2002 and 2003 peak malaria incidence in riverside Porto Velho, Rondonia, followed peak rainfall by one to three months (Gil et al 2007). Likewise, malaria on the Surinamese side of the Maroni River peaks with max monthly rainfall and river height (Hudson 1984).
How deforestation can impact malaria in the Amazon Basin
A study of deforestation and A. darlingi habitat carried out in the Peruvian Amazon shows strong connections between local environment and malaria. Controlling for human population density, sites with greater than 80 percent deforestation have mean biting rate 8.33
20 (7.86, 8.81) and sites with less than 30 percent deforestation have mean biting rate 0.03 (0.01, 0.08). Sites near deep water have mean biting rate of 4.8 compared to 1.9 for sites distant from deep water (Vittor et al 2006). The larval habitat survey showed the likelihood of finding of A. darlingi larvae doubles in breeding sites with less then 20 percent forest compared to sites with 20–60 percent forest and the likelihood jumps seven fold when compared to sites with over 60 percent forest. On average, sites harboring A. darlingi are less likely to be situated on varillal forests (forests with sandy soils), and more likely to be situated in sites with secondary growth, shrub and grass or crop land (Vittor et al 2009). The effects of microclimate (very localized changes in temperature of water bodies and relative humidity) were not studied. However, as research in Africa has shown, ambient temperature, larval habitat temperature and humidity depend on forest coverage (Afrane et al 2005, 2006, 2007).
Deforestation has two operational pathways to malaria outcomes. The first pathway is a direct effect on local conditions, i.e. the distribution and abundance of vector habitat and microclimate. If the pattern of frontier malaria continues, that is, “the opening of roads, a disorganized occupation, associated disturbance of the natural environment and a lack of infrastructure,” it is conceivable that malaria will continue its resurgence in the region (de Castro 2003; de Castro et al 2006). One estimate predicts that 25 percent of currently closed forests in Amazonia are potentially malarious, placing 11,654,151 people at risk of malaria in 2005 (Guerra et al 2006). The combined pressures of a growing roadway system and international markets of free-range beef and soybean have driven deforestation rates to their highest level yet in the last decade (INPE 2009; Soares-Filho et al 2004; Foley et al 2007).
21 The second pathway is indirect, as deforestation is likely to effect precipitation, aquatic habitats, flooding regimes, and other ecosystem services related to mosquito habitat suitability (Foley et al 2007). Land surface processes and atmospheric feedbacks suggest that like much like historic deforestation, future deforestation will, “likely be significant and a complex function of how much vegetation has been removed from that particular watershed and how much has been removed from the entire Amazon Basin” (Coe et al 2009). While there is strong evidence for a combined effect of deforestation and climate, at present, the uncertainty of land surface and climate models in combination with a lack of regional malaria modeling – specifically hydrological modeling – in the Basin makes composing synergistic links to malaria risk difficult.
Climate, landscape, and ecological risk factors of Lyme disease
On the surface, Lyme disease appears unrelated to malaria. There is one infectious agent, a bacteria called Borrelia burgdorferi senso lato and one vector species, a tick called Ixodes scapularis, in the North Central and Eastern United States, compared to the dozen or so mosquito species responsible for the majority of global malaria transmission. And the Ixodes tick appears to have a uniform, spatial and temporal ecological niche in the eastern United States. Using the tick as its vector, B. burgdorferi persists in an enzootic cycle with multiple small vertebrate hosts, each with distinctive ecological niches, and some more likely to infect a biting nymphal tick than others (Randolph 2004). In the case of malaria, there are multiple parasitic agents, numerous vectors, but just one host. However, what vector-borne diseases such as Lyme disease and malaria have in common are their ecological roots and intimate connections to the environmental landscape (Ostfeld et al 2005; Rogers and Randolph 2003).
22 Lyme disease is the most prevalent vector-borne infectious disease in the United States. Disease symptoms are variable, but, if it is not timely diagnosed and successfully treated, potential sequlae include persistent joint inflammation lasting months to several years, neurocognititive decline, and severe musculoskeletal pain. It is entrenched along the northeastern seaboard of the United States, but since the mid-1960s, Lyme disease has slowly expanded across the North Central United States to reach Minnesota and Wisconsin, states that are now highly endemic (Steere et al 2004). In 2005, Lyme disease incidence in Minnesota was 17.7 per 100,000 and 26.4 per 100,000 in Wisconsin, and nationally, between 1992 and 2006 250,000 cases were reported to the CDC (Bacon et al 2008). Environmental and ecological risk factors underpin Lyme disease transmission and ultimately determine human disease risk.
How biodiversity impacts the enzootic cycle
Humans are dead end hosts in the complex enzootic transmission cycle of B. burgdorferi. The larvae of the primary vector, Ixodes scapularis hatch in the summer. The following year they molt into a nymphal stage and take blood meals in early spring to mid-summer. Blood meals taken by the larvae or nymphs from an infected host subsequently initiate the horizontal transfer of B. burgdorferi. Once the tick becomes infected, a subsequent blood meal can transfer the spirochete bacteria to an uninfected host. Unlucky humans interject themselves into this tidy enzootic transmission cycle primarily when nymphs are taking blood meals. Often, a bulls-eye like rash will develop on the skin of an infected patient, and in most cases prompt antibiotic treatment results in full recovery. Meanwhile, the lifecycle of the tick vector continues, as the
23 nymphs again molt and emerge the following year as adults. Adult ticks usually feed on whitetailed deer, where they mate and lay eggs in the detritus of deciduous forests, which will hatch and begin the cycle again the following year (Steere et al 2004).
The biodiversity of hosts during the larval and nymphal tick phases of the tick life cycle regulate Lyme disease risk along with abiotic factors. Logically, areas abundant with ticks, competent reservoir hosts, white-tailed deer, and frequent human trafficking are most likely to have a high Lyme disease risk. But not all reservoir hosts are equal and their competency varies from one host species to the next, hence the biodiversity of vertebrate host species drives the enzootic cycle of B. burgdorferi. The biological players in the enzootic cycle thrive in specific landscapes and climates, or ecological niches. Continentally, strong NDVI signals and higher winter temperature together explain some variation in tick occurrence, and other studies show that the maintenance of tick populations is associated with 30 year average monthly maximum (+), minimum (+) and mean (–) temperature, and vapor pressure variability (+) (Estrada-Peña 2002; Estrada-Peña 1998; Brownstein et al 2003).
At the local and regional scales, other environmental variables are known to be associated with tick distributions. Agricultural lands and contiguous forests suppress tick presence, while forests edges, landscape slopes between zero and four degrees, sandy soils and lower elevation correlate with tick presence (Bunnell et al 2003; Das et al 2002). Seasonally, climate can also influence vector abundance. Lengthy summers correlate with higher tick numbers in the same year, and lower tick numbers in the following year, while cumulative annual rainfall is more
24 correlated with tick numbers of the following year (Jones and Kitron 2000). Deer populations have exploded in part because their habitat has grown, or more precisely, has re-grown. Complete deforestation of the North Eastern forests took place in the 1700 and 1800s, and nearly eradicated the deer population. Barbour and Fish call deer a ‘keystone host’ for I. scapularis populations, and point out the necessary spatial overlap of deer and I. scapularis populations (1993).
The most precise estimate of Lyme disease risk is the abundance of infected vectors (Ostfeld et al 2005; Barbour and Fish 1993). As mentioned earlier, beyond the overlapping distribution of the enzootic actors, the biodiversity surrounding each tick larvae or nymph creates an ecological disease process that regulates Lyme disease risk. Controlling for vector abundance, scientists have investigated other risk factors that influence the rate at which vectors become infected with B. burgdorferi. Their findings build on observations of the variability of reservoir competency, or the ability of a host species to harbor the bacterial spirochete and infect a biting tick. In laboratory settings, white-footed mice (Peromyscus leucopus) have 90 percent reservoir competency, or nine times out of ten a biting tick becomes infected by taking a blood meal from an infected mouse, the greatest competency of all small mammals and almost twice the next highest, the Eastern chipmunk (Tamias striatus) (LoGiudice et al 2003). A genotyping study suggests the white-footed mouse, chipmunk, short-tailed shrew (Blarina brevicauda), and masked shrew (Sciurus carolinensis) have equivalent host competency, but much lower competencies for raccoons, opossums, birds, deer, and the striped skunk (Brisson et al 2008). Because the tick is a nonspecific feeder, it has been proposed that community assemblages of
25 low competency hosts limit disease spread by reducing the number of blood meals taken from high competency hosts, such as mice (Ostfeld and Keesing 2000). In high biodiversity settings, the phenomenon, called the ‘dilution effect,’ decreases the likelihood that larval blood meals are taken from the ubiquitous white-footed-mouse (high reservoir competence), while increasing the likelihood that larval blood meals are taken from squirrels or raccoons (low reservoir competence). Thus, lower host biodiversity settings increase the rate at which the ticks become infected (Schmidt and Ostfeld 2001; LoGiudice et al 2003; Brisson et al 2008).
The sum realization of local observations over space and time explain the geographic extent of a vector-borne infectious disease. For instance, Ostfeld and Keesing propose the dilution effect limits Lyme disease transmission in the Southern United States, a region associated with higher species richness, especially of ground-dwelling birds that have poor reservoir competency (2000). In other words, greater biodiversity and species assemblages may be limiting the southern spread of Lyme disease. Recent research shows that climate limits the migration of Ixodes ticks into eastern Canada, and several climate change scenarios predict that ticks and Lyme disease will spread to the southeastern corner of Alberta across to eastern tip of Newfoundland by 2050 or 2080 (Ogden et al 2005; Ogden et al 2006). This finding followed longitudinal studies showing earlier spring and warmer winters were positively associated with annual tick abundance and Lyme disease incidence.
Other tick-borne pathogens reveal similar geographic limitations. An increase in incidence of tick-borne encephalitis in Sweden between 1960 and 1998 is attributed to global
26 warming trends that resulted in an earlier onset of spring and milder winters (Lindgren and Gustafson 2001). The joint distribution of Ehrlichia chaffeensis and Anaplasma phagocytophilum, the agents of human monocytotropic ehrlichiosis and human granulocytic anaplasmosis, respectively, is bounded in range by climate in southeastern United States and to the west by the dwindling forest cover of the Great Plains (Wimberly et al 2007).
The role of fragmentation
Fragmentation can change the ecologic balance of the enzootic cycle and increase the incidence of B. burgdorferi in I. scapularis (Allan et al 2003; Brownstein et al 2005). Forest fragmentation and suburbanization is the latest landscape trend across the United States; Brown et al. show the transition of land-use by mapping the change in exurban use between 1950 and 2000 (Figure 1.5) (2005). Research suggests the dilution effect is an ecological disease process related to forest fragmentation and that landscape metrics are able to detect this process. Landscape metrics quantify the amount and arrangement of landscape components, so they can quantify the amount of fragmentation along with other characteristics (Turner et al 2001). Loss of habitat and fragmentation are the main drivers of biodiversity decline. Species richness declines as the landscape and habitats becomes more and more fragmented (Mace 2005; Nupp and Swihart 2000). In the case of Lyme disease, holding other factors constant, a small isolated forest patch will have a larger population density of white-footed mice, and higher nymphal infection prevalence than a larger isolated forest patch (Brownstein et al 2005; Nupp and Swihart 1996, 1998; Allan et al 2003).
27 Some studies have included forest fragmentation to create maps of human Lyme disease risk. While the connection of forest fragmentation and the nymphal infection prevalence is well established, the connection to human disease risk remains controversial. For instance, human disease risk was not correlated with fragmented forests (measured by patch size and isolation) in an area surrounding Lyme, Connecticut, but fragmented forests (measured by percent herbaceous edge adjacent to forest) for about a dozen counties in Maryland were positively correlated (Brownstein et al 2005; Jackson et al 2006). This theoretical framework, that nymphal infection prevalence is correlated with human incidence, makes the assumption visitor frequency and characteristics are even, case reporting rates are uniform, and the home address of a case closely correlates to the geographic location of infection, contiguously across the study area (Ostfeld et al 2005). Future analysis will have to address these assumptions, and consider either manipulative or mensurative ecological experiments to establish strong evidence that Lyme disease in humans is associated with the ecological mechanism of fragmentation (McGarigal and Cushman 2002).
Summary In this literature review, I document known physical environmental and ecological dependencies of the agents and vectors of malaria and Lyme disease. This knowledge is currently based on a mixture of local and global observations. Patterns of both dependencies are well understood for isolated research locations where these and a variety of other infectious disease risk factors can be adjusted. The focus of this dissertation is to expand an integrated understanding of physical, ecological, and socioeconomic infectious disease risk factors to
28 larger, population health relevant, regional spatial scales.
29 References Abu-Raddad LJ, Patnaik P, Kublin JG. Dual infection with HIV and malaria fuels the spread of both diseases in sub-Saharan Africa. Science. 2006 Dec 8;314(5805):1603-6. Afrane YA, Lawson BW, Githeko AK, Yan GY. Effects of microclimatic changes caused by land use and land cover on duration of gonotrophic cycles of Anopheles gambiae (Diptera: culicidae) in western Kenya highlands. J Med Entomol. 2005 Nov;42(6):97480. Afrane YA, Zhou G, Lawson BW, Githeko A, Yan G. Effects of microclimatic changes caused by deforestation on the survivorship and reproductive fitness of Anopheles gambiae in western Kenya Highlands. Am J Trop Med Hyg. 2006;74(5):772-8. Afrane YA, Zhou G, Lawson BW, Githeko AK, Yan G. Life-table analysis of Anopheles arabiensis in western Kenya highlands: Effects of land covers on larval and adult survivorship. Am J Trop Med Hyg. 2007 Oct;77(4):660-6. Allan BF, Keesing F, Ostfeld RS. Effect of forest fragmentation on Lyme disease risk. Conserv Biol. 2003 Feb;17(1):267-72. Alves FP, Gil LHS, Marrelli MT, da Silva LHP. Asymptomatic carriers of Plasmodium spp. as infection source for malaria vector mosquitoes in the Brazilian Amazon. Population and Community Ecology. 2005;42(5):777-9. Bacon RM, Kugeler KJ, Mead PS. Surveillance for Lyme disease – United States, 1992-2006. Morbidity and Mortality Weekly Report. 2008 October 3, 2008;57(SS10):1-9. Barbour AG, Fish D. The Biological and social phenomenon of Lyme disease. Science. 1993 Jun 11;260(5114):1610-6.
30 Bøgh C, Clarke SE, Walraven GEL, Lindsay SW. Zooprophylaxis, artefact or reality? A pairedcohort study of the effect of passive zooprophylaxis on malaria in the Gambia. Trans R Soc Trop Med Hyg. 2002;96:593-6. Bomblies A, Duchemin JB, Eltahir EAB. Hydrology of malaria: Model development and application to a Sahelian village. Water Resources Research. 2008 Dec 31;44(12). Bouma MJ, Dye C. Cycles of malaria associated with El Niño in Venezuela. Jama. 1997 Dec 3;278(21):1772-4. Bouma MJ, Poveda G, Rojas W, Chavasse D, Quinones M, Cox J, et al. Predicting high-risk years for malaria in Colombia using parameters of El Niño southern oscillation. Trop Med Int Health. 1997 Dec;2(12):1122-7. Brisson D, Dykhuizen DE, Ostfeld RS. Conspicuous impacts of inconspicuous hosts on the Lyme disease epidemic. Proc R Soc B-Biol Sci. 2008 Jan 22;275(1631):227-35. Brown DG, Johnson KM, Loveland TR, Theobald DM. Rural land-use trends in the conterminous United States, 1950-2000. Ecol Appl. 2005 December 2005;15(6):1851-63. Brownstein JS, Holford TR, Fish D. A climate-based model predicts the spatial distribution of the Lyme disease vector Ixodes scapularis in the United States. Environ Health Perspect. 2003 Jul;111(9):1152-7. Brownstein JS, Skelly DK, Holford TR, Fish D. Forest fragmentation predicts local scale heterogeneity of Lyme disease risk. Oecologia. 2005;146:469-75. Bunnell JE, Price SD, Das A, Shields TM, Glass GE. Geographic Information Systems and spatial analysis of adult Ixodes scapularis (Acari: Ixodidae) in the Middle Atlantic region of the USA. J Med Entomol. 2003 Jul;40(4):570-6.
