Background
Global conflict, economic plight, and natural disasters interact to displace people on a large scale [
1]. Over recent years, an unprecedented increase in refugee and internally displaced populations has led to the largest number of displaced-persons ever on record [
2]. Refugee camps are often built in neighboring countries to provide temporary protection and relief for refugees. However, often these crises continue for years or decades and therefore require refugees to live for an indeterminate amount of time in temporary conditions [
3]. Because these refugee camps were established to provide temporary rather than permanent shelter, investments in infrastructure have not been prioritized, leading to overcrowding and poor sanitation, in turn increasing the risk of cholera and other infectious diseases [
4,
5].
Cholera, a waterborne intestinal infection, causes watery diarrhea and is transmitted through fecal contamination of water and food, as well as person-to-person contact [
6]. While the case fatality rate of cholera is low when treated, it can be up to 50% when untreated [
7]. Approaches to reduce cholera spread include improvements in water, sanitation and hygiene (WaSH). However, political and economic hurdles can make these longer-term, larger-scale improvements challenging. Vaccination can be used as an complementary approach that results in a substantial reduction in cholera transmission [
8,
9]. Further, cholera vaccines can induce herd protection, thereby reducing the risk of disease for both the vaccinated and unvaccinated segments of the population [
9,
10]. Although, the effects of vaccination may attenuate over time, recent data has shown that protection remains steady several years after vaccine is administered [
11]. Generally, vaccination campaigns can be implemented in conjunction with other health interventions or while more permanent preventative measures (e.g. WaSH) are being put in place.
Maela refugee camp is in northwest Thailand, 8 km east of Burma (Myanmar). In December 2009, its population was 40,009 individuals, who were mostly Burmese refugees. The distribution of ethnic groups across all Thai/Burmese border refugee camps was quite diverse, with ∼61
% being of Karen origin [
12]. Although cholera has been reported in Thailand in past years [
13], a review of the literature between 1982 and 2007 found only 860 cases reported in Thailand (population of over 60 million), with the majority occurring in the northern part of the country [
14]. While we do not have a denominator to determine an attack rate for these data, it is clear that compared to the outbreaks reported in the literature, the burden of cholera in Maela had been higher, with more than 1000 cases between 2005 and 2010 among a population of less than 50,000 [
15]. In Maela, the majority of individuals have access water and sanitation facilities; however, sociopolitical issues and the mountainous terrain prevent its maintenance or improvement. Therefore, a vaccination campaign was implemented as a critical addition to existing efforts to reduce cholera transmission in Maela. This campaign was the first use of the oral cholera vaccine (OCV) Shanchol in a stable refugee camp [
16].
An oral cholera vaccine, Shanchol, was prequalified by the World Health Organization (WHO) in 2011 [
17]. The vaccine is administered in two doses, 14 days apart. Efficacy estimates from randomized control trials for the full two-dose series vary by setting and age. A large-scale age-adjusted trial conducted in India found a two-dose efficacy of 65% after follow-up at 5 years [
11]. Shorter-term estimates from observational studies have found vaccine effectiveness to be as high as 86.6% after a follow up of 6 months in Guinea [
18]. Notably, effectiveness of a reactive vaccination campaign in Haiti was quite close to the efficacy estimate at 63% among individuals self-reporting vaccination [
19]. One-dose efficacy was estimated in a randomized trial in Bangladesh and was found to be 52% after 2 years follow up [
20]. Observational studies have estimated one-dose vaccine effectiveness to be between 32.5% [
21] and 87.3% [
9], though the lower bound was not found to be significant. Despite logistical challenges, cholera vaccination in refugee camps has been found to be feasible and acceptable [
16,
22,
23] and has been recommended as a potential key intervention to prevent and control cholera transmission in these settings.
