The global burden of cholera remains substantial. In 2005, 131,943 cases and 2,272 deaths were reported to the WHO, and recently major, sustained epidemics have been reported in West Africa [1]. These statistics are gross underestimates, as many cholera-endemic countries do not report cholera to the WHO, including Bangladesh, which has among the highest rates of cholera in the world. More realistic estimates of the global burden of cholera mortality place the figure at 100,000–150,000 deaths per year. This high burden occurs because cholera targets the most impoverished populations, which often lack access to centers that can appropriately administer life-saving rehydration therapy.

This continuing high burden highlights the need for interventions to prevent cholera. While improved water and sanitation constitute the ultimate basis for the prevention of cholera, this is a far-off goal for the impoverished settings in which cholera thrives. Vaccines constitute near-term options for cholera control. During the past 20 y, killed oral cholera vaccines (OCVs) have been shown to be safe and protective in populations with endemic cholera [2,3]. A large-scale community OCV vaccine trial in Vietnam showed that mass vaccination for endemic cholera is operationally feasible in a developing country setting [4]. Yet, because these vaccines have moderate levels of protective efficacy, they have not been routinely adopted as control measures for endemic cholera [5,6]. Recently, however, a reanalysis of a field trial of killed OCVs in Bangladesh demonstrated that these vaccines are capable of inducing herd protection of nonvaccinees, as well as enhanced protection of vaccinees, when even modest levels of vaccine coverage of the targeted population are attained [7]. These observations prompted us to re-examine the potential impact of use of these vaccines in mass immunization programs on the control of endemic cholera. In this paper, we use a large-scale stochastic simulation model to investigate potential control of endemic cholera with OCV.

In the mid-1980s, a randomized controlled vaccine trial with OCV in Matlab, Bangladesh, yielded 70% direct vaccine efficacy for up to two years [2,3]. We use information about Matlab, Bangladesh to construct a model of the population as it was in 1985, consisting of 183,826 participants. These individuals were mapped into families and families were distributed in “baris,” i.e., patrilineally related household clusters [7]. In the model, baris are further clustered into subregions of about 6 km² that are considered to be the geographic cholera transmission areas. The model (see Figures S1–S5 and Texts S1 and S2) represents the number of contacts that a typical person makes with sources of potential cholera transmission in the course of a day. The age and bari size distributions of the population are based on data from Ali et al. [7]. Women and children are assumed to come into contact with sources of infection in the subregion where they live, while working males are assumed to make contact with infective sources in the subregion where they live as well as where they work. The population structure and movement distance function are given in Text S1.

The modeled natural history of cholera is described in Figure 1. The distributions of the duration of the incubation and infectious periods and other important natural history parameters are taken from the cholera literature [8–10]. The model was calibrated to cholera illness incidence data from a large cholera vaccine trial in the Matlab field area of the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B), that took place from 1985 to 1989 [2,3,11]. OCV or placebo (killed E. coli) was offered to children 2–15 y of age and women over 15 y. Matlab has cholera transmission year around, but it generally experiences a large cholera epidemic from September through December and then a somewhat smaller epidemic from March through May [12]. Although the definition of endemic cholera may vary, we define this pattern of cholera transmission to be “endemic,” i.e., at least a low level of cholera all year round, with periodic larger epidemics. For our analysis we assess cholera risk in individuals residing in different subregions in the field trial area. Subregions are useful for cluster analysis because they are geographically discrete with local sources of water. Infected people within subregions can possibly transmit the disease person-to-person [13]; more commonly transmission occurs when feces enter local ponds where vibrios from the feces are amplified in organic material, especially plankton, during the epidemic periods [14]. We define an infection function (see Text S2) that gives each susceptible person's daily probability of infection from all possible sources of infection created by infected people excreting cholera vibrios into the environment or through more direct contact. This probability of infection is proportional to the number of vaccinated and unvaccinated people in the subregion where contact is specified to occur (see Text S2).

