Space Imaging and Prevention of Infectious Disease: Rift Valley Fever

Abstract

Rift Valley fever (RVF) is a mosquito-borne viral disease causing febrile illness and death in domestic livestock (cattle, sheep, goats) and humans. In Africa, RVF erupts following abnormally high rainfall and flooding. Remote sensing surveillance of vegetative growth could provide early warning, weeks to months in advance of RVF emergence, and thus permit intervention strategies to ameliorate and prevent this infectious disease. To act on this advance notice, however, public health officials must quantify the economic cost associated with the disease (in terms of losses to agriculture and international trade as well as human morbidity and mortality) and weigh the averted losses against the diversion of financial and public health resources dedicated to other major ongoing health needs, such as malaria and HIV/AIDS. Other complications include the accuracy of the predictions, the shelf life of vaccines, and the effectiveness of vector control strategies.

9.Commentary: An Emerging Approach

• Joshua Michaud 

1. International Development Program, Johns Hopkins University School of Advanced International Studies, Washington, DC, USA

   Joshua Michaud

Public health has always faced the classic conundrum of valuing prevention. How do you assign worth when the desirable product of public health activity is the absence of disease in a population? Policymakers and the public in general are much more likely to notice and understand how important public health can be when it fails as opposed to when it works, since the results of failure—the societal and economic toll of disease and mortality—are highly visible and relatively easy to catalog. The value of cases prevented and costs avoided, on the other hand, is harder to appreciate. Often, public health practitioners lament that many proven and inexpensive health interventions are ignored or underutilized because their full benefits are not understood. Although this may be true to an extent, the problem of underutilization also derives from the continuing inability of public health practitioners themselves to fully understand, quantify, and communicate the value of their work. Public health methods for valuation are imperfect, and not used widely. Practitioners are understandably focused on the health consequences of their work and tend not to dwell on its economic value.

This valuation deficit is certainly present in the important area of disease outbreak prevention and response. It is into this “work-in-progress” valuation of disease outbreak information that David Hartley introduces his chapter, “Space Imaging and Prevention of Infectious Disease: Rift Valley Fever.” Hartley’s chapter provides an overview of a promising public health application of remote sensing, and it quickly becomes clear that it is an initiation of a discussion about assigning value to information useful for public health action rather than a proposed set of methodologies for doing so. The chapter proposes, introduces, outlines, and hints at the potential for using evaluation of costs and benefits of predicting and preventing infectious disease using remotely sensed data. Having read the chapter and having some sense of the other kinds of public health applications for which remotely sensed data could be used, I find it hard to argue with the chapter’s primary conclusion—that the potential for linking satellite data with epidemiologic tools to design and implement predictive capabilities for disease outbreaks is excellent. Still, we are very much at the beginning of this conversation and the process of tool development in this area.

In this response to the chapter, my hope is to contribute to the dialog around remotely sensed data and its application to disease outbreak interventions by highlighting a few points from Hartley’s chapter, building on and supplementing these points, and extending the points by outlining an even broader scope of application for this nascent interdisciplinary work. Further, I point out a few limitations of the approach and points for consideration to be kept in mind as we move forward.

9.1.1 9.C.1.Preliminary but Promising Field

The chapter indicates the existing set of methodologies and scientific literature on application of satellite derived data to disease outbreaks remains quite limited. Hartley’s wording when reviewing the literature in the field is indicative: “no studies,” “studies would provide,” “new field,” “literature is silent” and “great promise”; all of these phrases are present because, while there is reason to believe something of value could be extracted here, much more work is needed to understand and place a value on the information being generated.

As both a researcher in the field and a sometime consumer of the kind of predictive data on RVF that has been produced in recent years, through researchers like Hartley and through NASA and the U.S. Department of Defense’s global emerging infection system program, I can personally attest that even the limited, preliminary kinds of predictions that have been made available in recent years have been valuable to U.S. policymakers, let alone those working in Africa and other locations to reduce the effects of RVF outbreaks in the field. The Department of Defense and others have worked with NASA to use remotely sensed data in East Central Africa, helping produce forecasts of likely RVF activity in the area that rely on monitoring the vegetation index to forecast areas where mosquito hatching, and possibly viral transmission, could occur over subsequent weeks and months. The Department of Defense analyses the remotely sensed data and makes intermittent forecasts of RVF transmission risks, making what might have been an academic exercise of providing proof of principle for the predictive capability of these models a true, operational program. I was a consumer of this information as part of my work with the National Center for Medical Intelligence, and I observed first-hand how outbreak risk forecasts for RVF had decisionmaking implications for Defense in regard to its deployments, field exercises, and force health protection efforts.