31 Carter R, Mendis KN. Evolutionary and historical aspects of the burden of malaria. Clinical Microbiology Reviews. 2002;15(4):564-94. Ceccato P, Ghebremeskel T, Jaiteh M, Graves PM, Levy M, Ghebreselassie S, et al. Malaria stratification, climate, and epidemic early warning in Eritrea. Am J Trop Med Hyg. 2007 December 1, 2007;77(6_Suppl):61-8. Charlwood JD. Biological variation in Anopheles darlingi Root. Mem Inst Oswaldo Cruz. 1996;91(4):391-8. Charlwood JD. Observations on the Bionomics of Anopheles darlingi Root (Diptera, Culicidae) from Brazil. Bull Entomol Res. 1980;70(4):685-92. Chaves SS, Rodrigues LC. An initial examination of the epidemiology of malaria in the state of Roraima, in the Brazilian Amazon Basin. Rev Inst Med Trop Sao Paulo. 2000;42(5):26975. Coe MT, Costa MH, Soares-Filho BS. The influence of historical and potential future deforestation on the stream flow of the Amazon River - land surface processes and atmospheric feedbacks. Journal of Hydrology. 2009;369(1-2):165. Coluzzi M. The clay feet of the malaria giant and its African roots: hypothesis and inferences about origin, spread and control of Plasmodium falciparum. Parassitologia. 1999;41:27783. Craig MH, Snow RW, le Sueur D. A climate-based distribution model of malaria transmission in sub-Saharan Africa. Parasitol Today. 1999 Mar;15(3):105-11.
32 Das A, Lele SR, Glass GE, Shields T, Patz J. Modelling a discrete spatial response using generalized linear mixed models: application to Lyme disease vectors. International Journal of Geographical Information Science. 2002 Mar;16(2):151-66. de Barros FSM, Honorio NA. Man biting rate seasonal variation of malaria vectors in Roraima, Brazil. Mem I Oswaldo Cruz. 2007 Jun;102(3):299-302. de Castro MC, Monte-Mór RL, Sawyer D, Singer BH. Malaria risk on the Amazon frontier. Proc Natl Acad Sci U S A. 2006;103(7):2452-7. Deane LM. Notas sobre a distribuição e a Biologia dos Anofelinos das regiões Nordestina e Amazônica do Brasil. Rev Serv Esp Saúde Publ. 1948;1:827-965. Deressa W, Ali A, Berhane Y. Household and socioeconomic factors associated with childhood febrile illnesses and treatment seeking behaviour in an area of epidemic malaria in rural Ethiopia. Trans Roy Soc Trop Med Hyg. 2007 Sep;101(9):939-47. Dev V, Bhattacharya PC, Talukdar R. Transmission of malaria and its control in the northeastern region of India. JAPI. 2003;51:1073-6. Dev V, Phookan S, Sharma VP, Anand SP. Physiographic and entomologic risk factors of malaria in Assam, India. Am J Trop Med Hyg. 2004;71(4):451-6. Duarte EC, Gyorkos TW, Pang L, ABrahamowicz M. Epidemiology of malaria in a hypoendemic Brazilian Amazon migrant population: a cohort study. Am J Trop Med Hyg. 2004;70(3):229-37. Edillo FE, Toure YT, Lanzaro GC, Dolo G, Taylor CE. Survivorship and distribution of immature Anopheles gambiae s.l. (Diptera: Culicidae) in Banambani village, Mali. J Med Entomol. 2004 May;41(3):333-9.
33 Edillo FE, Toure YT, Lanzaro GC, Dolo G, Taylor CE. Survivorship and distribution of immature Anopheles gambiae s.l. (Diptera: Culicidae) in Banambani village, Mali. J Med Entomol. 2004 May;41(3):333-9. Eisele TP, Lindblade KA, Wannemuehler KA, Gimnig JE, Odhiambo F, Hawley WA, et al. Effect of sustained insecticide-treated bed net use on all-cause child morality in an area of intense perennial malaria transmission in Western Kenya. Am J Trop Med Hyg. 2005;73(1):149-56. Estrada-Pena A. Geostatistics and remote sensing as predictive tools of tick distribution: a cokriging system to estimate Ixodes scapularis (Acari: Ixodidae) habitat suitability in the United States and Canada from advanced very high resolution radiometer satellite imagery. J Med Entomol. 1998 Nov;35(6):989-95. Estrada-Pena A. Increasing habitat suitability in the United States for the tick that transmits Lyme disease: A remote sensing approach. Environ Health Perspect. 2002 Jul;110(7):635-40. Foley JA, Asner GP, Costa MH, Coe MT, DeFries R, Gibbs HK, et al. Amazonia revealed: forest degradation and loss of ecosystem goods and services in the Amazon Basin. Frontiers In Ecology And The Environment. 2007 Feb;5(1):25-32. Gagnon AS, Smoyer-Tomic KE, Bush ABG. The El Niño Southern Oscillation and malaria epidemics in South America. Int J Biometeorol. 2002 May;46(2):81-9. Gallup JL, Sachs JD. The economic burden of malaria. Am J Trop Med Hyg. 2001 Jan-Feb;64(12):85-96.
34 Gil LHS, Alves FP, Zieler H, Salcedo JMV, Durlacher RR, Cunha RPA, et al. Seasonal malaria transmission and variation of Anopheline density in two distinct endemic areas in Brazilian Amazonia. J Med Entomol. 2003 Sep;40(5):636-41. Gil LHS, Tada MS, Katsuragawa TH, Ribolla PEM, da Silva LHP. Urban and suburban malaria in Rondônia (Brazilian Western Amazon) II. Perennial transmissions with high Anopheline densities are associated with human environmental changes. Mem I Oswaldo Cruz. 2007 Jun;102(3):271-6. Gilles HM, Warrell DA. Bruce-Chwatt's essential malariology. 3rd ed. London; New York: Arnold; 1993. Gilles MT, DeMeillon B. The Anophelinae of Africa south of the Sahara (Ethiopian zoogeographical region). 1968;54. Gollin D, Zimmerman C. Malaria: disease impacts and long-run income differences. Institue for the Study of Labor Discussion Paper. 2007(2997). Guerra CA, Gikandi PW, Tatem AJ, Noor AM, Smith DL, Hay SI, et al. The limits and intensity of Plasmodium falciparum transmission: implications for malaria control and elimination worldwide. Plos Medicine. 2008 February 2008;5(2):e38. Guerra CA, Snow RW, Hay SI. A global assessment of closed forests, deforestation and malaria risk. Annals Of Tropical Medicine And Parasitology. 2006 Apr;100(3):189-204. Guerra CA, Snow RW, Hay SI. Defining the global spatial limits of malaria transmission in 2005. Advances In Parasitology, Vol 62. San Diego: Elsevier Academic Press Inc; 2006. p. 157-79.
35 Hamoudi A, Sachs JD. The changing global distribution of malaria: a review: Center for International Development at Harvard University; 1999. Report No.: 2. Hay SI, Guerra CA, Tatem AJ, Noor AM, Snow RW. The global distribution and population at risk of malaria: past, present, and future. Lancet Infect Dis. 2004 Jun;4(6):327-36. Hay SI, Snow RW, Rogers DJ. Predicting malaria seasons in Kenya using multitemporal meteorological satellite sensor data. Trans Roy Soc Trop Med Hyg. 1998 JanFeb;92(1):12-20. Hudson JE. Anopheles darlingi Root (Diptera: Culicidae) in the Suriname rain forest. Bulletin Entomological Research. 1984;74:129-42. Jackson LE, Levine JF, Hilborn ED. A comparison of analysis units for associating Lyme disease with forest-edge habitat. Community Ecol. 2006;7(2):189-97. Jones CJ, Kitron UD. Populations of Ixodes scapularis (Acari: Ixodidae) are modulated by drought at a Lyme disease focus in Illinois. J Med Entomol. 2000 May;37(3):408-15. Kent RJ, Thuma PE, Mharakurwa S, Norris DE. Seasonality, blood feeding behavior, and transmission of Plasmodium falciparum by Anopheles arabiensis after an extended drought in southern Zambia. Am J Trop Med Hyg. 2007 Feb;76(2):267-74. Killeen GF, Fillinger U, Kiche I, Gouagna LC, Knols BGJ. Eradication of Anopheles gambiae from Brazil: lessons for malaria control in Africa? Lancet Infect Dis. 2002 Oct;2(10):618-27. Killeen GF, Seyoum A, Knols BGJ. Rationalizing historical successes of malaria control in Africa in terms of mosquito resource availability management. Am J Trop Med Hyg. 2004 Aug;71(2):87-93.
36 Kiszewski A, Mellinger A, Spielman A, Malaney P, Sachs SE, Sachs J. A global index representing the stability of malaria transmission. Am J Trop Med Hyg. 2004 May;70(5):486-98. Kleinschmidt I, Sharp BL, Clarke GPY, Curtis B, Fraser C. Use of generalized linear mixed models in the spatial analysis of small-area malaria incidence rates in KwaZulu Natal, South Africa. Am J Epidemiol. 2001 Jun 15;153(12):1213-21. Koenraadt CJM, Takken W. Cannibalism and predation among larvae of Anopheles gambiae complex. Medical & Veterinary Entomology. 2003;17:61-6. Laserson KF, Wypij D, Petralanda I, Spielman A, Maguire JH. Differential perpetuation of malaria species among Amazonian Yanomami Amerindians. Am J Trop Med Hyg. 1999 May;60(5):767-73. Le Seuer D. PhD Thesis: University of Natal; 1991. Lindblade KA, Walker ED, Onapa AW, Katungu J, Wilson ML. Land use change alters malaria transmission parameters by modifying temperature in a highland area of Uganda. Trop Med Int Health. 2000 April;4(4):263-74. Lindgren E, Gustafson R. Tick-borne encephalitis in Sweden and climate change. Lancet. 2001 Jul 7;358(9275):16-8. Livingstone FB. Anthropological implications of sickle cell gene distribution in West Africa. American Anthropology. 1958;60:533-62. LoGiudice K, Ostfeld RS, Schmidt KA, Keesing F. The ecology of infectious disease: Effects of host diversity and community composition on Lyme disease risk. Proc Natl Acad Sci U S A. 2003 Jan 21;100(2):567-71.
37 Mabaso MLH, Kleinschmidt I, Sharp B, Smith T. El Niño Southern Oscillation (ENSO) and annual malaria incidence in Southern Africa. Trans Roy Soc Trop Med Hyg. 2007 Apr;101(4):326-30. Mace G. Biodiversity. In: Carpenter SR, Pingali P, editors. Conditions and Trends Assessment of the Millennium Ecosystem Assessment. Washington, DC, USA: Island Press; 2005. Magris M, Rubio-Palis Y, Menares C, Villegas L. Vector bionomics and malaria transmission in the Upper Orinoco River, southern Venezuela. Mem I Oswaldo Cruz. 2007 Jun;102(3):303-11. Mahande A, Mosha F, Mahande J, Kweka E. Feeding and resting behaviour of malaria vector, Anopheles arabiensis with reference to zooprophylaxis. Malaria Journal. 2007 Jul 30;6. Maharaj R. PhD Thesis: University of Natal; 1995. Manguin S, Roberts DR, Andre RG, Rejmankova E, Hakre S. Characterization of Anopheles darlingi (Diptera: Culicidae) larval habitats in Belize, Central America. J Med Entomol. 1996;33(2):205-11. Martens P, Kovats RS, Nijhof S, de Vries P, Livermore MTJ, Bradley DJ, et al. Climate change and future populations at risk of malaria. Global Environ Chang. 1999;9:S89-S107. Martens WJM, Jetten TH, Focks DA. Sensitivity of malaria, schistosomiasis and dengue to global warming. Climatic Change. 1997 FEB;35(2):145-56. Martens WJM, Niessen LW, Rotmans J, Jetten TH, McMichael AJ. Potential impact of global climate-change on malaria risk. Environ Health Perspect. 1995 May;103(5):458-64. McGarigal K, Cushman SA. Comparative evaluation of experimental approaches to the study of habitat fragmentation effects. Ecol Appl. 2002 Apr;12(2):335-45.
38 Minakawa N, Munga S, Atieli F, Mushinzimana E, Zhou G, Githeko AK, et al. Spatial distribution of Anopheline larval habitats in Western Kenyan Highlands: effects of land cover types and topography. Am J Trop Med Hyg. 2005;73(1):157-65. Minakawa N, Sonye G, Mogi M, Githeko A, Yan GY. The effects of climatic factors on the distribution and abundance of malaria vectors in Kenya. J Med Entomol. 2002 Nov;39(6):833-41. Mouchet J, Nadire-Galliot M, Gay F, Poman JP, Lepelletier L, Claustre J, et al. Malaria in Guiana. II. The characteristics of different foci and antimalarial control. Bull Soc Pathol Exot Filiales. 1989;82(3):393-405. Munga S, Minakawa N, Zhou G, Githeko AK, Yan G. Survivorship of immature stages of Anopheles gambiae s.l. (Diptera: culicidae) in natural habitats in western Kenya highlands. J Med Entomol. 2007 Sep;44(5):758-64. Need JT, Rogers EJ, Phillips IA, Falcon R, Fernandez R, Carbajal F, et al. Mosquitos (Diptera, culicidae) captured in the Iquitos area of Peru. J Med Entomol. 1993 May;30(3):634-8. Nupp TE, Swihart RK. Effect of forest patch area on population attributes of white-footed mice (Peromyscus leucopus) in fragmented landscapes. Can J Zool-Rev Can Zool. 1996 Mar;74(3):467-72. Nupp TE, Swihart RK. Effects of forest fragmentation on population attributes of white-footed mice and eastern chipmunks. J Mammal. 1998 Nov;79(4):1234-43. Nupp TE, Swihart RK. Landscape-level correlates of small-mammal assemblages in forest fragments of farmland. J Mammal. 2000 May;81(2):512-26.
39 Ogden NH, Barker IK, Beauchamp G, Brazeau S, Charron DF, Maarouf A, et al. Investigation of ground level and remote-sensed data for habitat classification and prediction of survival of Ixodes scapularis in habitats of southeastern Canada. J Med Entomol. 2006 Mar;43(2):403-14. Ogden NH, Bigras-Poulin M, O'Callaghan CJ, Barker IK, Lindsay LR, Maarouf A, et al. A dynamic population model to investigate effects of climate on geographic range and seasonality of the tick Ixodes scapularis. International Journal for Parasitology. 2005 Apr 1;35(4):375-89. Olson SH, Gangnon R, Elguero E, Durieux L, Guegan JF, Foley JA, et al. Links between climate, malaria, and wetlands in the Amazon Basin. Emerg Infect Dis. 2009 Apr;15(4):659-62. Ostfeld R, Keesing F. The function of biodiversity in the ecology of vector-borne zoonotic diseases. Can J Zool-Rev Can Zool. 2000 Dec;78(12):2061-78. Ostfeld RS, Glass GE, Keesing F. Spatial epidemiology: an emerging (or re-emerging) discipline. Trends In Ecology & Evolution. 2005 Jun;20(6):328-36. Pascual M, Ahumada JA, Chaves LF, Rodo X, Bouma M. Malaria resurgence in the East African highlands: Temperature trends revisited. Proc Natl Acad Sci U S A. 2006 APR 11;103(15):5829-34. Patz JA, Strzepek K, Lele S, Hedden M, Greene S, Noden B, et al. Predicting key malaria transmission factors, biting and entomological inoculation rates, using modelled soil moisture in Kenya. Trop Med Int Health. 1998 Oct;3(10):818-27.