We base the indirect, overall, and total vaccine effectiveness on the reduction in infection rates when comparing the appropriate groups within a subregion with no vaccination to a comparable subregion with a fraction f > 0 of the population vaccinated (Figure 2). The direct effectiveness compares the vaccinated to the unvaccinated within a subregion (Figure 2). Since the vaccinated and unvaccinated in a particular subregion are subject to the same intensity of infection, the direct effectiveness should be independent of the vaccine coverage [15]. This condition held for the randomized vaccine trial; however, if in actual practice vaccinated people tend to have less risk of cholera infection than unvaccinated people due to factors other than vaccination itself, then the actual vaccine effectiveness measures would be attenuated. In addition, we assume that the vaccine induces immunity that results in a proportional reduction in the probability of infection per contact with an infectious source [15], i.e., a leaky vaccine. For our application, we further average over all the subregions within vaccination coverage strata.

Vaccines would need to contain antigens that are reactive or have cross-reactivity to current circulating strains of cholera in endemic areas [12]. These would include the Vibrio cholerae O1 El Tor biotype with both the Ogawa and Inaba serotypes, as well as the O139 serogroup that has been circulating since 1993.

Since we are modeling endemic cholera, we assume that there is significant partial population-level immunity due to past exposure to cholera. The actual level of such protection is complex and difficult to determine. Past analysis of cholera incidence in Matlab reveals that there is partial waning of population-level immunity over six months [12], although there is reason to believe that there is some partial long-term immunity as well [16,17]. Since we are modeling endemic cholera over a limited (one year) time horizon, we assume that infection in individuals results in complete immunity for the remainder of the simulation period, but this immunity would be partially lost with time. We also model populations that may not have as high a level of partial population natural immunity than that of Matlab. We do this by increasing the relative susceptibility of the simulation population in contrast to Matlab. For example, people in a population relatively new to cholera could have a susceptibility that is twice as high as that of Matlab per contact with an infectious source. In this case, the probability of infection per contact with an infectious source would be twice has high as that probability in Matlab. The relative susceptibility multiplier would be two in this case (see Text S3).

We calibrated the simulation model using cholera incidence data observed in the first year of the vaccine trial (Table 1) over a 180 d period in order to capture all the cholera transmission during the large annual cholera outbreak. This was done by varying the transmission probability, π (see Figure S6 and Text S2), such that the differences between the observed incidence rates and the simulated incidence rates in Table 1 were minimized. The estimated reproductive number was 5.0 with a standard deviation of 3.3 (see Text S2). A summary of the parameters and their baseline values are shown in Table S1. The vaccine coverage levels in the target population and the effective coverage in the entire population from the trial are summarized in Table 1. We assume that vaccinated people receive an effective regimen of two doses. The observed cholera incidence rates among the unvaccinated ranged from a high of 7.0 cases/1,000 over 180 d for the subregions with the lowest coverage in the target population, centered at 14%, to 1.5 cases/1,000 for the highest coverage, centered at 58%. The observed cholera incidence rates among the vaccinated ranged from a high of 2.7 cases/1,000 for the subregions with the lowest coverage to 1.3 cases/1,000 for the highest coverage. We set the vaccine efficacy (VE) for susceptibility to VE_S = 0.7 [2,3] and for infectiousness to VE_I = 0.5. The simulated incidence rates provided a good fit to the data based on a χ² goodness-of-fit test for frequency data (p = 0.84, 9 degrees of freedom). Figure 3A–3D show the number of cases over time comparing the unvaccinated to the vaccinated populations. Videos S1 and S2 show the spatial–temporal epidemics at different coverage levels. For effectiveness measures, we compare the intervention subregions to hypothetical subregions that receive no vaccine, i.e., f = 0. Table 2 shows the indirect, total, and overall effectiveness estimated by the model for possible coverage levels when comparing coverages in the entire population, 2 y of age and older, ranging from 10% to 90% compared to no vaccination. For example, the average indirect effectiveness, comparing a population with a coverage of 30% to one with no vaccination, is 70% (also see Figure S7). This indicates that on average, the cholera incidence among unvaccinated people in a population with 30% coverage would be reduced by 70% compared with a completely unvaccinated population. At this level of coverage, the total effectiveness of 90% indicates high protection for a vaccinated person in a population with 30% vaccination coverage, while the overall effectiveness of 76% indicates a good overall reduction in risk to the overall population. According to the model, around 40 cases of cholera are prevented per 1,000-dose regimens of vaccine at low coverage and 13 cases at high coverage. At coverage levels of 50% and higher, all levels of effectiveness exceed 85%, resulting in the nearly total control, i.e., an overall annual cholera incidence of about 1 case per 1,000 people, of cholera transmission.