As Hartley has effectively outlined, RVF makes an excellent case study for this kind of predictive modeling because of the unique relationship between certain environmental variables. Transmission of disease is highly associated with microenvironments, locations where certain combinations of temperature, rainfall, and vegetation in areas with dambos and dambolike terrain together come together with the mosquito vectors and distributions of animal and human hosts to create ideal RVF transmission pockets. Fortuitously, many of the associated environmental variables can be characterized from space using remote sensing. So, as Hartley shows, there is little doubt that through existing technology, remote sensing can be useful in the prediction of RVF disease outbreaks in East Central Africa, and perhaps in other locations as well.

As interesting and compelling as it is to elucidate an epidemiological connection between satellite data and RVF outbreaks, some additional questions for the purposes of understanding the value of information are, “What are the economic implications of generating RVF (and other infectious disease) predictions?” and, “Are gathering and analyzing that information worth the expense?”

In theory, linking the release of a predictive assessment of a future RVF outbreak with the actions of policymakers and farmers in the region should be measurable, along with an assessment of the costs, but in reality, gathering the wide set of information needed to make accurate cost-benefit calculations, especially in an area such as East Central Africa, is tremendously problematic. The cost of the disease is borne most acutely, one could argue, in the economic sense through the loss of income when farmers’ herds become infected. Certainly, the human toll can be significant, but much of the concern around RVF is centered squarely on its agricultural implications. For this reason, the United States is mainly concerned about RVF, I would say, as an agricultural importation threat. The costs of RVF to livestock farming and potential damage are very significant. For this reason, the Department of Agriculture considers RVF one of the most threatening “foreign animal diseases” out there, and that agency has performed some analyses of the potential economic repercussions should RVF be imported into the United States and cases of the disease be found in U.S. livestock. Local farmers in East Africa and other RVF-affected areas are no less cognizant of the potential losses associated with the disease. But, as Hartley touches on and I hope to indicate more fully in the next section, there is more than just the simple accounting of costs of illness and lost income to consider.

9.1.2 9.C.2.Economic Considerations for Prediction and Response

Rather than attempt to outline a full methodology for calculating the cost-benefit of disease prevention through RVF prediction, Hartley’s chapter is only able to sketch how such a calculation might be done. This is mainly a reflection of the lack of prior academic work and methodology relevant to this particular area. Hartley ably reviews several categories of costs and benefits that would have to be included in valuation calculations, which I will not repeat here. Rather, in this section, I would like to highlight additional considerations that would have to be incorporated in a full valuation model for the kind of remotely sensed work that the chapter characterizes. As Hartley partially recognizes, attempting to determine whether RVF surveillance and prediction are worthwhile requires not only understanding the costs of the animal and human disease burden and the expense of the sensing platforms and public health interventions and the like, but also disentangling a larger set of questions about incentives and externalities that are inherent in disease outbreak prediction, detection, and response. The following discussion centers around three public health functions in this area: surveillance (initial detection—or accurate prediction—of an outbreak), reporting (communicating the presence of an outbreak once it is detected), and response (implementing public health actions to stop transmission and reduce cases of the disease). The discussion is not limited to RVF alone, since it draws lessons from other infectious diseases; the points are applicable for RVF, but also more broadly for many kinds of outbreaks.

Surveillance: Is More Information Always Better? It may come as a surprise to learn that in the context of disease outbreaks, more surveillance information is not always better. “Better” here refers to “economically rational” for the actors involved in conducting surveillance. The reason more surveillance might not be better involves the built-in economic disincentives to infectious diseases that potentially leave some people, industries, and countries worse off with more information. Take a Kenyan farmer with livestock at risk of being infected with RVF. Were he to discover that his animals had been infected, many (or all) might be put down in an attempt to control the disease, or access to markets where the farmer might sell his animals or derived products might be restricted. Such actions might lead to a significant loss of income or even destitution that he would wish to avoid. If some of the animals did become infected and the farmer was unaware (either by chance or by choice), then he might still be able to extract some gain from selling the animals and avoid the potential loss of his entire herd and income. Given the choice of knowing or not knowing, he might prefer not knowing—in other words, he is disincentivized to participate fully in surveillance. The same logic goes for a methodology of prediction using satellites or any other tools. An area’s farmers may feel that by knowing about an impending epidemic in advance—one in which their livelihoods are guaranteed to suffer while the benefits of this knowledge are less certain—they could be worse off than not knowing.