In recent years, mathematical modeling has emerged as a useful tool in examining counterfactuals, evaluating intervention strategies, and assisting in policy decision-making [
24]. Modeling of cholera transmission, in particular, has been used to guide policy and planning decisions. For example, the US Centers for Disease Control and Prevention (CDC) used real-time modeling to predict the effects of vaccination during the 2010 cholera epidemic in Haiti and to anticipate the total numbers of cases and hospitalizations [
25]. More broadly, a wide range of cholera transmission models have been developed [
26‐
31], accounting for different mechanisms, including spatial dynamics [
29,
32,
33], age structure [
34,
35], environmental drivers [
36‐
39], and disease transmission characteristics such as proportion of asymptomatic individuals [
38], hyperinfectiousness [
40], dose response effects [
27,
29,
41,
42], and multiple transmission pathways [
28,
30]. One particularly relevant modeling analysis examined the impact of one compared with two doses of cholera vaccine in Haiti, Zimbabwe, and Guinea when supplies are limited [
43]. Another used an agent-based model to examine cholera transmission in a refugee camp setting [
44], and one mathematical modeling analysis included displaced-persons to examine cholera dynamics and rainfall in Haiti using a model accounting for environmental transmission [
39]. Finally, another used compartmental models to examine cholera transmission and vaccination–induced herd immunity among displaced persons in South Sudan [
45]. These models have been useful in explaining different drivers of transmission across settings and evaluating proposed interventions. Modeling of cholera vaccination in displaced person camp settings is growing area of research in need of more attention. A wide range of characteristics such as overcrowding, high levels of population mixing, limited access to clean water and medical care, and political barriers to different interventions make it necessary to develop specific models fit to real-world data to assess the effects of vaccination using population and outbreak data directly from displaced-person settings. These models can provide a useful additional evidence-base to help policymakers make decisions.
In this study, we used the Susceptible-Infectious-Water-Recovered (SIWR) modeling framework [
28], expanded to include two-stage vaccination and an adult/child age structure. The SIWR modeling framework accounts for both indirect transmission through environmental water sources as well as a direct pathway representing transmission through food, household water sources, and person-to-person contact. This model is an extension of the classical SIR model (with Susceptible, Infectious, and Recovered compartments to track the total number of individuals at different stages of the disease), with an additional water or environmental compartment representing the concentration of pathogen. The SIWR model has been applied to a range of cholera outbreaks as well as several theoretical studies [
28,
29,
32,
42,
46]. It has been integrated with a gravity model to determine how distance and population sizes affect the spread of cholera in Haiti [
32] and it was used to estimate the basic reproduction number (
\(\mathcal {R}_{0}\)) in a range of settings [
29,
30,
32]. Finally, age-structured SIWR models [
34,
47,
48] have been used to represent different transmission rates by demographic group.
The primary goal for this study was to develop a transmission model to evaluate a recent vaccine campaign in Maela refugee camp, and use this model to plan for future vaccine campaigns. The Maela refugee camp study setting is particularly relevant given the recent worldwide increase in total numbers of refugees—the approach considered here can be generalized to inform vaccine campaign planning in a wide range of contexts. Because our modeling analysis coincided with preparation and implementation of the cholera vaccination campaign in Maela, we had the opportunity to build and expand our model iteratively based on the vaccination campaign results and implementation. Our model was thus used real-time to help predict the outcome of the campaign, to forecast whether any cholera outbreaks would occur in the near future, and to evaluate the necessity of administering a booster campaign for the years following the campaign. This study illustrates how mathematical modeling can be used iteratively in intervention planning to inform policy and intervention decision-making.