From Table 1, we see that the simulated direct effectiveness at all coverage levels is estimated from the simulations to be about 66%, while the vaccine efficacy for susceptibility, VE_S is pre-set at 70%. This small underestimation is due to the fact that we model the vaccine effect to be a 70% reduction in the risk of infection per contact with an infective source, i.e., a leaky effect, but we use the risk ratio estimator of vaccine effectiveness over the entire cholera epidemic. We do this for the purpose of comparison, as this was the primary estimator used in the cholera vaccine trial in Matlab [2]. We have shown that an estimator based on the monthly hazard ratio gives a similar, but more accurate estimate of actual vaccine efficacy [3]. Also note from Table 1 that the observed estimate of the direct effectiveness is only 14% in the highest vaccination coverage category when it should be around 66%. This discrepancy probably reflects small sample bias due to the low cholera incidence in the highest vaccine coverage category.

Figure 4 shows the overall effectiveness estimated by the model for possible coverage levels in populations at different levels of relative susceptibility compared to Matlab. For populations that are 1.5 times as susceptible as those in Matlab, 50% coverage would still be sufficient to achieve an overall effectiveness of 80%. However, for populations that are 2–2.5 times as susceptible, 70% vaccine coverage would be necessary to achieve control.

Since vaccine efficacy can vary for different field settings and vaccines, a sensitivity analysis was carried out on the VE_S and the VE_I to determine the minimum efficacy needed to maintain control at the vaccine coverages explored above. The VE_S and VE_I were varied from 0.3 to 0.8 (Figures S8–S15). These analyses reveal that theVE_S could be as low as 0.5 and the VE_I as low as 0.3 to still maintain control of cholera, as long as the vaccine coverage were 50% or higher. Further sensitivity analysis on the vaccine coverage (Figures S13–S15) reveals that vaccine coverage should be at least 50% to maintain control.

Sensitivity analyses also were performed for values of the seasonal boost factor, the relative infectiousness of symptomatic infectives, and varied subregion sizes. The baseline epidemic with no vaccination was calibrated to the simulated cholera incidence data for Matlab with no vaccination. Our result that in populations like Matlab 50% vaccine coverage should be sufficient to control cholera remains valid for variation in the season boost factor (Figure S16) and relative infectiousness (Figure S17). However, we did see variation with respect to subregion size. For larger subregions (>6 km² each), 50% vaccine coverage was sufficient for control. But for very small subregions (0.04 km² each), the average overall vaccine effectiveness approaches 75% with a vaccination coverage of 70% (Figure S18). This lower effectiveness implies that random mass vaccination may not be effective in small subregions, and vaccination would have to concentrate on the small subregions where transmission is occurring. This result could be applicable to epidemic cholera in small, dense settings such as refugee camps where coverage would have to be high to control transmission.