Those kinds of disincentives for surveillance are not restricted to individual farmers in developing countries at risk for outbreaks. In the United States, for example, when birds illegally smuggled into the country, some of which might have come from geographic areas endemic for highly pathogenic H5N1 avian influenza (HPAI H5N1, another frightening “zoonotic” disease, or a disease animals that can affect humans), are intercepted, no laboratory testing of the birds is performed prior to culling them. The rationale for this is to avoid having to say that HPAI H5N1 has been found inside the borders of the United States. So far, the HPAI H5N1 virus has not been found in the United States, but detection of the virus would surely have major implications for the poultry and other industries because immediate trade restrictions and possibly panic might ensue. This is a missed surveillance opportunity put in place for economic concerns, and it indicates the power of an economic disincentive for more information about potentially deadly diseases.

In another example, there were similar difficulties in surveillance for bovine spongiform encephalopathy (BSE, or “mad cow disease”) in the United Kingdom, since farmers had little incentive to report suspected cases in their herds. In fact, the United Kingdom’s BSE inquiry report stated, “one reason why BSE was not picked up at a very early stage by the system was the lack of incentive for farmers to refer an isolated case of an unrecognized disease in their herd for laboratory investigation. Indeed, there was a positive disincentive, namely the cost of a post-mortem examination” (UK Government 2000). This obstacle often appears in the context of zoonotic infections that affect agricultural livestock because there are potentially large economic losses from the culling of sick and potentially sick animals. Such culling and destroying of livestock and the associated trade restrictions and lack of access to markets that usually coincide with an outbreak response can serve as a powerful disincentive for individuals (and sometimes whole towns or industries) from participating in disease surveillance. This disincentive for good surveillance information exists whether the surveillance is performed through diagnostic tests or through application of remote sensing data in a predictive climate-based model.

Reporting: Is Being Completely Transparent Always Rational? In a similar vein, there is commonly a disincentive to be fully open and honest about reporting detected outbreaks. Farmers who know their flocks are ill may avoid saying anything for fear of losing their income. Countries wishing to avoid economic damage sometimes downplay or fail to report disease outbreaks. In areas affected by HPAI H5N1, for example, poultry farmers are often reluctant to report cases of dead birds to health authorities for fear that officials will rob them of their livelihoods (and important sources of food) by culling their flocks. As one Nigerian poultry farmer stated, “If the government isn’t able to compensate me [sufficiently], why should I bother to report if my birds become sick? Wouldn’t I be better off just taking my chances?” (Bellagio Meeting 2006).

This reporting disincentive also plays out along international trade routes, motivating obfuscation by governments. A great hindrance to transparency and early disease detection internationally is the cost that an affected country faces when the rest of the world finds out about the outbreak. On learning about an outbreak, many times neighboring countries close borders, trading partners restrict or stop imports, and travel and tourism cease. These actions have real and sometimes very damaging effects on important industries or economic sectors within a country, and therefore there is a strong incentive for underreporting or not reporting at all (Cash and Narasimhan 2000). Economic costs can be significant, in particular if the infectious disease is linked to the agricultural export sector, as RVF often is.

There are many examples of this kind of negative trade consequence from reporting a disease outbreak. Some of the more commonly cited figures include the 1991 cholera epidemic in Peru, which is estimated to have cost the country more than $1.5 billion in lost exports and tourism (Knobler et al. 2006), and India’s 1994 outbreak of suspected plague, which likely cost the country an estimated $1.7 billion (WHO 2005). Thailand initially denied it had H5N1 avian influenza in its chickens, and Indonesia delayed reporting its first bird outbreaks of H5N1 (CNN 2004). Burma failed to report its first H5N1 bird cases when they occurred in 2004 (Beyrer 2006). China has reportedly covered up H5N1 outbreaks in its flocks multiple times. In 2006, the World Health Organization (WHO) accused the Chinese Ministry of Agriculture of “selectively reporting” outbreaks of H5N1 in its chickens and refusing to send samples from infected birds out for testing. At that time, the chief WHO representative in China stated, “It’s so sad that we haven’t got that [outbreak] information or those [H5N1] viruses from the Ministry of Agriculture … it’s really beyond comprehension to us” (CBC 2006). Once reporting of these bird outbreaks does occur, the economic consequences can be very painful. When Thailand’s troubles with bird flu became known, the resulting collapse in poultry exports cost it some $1 billion (Economist 2006). When Vietnam first reported the presence of H5N1 (the virus and the culling wiped out 17% of the country’s chickens in 2004), the outbreak and subsequent trade bans resulted in a loss of more than $83 million for this developing nation (Vietnam News Brief Service 2004). In 2003, when another pathogenic avian influenza subtype (H7N7) was found in poultry in the Netherlands, Belgium, and Germany, 28 million birds were culled and restrictions on trade in both poultry and swine (Dutch pigs were found to harbor evidence of infection) were enforced (Kimball 2005).

