A Very Brief Epidemiology of Infectious Diseases

Pathogens or microorganisms capable of causing infectious disease usually enter our bodies through the eyes, mouth, nose, or urogenital openings or through wounds or bites that breach the skin barrier. Some diseases spread via direct contact with infected skin, mucous membranes, or body fluids. Diseases spread that way include hepatitis B (blood, saliva, semen, and vaginal fluids), hepatitis C (blood), HIV/AIDS (semen and vaginal secretions, blood, breastmilk), Ebola (urine, saliva, sweat, feces, vomit, breast milk, semen), chickenpox (saliva, skin contact, breastmilk), gonorrhea (semen and vaginal secretions), and glandular fever (saliva). Pathogens can also be spread by indirect contact when an infected person touches a surface such as a doorknob, leaving behind microbes that are then transferred to another person who touches that surface and then touches his or her eye(s), mouth, or nose. Droplets spread by sneezes, coughs, or simply talking can transmit infection if droplets from one person come in contact with the mucous membranes of the eye(s), mouth, or nose of another person. Diseases spread that way include the common cold, meningococcal disease, and rubella. Influenza is spread by airborne droplet transmission but more commonly by indirect contact on surfaces. Acute lower respiratory tract infections, mainly pneumonia, are spread via droplets or are complications of influenza and are the leading cause of death in children under 5 years old, with low birthweight, malnutrition, and poor living conditions being contributing factors. Pathogens can also spread when residue from evaporated droplets or dust particles containing microorganisms is suspended in air for long periods of time. Diseases spread by airborne transmission include TB, measles, chickenpox, Hantavirus pulmonary syndrome, and legionnaires’ disease. Common vehicles, such as contaminated water, food, or soil, may spread pathogens. Rotavirus and Escherichia coli are the two most common organisms spread that way in low-income countries, causing moderate-to-severe diarrhea. Cholera, typhoid, and salmonella also enter the digestive system in this manner.

Living organisms, such as fleas, mites, and ticks—called vectors—can also transmit disease. The most common vector for human infection is the mosquito, which transmits malaria, West Nile virus, Chikungunya, dengue fever, yellow fever, and Zika. Other vector-borne diseases are plague (fleas) and Lyme disease (ticks). Helminths are parasitic worms that produce a high burden of disease in low-income countries and are most commonly transmitted through ingestion of contaminated soil or vegetables, drinking water, and raw or undercooked meat. The most serious helminth diseases are schistosomiasis, transmitted by snails living in contaminated water, and soil-transmitted helminthiasis, roundworm, hookworm, and whipworm.

The Epidemiological Model of Disease Transmission

The standard approach to economic evaluations of healthcare technology and interventions rarely works for infectious diseases (Brisson & Edmunds, 2003). Economic evaluations inform policymakers on welfare-maximizing policy interventions and technologies. A randomized or quasi-randomized research design is used to compare the benefits and costs of a group of individuals treated using an intervention with a group receiving the current standard of care. Estimates of costs and benefits may be projected into the future, and this is usually done by modeling closed cohorts, i.e., groups of individuals that are followed until a specific point in time or until they have all died. In the case of infectious disease, however, it is necessary to project the benefits and costs not only for the treated group but also for those groups not directly treated but indirectly affected via positive or negative externalities, now and in future periods. The closed cohort approach needs to be replaced by a dynamic model that can model the externalities.

The starting point for evaluating policy interventions for infectious diseases is the classical dynamic transmission model from mathematical epidemiology (Anderson, May, & Anderson, 1992; Keeling & Rohani, 2008). In a basic compartmental model of person-to-person transmission, individuals in the population are assigned to different subgroups or compartments, each representing a specific stage of the epidemic. The transition rates from one class to another are mathematically expressed as derivatives, hence the model is formulated using differential equations. In each time period, a group of susceptible individuals, S, is uninfected but at risk of getting infected with a certain probability that is endogenous, the “force of infection.” It depends upon the rate of contact between S and infected (and infectious) individuals, I, the transmission probability at each contact between S and I, and the prevalence of the disease in the population , I/N, at each point in time, with N denoting the total population. Infected individuals may recover from the disease or die with a certain probability in each time period. In the case of an incurable disease, most prominently HIV, the only possible outcome is (eventual) death, giving rise to the S-I model. Recovered individuals, R, may become immune to any future infection (examples are measles or mumps), which results in the S-I-R model. For some diseases, for example seasonal influenza, immunity wanes over time, resulting in the S-I-R-S model. If persons surviving the infection do not acquire immunity, they are at risk of catching the disease again and therefore re-enter S, resulting in the S-I-S model, such as for malaria or TB. Vaccinations move S directly into the group of R, but at less than 100% if they are only partially effective.