40 Poveda G, Jaramillo A, Gil MM, Quiceno N, Mantilla RI. Seasonality in ENSO-related precipitation, river discharges, soil moisture, and vegetation index in Colombia. Water Resources Research. 2001 Aug;37(8):2169-78. Quinnes ML, Suarez MF. Indoor resting heights of some anophelines in Columbia. J Am Mosq Control Assoc. 1990;6:602-5. Randolph SE. Tick ecology: processes and patterns behind the epidemiological risk posed by Ixodid ticks as vectors. Parasitology. 2004;129:S37-S65. Roberts RD, Alecrim WD, Tavares AM, Radke MG. The house frequenting, host seeking and resting behaviour of Anopheles darlingi in southeastern Amazonas, Brazil. J Am Mosq Control Assoc. 1987;3:433-41. Rogers DJ, Randolph SE, Snow RW, Hay SI. Satellite imagery in the study and forecast of malaria. Nature. 2002 Feb 7;415(6872):710-5. Rogers DJ, Randolph SE. Studying the global distribution of infectious diseases using GIS and RS. Nature Reviews Microbiology. 2003 Dec;1(3):231-7. Rogers DJ, Randolph SE. The global spread of malaria in a future, warmer world. Science. 2000 Sep 8;289(5485):1763-6. Roper MH, Torres RSC, Goicochea CGC, Andersen EM, Guarda JSA, Calampa C, et al. The epidemiology of malaria in an epidemic area of the Peruvian Amazon. Am J Trop Med Hyg. 2000 Feb;62(2):247-56. Rozendaal JA. Observations on the biology and behaviours of Anophelines in the Suriname rainforest with special reference to Anopheles darlingi Root. Ent méd et Parasitol. 1987;25(1):33-43.
41 Rozendaal JA. Relations between Anopheles darlingi breeding habitats, rainfall, river level and malaria transmission rates in the rain-forest of Surinam. Med Vet Entomol. 1992 Jan;6(1):16-22. Sawyer D. Frontier malaria in the Amazon region of Brazil: types of malaria situations and some implications for control. Symposio Sobre a Malaria; 1988; Brazil; 1988. Schmidt KA, Ostfeld RS. Biodiversity and the dilution effect in disease ecology. Ecology. 2001 Mar;82(3):609-19. Singer BH, De Castro MC. Agricultural colonization and malaria on the Amazon frontier. Population Health And Aging. New York: New York Acad Sciences; 2001. p. 184-222. Singh N, Sharma VP. Patterns of rainfall and malaria in Madhya Pradesh, central India. Annals Of Tropical Medicine And Parasitology. 2002 Jun;96(4):349-59. Snow RW, Craig M, Deichmann U, Marsh K. Estimating mortality, morbidity and disability due to malaria among Africa's non-pregnant population. Bull World Health Organ. 1999;77(8):624-40. Soares-Filho B, Alencar A, Nepstad D, Cerqueira G, Diaz MdCV, Rivero S, et al. Simulating the response of land-cover changes to road paving and governance along a major Amazon highway: the Santarém-Cuiabá corridor. Global Change Biology. 2004;10(5):745-64. Spielman A, D'Antonio M. Mosquito: a natural history of our most persistent and deadly foe. 1st ed. New York: Hyperion; 2001. Steere AC, Coburn J, Glickstein L. The emergence of Lyme disease. J Clin Invest. 2004 Apr;113(8):1093-101.
42 Strobel M, Lefait JF, Dedet JP. Paludisme à Plasmodium falciparum chez les Amérindiens Wayana de Guyane française. Méd Mal Inf. 1985;4:162-4. Tadei WP, Dutary Thatcher B. Malaria vectors in the Brazilian Amazon: Anopheles of the subgenus Nyssorhynchus (1). Rev Inst Med Trop Sao Paulo. 2000 March-April;42(2):8794. Thomas CJ, Hay SI. Global climate change and malaria - Reply. The Lancet Infectious Diseases. 2005 May;5(5):259-60. Tuno N, Githeko AK, Nakayama T, Minakawa N, Takagi M, Yan GY. The association between the phytoplankton, Rhopalosolen species (Chlorophyta; Chlorophyceae), and Anopheles gambiae sensu lato (Diptera: Culicidae) larval abundance in western Kenya. Ecol Res. 2006 May;21(3):476-82. Tuno N, Okeka W, Minakawa N, Takagi M, Yan GY. Survivorship of Anopheles gambiae sensu stricto (Diptera: Culicidae) larvae in western Kenya highland forest. J Med Entomol. 2005 May;42(3):270-7. Turner MG, Gardner RH, O'Neill RV. Landscape ecology in theory and practice: pattern and process. New York: Springer; 2001. Vittor AY, Gilman R, Tielsch J, Glass G, Shields T, Pinedo-Cancino V, et al. Linking deforestation to malaria in the Amazon: characterization of the breeding habitat of the principle malaria vector, Anopheles darlingi. Am J Trop Med Hyg. 2009. Vittor AY, Gilman RH, Tielsch J, Glass G, Shields T, Lozano WS, et al. The effect of deforestation on the human-biting rate of Anopheles darlingi, the primary vector of falciparum malaria in the Peruvian Amazon. Am J Trop Med Hyg. 2006 JAN;74(1):3-11.
43 WHO. Field guide for malaria epidemic assessment and reporting. Draft for field testing. Geneva: World Health Organization; 2004. WHO. Malaria early warning systems, concepts, indicators and partners. A framework for field research in Africa. Geneva: World Health Organization; 2001. WHO. World malaria report 2008. Geneva: World Health Organization; 2008. Wimberly MC, Yabsley MJ, Baer AD, Dugan VG, Davidson WR. Spatial heterogeneity of climate and land-cover constraints on distributions of tick-borne pathogens. Glob Ecol Biogeogr. 2007. Zhou G, Minakawa N, Githeko AK, Yan G. Association between climate variability and malaria epidemics in the East African highlands.[erratum appears in Proc Natl Acad Sci U S A. 2004 Sep 14;101(37):13694]. Proc Natl Acad Sci U S A. 2004 Feb 24;101(8):2375-80.
44 Figures Figure 1.1. Levels of global malaria endemicity based on parasite ratios (Hay et al 2004).
45 Figure 1.2. Global distributions of dominate or potentially dominate important malaria vectors (Kiszewski et al 2004).
46 Figure 1.3. El Niño months are labeled in orange and La Niña months are labeled in green. The grey lines are the standard errors around mean temperature in black. The blue line is precipitation.
47 Figure 1.4. a) Temperature distribution and areas with high incidence at high temperatures and b) location of county population centers with greater than 30˚C monthly mean temperature and greater than 0.01 malaria cases/person. a.
b.
48 Figure 1.5. Land use change in rural America 1950–2000 (Brown et al 2005).
49 Chapter 2
Paper #1: Links between Climate, Malaria, and Wetlands in the Amazon Basin
Sarah H. Olson, Ronald Gangnon, Eric Elguero, Laurent Durieux, Jean-François Guégan, Jonathan A. Foley, and Jonathan A. Patz. Emerging Infectious Diseases. 2009 April;15(4):65962.
Article Summary: In the Amazon Basin, the relationship between precipitation and malaria can change sign, depending on the geography and underlying landscape: regions with high wetlands show a negative relationship between precipitation and malaria, and areas with low wetlands show a variable relationship between precipitation and malaria.
Abstract Climatic changes are altering patterns of temperature and precipitation, potentially affecting regions of malaria transmission. Here we find that areas in the Amazon Basin with few wetlands show a variable relationship between precipitation and malaria, while areas of high wetlands and malaria incidence show a negative relationship.
50 Text Global models of malaria can be used to forecast the impact of climate change on malaria – a highly climate-sensitive disease, currently causing over one million deaths each year, predominantly in children. However, a limitation of these models is the application of a uniform malaria-precipitation relationship to geographically diverse regions (1, 2, 3). Moreover, the Millennium Ecosystem Assessment has called for a, “more systematic inventory, by region and country, of current and likely population health impacts of ecosystem change” (4).
Precipitation and surface hydrology are key factors in determining the abundance of Anopheles mosquito vectors. Mosquitoes require pools of water to complete their life cycle, and malaria models have estimated changing transmission by setting minimum levels of precipitation below which mosquito populations are (theoretically) suppressed by dry conditions. But employing a uniform hydrological threshold to malaria does not capture critical characteristics of the landscape, soil, and rainfall regime (intensity, frequency), which are all known to contribute to the abundance, persistence, and spatial distribution of mosquito habitat.
In the Amazon, the predominant vector of malaria is Anopheles darlingi. Short longitudinal studies show that human landing catches of A. darlingi, which breeds along edges and debris in partially sunlight clear pools, are closely associated with local malaria rates (5, 6). These observations establish that elevated biting rates are found in regions of elevated malaria risk. Human biting rates likewise correlate with larval abundances, abundance of larval habitat, and proximity of humans and larval habitat (7, 8).
51 Various local observations demonstrate the existence of different seasonal patterns of malaria. In a three-year study in Roraima, eight municipalities show increased risk of malaria during the middle of the dry season or shortly following the wet season (9). Other literature on seasonal patterns is limited to local and short (lasting less than three-years) longitudinal studies that lack statistical analysis. While differing seasonal patterns do emerge in graphs, the collage of different data sources makes formulating a cohesive picture of these patterns in the Amazon rather difficult.
At a more regional scale inter-annual climatological cycles such as El Niño events provide insights into lower frequency malaria patterns. El Niño events, caused by warming sea surface temperatures in the central tropical Pacific, in Columbia are associated with a warmer temperatures, higher dew points, and less precipitation and river discharge. These climatic changes are associated with increases in malaria in the second half of El Niño years and persist for the following year (10). Similarly, higher malaria incidence occurs in the year following an El Niño event in Venezuela and Guyana (11).
Here with monthly reports of malaria and precipitation from across the Brazilian Amazon, we demonstrate that malaria incidence and precipitation patterns vary across this large region, and are influenced by the extent of wetlands.
In our study, we used monthly reports of slide-confirmed malaria and population data taken from 434 counties (municípios) in the Brazilian Amazon between 1996 and 1999, during which
52 there was no coordinated national malaria interventions (12). To relate these to climate, we used monthly precipitation and temperature from the CRU TS 2.1 gridded climate data set (13). The patterns of malaria incidence and monthly rainfall for the states of Amazonas, Mato Grosso and Roraima are shown in Figure 2.1. To consider how the precipitation-malaria relationship depends on surface water conditions – including the extent of open water and wetlands – we used 100 m x 100 m maps from the JERS-1 Synthetic Aperture Radar satellite and calculated the percentage of maximum inundatable open water and wetland coverage in each county (Fig. 2.2C) (14). In this region, monthly temperatures fell between 24.6˚C and 29.4˚C (well within the range for optimal malaria transmission) for 95 percent (18,416/19,364) of the observations included in the analysis; temperature relationships are not shown here.
The evaluation of seasonal patterns requires a degree of comparability of the models across the various regions. If one allowed both the lag and the rainfall coefficient to vary across regions, it would be very difficult to make any meaningful comparisons across regions as neither the lag nor coefficient have consistent meanings across models. So, one must either fix the coefficient and vary the lags (more difficult) or fix the lag and vary the coefficients (relatively easy) to maintain any ability to interpret the results. The aim is not to create a highly predictive model, but to describe the variable patterns of malaria incidence and precipitation. We chose to fix the lag and vary the coefficients.
To assess the association between malaria incidence and precipitation data, we estimated the rate ratio of malaria incidence associated with a one standard deviation increase in monthly
53 precipitation (~14 cm) separately for each county using the following Poisson regression model, which includes a flexible temporal trend represented as a natural cubic spline with six degrees of freedom (Fig. 2.2A).
The (estimated) regression coefficients from the county-specific models were then modeled as a spatially smooth surface, a thin-plate spline. The degrees of freedom for the thin-plate spline were selected using generalized cross-validation (Fig. 2.2B).
Our analysis shows that the relationships between precipitation and malaria incidence in the Amazon are spatially varied, and even change sign. Positive correlations between monthly precipitation and malaria incidence (rate ratios greater than one) occur in the upland regions of the southwest and central Amazon Basin, while negative correlations between precipitation and incidence (rate ratios less than one) occur in the north, largely along the main waterways of the Amazon and the major wetland regions of the basin (Fig. 2.2). For a ~14 cm increase in monthly rainfall, the rate of malaria may double in the upland area, whereas the rate may decrease by up to 80 percent along the main Amazon channel. The p-values of the precipitation coefficient are between 0.0002 and 0.0009 along the main waterways and 0.004 and 0.10 in uplands areas.
54 We hypothesize that the switching of the malaria-precipitation relationship from positive to negative is related to the extent of open water and wetlands in the Basin. Mosquito habitat in wetlands or along large rivers may washout during months with high precipitation, whereas in areas with fewer wetlands, mosquito habitat is precipitation limited.
To test this hypothesis, we compared the malaria-precipitation association for 338 counties that reported at least 80 cases of malaria over the 48 months against the estimated percent of open water / wetland cover for each county (Fig. 2.2D). Interestingly, the precipitation-linked risk of malaria falls as the percentage of wetland in each county increases, in keeping with our original hypothesis, but in counties with low percentages of wetlands the risk of malaria is variable. These same counties have very low levels of transmission, which may explain the absence of positive relationships.
Mechanisms similar to our wetlands hypothesis are found in the literature. These studies propose that flooding events create new pools of water suitable for mosquito larvae as the water levels slowly recede from alluvial forests along the Rio Branco in Roraima and the Maroni River, which sits on the frontier of Suriname and French Guiana (6, 15). Our results suggest monthly precipitation along the Amazon can have strong positive and negative associations with malaria incidence.
Further research will need to address the limitations of this study. These include the short time frame of our study and the crude countywide approximation of percentage wetlands as an
55 exposure. The quality and reliability of the health data is a concern, but we were able to verify the distribution of null reporting was unbiased temporally and spatially. Moreover, this study does not quantify increasing malaria in response to increasing or decreasing precipitation or the impact of lag factors. Instead we choose to focus on the seasonality of these patterns until longer data series of malaria incidence and climate data are made available.
Our evidence suggests that malaria risk in the Amazon is driven by precipitation, but that relationship is variable (more precipitation, more/less malaria) in the uplands or a negative relationship (more precipitation, less malaria) in areas dominated by wetlands and large rivers.
Our findings demonstrate the need to account for local landscape characteristics, especially the extent of wetlands and open water, in regional- to global-scale projections of the effects of climate change on malaria. A better understanding of the climatic- and landscape-controls on malaria will improve our ability to assess health risks in a changing world.