Discussion
Using a mathematical model of cholera transmission in a refugee camp, we have shown the dramatic effect that vaccination can have on reducing the risk of cholera outbreaks in refugee settings. Our analyses suggest that pre-vaccination campaigns, even if not fully covering the whole camp population, can lead to substantial reductions in the number of observed cases in the event of an outbreak. Of course, the best pre-vaccination strategy depends on both one-dose effectiveness and the amount of doses available. If resources are limited and one-dose offers acceptable protection, single dose-based strategies would be preferred. But if there are enough vaccine doses available or the effectiveness of a single dose is questionable, then full vaccination strategies are better. We also found that reactive vaccination strategies can result in a moderate reduction in the number of infections. Further, the small differences in attack rates in our model results suggest that, at the best-fit parameters, vaccinating more individuals with at least one dose (even if the full vaccination sequence is not completed) may be nearly or more effective than vaccinating fewer individuals with a complete two-dose program assuming conservative vaccine effectiveness estimates. This consideration is particularly important if vaccination occurs later in the outbreak, where time is of the essence as the outbreak is already ongoing—the full two-dose sequence requires a delay between doses, hindering its effectiveness in reactive vaccination campaigns that begin later in an epidemic. Additionally, if logistical constraints limit the ability to follow up and provide patients with a second dose, a one-dose strategy may be preferable. However, the relative one vs. two-dose effectiveness, timing and number of doses available should all be considered. Finally, our projections of future outbreaks after the 2013 vaccination campaign in the Maela refugee camp suggest that vaccination may have prevented outbreaks in 2013 and 2014, as no cases were observed in either year, both in the model and during follow-up in the camp. This ability to consider counterfactual scenarios and generate projections highlights the potential of modeling to help guide real-time public health decision-making.
Our mathematical model of cholera transmission reproduces the dynamics observed in Maela’s 2010 cholera outbreak (Fig.
2). We conducted a theoretical exploration of the dynamics of our model to examine different vaccination scenarios with only 20,000 doses of Shanchol distributed in the camp to reflect potential public health intervention strategies. Doses were distributed proportionally among children and adults. These results provide insight into the most effective strategies for vaccination when logistics might impede complete coverage of a population with one or two doses. We found that the two-dose pre-vaccination strategy was marginally the most effective with an attack rate of 5.5 cases per 1000 people, but others provided comparable protection. On the other hand, the most effective reactive vaccination scenarios are the one- and two-dose strategies with attack rates of 5.9 cases per 1000 people (Table
3), while again others provided similar protection. All similarities between scenarios are the result of the conservative effectiveness estimates we used such that one-dose effectiveness is nearly exactly half of two-dose effectiveness.
Cholera vaccines provide protection for individuals who do not directly receive vaccination [
9,
10]. As with many transmission models, our analysis implicitly accounted for indirect protection (herd immunity). Specifically, the direct vaccination of individuals in the model reduces transmission in the population which in turn provides indirect effects protecting nonvaccinated individuals from becoming infected. To illustrate this, we calculated the number of cases that occur among nonvaccinated individuals in the two-dose pre-vaccination strategy with the VE used in our analysis and compared this to the number of cases that occur among nonvaccinated individuals in the same scenario when the VE was set to 0%. We found that when we set the VE to 0%, 308.4 cases occurred among nonvaccinated individuals, but when the VE was set to our values from the literature (see Table
1), only 220.9 cases occurred among nonvaccinated individuals. Thus, our model shows a herd protection effect for individuals who did not receive cholera vaccine.
For pre-vaccination, all scenarios can achieve large reductions in case counts. If one-dose effectiveness is low or uncertain, the two-dose scenario is preferable since it can guarantee larger or more certain reductions in case counts i.e., 44,000 doses (administered to 22,000 people) are sufficient to achieve <50 cumulative cases. On the other hand, if one-dose effectiveness is only ∼50
% that of two-dose effectiveness, 45,000 doses (administered to 45,000 people) are sufficient (see Fig.