More sophisticated models can accommodate transmission dynamics involving vectors, statistical uncertainty, changes in N (by allowing for births, natural deaths, or migration), and more complex natural histories for specific diseases, for example, incubation and latent periods, asymptomatic infection, infectious period with multiple stages, or treatment that is associated with reduced transmission risk, and virus-strain specific immunity. An infectious disease is in an endemic steady state when it is sustained in a population without external inputs and when, on average, each infected person is infecting one other person. Any less, and the disease will burn out. Any more, and the number of infected increases exponentially up to a maximum incidence, when many individuals in the population have either become infected, died, or recovered and acquired immunity, and there are not many susceptibles remaining. If the proportion of N that is immune exceeds a critical unique threshold called “herd immunity,” then the disease can no longer persist in the population. The ultimate goal of prevention interventions is to achieve herd immunity. The main objective of modeling is to predict the dynamics of the spread of the disease over time and space, specifically to project the number of people who will become infected and who will die under alternative policy interventions. The standard transmission model assumes that the behavior of individuals with respect to prevention and treatment choices is fixed—or exogenous—over the course of an epidemic and that individuals either behave in the same way or can be allocated into a small number of groups displaying different behavior (e.g., high- and low-risk groups). This assumption is unrealistic to economists.

Prevalence-Elastic Demand for Prevention: The Theory

In its basic form, an epidemiological model assumes that the force of infection is an increasing function of prevalence. The larger I/N (the prevalence of the disease in the population), the larger the number of susceptibles who become infected in the next period. Contact rate and transmission probability are assumed constant. Most epidemiological models assume that all individuals behave in the same way or that they belong in a few groups with the same behavior, for example, high-risk and low-risk individuals. Some started to challenge this assumption with the proposition that susceptibles display individual-specific risk-compensating behavior. This implies that an increase in the prevalence of infectious disease I/N induces an increase in self-protective behavior of individuals, defined as “prevalence-elastic demand for prevention” (Geoffard & Philipson, 1996, 1997; Philipson & Posner, 1993). As I/N increases, susceptible individuals face a greater risk of infection and hence increase their demand for prevention. Depending on the strength of this effect, the force of infection may be a decreasing function of I/N, contrary to epidemiological model predictions. This would occur through reductions in contact rate and/or transmission probability or an increase in prevention efforts (e.g., vaccination rates) that would move S into R at a higher rate.

Prevalence-elastic demand or risk compensation implies that behavior is endogenous to the disease dynamics. This may have important implications for the spread of epidemics and optimal policies (Philipson, 2000): first, the growth of infectious disease is self-limiting not only because the group of susceptibles depletes but also because it induces preventive behavior among susceptibles; second, since the decline of a disease discourages prevention, initially successful public health efforts make it progressively harder to eradicate infectious diseases, with quite profound implications for optimal control of epidemics. For example, municipal efforts to reduce the amount of stagnant water for mosquito habitats decimates vector populations and malaria prevalence, but the resulting reduced risk of infection could make individuals less inclined to sleep under insecticide-treated bednets (ITNs). The long-term net impact of the municipal intervention would be much smaller than expected.

Prevalence-elastic demand for prevention can lead to a “vaccination game” (Barrett, 2003; Bauch, Galvani, & Earn, 2003): when vaccine coverage level is sufficiently high to achieve herd immunity, a disease can be eradicated without vaccinating everyone. This positive externality of vaccination provides a completely self-interested individual with less incentive to vaccinate as coverage increases, since nonvaccinators can gain the benefits of herd immunity without the cost of vaccination. As such, indirect protection by vaccination generates discrepancies between individual and group interests. The behavioral response is stronger the higher the—real or perceived—costs of vaccination, all else being equal (Bauch & Earn, 2004). The magnitude of the elasticity predicts the extent of risk-compensating behavior. If demand is elastic (i.e., the elasticity is greater than 1), a percentage decline in prevalence will lead to a greater percentage decline in prevention efforts of individuals. Consequently, it will become progressively more expensive to achieve further decreases in prevalence and the disease may never be eradicated. Alternatively, an inelastic prevention demand (i.e., the elasticity is between 0 and 1), implies that as infection risk declines, individuals will reduce their preventive behaviors less than proportionally, allowing for the possibility that prevention may lead to the eradication of the disease. Precise estimates of the elasticity’s magnitude are therefore crucial to predict the effect of individual choices and the need for government intervention.