56 References 1. Craig MH, Snow RW, le Sueur D. A climate-based distribution model of malaria transmission in sub-Saharan Africa. Parasitol Today. 1999 Mar;15(3):105-11. 2. Rogers DJ, Randolph SE. The global spread of malaria in a future, warmer world. Science. 2000 Sep 8;289(5485):1763-6. 3. Guerra CA, Gikandi PW, Tatem AJ, Noor AM, Smith DL, Hay SI, et al. The limits and intensity of Plasmodium falciparum transmission: implications for malaria control and elimination worldwide. Plos Medicine. 2008 February 2008;5(2):e38. 4. MEA (Millennium Ecosystem Assessment). Ecosystems and human well-being: Synthesis. Washington, D.C.: World Resources Institute; 2005. 5. Gil LHS, Tada MS, Katsuragawa TH, Ribolla PEM, da Silva LHP. Urban and suburban malaria in Rondonia (Brazilian Western Amazon) II. Perennial transmissions with high anopheline densities are associated with human environmental changes. Mem I Oswaldo Cruz. 2007 Jun;102(3):271-6. 6. de Barros FSM, Honorio NA. Man biting rate seasonal variation of malaria vectors in Roraima, Brazil. Mem Inst Oswaldo Cruz. 2007;102(3):299-302. 7. Vittor AY, Gilman RH, Tielsch J, Glass G, Shields T, Lozano WS, et al. The effect of deforestation on the human-biting rate of Anopheles darlingi, the primary vector of falciparum malaria in the Peruvian Amazon. Am J Trop Med Hyg. 2006 JAN;74(1):3-11. 8. Vittor AY, Gilman R, Tielsch J, Glass G, Shields T, Pinedo-Cancino V, et al. Linking deforestation to malaria in the Amazon: characterization of the breeding habitat of the
57 principle malaria vector, Anopheles darlingi. Am J Trop Med Hyg. 2009 JUL;81(1):5-12. 9. Chaves SS, Rodrigues LC. An initial examination of the epidemiology of malaria in the state of Roraima, in the Brazilian Amazon Basin. Rev Inst Med Trop Sao Paulo. 2000;42(5):26975. 10. Poveda G, Rojas W, Quinones ML, Velez ID, Mantilla RI, Ruiz D, et al. Coupling between annual and ENSO timescales in the malaria-climate association in Colombia. Environ Health Perspect. 2001 May;109(5):489-93. 11. Gagnon AS, Smoyer-Tomic KE, Bush ABG. The El Niño Southern Oscillation and malaria epidemics in South America. Int J Biometeorol. 2002 May;46(2):81-9. 12. PAHO Roll Back Malaria Initiative in the Rainforest Region of South America. Cartagena: Pan American Health Organization (2000). 13. Mitchell TD, Jones PD. An improved method of constructing a database of monthly climate observations and associated high-resolution grids. International Journal of Climatology. 2005;25(6):693-712. 14. Hess LL, Affonso AA, Barbosa C, Gastil-Buhl M, Melack JM, Novo EMLM. Basinwide Amazon wetlands mask, 100 m, version Aug04 [map]. Available from: http://lba.cptec.inpe.br/ 15. Rozendaal JA. Relations between Anopheles darlingi breeding habitats, rainfall, river level and malaria transmission rates in the rain forest of Suriname. Medical & Veterinary Entomology. 1992;6:16-22.
58 Figures
Figure 2.1. Malaria incidence per 1,000 (black) and mean monthly precipitation (blue) for the states of Amazonas (A), Mato Grosso (B), and Roraima (C) is graphed along with the occurrence of La Niña (orange) and El Niño (red) events.
59 Figure 2.2. (A) Map of risk ratios for malaria incidence for one standard deviation (~14 cm) change in monthly precipitation (January 1996–December 1999) plotted at each county seat of government. Red dots indicate a reduced risk of malaria for a ~14 cm increase in monthly precipitation, whereas blue dots indicate an increased risk of malaria with increased precipitation. (B) Spatially smoothed risk ratios for ~14 cm changes in monthly precipitation. (C) Map of the percent wetland for each county analyzed in the Amazon Basin (shades of blue), counties without wetlands data (yellow), and counties fewer than 80 total cases (grey). Wetland color corresponds to values in the box plot at right. (D) Risk ratios of malaria incidence for ~14 cm changes in monthly precipitation by percentage wetland cover. Box width is proportional to the number of counties in each box.
60
61 Acknowledgements We acknowledge funding support from the NASA LBA-ECO program and a NSF-FACE exchange grant. We also thank the National Science Foundation International program (OISE0623583) and Franco-American Cultural Exchange (FACE) for financial support as well as Enrique Loyola (PAHO) for assistance with the malaria data.
Biographical Sketch Sarah Olson is a dissertator working towards a joint PhD in Population Health from the School of Medicine and Population Health and in Environment and Resources from the Nelson Institute at the University of Wisconsin-Madison. Her research seeks to understand regional landscape and climate linkages in the ecology of vector-borne infectious diseases.
62 Chapter 3
Paper #2: Deforestation Links to Malaria in the Amazon
Sarah H. Olson, Ronald Gangnon, Guilherme Silveria, Jonathan A. Patz. In preparation for submission to Emerging Infectious Diseases.
Abstract Malaria is the most prevalent vector borne disease in the Amazon. We use 2006 malaria reports for health districts, collected by the Programa Nacional de Controle da Malária (PNCM), to determine if deforestation is associated with malaria incidence in the municipality of Mâncio Lima, Acre State, Brazil. Cumulative percent deforestation was calculated for the spatial catchment area of each health district using 60 m x 60 m resolution classified imagery and statistical associations were identified with univariate and multivariate general additive models adjusted for spatial effects. We show malaria incidence is positively associated with greater changes in percent cumulative deforestation within the health districts. After adjusting for access to care, health district size and spatial trends, we show that a six percent, or one standard deviation, change in deforestation between 1997 and 2006 is associated with a 39 percent increase of malaria incidence.
63 Introduction Malaria risk in the Amazon and around the malaria belt is an integrated mix of environmental and sociodemographic risk factors (1, 2, 3). Despite over half a century of malaria control efforts from 1997 to 2006 there was on average 500,000 annual confirmed cases (4, 5). Meanwhile, the World Health Organization (WHO) estimates that about one in three malaria cases in Brazil are not reported, and cases in 2006 actually totaled 1.4 million, representing nearly half of all malaria cases in the Americas (6). The vast majority of malaria cases in Brazil are occurring in the Amazon Basin, where logging rates between 1999 and 2002 ranged from 12,000 to 20,000 square kilometers per year, the sum of which would cover the country of Denmark (8). The main vector of malaria in the Amazon, Anopheles darlingi, seeks out larval habitat in partially sunlit areas, with clear water, neutral pH, and aquatic plant growth and it is notably present and/or more abundant in altered landscapes (7, 9, 10). In Peru, the Vittor et al. study suggests A. darlingi is seldom observed in standing water bodies within undisturbed forests because they are shaded and more acidic, and that these forests remain abundant and rich in mosquito species that do not transmit malaria (9, 11). Human altered landscapes provide a milieu of suitable larval habitats for A. darlingi, including road ditches, dams, mining pits, culverts, vehicle ruts, and areas of poor clearing. The characteristics of A. darlingi's preferred habitat and studies of human and entomological malaria risk suggest deforestation and land clearing contribute to the dynamic malaria patterns along the frontier of settlement. Frontier malaria theory explains this pattern in
64 new settlements as an initial epidemic that abates to persistent low incidence and eventually eradication as the result of changing social, ecological, and environmental relationships (12). For instance, between 1985 to 1995, malaria risk in Rondônia increased during the initial colonization phase due to ecosystem transformations that promoted larval habitats and then gradually subsided as the urban landscape expanded, agriculture became established, settlers became more knowledgeable, access to health care increased, house construction improved, and suitable larval habitats declined, until finally malaria risk was mostly linked to human behavioral factors (13, 2). In the Peruvian Amazon, along the Iquitos-Nauta road corridor, other research showed entomological risk factors of mosquito biting rate and larval count increase with more deforestation. Adjusting for population density, sites with greater than 80 percent deforestation had a significant mean biting rate of 8.33 and sites with less than 30 percent deforestation had mean biting rate of 0.03 per night (11). Furthermore, the likelihood of finding of A. darlingi larvae doubled in breeding sites with less then 20 percent forest compared to sites with 20–60 percent forest and it jumped seven times when compared to sites with over 60 percent forest (9). In this study, we examine the association of deforestation and malaria at the level of health districts (localidade) using a uniform surveillance tool implemented in 2003 by the Brazilian Ministry of Health's Programa Nacional de Controle da Malária (PNCM). This nationally standardized system covers 5.1 million km2 of the malaria belt and reports monthly malaria statistics for over 7,000 health districts. The surveillance system employs a 40-item questionaire that includes items concerning patient demographics, diagnosis, and area of
65 residence (14). The spatial, temporal, and overall quality of this surveillance program combined with spatial mapping presents an opportunity to identify ecological risk factors within an extensive existing surveillance network. Our hypothesis is that deforestation is positively associated with higher malaria risk in health districts in Mâncio Lima, Acre State, Brazil (Fig 3.1, Fig 3.2). We also examine the associations of socioeconomic and demographic factors, including age, access to care, method of surveillance, sex, and malaria type.
Materials and Methods Study Area. Mâncio Lima (4,672 km2) is situated in Acre State and is the westernmost county in Brazil, sharing a boarder with Peru to the west and Amazonas state to the north. Between 2000 and 2008, the population of the county increased 30 percent from 11,095 to 14,387. The county has four percent more men than women and a mixture of rural (48 percent) and urban (52 percent) households (15). The 67 percent of the territory that is considered uninhabited is made up of the Nukini and Poyanawa Indigenous Reserves and a portion of the Serra do Divisor National Park. The rural economy is based on agriculture and manoic flour production, and there are no areas licensed for mining exploration in the county (16, 17, 18). Mâncio Lima has an average monthly precipitation range from four to 23 cm and an average monthly temperature range from 19 to 32 ˚C (19). The city of Mâncio Lima is the administrative and main population center, which is connected via highway to Curzeiro do Sul, 24 km to the east. In 2006 Cruzeiro do Sul and Mâncio Lima ranked second and fourth highest, respectively, for malaria risk and combined they reported 12.5 percent of all malaria cases in Brazil (20).
66 Health Data Since 2003 the Brazilian Ministry of Health has administered a uniform malaria surveillance program under the National Malaria Control Program (PNCM). All suspected malaria cases, collected from both passive and active surveillance, are slide confirmed and reported by local health districts (localidades), which are often points of care. For each case, the survey tool records date, age (less than 10 years), sex, whether or not care was received within 48 hours of symptom onset, malaria type (vivax or falciparum, and here we classified mixed infections as falciparum), and method of surveillance (passive or active). In addition, the malaria case report form includes voluntary questions on education level and occupation type (14). We filtered for patient cases that reported to be residents of the health district to which they presented and extracted monthly and annual percentages of these records from the Information System of Epidemiological Surveillance of Malaria (SIVEP-MALÁRIA) for the county of Mâncio Lima using Tableau 4.0 and Excel v11.3. Remote Sensing In 2006 health district boundaries in Mâncio Lima were initially sketched by health district field staff and then mapped in real time with a GPS Garmin 12XL. Then Track Maker 13.0 and ArcView 3.2. software was used to convert the paths into 54 health district polygons and the population of each health district was enumerated. Next, the geographical data of each health district was linked to the SIVEP-MALÂRIA data. The uninhabited portion of the county, including Indigenous Reserves and the National Park, was divided into three geographical areas and excluded from data analysis (21).
67 Classified deforestation estimates at 60 m x 60 m resolution from 1997, and 2000–2006 were downloaded from the Determining Deforestion in the Amazon Program (PRODES) in the National Geographic Research Institute Instituto (INPE) (22) (Figure 3.2). PRODES processes photographic images and Landsat TM imagery acquired at 30 m x 30 m resolution and is considered the gold standard reference for spatial deforestation data (23, 24). The classification of deforestation in PRODES is cumulative; once a unit is deforested it does not revert back to forest (25). Subsequently, in our analysis we do not consider the effects of regrowth. ArcMap v9.3 was used to calculate the spatial center point of each health district and the amount of deforestation observed in 1997 and 2000–2006. Analysis and Modeling Explanatory variables from the SIVEP-MALÂRIA database and deforestation data were mapped on the health district shape file. We converted the explanatory variables into z-scores, or units of standard deviation. The best estimator of deforestation was selected to minimize the AIC of a negative binomial generalized additive model that adjusted for spatial trends (26). The deforestation estimators examined included 1997 and 2000–2006 measures of absolute deforestation, cumulative percent deforestation, and percent deforestation change. In the same fashion, we individually examined the social and demographic malaria risk factors of access to care, surveillance type, sex, age, and malaria type. When we expanded to a multivariate model, we intentionally retained the most predictive ecological variable, percent deforestation change from 1997 to 2000, and retained percent of cases with access to care within 48 hours of symptom onset to adjust for at least one social demographic variable. Other variables and interactions were examined in a stepwise fashion based on minimization of AIC. We present a final model of
68 malaria incidence and percent deforestation adjusted for spatial trends, percent access to care, area of the health districts, and inter-health district variability. A p-value less than 0.05 was considered significant. Maps, statistical analysis, and figures were completed in R v2.9.2 and Adobe Illustrator v10.0.3 (27).
Results Fifty-four health districts occupy 1270 km2 (27 percent) of Mâncio Lima and provide free access to malaria diagnosis and treatment. The spatial layout of health districts reflect the settling of the population along two dominant river channels and in the urban zone around the city of Mâncio Lima. In 2006, the health districts report a total of 15,437 slide confirmed malaria cases, a mixture of both falciparum (41 percent) and vivax (59 percent) malaria. The majority of malaria patients across health districts are males (56 percent) above the age of ten (72 percent), identified by active surveillance (65 percent) and receive access to care within 48 hours of symptom onset (71 percent). The average incidence rate of the malaria epidemic is 1.16 cases/person, but within individual districts the incidence is 0.4–12 cases/person (Fig 3.1, Fig 3.3). Our statistical analysis of voluntary categorical questions on education level and patient case activities within the last two weeks was precluded due to insufficient response rates. Population distribution, access to care, and malaria incidence maps are shown in Figure 3.4. Baseline deforestation in 1997 concentrates within and near the city of Mâncio Lima, with varying degrees of deforestation present in the health districts along the riverways. The most deforestation change between 1997 and 2006 locates just west and south of the city. Over this period, percent deforestation in health districts increases on average 6.6 percent up to a
69 maximum of 26 percent between 1997 and 2006. The standard deviation of this increase is 5.9 percent. Notably, a large wetland area to the northeast of the city limits the amount of land clearing taking place in that area (Fig 3.4). The univariate analysis adjusts for variability between the health districts and spatial trend (Table 3.1). We show the influence of ecological deforestation and social demographic risk factors on malaria incidence. Percent deforestation 1997-–2000 is the most predictive of malaria risk in the health districts based on AIC. Health districts deforested 4.3% (one standard deviation) from 1997–2000 are associated with a 33 percent [95 percent confidence interval; 1.12–1.58] increase in malaria risk. Historic baseline deforestation in 1997 is not significant, but malaria risk and percent deforestation from 1997–2002, 1997–2001, and 1997–2000 is significant and positively correlated. More recent percent deforestation changes between 2001 and 2006 are not associated with malaria risk, along with 1997 and 2000–2006 measures of absolute deforestation and cumulative percent deforestation. On the edge of significance in the univariate analysis, the risk of malaria increases 27 percent [0.97–1.66] when active surveillance increases by 19 percent within a health district and 18 percent [0.87–1.59] when the percentage of cases obtaining care within the first 48 hours of symptoms improves by 14 percent. The spatial area of health districts is also nearly significant as the relative malaria risk is 1.20 [0.97–1.48] for a 32 km2 increase in health district size. Associations with malaria risk based on age, sex, malaria type, or the size of the health districts are not significant (Table 3.1).