The results of this modeling study indicate that 50% coverage with OCV could control cholera transmission in endemic areas such as Matlab, Bangladesh through a combination of direct and indirect effects.. At this level of coverage, the model predicts an 89% reduction in cholera cases even among the unvaccinated, and a 93% reduction overall in the entire population. These results would apply only where cholera is endemic and population levels of immunity are relatively high. According to our simulations, areas where susceptibility is 2–2.5 as great as Matlab would need to have vaccine coverage of at least 70% to achieve cholera control. Since vaccine-induced protection with current OCV begins to wane after about two years, populations would have to be vaccinated biennially. This could be done in advance of the cholera season for regions that have clear seasonality. Alternatively, environmental predictors of cholera outbreaks could be sampled in regions where such a prediction capacity exists, and then vaccination could take place in advance of expected outbreaks. In environmental studies of cholera outbreak predictors in rural Bangladesh, increases in cholera incidence can be predicted several weeks in advance by water temperature, water depth, rainfall, conductivity, and copepod counts [14]. Such predictors have been developed for Bangladesh and parts of Latin America, but also need to be developed for other regions with substantial risk, especially in Africa. Rapid mass vaccination could take place after such predictors indicate that outbreaks are likely. Further research needs to be carried out on environmental predictors of cholera incidence to make such an assessment reliable. For endemic cholera, population-level immunity is relatively high, making control possible with relatively low vaccine coverage levels. For epidemic cholera, where population level immunity may not be high, rapid vaccination could also be beneficial, but further study would be required to determine the higher coverage levels necessary to obtain substantial indirect protection.

The results of our analysis are based on a mathematical model closely calibrated to cholera incidence data from a large-scale cholera vaccine trial in Matlab, Bangladesh. We would hope that the model captures the dynamics of endemic cholera sufficiently to allow us to evaluate the effectiveness of cholera vaccination at different coverage levels in other settings. However, a major limitation of this research is that the results are only as good as the data the model is based on and the veracity of the modeling assumptions. Verification of the results needs to take place through further community cholera vaccine trials with different coverage levels in a variety of geographic locations. A number of such vaccine trials are currently being planned. The modeling methods presented here can be used to help guide the design, analysis, and interpretation of such trials. Based on the results of such trials, such a model can be used to aid in designing successful cholera vaccination strategies. A second limitation of the model is that we have not explicitly modeled the level of natural immunity in populations with endemic cholera, but rather we alter the per-contact transmission probability to reflect different levels of immunity. This means that we can only approximate the effects of such immunity. More study is needed in this area.

We have not explicitly addressed how to target vaccines during a mass cholera vaccination campaign. In endemic cholera regions, partial immunity tends to increase with increasing age, as partially reflected in higher cholera incidence rates among the young [6,10]. In addition, it could take up to a month to vaccinate large populations. Although we would recommend vaccinating all age groups 2 y and over, with limited quantities of vaccine or time for administration, we would recommend the targeted vaccination of children, especially the very young.

The present analysis implies that mass immunization with killed, whole-cell based oral cholera vaccines could possibly confer a major protective impact against cholera in an endemic setting, even with modest levels of vaccine coverage, due to the combination of direct and indirect vaccine protective effects. It might, however, be questioned whether mass immunization with such vaccines is logistically feasible and affordable. Feasibility of delivery of two-dose regimens of B subunit–killed whole cell or whole cell-only oral cholera vaccines has been demonstrated in a stable refugee setting and a densely populated urban area in sub-Saharan Africa [18,19], as well as in an urban setting in Asia [4]. Although the cost of the B subunit–killed whole cell vaccine, which is produced by an international producer, may currently impede its use in developing country settings, we believe that a low-cost vaccine could be made available in the near future, pending regulatory issues. For example, the cost of purchase and delivery of a two-dose regimen of a currently available oral killed whole cell-only vaccine produced in Vietnam was only US$0.89 in Vietnam [4]. There is currently only one oral cholera vaccine, Dukoral, a B subunit–killed whole cell vaccine, in production and internationally licensed for people 2 y of age and older. This vaccine could be used in mass vaccination campaigns such as that undertaken against epidemic cholera in Mozambique in 2003–2004, with or without targeting of young children. The success of the vaccination campaign in Mozambique against a cholera epidemic in an endemic region demonstrates that cholera could be controlled with a relatively modest financial investment and an organized global surveillance and vaccination program. However, the cholera situation in particular regions should be carefully analyzed and modeled to investigate the coverages and vaccination strategies that would work best.