There is growing recognition of the importance of incorporating endogenous behavior change into transmission models (Ferguson, 2007). A number of theoretical models (“econ-epi models”) have been developed that attempt this integration, and this exciting multidisciplinary endeavor is an ongoing area of research (Manfredi & D’Onofrio, 2013; Verelst, Willem, & Beutels, 2016; Weston, Hauck, & Amlôt, 2018). However, it has proven difficult to generate empirical estimates of endogenous self-protective behavior. These would be needed to apply theoretical simulations to the real world and develop valid models for predictions and informing policy. At least two problems plague empirical studies of the behavioral response to changes in prevalence: misreporting of preventive behavior and reverse causality between the demand for prevention and prevalence. Self-protective behavior is difficult to measure and is often based on self-report, which usually overstates preventive behavior. There is evidence that people underreport their risky sexual behavior to conform with social norms (Minnis et al., 2009). In the area of HIV/AIDS, studies have relied on condom use (Ahituv et al., 1996; Anglewicz & Clark, 2013; Godlonton & Thornton, 2013; Oster, 2012; Young, 2007), rate of partner change (Auld, 2006; Godlonton & Thornton, 2013; Lakdawalla, Sood, & Goldman, 2006; Oster, 2012), and measures that are imperfect proxies for unprotected sex, including contraceptive use (Magadi & Agwanda, 2010), abortion rate (Medoff, 2012), and fertility (Boucekkine, Desbordes, & Latzer, 2009; Young, 2007). Recent studies attempt to correct for underreporting of risky sexual behavior by using common sexually transmitted infections (STIs) as biomarkers in combination with statistical adjustments (Norwood, Hughes, & Amico, 2016), but this research has yet to be adopted by empirical studies of risk-compensating behavior. The second problem that is affecting econometric modeling of the demand for prevention is reverse causality. It is tricky to estimate the impact of prevalence on preventive behavior if the same behavior in previous periods has contributed to spread of the disease. This is likely to bias estimates upward unless appropriate research designs or econometric methods are employed (Oster, 2012).

Prevalence-Elastic Demand for Prevention: The Evidence

Empirical studies are dominated by HIV/AIDS. Of 10 empirical studies on HIV, 7 found evidence that behavior is reactive to prevalence or objective infection risk (Guillon & Thuilliez, 2015). However, only four studies produced estimates of prevalence elasticity that could be directly used in dynamic transmission models. The others proxied for prevalence, risk behavior, or both. A 1% increase in AIDS prevalence implies an 8.48% increase in the probability of condom use among young heterosexual men and women, clearly an elastic response (Ahituv et al., 1996). Among men who have sex with men in San Francisco, however, it results in a 0.46% decrease—an inelastic response—in the rate of partner change but almost no change in the propensity to participate in sexual activities (Auld, 2006). Elasticity is higher for groups classified as “low”-risk activity types (Auld, 2006). Among women, a 1% prevalence increase implies a 0.7% decrease in abortion rates in state-level data (Medoff, 2012). Lakdawalla et al. (2006) addressed the issue of reverse causality between demand for prevention and prevalence with instrumental variable estimation. They used the number of infected individuals under treatment as a proxy for HIV prevalence (and thus infection risk) and instrumented it by the generosity of Medicaid eligibility rules. This rests on the assumption that treatment increases the number of infected (and infectious!) individuals on treatment (at the time, the potential preventive effect of antiretroviral therapy [ART] was not known). Lakdawalla et al. (2006) found that only low-risk uninfected individuals seemed to react to the increase in HIV prevalence—and alleged infection risk—by reducing the number of their sex partners, but the effect was very small.

A study on couples in Malawi found that a partner’s HIV test result (as an objective measure of risk) did not influence subsequent condom use for either men or women, but the results are potentially affected by non-random selection into testing (Anglewicz & Clark, 2013). A study on Kenyan women found no statistically significant association between women’s fertility and local HIV prevalence (Magadi & Agwanda, 2010). Two country-level studies confirm this finding for sub-Saharan countries (Boucekkine et al., 2009; Young, 2007). A study on men and women in Malawi proxied local HIV prevalence with the percentage of community members who learned their HIV status after a testing campaign (Godlonton & Thornton, 2013). The rationale is that people tend to overestimate the prevalence of HIV and would, following a testing campaign in their village, revise these estimates downward. This decrease in risk perception would then lead to an increase in risk behaviors. The study found evidence for this: after 10% more community members had learned their HIV results, individuals 2 years later were 38% less likely to have used a condom with the current or last three partners. One study addressed the problem of reverse causality using instrumental variables regression with distance to Congo (the viral origin of HIV) as instrument (Oster, 2012). A doubling of HIV prevalence leads to a 1.8% decline in the chance of having multiple partners and an approximate 2% decline in having multiple partners without condom use for married individuals. The behavioral response is lower among unmarried individuals, a counterintuitive result.