70 The multivariate analysis shows a 4.2 percent increase in percent deforestation between 1997 and 2000 associates with a 48 percent [1.26–1.75] increase in malaria risk after adjusting for access to care and the spatial area of the health districts (Table 3.1). Figure 3.5 shows the interaction and joint relative risk of percent access to care and the spatial area on malaria incidence within each health district adjusted for percent deforestation 1997–2000. In Mâncio Lima, higher percent access to care decreases malaria risk for health districts under 23.4 km2 (mean value). The pattern of relative risk in health districts of larger size is less obvious, due to a shortage of observations.
71 Discussion We base our investigation of environmental and sociodemographic malaria risk factors on an existing surveillance system and estimate the relative risk of these factors at the resolution of health districts. Malaria surveillance in Brazil is unprecedented in scale and uniformity. Focusing on one county that has been linked to GIS data at the health district level, we report the characteristics of the health districts, map the distribution of risk factors, and find significant associations of deforestation and malaria incidence. Adjusting for population, access to care, and district size, we find malaria risk increases nearly 50 percent in health districts when four percent of the area is deforested during the first few years of the previous decade (1997–2000). Our approach shows relative associations of malaria incidence and deforestation patterns across space, rather than a trend of malaria incidence and deforestation across time. By limiting the analysis to 2006, we standardized for annual regional varability in temporal risk factors, such as climate and intervention measures. However, given the cross sectional design, the association of malaria incidence to prior deforestation does not necessarily imply a causal trajectory of increased deforestation and elevated malaria incidence. The study takes place during the peak of a malaria epidemic under a surveillance system that captures roughly 30 percent of all cases. During this epidemic, we find the univariate models predict higher malaria risk is associated with more active surveillance and access to care. This seems counterintuitive, as active surveillance generally identifies cases quickly leading to faster treatment and lower disease risk, and access to care is a reliably associated with lower disease rates. For example, in the Indian state of Assam, malaria incidence is consistently lower
72 in villages within 5 km of health care facilities (28). The univariate relative risks suggest more active surveillance and access during the epidemic identified cases that normally would have gone unreported. The significant interaction of health district size and access to care improved the performance of the multivariate model for percent deforestation 1997–2000, health district size, and access to care. The interaction and joint relative risk shows higher surveillance in health districts 23 km2 is protective against malaria risk, after adjusting for percent deforestation. Landscape establishes local ecology and biodiversity, and our results confirm an association of cleared land with higher malaria risk. This association is identified in previous research, but here we link the observation to an existing malaria surveillance program. Vittor et al. propose an entomological mechanism for this process by showing larval A. darlingi are less abundant in forested areas versus deforested areas (9, 11). Mosquito survival can depend on slight variations in temperatures, humidity, and sunlight as a result of deforestation (19, 30). In the Amazon, we find human malaria risk is significantly associated with the percentage change in deforestation of health districts in the past decade, and more specifically with deforestation five to ten years prior. There is not an association with deforestation later or earlier than that time frame. Vittor et al document significant increased odds of finding larvae in shrub versus forested land. Shrubbery develops five years post deforestation and becomes classified as secondary growth approximately 15 years post deforestation (Vittor et al 2009). Combined our results help resolve a landscape-based timeframe of risk. From an entomological perspective, the most suitable ecological niche for A. darlingi apparently occurs five to ten years following deforestation. Moreover, both Vittor and our observation agree with one premise of the frontier
73 malaria theory, which proposes land cleared prior to the last decade will have transitioned beyond the high malaria risk phase (12). The study is limited by several factors. The malaria data is based on annual percent measures derived from the Brazilian Ministry of Health malaria surveillance questionnaire. Each health district becomes the unit of analysis, so we are unable to adjust for risk factors at the individual case level. The data structure can be restrictive, age, for example is a continuous variable that is transformed into a binomial outcome, either above or below the age of ten. Insufficient reporting on voluntary portions of the survey, such as occupation, principal activity in the last two weeks, and education limit our ability to adjust for socioeconomic drivers. Moreover, we do not know the frequency of double reporting, but we have filtered the data for only those cases reporting to reside within the health district they present. In the absence of current immigration information, we cannot know whether the higher rate of cases associates with land cover change or is a result of recent migrants with naïve knowledge of malaria settling in the deforested health districts. We do know that each case reported living in the health district to which he/she presented. Several observations suggest that migration is an unlikely factor in explaining malaria patterning in this study site. In 2006 there were a total of 750,000 emigrants living in the Northern Region, which encompasses the states of Acre, Amazonas, Anapá, Roraima, Rondônia, Pará, and Tocantins. This represents just four percent of all emigrants to new regions within Brazil based on place of birth. More locally in 2000, Máncio Lima recorded an influx of just 29 emigrants age five and above since 1995 from
74 areas outside of Acre, or just 0.2 percent of all migration to Acre (IBGE 2000). These trends suggest a minimal amount of migration to Mâncio Lima occurred before 2006 (15). An emerging local aquaculture industry in Mâncio Lima sponsored by the Acre Office of Support for Micro and Small Business (Sebrae/AC) is an important concern that might also be correlated with the deforestation patterns in health districts. Vittor et al. show that ponds, wells or fish farms greater than 50 meters in circumference significantly increase the abundance of A. darlingi larvae (9). In 2003, 20 farmers with Sebrae extension support formed an aquaculture association of Mâncio Lima. The ponds ranged size from five to 175 ha and the first despesca, or harvest, of fish by draining the ponds, recovered 3,500 kg of fish valued at R$15,000. By May 2005 the second despesca yielded 5, 200 kg of fish and R$30,800, and by the third despeca in April 2006 the farmers produced 7,300 kg of fish at a R$56,400 profit. Most recently, the despeca in April 2007 jumped to 30,000 kg valued at R$ 216,000 (31). Taking the 2007 harvest at a yield obtained in the neighboring state of Amazônia of ~70 kg/ha/yr, suggests there was ~430 ha of aquaculture in Mâncio Lima (32). Fish farms are often located in degraded and deforested lands, yet this practice maybe leading to more mosquito larval habitat and higher malaria incidence. Further investigation is needed to differentiate deforestation from the effects of fish farming. Our models assume environmental exposures occur in the health district in which a case claims residency. If individuals sleep and work in different areas we cannot directly associate exposure with environmental variables within the health district. However, the diurnal biting pattern of A. darlingi, which generally peaks in the evening and sometimes in the early morning,
75 means most exposure will occur near the home (11, 33, 7). Additionally, we are not able to adjust for the presence or absence of the agent, plasmodium sporozoites, yet the county was saturated with malaria at the peak of an epidemic – nearly 2 cases/person – increasing the probability of widespread malaria exposure. In a scenario where A. darlingi mosquitoes are very abundant but the parasite is absent, once the malaria sporozoite is introduced, malaria should spread. We show the framework of health districts can link landscape and disease risk, but the overall generalizability of our findings is limited. In the Amazon, patterns of malaria and relative malaria risks are known to change from one community to the next. We find age and sex are not associated with malaria risk in Mâncio Lima. Duarte et al. finds men carry double the risk of women in some communities, and in others gold miners have three times the risk of urban residents (31). In yet another community Roper et al. finds no age-specific, occupational or gender risks, but activities such as strolling outdoors after six pm, waking before six am for adults, and children attending church services in the evening are significantly associated with malaria risk (35). Even though eradication of malaria is a reemerging priority of the global health community, there is no spatially standardized approach to monitor patterns of malaria at the clinic or treatment unit (36). At present, policy makers and epidemiologists continue to speculate about the regional and local variation of malaria and malaria risk factors. But policy makers also know that, "policies are sometimes applied more broadly than appropriate to large regions when it may actually only be relevant to a particular setting within the region…(and) policies often need to be
76 specific to be useful" (36). Currently, beyond the evidence presented in our study and others in Peru and Rondônia, the significance and geographical extent of the malaria incidence and deforestation process is unknown. In sum we show focused monitoring and high resolution spatial mapping of health districts can identify ecological associations between malaria incidence and deforestation. Other human health and ecology linkages may be discernable with similar high resolution and spatially explicit data.
77 References 1. Gurgel HdC. Paludisme et dynamiques environnementales dans l'Etat du Roraima au Brésil: Université Paris X Nanterre; 2006. 2. de Castro MC, Monte-Mór RL, Sawyer D, Singer BH. Malaria risk on the Amazon frontier. Proc Natl Acad Sci U S A. 2006;103(7):2452-7. 3. Packard RM. The making of a tropical disease: a short history of malaria. Baltimore, Md.: Johns Hopkins University Press; 2007. 4. Ministério da Saúde Brasil. Situação Epidemiológica da Malária no Brasil, Ano de 2007. Brasília: Ministério da Saúde; 2008. 5. Silveira AC, de Rezende DF. Avaliação da da estratégia global de controle integrado da malária no Brasil: Organização Pan-Americana da Saúde; 2001. 6. World Health Organization. World malaria report 2008. Geneva: World Health Organization; 2008. 7. Charlwood JD. Biological variation in Anopheles darlingi Root. Mem Inst Oswaldo Cruz. 1996;91(4):391-8. 8. Asner GP, Knapp DE, Broadbent EN, Oliveira PJC, Keller M, Silva JN. Selective logging in the Brazilian Amazon. Science. 2005 Oct 21;310(5747):480-2.
78 9. Vittor AY, Gilman R, Tielsch J, Glass G, Shields T, Pinedo-Cancino V, et al. Linking deforestation to malaria in the Amazon: characterization of the breeding habitat of the principle malaria vector, Anopheles darlingi. Am J Trop Med Hyg. 2009. 10. Tadei WP, Thatcher BD, Santos JM, Scarpassa VM, Rodrigues IB, Rafael MS. Ecologic observations on anopheline vectors of malaria in the Brazilian Amazon. Am J Trop Med Hyg. 1998 Aug;59(2):325-35. 11. Vittor AY, Gilman RH, Tielsch J, Glass G, Shields T, Lozano WS, et al. The effect of deforestation on the human-biting rate of Anopheles darlingi, the primary vector of falciparum malaria in the Peruvian Amazon. Am J Trop Med Hyg. 2006 JAN;74(1):3-11. 12. Sawyer D. Frontier malaria in the Amazon region of Brazil: types of malaria situations and some implications for control. Symposio Sobre a Malaria; 1988; Brazil; 1988. 13. Singer BH, De Castro MC. Agricultural colonization and malaria on the Amazon frontier. Population Health And Aging. New York: New York Acad Sciences; 2001. p. 184-222. 14. Ministério da Saúde Brasil, Secretaria de Vigilância em Saúde (MS/SVS). Programa Nacional de Prevenção e Controle da Maária PNCM/Ministério da Saúde, Secretaria de Vigilância em Saúde. Brasília: Ministério da Saúde; 2003. 15. IBGE (Fundação Instituto Brasileiro de Geografia e Estatística). Censo Demográfico: Brasil, 2000. Rio de Janerio: Departamento de População e Indicadores Sociais; 2000. 16. Toni F. Forest management in Brazil's Amazonian municipalities. In: Ferroukhi L, editor.
79 Municipal forest management in Latin America. San José, Costa Rica: Center for International Forestry Research (CIFOR) and the International Development Research Centre (IDRC); 2003. 17. Superintendência da Zona Franca de Manaus (SUFRAMA). Potentialities of the State of Acre. 2009 October 14, 2009 [cited; Available from: http://www.suframa.gov.br/publicacoes/potencialidades/ingles/Acre/acre.htm] 18. Barreto P, Souza CJ, Anderson A, Salomão R, Wiles J, Noguerón R. Human pressure on the Brazilian Amazon. State of the Amazon. 2005 May 2005:1-6. 19. IBGE (Fundação Instituto Brasileiro de Geografia e Estatística). Perfil dos municípios brasileiros: Pesquisa de informações básicas municipais 1999. Rio de Janerio: Departamento de População e Indicadores Sociais; 2001. 20. Ministério da Saúde Brasil, Secretaria de Vigilância em Saúde (MS/SVS). Relatórios técnicos do SIVEP-malária e SIH/SUS no perído janeiro-outubro e comparativo dos anos 2005 e 2006. Brasíl: Divisão do Pograma de Controle da Malária no Brasil; 2006. 21. Macário EM, Dimech GS, Araujo WN, Ladislau JLdB, Braz RM, Ramalho WM. Uso do Sistema de Informação Geográfica na Vigilância da Malária. I Encontro científico do programa de treinamento em epidemiologia aplicada aos serviços do SUS; 2007; Brasilia; 2007. 22. Instituto Nacional de Pesquisas Espaciais (INPE). Projeto PRODES: Monitoramento da Floresta Amazônica Brasileira por Satélite. Instituto Nacional de Pesquisas Espaciais (INPE), São José dos Campos, São Paulo, Brasil; 2009.
80 23. Asner GP, Broadbent EN, Oliveira PJC, Keller M, Knapp DE, Silva JNM. Condition and fate of logged forests in the Brazilian Amazon. Proc Natl Acad Sci U S A. 2006 Aug 22;103(34):12947-50. 24. Câmara G, de Morrisson Vaeriano D, Soares JV. Metodologia para o Cálculo da Taxa Anual de Desmatamento na Amazônia Legal. São José dos Campos: Instituto Nacional de Pesquisas Espaciais (INPE); 2006. 25. Motta Md, Cordeiro JPC, Valeriano DM. Using LEGAL - Map Algebra - as a tool to support estimation of Amazonian Deforestation. São Paulo, Brazil: Instituto Nacional de Pesquisas Espaciais (INPE); 2004. 26. Akaike H. A new look at the statistical model identification. IEEE Transactions on Automatic Control. 1974;19(6):716-23. 27. R: The R Foundation for Statistical Computing. Version 2.9.2. Copyright 2009. 28. Dev V, Bhattacharya PC, Talukdar R. Transmission of malaria and its control in the northeastern region of India. JAPI. 2003;51:1073-6. 29. Tuno N, Githeko AK, Nakayama T, Minakawa N, Takagi M, Yan GY. The association between the phytoplankton, Rhopalosolen species (Chlorophyta; Chlorophyceae), and Anopheles gambiae sensu lato (Diptera: Culicidae) larval abundance in western Kenya. Ecol Res. 2006 May;21(3):476-82. 30. Afrane YA, Lawson BW, Githeko AK, Yan GY. Effects of microclimatic changes caused by land use and land cover on duration of gonotrophic cycles of Anopheles gambiae (Diptera: culicidae) in western Kenya highlands. J Med Entomol. 2005 Nov;42(6):97480.
81 31. Duarte RBdA. Histórias de sucesso: agronegócios: aqüicultura e pesca. Brasília: Sebrae; 2007. 32. Federação das Indústrias do Estado do Amazonas (FIEAM). Piscicultura. 2009 October 13, 2009 [cited; Available from: http://www.fieam-amazonas.org.br 33. Tadei WP, Dutary Thatcher B. Malaria vectors in the Brazilian Amazon: anopheles of the subgenus Nyssorhynchus (1). Rev Inst Med Trop Sao Paulo. 2000 March-April;42(2):8794. 34. Duarte EC, Gyorkos TW, Pang L, Abrahamowicz M. Epidemiology of malaria in a hypoendemic Brazilian Amazon migrant population: a cohort study. Am J Trop Med Hyg. 2004;70(3):229-37. 35. Roper MH, Torres RSC, Goicochea CGC, Andersen EM, Guarda JSA, Calampa C, et al. The epidemiology of malaria in an epidemic area of the Peruvian Amazon. Am J Trop Med Hyg. 2000 Feb;62(2):247-56. 35. Roll Back Malaria Partnership (RBM). The global malaria action plan for a malaria-free world. Geneva: World Health Organization; 2008.
82 Table Table 3.1. Summary statistics of risk factors, and relative risks (RR) and 95 percent confidence intervals for the univariate and multivariate negative binomial generalized additive models with integrated smoothness estimation of spatial correlation. The standard deviation (SD) is used as the unit of analysis for all risk factors and the AIC values for the univariate and multivariate models are shown.
