Contact structure, mobility, environmental impact and behaviour: the importance of social forces to infectious disease dynamics and disease ecology

Abstract

Human factors, including contact structure, movement, impact on the environment and patterns of behaviour, can have significant influence on the emergence of novel infectious diseases and the transmission and amplification of established ones. As anthropogenic climate change alters natural systems and global economic forces drive land-use and land-cover change, it becomes increasingly important to understand both the ecological and social factors that impact infectious disease outcomes for human populations. While the field of disease ecology explicitly studies the ecological aspects of infectious disease transmission, the effects of the social context on zoonotic pathogen spillover and subsequent human-to-human transmission are comparatively neglected in the literature. The social sciences encompass a variety of disciplines and frameworks for understanding infectious diseases; however, here we focus on four primary areas of social systems that quantitatively and qualitatively contribute to infectious diseases as social–ecological systems. These areas are social mixing and structure, space and mobility, geography and environmental impact, and behaviour and behaviour change. Incorporation of these social factors requires empirical studies for parametrization, phenomena characterization and integrated theoretical modelling of social–ecological interactions. The social–ecological system that dictates infectious disease dynamics is a complex system rich in interacting variables with dynamically significant heterogeneous properties. Future discussions about infectious disease spillover and transmission in human populations need to address the social context that affects particular disease systems by identifying and measuring qualitatively important drivers.

This article is part of the themed issue ‘Opening the black box: re-examining the ecology and evolution of parasite transmission’.

1. Introduction

With over a million reported cases between 1982 and 1988, the Brazilian state of Rondônia became home to one of the largest outbreaks of malaria in human history [1]. Understanding this epidemic requires us first to answer the question of how a million landless poor people from the coastal cities of Brazil ended up in the Amazon rainforest, immunologically naive and actively, if inadvertently, constructing prime habitat for the malaria vector, Anopheles darlingi mosquitoes. The drivers behind the Rondônia malaria epidemic of the 1980s were complex and multifaceted and not limited to the ecological system of the mosquito vector and the landscape, nor to the transmission system of the parasite and human host. There were important interactions between these and other factors including the politics of the time, the economy of distant urban Brazil, and the standard practices by the migrant population of home and road construction, subsistence agriculture and deforestation. The complexities of the Rondônia epidemic represent the norm for real-world disease systems, not the exception. Disease ecologists should therefore be prepared to address and study the importance of such drivers and their interactions in order to better understand the spread of infectious disease.

The contemporary epidemiological environment is very different from the one in which epidemiology originally developed [2]. As many of the exposures are ubiquitous and pathways to pathology more diffuse, the epidemiologist's task has changed from simply mapping well-defined exposures onto specific disease outcomes, to one of understanding how contextual features—ostensibly external to the process of pathogenesis—create opportunities for the progression of disease [3]. McMichael [4] notes that the methodological individualism, the focus on the individual as the primary unit of analysis for social causation [5], which has characterized modern epidemiology, has led to less concern with social, structural and historical influences on health outcomes. McMichael's description highlights the structural tendencies of infectious disease in a changing field: ‘modern epidemiology is thus oriented to explaining and quantifying the bobbing of corks on the surface waters, while largely disregarding the stronger undercurrents that determine where, on average, the cluster of corks ends up along the shoreline of risk’ [4, p. 634]. Risk-factor studies, a mainstay of epidemiology research methods, compare the risk for disease between groups within the same community to identify associations with exposures that are potential causes of the disease. A major limitation of this approach is that by focusing only on community members, it fails to link community risk to the broader political and social context within which the community lies. Although research on the emergence and spread of infectious disease has traditionally focused on the biological mechanisms of transmission, and more recently the importance of the environmental context of the pathogen and reservoir, the social undercurrents that inform human exposure and transmission have been comparatively neglected.

Social scientists have previously suggested that greater attention be paid to contextual features of epidemiological systems [6,7]. For example, the field of social epidemiology has specifically studied the role of social systems in the health and spread of infectious disease in human populations [8,9]. We follow in this tradition, but focus on issues that have particular bearing on coupled human and natural systems (i.e. social–ecological systems), influence disease dynamics, and on which recent progress in new methods, instruments or computation has been made. In a framework that views the transmission patterns of infectious disease as a complex system, ignoring the social contributions to disease dynamics and the importance of heterogeneity jeopardizes the ability to fully understand the drivers of transmission, to make accurate predictions, and to develop effective prevention, control and eradication strategies [10]. Disease ecology and the social sciences provide new and increasingly powerful tools to better understand these dynamics and to study infectious disease as a system, including the multifaceted and heterogeneous dynamics introduced by the social context. In this review, we limit ourselves to infectious diseases of humans, be they parasitic, bacterial or viral, directly or indirectly transmitted.