The findings may not be applicable to other diseases or to subgroups of the population with systematically different risk behavior. In many regions of sub-Saharan Africa, the HIV epidemic is accelerated by commercial sex work, but the private market may dampen the self-limiting effect of prevalence-elastic demand. Consistent with the theory, there is evidence that in areas with a low prevalence of STIs, the premium for unprotected sex is small (Arunachalam & Shah, 2013). However, the premium for non-condom sex increases with STI prevalence. The private market compensates sex workers for the increase in risk, and the numbers of risky sex acts do not decrease as strongly as economic theory would predict. This would imply that commercial sex work, a major driver of STIs in many countries, would work against the dampening effect of prevalence-elastic demand. For other diseases, an early U.S. study found that prevalence of measles in the respondent’s state of residence reduces the age in months at which first measles vaccination occurs (Philipson, 1996). Malaria prevalence has a positive effect on bednet usage across nine countries in sub-Saharan Africa, but it is inelastic, with values ranging from 0.42 for adult women to 0.59 for older children (Picone, Kibler, & Apouey, 2017). Adherence to therapy for asymptomatic patients with latent TB infection in the United States was found to be prevalence-elastic (Fluegge, 2015). Vaccination against human papilloma virus did not respond to changes in cervical cancer prevalence, possibly because of lack of awareness of the virus and its causal link to cervical cancer (Staben, 2016).

Other Determinants of Self-Protective Behavior

Many other factors affect the demand for prevention, most of which are not generic to infectious disease. In a review, Dupas (2011) lists financial constraints, poor delivery of preventive services, lack of information, or more subtly wrong information provided to the wrong target groups, lack of education, social learning or peer effects, and present bias or competing mortality risks whereby people who expect to die young from other conditions have weak incentives for prevention. Related to this point, the availability of effective treatment reduces the demand for prevention because it lowers the expected costs of being sick. An example is the introduction of highly active ART for the treatment of HIV/AIDS, which reduced the implicit price of risky sex (Chan, Hamilton, & Papageorge, 2016; Mechoulan, 2007). A new technology or intervention that reduces transmission risk may decrease prevention efforts among susceptibles. This risk compensation is problematic if transmission risk is not reduced to zero or if the behavioral adjustment overcompensates. There is evidence that pessimistic expectations about the future of the epidemic could induce more risky behavior in the present (Auld, 2003). If future infection risk increases, then there is less “reward” for preventive behavior today. Fatalism may even result in counterintuitive behavior change. Increased HIV infection risk creates incentives for people with low sexual activity to reduce their activity but may make high-activity people fatalistic, leading them to reduce their activity only slightly, or actually increase it (Kremer, 1996). This is because highly sexually active people are so likely to be infected by their inframarginal partners that their marginal probability of infection from an additional partner changes little. If high-activity people reduce their activity by a smaller proportion than low-activity people, the composition of the pool of available partners will worsen, creating positive feedbacks and even increasing transmission. Last, beliefs about one’s disease status may determine preventive choices. For example, people with higher beliefs of being HIV-negative might be more likely to have multiple partners, although that has been contested (for a review, see Gong, 2015).

In summary, theoretical economic models show that demand for prevention is prevalence-elastic. Relatively few empirical studies have studied this phenomenon. Investigations would find it difficult to convincingly overcome the methodological and practical difficulties that plague this research question. If studies find an association as predicted, most estimate an inelastic response. However, some studies, particularly for low-income countries, find no association. We now move from individuals’ preventive choices about their own infection risk to preventive choices affecting others’ infection risk.

Prevention Decisions of Infected Individuals: The Power of a Test

Altruism in prevention decisions among infected individuals requires that individuals be able to make decisions and that they be aware of their infectious status. Often this is not the case, as certain infectious diseases are asymptomatic for some individuals or at certain stages of the disease (e.g., dengue and HIV/AIDS). Empirical studies on altruism among infected individuals usually focus on behavior change after learning about one’s infectious status via a diagnostic test. Many have found that people who learn they are HIV positive reduce behavior that puts others at risk, especially in longstanding partnerships, confirming they behave altruistically (Gersovitz, 2011).