Univariate*
Multivariate*
Variable
Mean value
SD (unit)
RR
95% CI
AIC
Environmental % Deforested in 1997 % Deforest 1997–2006 % Deforest 1997–2002 % Deforest 1997–2001 % Deforest 1997–2000 % Deforest 2001–2006 Social demographic
40.2% 6.6% 3.2% 2.7% 2.3% 3.4%
32.5% 5.9% 4.3% 4.2% 4.3% 3.6%
0.84 1.28 1.13 1.32 1.33 1.03
0.64–1.09 1.07–1.52 1.11–1.56 1.11–1.57 1.12–1.58 0.85–1.23
671 667 664 664 664 673
64.9% 70.6% 27.6% 55.9% 41.3%
19.3% 13.5% 9.3% 7.8% 10.1%
1.27 1.18 1.18 1.07 1.11
0.97-1.65 0.87-1.59 0.94-1.46 0.88-1.31 0.87-1.41
670 671 670 672 669
23.4
32
1.20
0.97-1.48
% Active Surveillance % Access to care < 48 hrs % Cases < 10 yrs % Cases male % Falciparum cases Spatial Area (km2) Interaction (area X access to care) *Models adjusted for spatial trend
RR
95% CI
1.48
1.26–1.75
0.92
0.72–1.17
1.26
1.06–1.49
1.20
1.05–1.39
AIC
656
83 Figures Figure 3.1. Mâncio Lima is the western-most county in Brazil. The 2006 malaria incidence is mapped according to health districts (n=54).
84 Figure 3.2. Deforestation map of Mâncio Lima based on PRODES 60 m x 60 m classified satellite imagery. The health districts are outlined in black. Baseline deforestation that occurred in 1997 is orange, deforestation that occurred between 1997 and 2006 is light brown, nonforested land is blue, and forested land is green.
85 Figure 3.3. Box and whisker plots of slide confirmed malaria cases by health districts in Mâncio Lima from 2003–2008. Error bars indicate interquartile ranges, and thick horizontal bars indicate the median.
86 Figure 3.4. Cloropleths of selected malaria risk factors for health districts in Mâncio Lima. A) Resident population in health districts in 2006. B) Percent of slide confirmed malaria cases receiving access to care within the first 48 hours of symptom onset in 2006. C) The percent of 1997 deforestation in each of the health districts calculated from 60 m x 60 m resolution classified PRODES data. D) Cumulative percent change in deforestation by health district from 1997 to 2006. Uninhabited areas are excluded from the analysis (blue).
87 Figure 3.5. Joint relative risk plot of access to care and health district spatial area. Contour lines indicate the joint relative risk of standard deviation changes in percent access to care and health district spatial area. Open circles are the observed percent access to care and health district spatial area size data pairs for the 54 health districts. The contour line increment of relative risk is 0.2, increasing with the shading from red to white.
88 Acknowledgments
Disclaimers This study was supported by the National Aeronautics and Space Administration (NASA) LBAECO program.
Biographical sketch Ms. Olson is working toward a joint PhD from the University of Wisconsin-Madison. Her course of study combines a degree in population health from the School of Medicine and Population Health and a degree in environment and resources from the Nelson Institute. Her research addresses regional landscape and climate links in the ecology of vector-borne infectious diseases.
89
Chapter 4
Paper #3: Landscape Ecology influences Lyme Disease Vector Abundance in the MidAtlantic Region, USA
Sarah H. Olson, Jeffrey A. Cardille, Murray K. Clayton, Joseph E. Bunnell, Scott Heckle, Jonathan A. Patz. Submitted.
Abstract Landscape patterns have been linked to the ecology of Lyme disease at local scales but have not taken regional landscape characteristics into account. We now have augmented the predictability of Lyme disease risk by creating a regional model of the adult tick vector, Ixodes scapularis for Delaware and portions of New Jersey, Pennsylvania, Maryland, and Virginia. The model combines an extensive two-year tick survey with measures of landscape patterning taken from remotely sensed images. We calculated the composition and configuration of forest and deciduous land cover types for 65,000 1.1 km2 non-overlapping landscapes in the Mid-Atlantic region with METALAND software based on the 1992 National Land Cover Data. Next we consolidated tick abundance survey data into 175 sites, linking them to landscapes, and created a risk map by interpolating a log-linear regression across the region. Our results show, for the first time, that contiguous landscape patterning and intact deciduous forests can significantly reduce
90 the abundance of ticks across the Mid-Atlantic. In landscapes with deciduous forest, the number of adult ticks declines as forest patch size increases. These findings support the connection between disease risk and land cover patterns and provide further rationale for land conservation in deciduous forests as an important Lyme disease intervention measure.
91 Introduction The national annual incidence of Lyme disease increased nearly 200 percent between 1992 and 2005 and it remains the most common vector-borne disease in the US (CDC 2007). Since its original characterization in the mid 1960s, the geographic range of the disease progressively rippled away from its northeastern epicenter. These two characteristics, recent emergence and high incidence, make Lyme disease a useful model of zoonotic disease spread. Detailed indicators of landscape structure and arrangement to predict geographic distribution and seasonality of disease are an advantage of local studies of Lyme and other vector-borne diseases while broader regional studies, due to their scale, have not traditionally used indicators of landscape structure, instead they rely on coarser indicators of temperature, precipitation, and seasonality (NDVI) (Rogers and Randolph 2003). Few studies link measures of landscape structure and fragmentation, or landscape metrics, to vector-borne disease processes at regional geographic scales.
Dynamic linkages of environmental and ecological systems characterize the elements of Lyme disease transmission. The larvae of the primary vector, Ixodes scapularis, hatch in the summer and become reservoirs of the infectious agent, Borrelia burgdorferi, if they feed on an infected host. Larvae molt into nymphs and can transmit the disease during a second blood meal that is taken in the following spring. Adult ticks will have had an additional opportunity to acquire and transmit the spirochetes, but may be less likely to transmit the agent to humans than nymphs because of their larger size. Small mammal populations also alter Lyme disease risk. Small mammal reservoir competency, or the ability of a species to infect a biting larva, varies
92 from species to species. In the northeast, white-footed mice (Peromyscus leucopus), eastern chipmunks (Tamias striatus), short-tailed (Blarina brevicauda), and masked shrews (Sorex cinereus), are the reservoir for 80–90 percent of infected nymphs (Brisson et al 2008). Because the tick is a nonspecific feeder, diverse community assemblages apparently reduce disease spread by reducing the number of blood meals taken from more competent reservoirs. So in high biodiversity settings, the phenomenon, called the ‘dilution effect,’ decreases the likelihood that larval blood meals are taken from an animal with high reservoir competence, while increasing the likelihood that larval blood meals are taken from species with lower reservoir competence, such as raccoons or squirrels (Schmidt and Ostfeld 2001; LoGuidice et al 2003: Brisson et al 2008).
Research suggests the dilution effect is a localized ecological disease process related to forest fragmentation. In smaller forest patches the population density of white-footed mice, a highly competent reservoir of Lyme disease, nymphal infection prevalence (NIP), and nymph density increase exponentially (Nupp and Swihart 1998; Allan et al 2003). Landscape ecology, the study of landscape structure and function, provides tools to link small-scale ecological processes to larger spatial scales. The structure and function of land cover pattern and composition can be measured and compared across different spatial scales with landscape metrics, which quantify the amount and arrangement of landscape components. Translating information about landscape characteristics across scales is important because often decreased habitat patch size correlates with reduced species richness, but little is known about the
93 generalizeability of this process and its effect on the disease buffering ability of a regional ecosystem.
In addition to the dilution effect, other environmental and ecological processes may relate Lyme disease to forest fragmentation. The tick disease vector is linked to numerous environmental and ecological systems including soil characteristics, including soil order and texture, bedrock, land cover and forest type, sandy soil, and low elevation (Bunnell et al 2003; Jones and Kitron 2000; McCabe and Bunnell 2004). Deciduous forests have edaphic characteristics that support higher levels of ticks than evergreen forests (Guerra et al 2001; Bunnell et al 2003). In townships surrounding Lyme, Connecticut, landscape metrics of fragmentation are positively associated with tick density and NIP (Brownstein et al 2005). Smaller (less than 2.5 ha) deer exclosures had greater densities of ticks and within those exclosures, greater densities in the center (Perkins et al 2006). These varied findings suggest that there may be multiple forest fragmentation-linked ecological processes related to Lyme disease.
If landscape based indicators of vector-borne disease are to inform broader environmental health policies, they will require validation beyond the local level, or greater than ~20,000 km2, of analysis. For insight into the regional distribution of different landscape patterns associated with adult tick abundance, we use new landscape metric software to look closely at the structure and arrangement of forest landscapes. We test the hypothesis that landscape metrics quantify the abundance of adult ticks at this scale by surveying a small set of landscape metrics as indicators of habitat suitability for ticks. Our aim is to characterize the landscape composition and
94 configuration fingerprint that corresponds to the abundance of adult ticks across the Mid-Atlantic region.
Methods The tick sampling data provided by Bunnell et al. (2003) were collected between the months of September and December in 1997 and 1998. Briefly, the protocol consisted of 15 minute flagging bouts (quadrats, or transects) at non-random sites throughout the Mid-Atlantic state region that included state parks, forests, and large open areas. A total of 663 mature adult I. scapularis ticks were collected over six months and 320 transects.
The 1992 National Land Cover Data (NLCD) identifies 21 different land cover types at 30 m resolution within the study region (~37.2º to 40.5º N and 74.7º to 77.8º W), which covers approximately 82,500 km2 (Vogelmann et al 2001). METALAND (Cardille et al 2005) used the NLCD and FRAGSTATS 2.0 (McGarigal et al 2002) to calculate a suite of landscape metrics for 65,000 non-overlapping 1.1 km2 landscapes within the study area. One projection of the NLCD was used to obtain the 21 land cover type landscape metrics and the deciduous class metrics, and a second, forest vs. non-forest projection was used to obtain the forest vs. non-forest landscape metrics. Within the forest and non-forest classification, the forest class includes deciduous, evergreen, and mixed forest cover types.
METALAND matched tick-sampling sites to 1.1 km2 landscapes. Landscapes without forestland were excluded from regression analysis. Also excluded, based on the ‘rule of thumb’,
95 were landscapes with median or mean forest patch size greater than 77 hectares (ha), an area equivalent to a forest patch covering 67 percent of the landscape area. Ideally, the largest patch size should be two to five times smaller than the extent to avoid underestimating the size of patches given a fixed window size (O’Neill et al 1996). Sampling intensity was recorded as the number of transects within a landscape and incorporated as a control variable in all models.
Landscape metric variables related to hypothesized measures of tick habitat suitability were included in a standard statistical summary and the first modeling step. The hypothesized measures initially included in the model were mean and median patch size (ha), edge (m) and percent cover of forest and deciduous land cover. This limited set of metrics was selected based on ease of interpretation and recognition that landscape metrics of fragmentation are often highly correlated. To accommodate the numerous zero tick counts we selected a log-linear Poisson model. From these parameters stepwise regression reduced the number of parameters in the original model by removing parameters that did not significantly improve the description of tick count variability. Stepwise regressions allow parameters to enter or exit the model such that the Akaike Information Criterion (AIC) is minimized. This index limits model saturation by penalizing the log likelihood statistic as the degrees of freedom are reduced. An interaction term of mean forest patch and median forest patch that reduced the residual deviance and did not patently change any of the predictor coefficients was added to the final model. Either percent forest or deciduous cover was included in trial models to control confounding by forest amount on configuration (Turner et al 2001).
96 Spatial autocorrelation of residuals was assessed with variograms and all statistical analysis was performed using R software and packages (The R Foundation for Statistical Computing, Copyright 2005, Version 2.2.1). P-values in all models are adjusted for overdispersion in the dependent variable. The model was projected across the entire study region by assuming one survey occurred in each landscape, which is the median number of times the 175 landscapes were surveyed.
Results The 320 individual tick transects were located in a total of 201 1.1 km2 landscapes. Of the 201 landscapes, 19 did not contain forestland and seven landscapes with vary large forest patches broke the ‘rule of thumb’, reducing the sample size to 175. Histograms of the response and explanatory variables indicate non-normal distributions and confirm the choice of log-linear regression in the modeling analysis (Figure 4.1). The majority of landscapes sampled contained less than 50 percent deciduous forest and had median forest patch size less than 5 ha. The low median and mean forest patch areas are caused by large numbers of singlets (0.09 ha) or isolated pixels in the landscapes. Although the original tick survey did not employ random sampling techniques, the distributions of landscape metrics for the surveyed sites are very similar to the overall distributions of landscape metrics.
The final stepwise regression model had significant p-values for the sampling intensity, mean forest patch, and percent forest and deciduous cover (Table 4.1). Increasing mean or median patch size correlated with lower tick abundances, and percent forest and deciduous cover
97 had positive and negative relationships respectively with tick abundance. A 25 percent increase in percent deciduous forest would translate into a 35 percent reduction in ticks if all other parameters could be held constant. By calculating the 95 percent confidence intervals we rejected the null hypothesis that the coefficients for mean forest patch [0.88, 0.99], percent forest cover [1.01, 1.05], and percent deciduous cover [0.96, 0.99] are zero. Maintaining the same percentage of forest, the model predicted that within an approximately one kilometer square landscape, the odds of finding an adult tick is reduced seven percent for an increase of one hectare in the mean forest patch area. Figure 4.2 shows how patch sizes can vary even when percent forest is unchanged. The metrics of forest edge and deciduous forest edge were removed from the model by stepwise regression. Variograms of all model residuals did not show evidence of spatial autocorrelation. Observed versus the predicted tick counts (n=175) of the model have a correlation of 0.60 (Figure 4.3).
Within the Mid-Atlantic, estimates of tick abundance were calculated for 62,827 1.1 km2 landscapes in the study area. Estimates ranged between zero and 36 ticks. The adult tick map (Figure 4.4) predicts lower levels of ticks in the northwest, near South Mountain, which traverses the border of Pennsylvania and Maryland. The grey patches in this region, which follow the Blue Ridge Mountains, contained deciduous or forest patches greater than 77 hectares, and were excluded from predictions. Southeast New Jersey and the southern tip of Maryland have the highest predicted level of ticks within this study region. The model was used to estimate tick abundances for different landscapes in Figure 4.5.
98 Discussion We have demonstrated the utility of readily available measures of land cover pattern and composition to describe how landscape metrics relate to adult tick abundance across a large region endemic for Lyme disease. Our findings suggest reductions in forest patch size and coverage of deciduous forests are associated with elevated adult tick abundance across much of the Mid-Atlantic. The model, based solely on landscape metrics, explained over a third of the variability in adult tick abundance; if coupled to other known risk factors, such as NIP, deer population density, and human habitation, use of landscape metrics at a regional scale may enhance Lyme disease risk prediction.
METALAND generated the regional landscape metric distributions and facilitated the linkage of landscape metrics to regional adult tick abundance. In most studies landscape metrics are applied to specific field sites or local scales but are not standardized across large areas. That is, unlike previous analyses of landscape fragmentation, we were able to contextualize each of the survey sites to all landscapes within the region. Subsequently, we are able to rule out selection bias in our surveyed landscapes by comparing the distribution of the surveyed data against the complete set of landscapes. The extensive METALAND database of metrics for each landscape within the region greatly simplified the final step of projecting the model.