Response: Can Anything Be Done About It? Even when a disease outbreak can be detected early and reporting does occur, there might exist a gap between what should be done to implement an ideal public health response, and what can be done given what a country, region, or local community can do or is willing to do. Information that is not “actionable” may not be valuable. It does no good for a country to know where and when an outbreak is occurring if it does not possess the ability, or the willingness, to respond. In such a case, the information would have been generated just for information’s sake, not for policy action. Again, such a restriction on the value of outbreak information applies equally to confirmation in the form of a diagnostic test result, or a trusted prediction based on satellite data.

Therefore the links among surveillance, reporting, and response capacity are critical, and we should not emphasize more and better data when the relevant actors cannot implement or improve policies with that information. One of these activities without the others provides limited or no benefit; all must be provided. Clearly, disease outbreaks are prone to collective action problems, since “rational” action by individuals and governments protecting their own economic interests can lead to overall irrational outcomes, such as worse outbreaks, greater health consequences, and more interruptions of trade and economic activity. Valuing the information contained in the prediction of RVF outbreaks through remote-sensing data would have to take into consideration these characteristics. Could an accurate RVF outbreak prediction actually make farmers in the targeted area worse off economically because of preemptive trade bans or other damaging actions? Is this risk worth it if the outbreak likely cannot be contained, given weak public health capacity? What are the optimal outcomes for all parties involved, economically speaking? Hartley hints at these complications, but it is worthwhile to highlight them more clearly.

9.1.3 9.C.3.Broader Potential for Environmental Observation and Disease Prediction

Although Hartley’s chapter focuses on RVF, a subtext here is that similar techniques and methodologies could perhaps be applied to other infectious disease threats. Certainly the literature is already relatively rich with studies examining the relationship between environmental variables and disease epidemics (Kelly-Hope and Thomson 2008; Harvell et al. 2002).WHO in 2005 identified 14 infectious diseases it classified as potential candidates for environmentally based “early warning systems,” a list that includes RVF, malaria, dengue fever, cholera, meningococcal meningitis, and influenza (WHO 2005). All of the 14 diseases are affected to some extent by the environment, but each to a unique extent, such that variables strongly associated with one may not be associated with others. In addition, many factors besides the environment must be taken into consideration when judging the transmission of these pathogens—everything from geographic variations in endemicity to human and vector behavior, to varied and changing control measures, to dynamic immune states, and other measures that may be unknown or not measurable.

In the case of RVF, the disease’s very direct link to the environmental conditions that favor mosquito breeding in dambos (precipitation, temperature, and other factors that can be measured through satellite monitoring) make it a good candidate for forecasting. For other diseases, environmental variables serve as drivers of disease transmission but are only relatively minor contributors to the overall set of factors that determine when and where disease outbreaks emerge and spread. Thus, among those infectious diseases linked to environmental factors, RVF in parts of East Africa may in fact be the lowest-hanging fruit of remote sensing–based disease prediction. Extending the prediction technique beyond RVF, while possible and worth pursuing, might be more involved, less accurate, and potentially more costly.

Dengue is sometimes referenced as a disease that might be predicted based on environmental factors. Just as in the case of RVF, breeding and activity of mosquito vectors are influenced by temperature and rainfall, but complications in the ecology of dengue transmission make it a bit more unpredictable. This is especially true in the case of the dreaded and explosive “urban” dengue outbreaks, because the drivers of these types of epidemics, which are becoming more common in many cities of tropical developing countries, are heavily based on human behavior rather than on strictly environmental factors. Urban dwellers who leave open containers of water or fail to clear stagnant puddles create accommodating habitats for mosquito breeding whether it has rained recently or not. The complexities of human immunity to dengue’s multiple serotypes are not fully understood, also making clear prediction more difficult. These additional factors have made dengue a more difficult target for environmental modeling and linking to remotely sensed data. Other diseases bring their own complications: the link between the environment and plague, for example, is moderated through the activities not only of the vector that transmits the bacterium, but also the rodent hosts of that vector; this and plague’s highly focal natures takes prediction from environmental observations several steps further away from a direct causality.