This finding is put into question by Boozer and Philipson (2000), who established the importance of prior belief about one’s disease status when investigating the impact of test results on behaviors. Only individuals who believed they were HIV negative reduced their risky behavior upon learning about their HIV-positive status, whereas individuals who believed they were infected with HIV—and received confirmation with their test result—did not change their behavior. However, empirical analyses of the impact of testing on preventive behavior are complicated by selection bias; unobserved characteristics are associated with an individual’s propensity both to get tested and to change behavior once learning about his or her status. Studies have attempted to overcome this using randomized study designs or econometric modeling. HIV-positive individuals who had just learned about their status were three times more likely to purchase condoms 2 months later than HIV-positive individuals who had not learned about their status, but the effect was small as they purchased only two additional condoms on average (Thornton, 2008). A decrease in the perceived probability of being HIV positive from 10% to 0% increased the probability of engaging in extramarital affairs from 8.3% to 14.1% (Paula, Shapira, & Todd, 2014). These studies relied on self-report, which may have biased their results. Two randomized studies that used STIs as markers of risky sex confirmed that individuals who learned they were HIV positive increased their risky sexual behavior, an effect that was even stronger among individuals surprised by a positive test result (Baird, Gong, McIntosh, & Özler, 2014; Gong, 2015). This seems to confirm the findings by Boozer and Philipson (2000) and suggests that the benefits of risky sex are high and altruism toward sexual partners low for some infected individuals in the countries where the studies were conducted (Malawi, Kenya, and Tanzania).

In their most basic form, epidemiological models of STIs assume that infected and uninfected individuals match at random. The economic theory of matching markets, however, predicts that low-risk individuals are more likely to match with other low-risk individuals, and high-risk individuals with other high-risk individuals. The complementarity in health status that generates assortative matching stems from the fact that low-risk individuals have more to gain by the choice of low-risk partners than do high-risk individuals. Assortative matching has the important implication that disease growth is slower than in the random matching case considered by epidemiological analysis. Dow and Philipson (1996) analyzed data from San Francisco and found that, on average, HIV-positive individuals are more than twice as likely as HIV-negative individuals to have positive partners, and the estimated incidence reduction implied by such matching is about one-third.

Treatment Decisions of Infected Individuals

Infected individuals may impact the infection risk of others by choosing to get treatment, with resulting positive or negative externalities. The direction and size of the externalities are highly dependent on the clinical features of the disease and the treatment. Treatment may confer positive externalities if it decreases the force of infection. This can occur because of an increase in recovery rate, i.e., a decrease in the pool of infected I/N, or a decrease in transmission risk if it renders infected individuals less infectious. Treatment may not lead to recovery but increase the longevity of infected individuals. It then confers a negative externality by increasing I/N without a reduction in the transmission risk. Treatment creates positive benefits for the infected individual but generates potential costs for susceptibles who might be infected by treated individuals who remain infectious. A few studies speculated whether improved access to, and efficacy of, ART was linked to the increase in HIV incidence at the same time (Lakdawalla et al., 2006; Skåtun, 2003). These findings may be superseded by the discovery that ART reduces transmission risk. However, it is unclear as yet whether this actually applies under real-world conditions where treatment adherence is imperfect. If it does, it would make HIV treatment a means of prevention.

Antimicrobial resistance is a negative treatment externality associated with the use of antimicrobials in treating or preventing infections in humans and animals. The current stock of antimicrobials can be interpreted as a depletable natural resource. The effects of antimicrobial use in terms of resistance are unlikely to be felt directly by either the consumer or the supplier of treatment (and hence they have no incentive to reduce the use of antimicrobials) but will affect the overall welfare of the community. Patients infected with a resistant microorganism are less likely to recover from infection with the first antimicrobial used in treatment and have a greater likelihood of premature death. Patients may require both extra investigations and extra treatment (usually more expensive), and for some patients a cascade of antimicrobial drugs will be tried before one is successful in eradicating the infection. This may result both in longer hospital stays and longer periods of time away from work. The costs of AMR in the United States have been estimated at $55 billion ($20 billion in health service costs and $35 billion in lost productivity) per year overall (Smith & Coast, 2012). However, the costs of resistance could be much higher than these estimates suggest (Smith & Coast, 2013). Many antibiotics are given as prophylaxis, and infection after standard procedures (with associated morbidity and mortality) will be much more common in a world without effective antibiotics, a cost rarely considered in current estimates. Moreover, anticipation of untreatable infections could change treatment decisions; for example, clinicians may be less inclined to choose surgical treatment, with associated adverse outcomes for patients. Such adaptive behavior was found in German intensive care units; demand for reserve antibiotics grew in expectation of increasing resistance against the first-line therapy (Heister, Hagist, & Kaier, 2017).