Notably, “pattern analysis techniques are most useful when applied and interpreted in the context of the organism(s) and ecological processes of interest, and at appropriate scales” (Gustafson 1998). In the case of Lyme disease, tick abundance and infection prevalence depend
99 of several ecological processes that appear linked to forest fragmentation. Here we describe how forest fragmentation in the Mid-Atlantic leads to greater abundance of adult ticks. This process may relate to habitat preference of the adult tick or habitat preferences of deer, which disperse adult ticks. Simultaneously, processes linked to small mammal assemblages may be ongoing and studies indicate higher densities of tick nymphs and larvae as well as elevated NIP are associated with fragmented deciduous forests (Horobik et al 2006; Brownstein et al 2005).
Deciduous forest cover is an essential component of Lyme disease ecology. Our regional model shows that controlling for forest fragmentation, expanded deciduous forest coverage decreases adult tick abundance. Using the same data set, Bunnell et al. found abundance increased for transects within zero to 30 meters of deciduous forest (2003). Other localized studies suggest different relationships between forest cover, vector populations, and risk factors. In Dutchess County, New York, nymph density and NIP is higher inside forests compared to forest edges and fields (Horobik et al 2006). However, in Maryland, the number of adult ticks found on deer is lower for deer captured inside forests than along forest edges (Das et al 2000).
Fragmentation-linked ecological processes related to Lyme disease may vary from place to place. Allan et al. show NIP increases with decreasing patch area but no relationship between patch size and density of larval ticks (2003). This suggests that small animal assemblages or the ‘dilution effect’ is the dominant fragmentation-linked ecological process at their study location. Other evidence indicates a variable impact of fragmentation-linked ecological processes on human incidence of Lyme disease. Lyme disease incidence was not correlated with fragmented
100 forests (measured by patch size and isolation) in an area surrounding Lyme, Connecticut, but fragmented forests (measured by percent herbaceous edge adjacent to forest) was positively correlated to human risk for a dozen counties in Maryland (Brownstein et al 2005; Jackson et al 2006).
It is important to note that as each study uses different spatial units of analysis, grain, and extent, even identical metrics such as forest edge or patch size may not be comparable from one study to the next (Wu 2004). Hence, comparisons between different metrics are troublesome and more so for comparisons across study designs. This may explain why forest edge and deciduous forest edge were less predictive in explaining tick abundance in our study design than in other study designs.
Although our study is able to track adult tick abundance, we are not able to relate landscape level processes to actual human health risk and outcomes, as nymphal infected prevalence is regarded as the main risk. Our aim was to determine regional patterns, thus we were restricted to a timeframe for which a regional high-resolution map could be aligned spatially and temporally with a tick sampling data set. A full exploration of causality between tick abundance and incidence of Lyme disease would necessitate a longitudinal study of human behaviors, identification of sites of tick acquisition and landscape change, and NIP.
Landscape ecology provides a conduit to extrapolate small-scale observations to landscape patterns of vector-borne disease risk elements. We show that forest habitat
101 preservation – displayed by landscape metrics – can be a useful and easily accessible indicator of adult tick abundance and, thereby, a useful tool for prevention.
Acknowledgements We thank the tick survey collection crews in 1997 and 1998.
102 References Allan B.F., Keesing F. and Ostfeld R.S. 2003. Effect of forest fragmentation on Lyme disease risk. Conservation Biology 17: 267-272. Brisson D., Dykhuizen D.E. and Ostfeld R.S. 2008. Conspicuous impacts of inconspicuous hosts on the Lyme disease epidemic. Proceedings Of The Royal Society B-Biological Sciences 275: 227-235. Brownstein J.S., Skelly D.K., Holford T.R. and Fish D. 2005. Forest fragmentation predicts local scale heterogeneity of Lyme disease risk. Oecologia 146: 469-475. Bunnell J.E., Price S.D., Das A., Shields T.M. and Glass G.E. 2003. Geographic Information Systems and spatial analysis of adult Ixodes scapularis (Acari: Ixodidae) in the Middle Atlantic region of the USA. Journal of Medical Entomology 40: 570-576. Cardille J., Turner M.G., Clayton M. and Price S. 2005. METALAND: Characterizing spatial patterns and statistical context of landscape metrics. Bioscience 55: 983-988. Center for Disease Control (2007). Lyme disease tables. Available from http://www.cdc.gov/ncidod/dvbid/Lyme/ld_statistics.htm (accessed August 2008) Das A., Lele S.R., Glass G.E., Shields T. and Patz J. 2002. Modelling a discrete spatial response using generalized linear mixed models: application to Lyme disease vectors. International Journal of Geographical Information Science 16: 151-166. Guerra M.A., Walker E.D. and Kitron U. 2001. Canine surveillance system for Lyme borreliosis in Wisconsin and northern Illinois: Geographic distribution and risk factor analysis. American Journal of Tropical Medicine and Hygiene 65: 546-552.
103 Gustafson E.J. 1998. Quantifying landscape spatial pattern: What is the state of the art? Ecosystems 1: 143-156. Horobik V., Keesing F. and Ostfeld R.S. 2006. Abundance and Borrelia burgdorferi-infection prevalence of nymphal Ixodes scapularis ticks along forest-field edges. EcoHealth 3: 262-268. Jackson L.E., Levine J.F. and Hilborn E.D. 2006. A comparison of analysis units for associating Lyme disease with forest-edge habitat. Community Ecology 7: 189-197. Jones C.J. and Kitron U.D. 2000. Populations of Ixodes scapularis (Acari: Ixodidae) are modulated by drought at a Lyme disease focus in Illinois. Journal of Medical Entomology 37: 408-15. LoGiudice K., Ostfeld R.S., Schmidt K.A. and Keesing F. 2003. The ecology of infectious disease: Effects of host diversity and community composition on Lyme disease risk. Proceedings of the National Academy of Sciences of the United States of America 100: 567-571. McCabe G.J. and Bunnell J.E. 2004. Precipitation and the occurrence of Lyme disease in the northeastern United States. Vector-Borne and Zoonotic Diseases 4: 143-148. McGarigal K., Cushman S.A., Neel M.C. and Ene E. 2002. FRAGSTATS: Spatial Pattern Analysis Program for Categorical Maps. University of Massachusetts, Amherst. Nupp T.E. and Swihart R.K. 1998. Effects of forest fragmentation on population attributes of white-footed mice and eastern chipmunks. Journal of Mammalogy 79: 1234-1243.
104 O'Neill R.V., Hunsaker C.T., Timmins S.P., Jackson B.L., Jones K.B., Riitters K.H. and Wickham J.D. 1996. Scale problems in reporting landscape pattern at the regional scale. Landscape Ecology 11: 169-180. Perkins S.E., Cattadori I.M., Tagliapietra V., Rizzoli A.P. and Hudson P.J. 2006. Localized deer absence leads to tick amplification. Ecology 87: 1981-1986. Rogers D.J. and Randolph S.E. 2003. Studying the global distribution of infectious diseases using GIS and RS. Nature Reviews Microbiology 1: 231-237. Schmidt K.A. and Ostfeld R.S. 2001. Biodiversity and the dilution effect in disease ecology. Ecology 82: 609-619. Turner M.G., Gardner R.H. and O'Neill R.V. 2001. Landscape ecology in theory and practice: pattern and process. Springer, New York. Vogelmann J.E., Howard S.M., Yang L.M., Larson C.R., Wylie B.K. and Van Driel N. 2001. Completion of the 1990s National Land Cover Data Set for the conterminous United States from the Landsat Thematic Mapper data and ancillary data sources. Photogrammetric Engineering and Remote Sensing 67: 650-662. Wu J.G. 2004. Effects of changing scale on landscape pattern analysis: scaling relations. Landscape Ecology 19: 125-138.
105 Table Table 4.1. Log-linear model of tick abundance. Variable
Estimate
p-value
Intercept
0.109
0.68
Sampling intensity
0.387
0.00
1.35–1.60
Mean forest patch size (ha)
-0.0665
0.04
0.88–0.99
Median forest patch size (ha)
-0.0683
0.24
0.83–1.04
Percent forest
0.0285
0.00
1.01–1.05
Percent deciduous forest
-0.0171
0.02
0.96–0.99
Mean X Median forest patch
0.0020
0.05
0.99–1.00
Null deviance
958.27
Residual deviance
629.84
95% CI___
106 Figure legends Figure 4.1. Explanatory and response variable histograms are displayed for 175 tick survey sites (a) and the approximately 62,827 landscapes (b) within the Mid-Atlantic region.
Figure 4.2. The example landscapes depicted here both contain 40 percent forest cover (green). Under the four-neighbor patch definition, (a) has a mean patch size of 1.4 ha and (b) has a mean patch size of 3.5 ha. According to the model and controlling for percent forest, (b) will have 13 percent reduced tick populations.
Figure 4.3. Plot of the observed and Poisson model predicted tick counts with a y=x plotted line. The correlation of the observed to predicted is 0.60.
Figure 4.4. This map estimates the relative abundance of the adult ticks that transmit Lyme disease over the region 37.2º N to 40.5º N and 74.7º W to 77.8º W. It is based on the interpolation of a log-linear Poisson model that uses four descriptors of landscape configuration. Colors correspond to a prediction of adult tick abundance.
Figure 4.5. Tick abundance is estimated using the log-linear Poisson model for ten landscapes from within the region. The density plot shows where these estimates are sampled from the regional distribution of tick abundance landscapes.
107 Figures Figure 4.1a.
108 Figure 4.1b.
109 Figure 4.2.
110 Figure 4.3.
111 Figure 4.4.
112 Figure 4.5.
113
Chapter 5
Conclusion
Overview and relevance of results
Omnia vivunt, omnia inter se conexa. Everything is alive, everything is interconnected. ~ Cicero
The science of vector-borne disease ecology has traditionally rested on disparate spatial platforms, the explicitly local and the equivocally global. From the time Ronald Ross discovered mosquitoes transmit malaria parasites in the late nineteenth century until Lysenko published the first global malaria endemicity maps in 1968, the field of vector-borne disease ecology focused on small landscapes (Lysenko 1968; Hay et al 2004; Boyd 1949). One classic example of local disease ecology in Trinidad, British West Indies, connected the ecological dots between an agricultural shift to cocoa plantations and a malaria outbreak. Cocoa plants require shade, so the plantations also cultivated tall Erythrina tree species. Bromeliads grew on these shade trees and their water reservoirs provided a breeding site for the efficient malaria-transmitting Anopheles bellator mosquito. This cascade of ecological relationships resulted in a malaria epidemic, which finally abated when the bromeliads were cut down from the shade trees (Downs et al 1946). This type of intense, ecologically rich, local research was largely abandoned when DDT
114 (dichlorodiphenyltrichloroethane, a synthetic pesticide) was introduced and appeared to be the panacea for global mosquito-borne infectious diseases. Since the DDT campaign waned the WHO Global Eradication of Malaria Program was declared dead in 1972 and up until more recent times, research and surveillance at the local ecological scale remained spotty. What expanded was more generalized surveillance and research at much larger scales that neglected environmental factors (Desowitz 1991; Ministério da Saúde Brasil 1966; Lysenko 1968; Hudson et al 1984). Examples of typical regional scale malaria surveillance in Brazil and French Guiana are shown in Figure 5.1 and 5.2 (Brasil Ministério da Saúde 1966; Juminer et al 1981).
Like malaria research, Lyme disease ecology has been approached at different scales. Since its discovery in 1975, scientists have identified both specific local environmental drivers as well as larger global-scale processes that define the geographic boundaries of Lyme disease transmission. Research has since attempted to understand the processes that link the multi-scale drivers.
Methods from landscape ecology and spatial epidemiology are central to this dissertation and critical elements of future vector-borne disease health policy. Together, they provide conceptual approaches to link local observations to regional patterns. Landscape ecology can show how the ecological composition (e.g. the type, amount, and arrangement of biotic and abiotic landscape elements) can define the geographical distribution and extent of disease transmission. Ostfeld and colleagues argue, “the impacts of landscape structure on epidemiological processes have so far been neglected,” and that spatial epidemiology, “has only
115 rarely been incorporated into disease studies” (2005). They conclude by suggesting, “that greater incorporation of explicit landscape approaches would improve our understanding and prediction of disease risk” (2005).
Choice of spatial scale in vector-borne infectious disease research affects the applicability of the study’s conclusions to health policy. On the one hand specific local observations are not generalizable to areas outside the study region, and on the other hand broad global parameters cannot be accurately down-scaled to local areas. This impasse is strongly evident in the recommendations from recent assessments of climate change impacts on human health. The human health chapter in the 2007 Intergovernmental Panel on Climate Change found a lack of, “region-specific projections of changes in exposures of importance to human health,” and also noted, “the difficulty of generalizing health outcomes from one setting to another, when many diseases (such as malaria) have important local transmission dynamics that cannot easily be represented in simple relationships” (Confalonieri et al 2007). Subsequently, in a roadmap for applied research on climate change and human health, there is a call, “for improved policyrelevant risk assessment, building a stronger bridge between evaluation of the existing health risks … in the context of other drivers” (Campbell-Lendrum 2009). The field of vector-borne disease ecology is facing the same challenge, namely finding the balance between local, regional, and global-scale research that will produce disease ecology information that is most relevant and significant for health policy.
116 This dissertation begins to bridge vector-borne disease risk with underlying environmental conditions by explaining how the environment affects malaria and Lyme disease transmission at policy-relevant scales. Advanced landscape imagery from remote sensing, expanded weather monitoring, and improved climate data were used to study the environmental drivers of vector-borne disease transmission. The next generation of spatial information and analytical tools that build on these technologies will continue to narrow the gap between local and global disease risk factors, define the limits of disease transmission, and provide a stage for cross-disciplinary analysis and ‘translational’ research (Kitron 2000; Rogers and Randolph 2003; Randolph 2009). My results scientifically document an understanding of vector-borne disease landscape patterns in three unique settings. Below I summarize the main conclusions and the contributions of each chapter to the field of vector-borne disease ecology.
•
Chapter two uncovers a seasonal relationship of malaria and precipitation for the Brazilian Amazon. We demonstrate how regional scale models based on high quality surveillance data help characterize vector-borne infectious disease risk. Our results show that the seasonal association between rainfall and malaria incidence is not globally uniform, and not even uniform within the Amazon region. This regional model is one of, if not the first, regional malaria climate model in Brazil. Moreover this model shows that malaria incidence peaks six months after the height of the rainy season in wetland areas, an important finding that will help guide intervention efforts. The study uses a novel statistical mapping approach to present the relationship of precipitation and malaria seasonality using risk ratios for each county. This study introduces a new regional
117 mechanism of malaria disease ecology; wetlands may bring about malaria seasonality across the Basin by limiting larval habitat during the flooded period following the rainy season. Such models will help improve predictions of climate and environmental change impacts on malaria in the Amazon.
•
Chapter three assesses the association between deforestation and malaria in the western Brazilian Amazon. We discover a strong association between landscape and vectorborne disease risk using high spatial resolution health data. At the spatial resolution of health clinics we discern an ecological mechanism between increasing recent deforestation and elevated human malaria incidence. This finding supports and quantitatively expands on prior research that proposed an association between deforestation and entomological risk. The results suggest high-resolution surveillance data can be used to estimate the relative contribution of risk factors and different intervention methods. This study also underscores data collected from an existing surveillance framework in combination with remote sensed imagery can be used to understand relationships between, and the joint influence of, both environmental and socio-demographic risk factors.
•
Chapter four explores the scalability of a vector-borne ecological disease process. We show the first application of a tool called METALAND that both measures and contextualizes landscape metrics, including those of fragmentation, for an entire region. Our study finds contiguous forest plots are associated with fewer adult ticks than more
118 fragmented forest plots in the Mid-Atlantic United States. This study emphasizes that landscape ecology can assist vector-borne disease ecologists who want to extrapolate local observations linked to landscape patterning to a larger region.