We lay out broad themes from the social sciences that have particular bearing on infectious disease dynamics to lend credence to the importance of social systems and the social sciences that study them. Two key contributions from the social sciences to disease ecology that we will underscore here are: (i) the tenant that human behaviour, decisions and patterns are key components of complex infectious disease systems and central to disease prediction and control, and (ii) the social sciences have methods and tools for measuring social heterogeneity, a fundamental driver of the spread of infectious disease.

2. Infectious diseases are complex systems

The components, agents and drivers involved in how infectious diseases spread through a population constitute a complex system, defined by coupled ecological, transmission and social dynamics. A complex system, or complex, adaptive system, is a system made up of strongly interconnected elements where the dynamics of these elements are frequently nonlinear and the connections between elements are themselves nonlinear [11]. The system dynamics are not understandable by simply summing the effects from all the components; the interactions are dynamically important. Features of complex systems can include feedback loops, a strong sensitivity to initial conditions, multiple equilibria or basins of attraction, and the propensity for oscillatory and sometimes chaotic behaviour. An important consequence of such nonlinear interactions is that complex systems can be difficult to predict and control. Even when the behaviour of the system over short timescales appears quite regular, long-term prediction can still be difficult. This is why iterative, updating processes, such as those used by adaptive management, have been developed to navigate uncertainty in long-term system dynamics [12,13]. The ability to predict system outcomes is predicated on detailed knowledge of the overall system. Thus, failure to measure and integrate information from the ecological, social and transmission elements of an infectious disease, depending on the disease and its context, may lead to highly inaccurate and potentially misleading predictions, even in the short term. We shall argue that integrating these three, frequently separated, components of the larger infectious disease system is essential for gaining purchase on the generalization of phenomena, preventing and mitigating emerging infectious diseases, and eradicating and controlling existing ones.

The component systems of infectious diseases—social, ecological and transmission—all contain substantial nonlinearities. The epidemiological formalism of compartmental models has a fundamental nonlinearity at its core, namely the multiplicative interaction between the infectious and susceptible fractions of the population [14,15]. The other systems also potentially contain nonlinearities in the form of competition, density- or frequency-dependent migration [16], social interaction [17] or economic growth. Bonds and colleagues (e.g. [18–20]) present the integration of systems leading to economic growth as a promising framework for understanding infectious disease dynamics as components of broader complex systems. Furthermore, following on from the econometric argument of Sachs & Malaney [21], Bonds et al. [18] show how the feedback between a simple coupled model of an infectious disease and economic growth can produce poverty traps, vicious cycles of positive feedback in which infectious disease and poverty mutually reinforce each other. This result and the more general results of Ngonghala et al. [20] are early examples of complex-systems models that could inform the next generation of strategies for prediction, control and eradication of infectious disease.

Complexity science is concerned with interactions of elements of a system that are qualitatively important to the system dynamics [11], meaning they have structural or phenomenological impact on key components, variables or outputs. The focus then is on identification of key, qualitatively important variables and the characterization of dynamic phenomena, rather than incorporating or modelling all components that may influence the quantitative ends. Understanding which variables and elements of the system can be represented by homogeneous parameters and which may contain heterogeneities that, when incorporated, qualitatively influence the system is important for selecting variables for inclusion in mathematical models and for empirically quantifying them.

3. Heterogeneity drives epidemics

Since the pioneering work on HIV transmission by Anderson et al. [22], we have known that behavioural heterogeneity can drive both the timing and size of epidemics as well as determine differences in risk for different subpopulations. For example, in transmission-dynamics models for HIV that are structured by people's degree of sexual activity, the basic reproduction number (R₀) increases proportionally with the variance in sexual contacts. Therefore, epidemics of sexually transmitted infections in communities with heterogeneous sexual contact behaviours have lower epidemic thresholds, lower endemic equilibria and higher critical vaccination thresholds than those among communities where such risk behaviours do not differ across the population. This insight into the importance of heterogeneity has been generalized to other disease systems, such as those for vector-borne pathogens [23–25].