In summary, economic theory predicts that individuals’ choices regarding prevention generate externalities. This would mean that individuals are not altruistic because they do not—or do not sufficiently—take into account the effects of their actions on the infection risk or treatment outcomes of others, either negative or positive. In the case of flu vaccination, there is evidence that some individuals are altruistic and do take others into account, but this depends on circumstances. Evidence on whether infected individuals are altruistic is conflicting, and there is even reason to believe that they may reduce preventive efforts on learning about their disease status. Treatment may generate positive externalities if it reduces transmission risks or the length of time individuals are infected but negative externalities if it leaves infected individuals infectious. AMR is a negative externality imposed by infected individuals or food producers of the present onto infected individuals of the future.

Control of Infectious Disease

Global Eradication: Is It Worth It?

The discussed policies have achieved, at best, elimination of an infectious diseases in a particular country or region and as a second-best, control, whereby incidence of disease is reduced below a critical level but not to zero. The global eradication of an infectious disease is an extreme—indeed, a singularly ambitious—policy goal. Eight attempts have been made to date to eradicate infectious diseases: two successful programs targeting smallpox and rinderpest (an animal disease); four ongoing programs targeting poliomyelitis, yaws, dracunculiasis, and malaria; and two former programs targeting hookworm and yellow fever. The vector control campaigns of the 1940s and 1950s virtually eliminated yellow fever everywhere except Africa. When the disease subsided in the Americas, funding for mosquito control was reduced and resulted in a recovery of the mosquito populations. This may have contributed to recent localized outbreaks of yellow fever, the increase in the incidence of dengue over the past decades, and localized outbreaks of Chikungunya and Zika, all diseases transmitted by the same mosquito.

An important question is whether a country should push for and support international efforts to achieve global eradication, aim for elimination within its borders, or attempt optimal control, which involves moving into and sustaining a steady state with a positive level of infection. It is very difficult to identify the welfare-maximizing policy, and recommendations need to rely on projections of uncertain benefits far into the future. If a disease is already controlled at very high level, for example by vaccination, then a steady state with positive level of infections is maintained at comparably high costs, and just a slight increase in the vaccination rate would cause the disease to be eliminated or eradicated. Eradication would increase costs in the short run, and the marginal costs of the last prevented case are probably very high. If public vaccination policy is (partially) crowded out by market behavior under the assumption that the private demand for vaccination is prevalence-elastic, then economic theory suggests that eradication may only be achievable as time goes to infinity (Geoffard & Philipson, 1997).

The main benefit of eradication does not lie in the few additional infections averted but in making the pathogen disappear. Eradication would avoid the need ever to invest in prevention and surveillance of the disease in the future—the “eradication dividend,” an enormous economic benefit. A very high level of control is therefore unlikely to be optimal. The optimal policy will require either a low level of control or eradication (Barrett, 2007). In the case of smallpox, there is evidence that eradication was cost-saving and led to large economic benefits due to avoided vaccination costs, at least when considering only the incremental costs needed to eliminate smallpox from the remaining endemic countries at the time the decision was made to eradicate (Fenner, Henderson, Arita, Jezek, & Ladnyi, 1988). Smallpox was the ideal candidate for eradication due to its clinical characteristics. Unfortunately, the eradication of other diseases is likely to be more difficult and less attractive in benefit–cost terms. The probability that the elimination of malaria would be cost-saving over 50 years was estimated at a range from 0% to 42%, based on data from five sites (Sabot et al., 2010). The threat of bioterrorism weakens the economic case for eradication. Countries may feel the need to continue to vaccinate, even if at a relatively low level, or to stockpile vaccine and prepare for emergency distribution.

Eradication requires strong international cooperation, and it is a “game,” because some countries may be willing to eliminate the disease within their borders only if assured that all others will eliminate the disease within their borders. International financing is also a game, because each country would rather free-ride than contribute (Barrett, 2013). If eradication fails, much of the money spent will have been wasted. If it succeeds, the world will reap the dividend.