Study limitations, strengths and weaknesses Acausality, unmeasured socioeconomic drivers, and reporting bias are three caveats to the results presented in this dissertation. Proving causality in epidemiology is a high bar to achieve, requiring accumulating extensive evidence and/or in an ideal world, a randomized controlled trial. My analytical approach shows relative associations of vector-borne disease risk and environmental patterns of change across space, as opposed to temporal trends. The crosssectional basis of these models means one cannot impose a causal trajectory between a change in an environmental risk factor and an increase in vector-borne disease risk. Likewise, the crosssectional design means the association cannot be used to predict changes in disease risk for individual spatial units, as the models describe the sum of risk across every spatial unit. A subtle benefit of the approach is its implicit adjustment for annual variability of any temporal risk factors, such as climate, evenly distributed across the study region.
In complex disease systems, modelers need adequate control of differences over space and time in risk factors. An epidemiologist should adjust for all known risk factors, such as socioeconomic status, yet the spatial and temporal datasets may simply not exist. The task of identifying and modeling known risk factors for a global static state is problematic, and more so over space and time. The models that do forecast contemporaneous and future disease risk are
119 dependent upon the validity and reliability of spatial and temporal risk factors. The scale and inherent data limitations often mean appropriate adjustments for individual risk factors cannot be made and the resulting models are underspecified.
Limited data increases the difficulty of untangling the interwoven risk factors, especially in complex disease systems. Take for example our study of deforestation and malaria linkages in the Amazon. There are two issues to consider because we have omitted the risk factor of immigration of a susceptible population. First, if the omitted variable, such as migration, happened to be the real driver of disease risk, policies regulating deforestation would be ineffective. While we could not directly adjust for the influence of migration we were able to identify national and local trends that make us believe migration alone is unlikely to explain malaria variability at our study site. In 2006 there were 750,000 emigrants who were not born in the Northern region, living in the Northern Region, which encompasses the states of Acre, Amazonas, Anapá, Roraima, Rondônia, Pará, and Tocantins. This represents four percent of all emigrants to new regions within Brazil based on place of birth. More specifically within Máncio Lima, the 2000 census showed an influx of 29 migrants age five and above since 1995, representing 0.2 percent of all migration to the state of Acre (IBGE 2000). A related second issue is reverse causation. In the migration example, suppose biting mosquitoes displace the local population to an area of lower risk, leaving the new susceptible arrivals deforesting land along the edge of town. Without temporal and spatial information of migration, our study would falsely conclude deforestation causes malaria risk, when in fact malaria risk is causing deforestation (Pattanayak and Yasuoka 2007).
120 The overall influence of non-environmental variables is not trivial. WHO estimates 76 percent of global disease burden and 77 percent of global disease mortality is not attributable to environmental factors (Prüss-Ustün and Corvalán 2006). An omitted variable, such as migration, may lead to biased inferences but conversely, if the correlation holds over space and time with the changing environmental risk factor, an environmental risk factor may become an important proxy measure of a socioeconomic driver, just as recent satellite lights at night data were shown to correlate with economic activity and growth (Doll et al 2006; Henderson et al 2009). Regional modelers could use the correlation to their advantage. For example, if there is a true correlation between migration and deforestation, and migration is the main driver of the two, deforestation may be the easier variable to monitor over space and time.
Reporting bias is an important concern because of limited spatial socioeconomic data and because the observed cases may be a biased representation of human risk. Figure 5.3 shows an ideal progression of spatial disease knowledge, from environmental factors that define the vector and host distributions to areas of human risk, actual cases, and then finally the subset of cases that are reported (Ostfeld et al 2005). In practice, however, the location of infection may not correspond with the location of reported cases. This is a particularly important issue for Lyme disease research where the best health data available on exposure is often the home address of the case, which directly implies the environmental exposure associated with a case is the house and environs and excludes the possibility that infection may have occurred many miles away (Glass et al 1995). Studies can be biased if the environmental exposure and infection are not spatially matched or adjusted to disease reports.
121 Reflections and recommendations for future research Vector-borne infectious diseases are rooted in a landscape of complex social and ecological relationships. We are only beginning to quantify the consequences of land use and climate patterns on biodiversity, vector-borne diseases, and human health. Ahead there are many opportunities to improve understanding of vector-borne disease transmission towards identifying upstream, sustainable interventions. The spatial stage will continue to be a powerful tool central to assessments of global, regional, and local changes on human health.
•
Implement uniform and spatially explicit infectious disease surveillance. Consistent, uniform health reporting is more the exception then the rule, particularly in the global South. With respect to malaria, WHO recognizes, “the potential of existing health information is not being fully exploited,” because, “many organizations use different indicators and methodologies” (WHO 2009). The inadequate level of health information greatly limits the ability of the malaria eradication effort to monitor spatial patterns of incidence, socioeconomic and environmental drivers, and relationships of endemicity and sustained eradication. I show that environmental and social drivers are significant at different health reporting scales in Brazil. Spatial data at the very least needs to provide the geographic coordinates of the health center where a case report is filed. Disease relevant socioeconomic and environmental conditions should be recorded in a survey attached to the case report form. In addition to filling out case reports, the health center needs to be charged with enumerating the population it serves, or a surrogate measure thereof. Cost and resource barriers are surmountable with cell phones and Internet
122 technology, which are becoming globally ubiquitous (McElroy 2009). Health surveillance improvement should target priority countries with the greatest burden of vector borne disease. Timely health information can assist in rapidly assessing trends and relating socioeconomic and environmental variables to incidence data, a necessary step for developing and targeting appropriate interventions.
•
Incorporate comprehensive landscape approaches into studies of vector-borne disease. As demonstrated, spatial epidemiology is a powerful framework for cross-disciplinary analysis and translational research, especially at regional scales. At regional scales, epidemiologists find they must reconnoiter with the role of spatial environmental relationships in their models along with socioeconomic and human behavioral risk factors. The need for a comprehensive landscape approach is demonstrated by reviewing recent research on the impact of climate change on malaria. These studies have focused on rising temperature and malaria incidence to the exclusion of considering the roles and interactions of changing land use and land cover, precipitation regimes, and social and political environments (Rogers and Randolf 2006). Overall, more research is needed to characterize the full spectrum of disease transmission risk factors at an intervention deliverable scale.
•
Investigate non-linear effects of climate, environmental, and social change on vector borne diseases. Cross-disciplinary assessments are important to understand the synergistic effects, feedbacks, variability, and thresholds within vector borne disease
123 transmission systems. In the Amazon, coupled land-climate models show compelling feedbacks and synergistic effects of deforestation, precipitation, and stream flow, factors already associated with malaria transmission (Coe et al 2009). For malaria alone, there may also be synergistic effects of HIV infection rates, population demographics, drug resistance, and socioeconomic factors. And there may be more synergistic effects of variability and thresholds, the influence of which has mostly been viewed in terms of isolated variables. Examples of variable environmental factors important to malaria transmission include diurnal temperature fluctuations that are known to effect the extrinsic incubation period, and daily temporal and spatial hydrological variability that are necessary for village scale malaria forecasting (Paaijmans et al 2009, Bomblies et al 2009). A threshold is an important leverage point in disease transmission. The vaccination phenomenon of herd immunity is an example of an epidemiological threshold, but thresholds also exist in ecological and biogeochemical cycles. Similar to herd immunity, in the case of malaria, optimal percentages of insecticide-treated nets for different distribution strategies result in effective malaria control at less cost (Killeen et al 2007). Identification of ecological vector-borne disease transmission thresholds, such as critical levels or arrangement of land clearing in the Amazon, can likewise optimize population health measures.
•
Identify and address issues of ‘ideal-scale’ disease ecology. The advent of remote sensing technology has changed the way we look at the world, and opened up new ways to understand earth systems and ecological processes. These new resources are allowing
124 disease ecologists to take observations of local disease mechanisms to new and larger spatial and temporal scales (Herbreteau et al 2005). Globally there is an urgent demand for disease ecologists to not just identify larger-scale patterns of climate, environmental and social risk factors, but to identify and address the scales that are relevant to population health outcomes and policy, or ‘ideal-scales’. Pascual and Dobson point out that, “we should be cautious about the suggestion that appropriate larger scales will always resolve the problem of local variability and present strong linear associations…(because) public health measures might require predictions not only at national and regional scales, but also at a variety of smaller scales” (2005). Vector-borne disease ecologists need to research and assess how processes on the socioeconomic, physical, and ecological landscape are scalable – both in time and space – and relevant to the scale of public health policy and interventions.
Summary and policy synthesis The key findings of this thesis address the human health related knowledge gaps and policy needs raised by the 2005 UN Intergovernmental Panel on Climate Change (IPCC) and the 2007 UN Millennium Ecosystem Assessment (MEA) reports (Confalonieri et al 2007; Corvalan et al 2005). The IPCC report finds a lack on knowledge about what health outcomes are attributable to climate or climate change. I show malaria seasonality in the Amazon Basin is not driven by temperature but precipitation, a correlation that is further mitigated wetlands (Chapter 2). Relationships between ecological changes and human health need to be quantified according to the MEA report. I identify a variety of new ecological relationships between malaria,
125 wetlands (Chapter 2), and deforestation (chapter 3), as well as an association of forest fragmentation and adult tick abundance in the Mid-Atlantic USA (Chapter 4). Finally both the IPCC and MEA report have identified a gap in region-specific projections of changes in exposures of importance to human health, and knowledge about the extent of these patterns. I delineate a Amazon region specific malaria precipitation pattern that is linked to wetlands (Chapter 2), reproduce the deforestation and malaria link on a new expandable population health focused platform (Chapter 3), and regionalize the role of forest fragmentation in Lyme disease transmission (Chapter 4).
The implications of my research for climate change and conservation policy as well as decision makers are broad and narrow. Narrowly the precipitation and malaria story can help public health officials improve the timing of malaria information and intervention campaigns, and the deforestation and malaria story informs conservationists how to minimize malaria incidence, both with respect to the 5-10 year lag and the structure of deforested regrowth. The precipitation story is ready to inform policy and decision makers, while the deforestation and Lyme disease story require more documentation. At the broad scale, I document strong evidence of underlying ecosystem services that contribute to sustainable human health. This knowledge strengthens the connection of human health to a landscape undergoing physical and ecological changes.
126 References Bomblies A, Duchemin J-B, Eltahir EAB. A mechanistic approach for accurate simulation of village scale malaria transmission. Malar J. 2009;8(233):1-12. Boyd MF. Epidemiology of malaria: factors related to the intermediate host. Malariology; a comprehensive survey of all aspects of this group of diseases from a global standpoint. Philadelphia: Saunders; 1949. p. 551-607. Campbell-Lendrum D, Bertollini R, Neira M, Ebi K, McMichael A. Health and climate change: a roadmap for applied research. Lancet. 2009 May 16;373(9676):1663-5. Coe MT, Costa MH, Soares-Filho BS. The influence of historical and potential future deforestation on the stream flow of the Amazon River - Land surface processes and atmospheric feedbacks. Journal of Hydrology. 2009;369(1-2):165. Confalonieri UE. Human Health. In: Parry ML, Canziani OF, Palutikof JP, van der Linden PJ, Hanson CE, editors. Climate change 2007: Impacts, adaptation and vulnerability, Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge, UK: Cambridge University Press; 2007. p. 391-431. Corvalan CF, Woodward A. Chapter 15: Responses working group: consequences and options for human health. Millennium Ecosystem Assessment; 2005. Desowitz R. The malaria capers: more tales of parasites and people, research and reality. New York: W. W. Norton & Company, Inc.; 1991. Doll CNH, Muller J-P, Morley JG. Mapping regional economic activity from night-time light satellite imagery. Ecological Economics. 2006;57:75-92.
127 Downs WG, Pittendrigh CS. Bromeliad malaria in Trinidad, British West Indies. Am J Trop Med Hyg. 1946;26:47-66. Glass GE, Schwartz BS, Morgan JM, Johnson DT, Noy PM, Israel E. Environmental RiskFactors for Lyme-Disease Identified with Geographic Information-Systems. Am J Public Health. 1995 Jul;85(7):944-8. Hay SI, Guerra CA, Tatem AJ, Noor AM, Snow RW. The global distribution and population at risk of malaria: past, present, and future. Lancet Infect Dis. 2004 Jun;4(6):327-36. Henderson JV, Storeygard A, Weil DN. Measuring economic growth from outer space: The National Bureau of Economic Research; 2009. Herbreteau V, Salem G, Souris M, Hugot J-P, Gonzalez J-P. Sizing up human health through remote sensing: uses and misuses. Parassitologia. 2005;47:63-79. Hudson JE. Anopheles darlingi Root (Diptera: Culicidae) in the Suriname rain forest. Bulletin Entomological Research. 1984;74:129-42. IBGE (Fundação Instituto Brasileiro de Geografia e Estatística). Censo Demográfico: Brasil, 2000. Rio de Janerio: Departamento de População e Indicadores Sociais; 2000. Juminer B, Robin Y, Pajot FX, Eutrope R. [Malaria pattern in French Guyana (author's transl)]. Med Trop (Mars). 1981 Mar-Apr;41(2):135-46. Killeen GF, Smith TA, Ferguson HM, Mshinda H, Abdulla S, Lengeler C, et al. Preventing childhood malaria in Africa by protecting adults from mosquitoes with insecticide-treated nets. Plos Medicine. 2007 Jul;4(7):1246-58. Kitron U. Risk maps: Transmission and burden of vector borne diseases. Parasitol Today. 2000 Aug;16(8):324-5.
128 Lysenko AJ, Semashko IN. Moscow: Academy of Sciences; 1968. McElroy P. Zanzibar: beyond malaria control. President's Malaria Initiative. 11/04/09 http://www.fightingmalaria.gov/countries/profiles/zanzibar.html Ostfeld RS, Glass GE, Keesing F. Spatial epidemiology: an emerging (or re-emerging) discipline. Trends In Ecology & Evolution. 2005 Jun;20(6):328-36. Paaijmans KP, Read AF, Thomas MB. Understanding the link between malaria risk and climate. Proc Natl Acad Sci U S A. 2009 Aug 18;106(33):13844-9. Pascual M, Dobson A. Seasonal patterns of infectious diseases. Plos Medicine. 2005 Jan;2(1):1820. Pattanayak SK, Yasuoka J. Deforestation and malaria: revisiting the human ecology perspective. In: Colfer CJP, editor. People, Health, and Forests: A Global Interdisciplinary Overview: Earthscan; 2008. Prüss-Üstün A, Corvalan C. Preventing disease through healthy environments. Towards an estimate of the environmental burden of disease. Geneva, Switzerland: World Health Organization (WHO); 2006. Randolph SE. Perspectives on climate change impacts on infectious diseases. Ecology. 2009 Apr;90(4):927-31. Rogers DJ, Randolph SE. Studying the global distribution of infectious diseases using GIS and RS. Nature Reviews Microbiology. 2003 Dec;1(3):231-7. Rogers DJ, Randolph SE. Climate change and vector-borne diseases. Advances In Parasitology, Vol 62. San Diego: Elsevier Academic Press Inc; 2006. p. 345-81. WHO (World Health Organization). Surveillance, monitoring and evaluation. 2009. 11/04/09
129 http://apps.who.int/malaria/monitoringandevaluation.html
130 Figures Figure 5.1. Map of annual malaria incidence 1966 in Brazil (Brasil Ministério da Saúde 1966).
131 Figure 5.2. Map of malaria risk areas circa 1980s in French Guiana. ‘Zone de maintien,’ is the endemic area of transmission, ‘zone d’attaque,’ is the area of epidemics, and, ‘zone de consolidation,’ is the area of mixed transmission (Juminer et al 1981).
132 Figure 5.3. Conceptual model of spatial epidemiology in practice (Ostfeld et al 2005).