Heterogeneity in transmissibility has received substantial attention since the SARS outbreak of 2002–2003 when certain infected individuals caused a disproportionately high number of secondary infections [26,27]. Greater than 70% of the infections in both the Hong Kong and Singapore outbreaks were due to ‘super-spreading events,’ and this highly skewed distribution was of fundamental importance to the dynamics of SARS outbreaks [26,28]. The super-spreader concept can, in part, be explained by pathogenicity as these super-spreaders may shed more virus than other carriers, but social factors can also be important. Within any community exists a wide distribution of contact events driven by a variety of individual motivations and preferences, social structures and geographical features [29–33]. Social heterogeneity can also be important in the distribution of individual contacts [29], the types and rates of human mobility through the environment [34], the behaviours affecting exposure to disease [6], and inequalities in access to health services or materials [20]. Understanding the range of individual and collective structure, movement, environmental impact and behaviour provides important insights into the dynamics of the disease system as a whole. The two themes of complexity and social heterogeneity as discussed above serve as the lens through which we view the contributions of the social system and social science research methods on the understanding of infectious disease dynamics.

4. Social forces impact infectious disease dynamics and control

At the most fundamental level, every disease transmission event, whether from animal to human, human to human or environment to human, has an ecological context and a social driver that leads to contact with the pathogen or its host. We highlight four areas of the social sciences that underscore the significance of the social system to disease emergence, propagation and dynamics, and demonstrate that the integration of the social system with the transmission and ecological systems can bring about a deeper, more holistic understanding of disease dynamics. These four areas are: (a) social mixing and structure, (b) space and mobility, (c) geography and environmental impact, and (d) behaviour and behaviour change. These areas do not present a comprehensive review of social factors associated with infectious disease, rather they provide exemplars of the role of social sciences in disease ecology and their mathematical importance.

(a). Social mixing and structure

The structure of social relationships, who is in contact with whom, and how and at what rate they come into contact, or mix, can significantly influence disease dynamics and equilibria [14,35]. The classical ordinary differential equation-based compartmental models of infectious disease dynamics assume that the population is ‘well-mixed,’ that is, that the risk structure can be approximated by a bilinear mass-action term. This assumption is sometimes appropriate, particularly for infections transmitted by casual contact like influenza [36]. However, starting in the 1980s, work by epidemiologists such as Anderson & May (e.g. [14,35]) has shown that the assumption of mass-action may not be warranted in all cases and that various forms of heterogeneous mixing often fit empirical data better. This is because human populations are not homogeneously mixed; instead, they typically exhibit community structure and assortativity of interactions. People tend to concentrate their contacts within smaller clusters embedded within larger networks and have preferential contact with people with particular demographic characteristics. For example, molecular epidemiological investigation has revealed significant ethnic clustering of tuberculosis (TB) infection in heterogeneous urban populations [37,38], suggesting that ethnic clustering may be driving TB transmission dynamics [39,40].

Graph-theoretic models of disease contact networks take the idea of heterogeneous mixing to its logical extent. These network models have made progress in representing realistic social interactions and their consequences for infectious disease transmission by embedding structure into the representation of the community and relaxing homogeneous mixing assumptions [41]. Social network analysis has given us tools for studying contact structure and its influence on infectious disease dynamics with important implications for control and eradication. Some examples we will discuss here are modelling heterogeneous contact degree distributions, targeted vaccination strategies of central individuals and techniques for empirical data augmentation.

Heterogeneous structural network features play a key role in epidemic outcomes and can be leveraged to improve infection control. For example, degree-based vaccination, the strategy of vaccinating the individuals with the most contacts, has been shown mathematically to be an effective intervention technique [42]. If the structure of the network is unknown, however, vaccinating a random contact of a randomly chosen node in a network (i.e. acquaintance immunization) is more likely to remove the higher-degree nodes than simple random vaccination [43]. This is because a random contact of an individual is, on average, more central to the network than the individual selected [44], a structural manifestation of individual heterogeneity in the propensity to form contacts [45]. If individuals tend to concentrate most of their contacts within communities embedded within a larger network, Salathé & Jones [46] show that preferentially vaccinating individuals with high betweenness centrality, a measure of how many network paths pass through an individual, reduces the final size of epidemics on networks more than a variety of alternative strategies, including degree-based vaccination. These conclusions have important implications for control, especially when vaccine supply is limited or rapid action is necessary.