Perhaps the biggest prize (and the one with the largest potential benefit) in the outbreak prediction field is malaria. This mosquito-borne parasitic infection is highly endemic in many countries around the globe and remains one of the leading killers of children in low-income countries. Currently, the prediction of malaria outbreaks through use of environmental variables is fraught with complications and confounders. Nonclimatic factors such as population immunity levels, nutrition status of a population, the state of control measures at local levels, the use of antimalarial drugs, and the pattern of drug resistance in circulating malaria strains strongly influence the environment-malaria link. These difficulties have not prevented research and development of climate-based malaria predictions, however. In fact, multiple studies of the relationship between climate factors such as El Niño–Southern Oscillation (ENSO) cycles or changes in temperature or rainfall and malaria transmission have shown there can be a relationship (Ebi 2009; Githeko and Ndegwa 2001), but to put it kindly, the evidence is mixed and highly contingent on the specific circumstances, location, and time.

In the background of those analyses of climatic variables and disease we have the looming shadow of global climate change and its possible effects on infectious disease. If prediction can be somewhat successful on the small scale of weeks and months, how successful can we be on a longer scale? Can we predict disease transmission patterns years in advance once we know what the climate will look like in the future? Current conventional wisdom in public health holds that many diseases that had previously been circumscribed to poor tropical areas of the world, driven by an ever-warming climate, will expand their reach into geographical areas and populations where they had previously not been found; malaria is typically held up as a prime example. But the link between climatic variables and disease transmission actually becomes more tenuous the larger the geographic and temporal scale over which one attempts to predict (Lafferty 2009). It is precisely because the nonclimate variables associated with transmission become so heterogeneous and difficult to model over large areas and longtime scales that the probability of accurate prediction becomes very small, and adherence to simple cause and effect becomes problematic.

Several examples indicate how far we have to go to make long-term predictions of disease transmission based on climate. A recent article by Gething et al. (2010) in Nature deftly points out the problems with blindly ascribing increases in malaria with climate change by comparing the best estimates for the effect size of climate change on malaria transmission compared with the effect sizes of different control and treatment measures. The authors’ bottom line is that a warming planet over a long time frame has the potential to affect transmission, but the effects of available control measures and treatments dwarf the effects predicted from climate. In other words, climate effects are drowned out by control effects (not to mention other nonmeasurable effects, such as general development), and predictions are often based on the erroneous assumption that control and treatment measures and technologies won’t change over the large time scales in these analyses. In another example, researchers in Australia concluded that as parts of that country become drier with climate change, the risk of dengue might actually increase as people hoard water in water tanks that would increase mosquito breeding (Kearney et al. 2009). This is a counterintuitive result, as one might assume that mosquito activity would most likely decrease—and disease transmission with it—as the environment becomes drier. Finally, it is worthwhile to note that two adjacent areas with equivalent climates can have dramatically different transmission patterns for some diseases that have been linked to climate. Researchers have examined the areas that straddle the U.S.-Mexico border and found that between 1980 and 1999 there were more than 62,000 reported cases of dengue on the Mexican side of the border (likely an underestimate), while on the U.S. side there were just 64 cases. The difference is mostly explained through differences in living standards between the United States and Mexico (Brunkard et al. 2007). So, making the link between observed information and disease is not linear and highly determined—many disease processes are complex and resist simple cause-and-effect explanations.

9.1.4 9.C.4.Concluding Remarks

Hartley’s chapter is a valuable review of the possibilities and the obstacles of making predictions about infectious disease outbreaks using climate observations. Certainly, as the chapter indicates, there is reason to believe that real associations have been discovered and that true predictive associations can be made between earth observations and disease transmission. As this commentary has attempted to indicate, the strength of these associations is highly dependent on the disease in question, the geographic and temporal scales involved, and the available data and understanding of disease processes. What is lacking, as is made abundantly clear in the chapter, is proven methodologies or sets of tools that can be applied to valuing the predictive disease work and that take into consideration the complications and externalities associated with transmission of diseases like RVF. Collection of better data along with the development of more robust epidemiologic and economic models of disease prediction would go a long way to bringing us closer to understanding, and ultimately assigning the proper value to, these efforts.