Vital processes associated with infectious disease can alter heterogeneity and strategies for control. For example, death can remove individuals in a structurally important position in the network of disease dissemination, and death is not a random process. Frequently, the most at-risk and connected individuals in a population will be the first to exit the susceptible population either through death or acquired immunity, particularly for emerging infectious diseases. This non-random removal of high-risk individuals is known as ‘network frailty’ [47]. For example, community randomized sexually transmitted infection (STI) treatment trials in the 1990s were conducted in three East African localities: Mwanza District, Tanzania and Rakai and Masaka, Uganda. While improved STI control substantially reduced HIV-1 incidence in Mwanza, it had no effect in either Rakai or Masaka. Detailed microsimulation models suggested that contact network frailty due to the mature epidemic in the two Ugandan localities, and in contrast with the more recent Mwanza epidemic, diluted the effectiveness of STI treatment on HIV-1 incidence [48].

Direct incorporation of local community structure in the estimation of key parameters in outbreak settings has been made possible through approaches such as Markov Chain Monte Carlo (MCMC) simulations. For example, Cauchemez et al. [49] used detailed data on social networks within classrooms and households to estimate the relative contributions of within-class, within-grade and within-school transmission during an outbreak of pandemic H1N1 influenza in Pennsylvania. By using data augmentation approaches, the authors were able to incorporate uncertainty in unobserved events (e.g. infection dates, which are nearly always unobserved) in parameter estimates. Similar approaches were used to understand the relative contributions of community-based, healthcare-based and funeral-based transmissions during the recent Ebola outbreak in West Africa [50]. The incorporation of local community structure into outbreak models has also been used to identify highly localized transmission during a chikungunya outbreak in Bangladesh with significant increased risk of infection in women, perhaps due to greater time spent at home [51]. Such inferential approaches also allow the quantification of the impact of interventions on transmission, providing a key measure of efficacy for public health officials [49,50].

(b). Space and mobility

Properties of structure and mixing may be tied to the location in which individuals live, work or conduct a variety of social activities at a variety of scales—from small-scale, local space and mobility to large-scale patterns of place and migration. Embedded in the locations of individuals are patterns of spatial clustering that can facilitate or inhibit transmission. For example, Arthur & Diamond [52] suggest that subsistence activities led to dispersed Navajo settlement patterns which may explain why Navajos suffered fewer losses to smallpox than did the nearby Hopi, who lived in more spatially dense communities. While households may be distinct or separate, within each household there are tightly knit clusters of people, and this has been shown to be important to large-scale spatial epidemic models and to vaccination and control strategies [53,54]. Geography can also be important for how epidemics ‘travel’ over space and through time and may create cycles of alternating amplification and dissemination [55,56].

Individuals may carry infectious diseases as they move across a landscape, including various types of movement, from long-range, annual migration to fine-scale mobility. Indeed, studies of migration—and mobility more generally—are key to understanding risk of infection and the dissemination of epidemics [35,57]. Ferrari et al. [58,59] present evidence that irregular measles periodicity in Niger—in stark contrast with the regular periodicity in historical Europe [55]—is driven by the yearly rural–urban circular migration related to the agricultural cycle. On a smaller scale, Stoddard et al. [57] estimate a value of R₀ for dengue in Iquitos, Peru nearly three times greater when the fine-scale movement of people is considered than when vector exposure is assumed to only occur at households. A key barrier to better understanding the role of human mobility in disease ecology has been the lack of data that describe movement, except in these detailed, and often small-scale, observational studies. However, increasing research access to call detail records, which contain the location of the cellphone tower used in each cell phone call, have provided a new method for quantification of population flows within a country [60].

Human mobility, as inferred from cell phone usage data, has explained patterns of malaria and rubella disease in Kenya, as well as dengue in Pakistan [61–63]. In addition to call detail records, which sometimes do not capture the most remote and poorest individuals, new analytical methods using high-resolution satellite imagery have captured density and movement of human populations [64]. Night-time satellite imagery capturing emitted light has revealed seasonal migration of individuals, and models with these data incorporated confirm the epidemiological importance of economically driven human movement [65]. On a smaller scale, geographical positioning trackers that document the movement of individuals, such as in villages or at schools, were able to reveal transmission patterns for influenza [36]. Using detailed household survey and real-time PCR testing, Padmanabha et al. [33] show that micro-bursts of dengue infection in Colombia are driven by movements that lead to frequent introductions into heterogeneous urban populations. These new methods for approximating density and movement of humans across space can be applied to monitoring risk of disease and informing effective strategies for targeting disease prevention initiatives [66].

(c). Geography and environmental impact

While methods for capturing the fine-scale movement of individuals across space remains a new frontier of research, large-scale changes to landscapes and anthropogenic environmental impact have long been recognized as a driver of emerging infections [67–70]. Changes to the environment can often bring people into contact with novel pathogens by increasing the overlap of human and animal ranges through the human–wildlife interface [71]. The archetypal example of this contact and transmission process is the killing and butchering of chimpanzees for bushmeat, which is thought to have led to the spillover of HIV-1 into human populations [72]. Further examples of this interaction include the emergence of SARS, which resulted from wild animal markets in Guangdong Province, China [73], and the transmission of Nipah virus from bats to people in Bangladesh. Nipah transmission is primarily driven by human harvesting and consumption of date palm sap, an agricultural product that would not be available to bats without human-induced changes to the landscape [74,75].

These interactions between people and wildlife have led to more general assertions about the effects of biodiversity on exposure to infectious diseases. The dilution-effect hypothesis attributes increased infection risk to decreases in biodiversity, which implies that conservation of natural landscapes that serve as habitats to endangered species may reduce infectious disease risk to humans [76]. However, evidence also supports the amplification effect, which points to tropical regions as the source of a disproportionate number of infectious diseases due to their biodiversity, presence of viable vectors and a number of species known to be reservoirs of pathogens that can infect humans [77–80]. While these two perspectives both have supportive evidence, when and why different levels of biodiversity, host composition or changes to landscape depress or attenuate disease is probably linked to human interaction with different landscape types that are in different stages of transition. Therefore, the history and nuance of human land use in an area may have more impact on whether or not decreasing biodiversity results in infectious disease emergence.

Developing countries in tropical regions are places of rapid human population growth [81,82], frequently accompanied by urbanization, development and economically driven industries and activities. These processes frequently change accessibility of localities and populations, increasing mixing between previously disconnected groups and affecting the viable transmission of infectious agents. Ongoing trends in rural–urban migration and the proliferation of cheap means of increasing mobility, as with the availability of motorcycles [83], suggest that contact heterogeneity at the landscape level is likely to intensify even as people make further inroads into formerly wild or sparsely populated spaces such as tropical forests. Markets may incentivize bushmeat and wildlife trade in these areas as well, which can increase people's contact with potential reservoirs [56,84,85] and promote conditions favourable to infectious disease transmission [86]. Accessibility, mobility and connection between disparate communities can be affected by new development projects. The construction of a new road in rural Ecuador, for example, increased the incidence of enteric infections, and villages with greater access to the new road were eight times more likely to become infected [87]. Urbanization has also been cited as a key contributor to the rapid expansion of dengue virus across global tropical and subtropical regions over the past 40 years [88]. The influx of human populations into urban centres has brought people in increasing contact with Aedes aegypti mosquitoes, which have become fully adapted to urban environments.

(d). Behaviour and behaviour change

Individual- and community-level behaviours impact the contact rate and the contact network for susceptible and infectious individuals, the transmission probability for each type of contact, and the duration of time an infectious person interacts with other members of the community before becoming isolated and removed from circulation. While some studies have examined the ability of mathematical models to characterize behaviour and its role in shaping disease dynamics (see Funk et al. [89] for a review), disease modellers typically neglect behavioural mechanisms in their models [90] due to difficulties of tractability, measurement or expertise. Consider the Ebola virus disease (EVD) epidemic of West Africa in 2014–2016. In September 2014, the Centers for Disease Control and Prevention (CDC) forecasted 1.4 million cases of EVD in Liberia and Sierra Leone by 20 January 2015 [91]. In fact, by 20 January 2015, there were fewer than 20 000 confirmed and suspected cases in Liberia and Sierra Leone [92]. Importantly, the forecast assumed that no changes in individual behaviours and community health decisions would take place. Hewlett & Amola [93] present evidence that Ebola-affected communities, in fact, typically adopt protective measures that change the outcome of epidemics. This disconnect between the epidemic modelling and the reality of how humans and their institutions react to a changing infectious disease epidemic illustrates the importance of incorporating behavioural research into epidemic forecasting efforts.

As individuals and communities assess and reassess how their behaviours affect their risk of exposure to a disease, behaviours are subject to modification. Assuming that individuals act in their own self-interest, behaviour change occurs when the health consequences of risky behaviour become greater than the benefits and reverses when the change in behaviour is believed to no longer be necessary [94]. For example, demand for condoms in the USA was responsive to local prevalence of HIV/AIDS in the 1980s and early 1990s [95], indicating that a perception of personal risk was elevated in accord with the spread of the disease and led to a specific protective behaviour. Similarly, the age of first measles vaccination in the 1980s in the United States was negatively associated with measles prevalence [96]—parents were less likely to vaccinate their children on time when they perceived the risk of measles to be low. Thus, behaviour, rather than a static condition of the social context, is dynamic and responsive to change.

Changing behaviour as an externally driven intervention for epidemic control is best achieved through cooperation between public health and the leaders and members of affected communities and respect for cultural norms and perceptions. During the Ebola outbreak in West Africa, early interventions to interrupt transmission targeted traditional burial practices often without meaningful dialogue with communities to problem-solve and build trust [97]. The result of this approach was an increase in infectious disease transmission when community members rejected the intervention and shunned treatment centres. Alternative strategies should instead utilize social science and anthropological methods that explicate community behaviours and address the community's understanding of the cause of transmission. To prevent transmission of Nipah virus in Bangladesh, for example, investigators first sought to understand existing community practices around date palm sap collection and discovered some residents already used cheap and effective bamboo skirts to prevent bats from contaminating the date palm sap, which increased the value of the agricultural product [75]. Subsequent interventions could then be designed that leveraged or took into account local perceptions and existing practices [98]. Social science methods have also been used to better understand behaviours that may drive super-spreader events of Nipah virus [99] and the cultural and logistical constraints around changing patient–caregiver interactions to prevent person-to-person spread [100]. Response to outbreaks of Nipah virus in Bangladesh routinely involve social scientists and anthropologists to help integrate biomedical models of infection transmission with local understanding to increase the efficacy of public health intervention messages [101].

5. Conclusion

This review has focused on two broad points: (i) infectious diseases are embedded within complex systems composed of interdependent ecological, transmission and social components, and (ii) social heterogeneity drives epidemics, the measurement of which is a fundamental contribution of the social sciences. We suggest that infectious diseases can usefully be seen in the framework of complex systems science and, following the weight of evidence from theoretical epidemiology, we argue that heterogeneity of the social context significantly impacts infectious disease dynamics. The identification and measurement of social heterogeneity (e.g. of behaviour, mobility, structure, density, etc.) is really the substance of the empirical social sciences. Harnessing the methodological tools, theoretical perspectives and general knowledge base of the social sciences will greatly increase the sophistication—and predictive power—of models of disease emergence, control and eradication [102]. Equally important from the perspective of collaborative buy-in from social scientists, the fact that these topics are essential to understanding disease emergence incentivizes professional collaborations, although interdisciplinary and multidisciplinary funding opportunities remain limited.

Disease emergence and the spread of epidemics rely on the behaviours of individual people, the social structures in which they are embedded and the political economy that makes particular social outcomes more likely for some people than for others. We have chosen four particular areas of study for this paper—structure and mixing, space and mobility, geography and environmental impact, and behaviour and behaviour change. We acknowledge there are many other social science topics in public health, psychology, economics and epidemiology that are important for understanding the dynamics and control of infectious disease. Social inequality [6,103], psychosocial stress and social support [104], to name just a few, all have known importance to community health and infectious disease. However, the subjects we review provide examples where social questions can be and have been incorporated into transmission-dynamics models to make novel predictions or help with the interpretation of complex model outputs.

The overarching themes of this paper and many of the topics we discuss are exemplified in our original case example of malaria in Rondônia. Government policies incentivized migration to the Amazon frontier, leading to extensive informal migration and subsequent deforestation [1]. Clearing for agriculture increased forest edge and produced ideal Anopheles darlingi breeding habitat [105]. High human population density allowed transmission to remain stable in a location that had not previously had extensive stable transmission and led to an outbreak unprecedented in scope. This case is a complicated story of human contact, space and mobility, environmental change and social behaviour, illustrating how a complex system with interacting social and ecological drivers can lead to real public health consequences.

Authors' contributions

R.F.A., E.S.G., H.S., L.S.P.B. and J.H.J. conceived the study. R.F.A., E.S.G., H.S., L.S.P.B. and J.H.J. wrote the manuscript and all contributed to the editing and approval of the final draft.

Competing interests

We have no competing interests.

Funding

R.F.A. was supported by the NSF GRFP. E.S.G. is grateful for support received from the Research and Policy for Infectious Disease Dynamics (RAPIDD) programme of the Science and Technology Directorate, US Department of Homeland Security, and the Fogarty International Center, NIH. ICDDR,B is grateful to the Governments of Bangladesh, Canada, Sweden and the UK for providing core/unrestricted support. H.S. was partially supported by grant no. R01AI102939-01A1. L.S.P.B. and J.H.J. were partially supported by grant no. R01AI098420.