5 for details). In the reactive vaccination scenarios, two doses may be preferable if one-dose effectiveness is low or uncertain. Further, we see a crucial interplay between timing and vaccine effectiveness on transmission. Past a certain point in the outbreak, one-dose effectiveness does not substantially change the total case counts. For instance at day 50 in the one-dose scenario, the largest potential difference in cases is 65. (total cases: 395.3 and 333.2 for a one-dose effectiveness of 0% and 63%, respectively) while at day 20 the largest potential difference in cases is 202.2 (total cases: 395.5 and 202.3 for a one-dose effectiveness of 0% and 63%, respectively). Additionally, when one-dose effectiveness is sufficiently high, the one-dose strategy is preferable, with one-dose able to achieve the lowest case counts during an idealized situation in which there is no delay in reactive vaccination administration and the one-dose effectiveness is nearly equal to the two-dose effectiveness. Although this scenario may not be possible in real-world settings, it underscores the necessity of considering delays and relative one vs. two-dose effectiveness in reactive vaccination campaigns. For the one-dose scenario considered here, more individuals are given vaccine over an 8 day campaign resulting in more people having some protection early in the outbreak. On the other hand, in the two-dose scenario individuals who have been vaccinated have a higher level of protection, but achieving the same amount of population level protection will take a longer period of time i.e., two 4 day campaigns occurring 14 days apart (see Fig.
7 for details). The results of the forecasting analysis for 2013 show that regardless of population level immunity, the OCV campaign using a partially mixed strategy (coverage levels shown in Additional file
1: Table A4) prevents a majority of outbreaks that might otherwise have occurred, shown in Fig.
8 and Additional file
1: Figure A8. Additionally, even if we assume a fully susceptible population prior to the 2013 OCV campaign, the vast majority of post-OCV campaign runs in 2014 still result in no outbreak. Indeed, the median of total cases across all runs is <1 (see Table
4 and Additional file
1: Figure A8). This suggests that an introduction of cholera would likely not have resulted in a significant outbreak even given a conservative assumption about the population level of immunity. Thus, it was determined that there was no need for a booster campaign in Maela (and indeed there was no cholera outbreak that year).
A key limitation of our model is the uncertainty in parameter values. Because we fit our model to Maela data from 2010, certain parameter values (e.g., birth and death rates) might not be completely representative of Maela in 2013 and 2014. However, to assess the uncertainty in our parameter values, we conducted global sensitivity analyses using LHS and assessed both practical and structural identifiability of the model. The quantitative results are heavily dependant on the vaccine effectiveness estimates however, we chose the lower bounds of values from recent studies to ensure that the model generated conservative results. Furthermore, the qualitative conclusions i.e. the interplay between vaccine timing and vaccine effectiveness will occur for most realistic effectiveness estimates. We also examined the effects of reducing child VE estimates and found that this did not substantially change any of our conclusions. Other weaknesses of this analysis are inherent in the model assumptions. For example, one key assumption for the vaccination scenario simulations is that there is no waning immunity because we are simulating over a short time course. We do, however, incorporate waning immunity into the forecasting analysis. Another assumption is that each infectious individual is equally infectious regardless of previous vaccination, cholera exposure, or time since infection. In any case, this assumption would result in an overestimate of the total number of cases because we are ignoring the fact that individuals who have been vaccinated might be less infectious. We are also ignoring hyperinfectiousness, as incorporating this into our model as including it would require tracking of pathogen through the human host and a shorter time scale because it decays after 18 hours [
78]. Although we are not explicitly modeling asymptomatic infections or errors in disease reporting, these are accounted for in the scaling factors,
ka for adults and
kc for children. Additionally, in our analyses we used a deterministic model which will not capture stochastic fluctuations. Stochasticity may play a role, particularly in the early phase of an outbreak following a new introduction of cholera—our model may therefore miss some of the stochastic die-out of epidemics in our forecasts, which again, would lead to an overestimate of the number of cases. We assume that the mortality rate for individuals with cholera is the same as that for individuals without cholera. Although this is likely not true, the data we fit our model to did not have a sufficient number of deaths to calculate case fatality rates for diseased compared with non-diseased population groups (there was only one death among individuals with cholera). As individuals in Maela have easy access to medical care, they are likely to be treated and have a high rate of survival. For the sake of parsimony [
79], we are also assuming that human-human transmission parameters are equal across demographic groups and separately human-water transmission parameters are equal since we obtained similar fits for a range of
β values in the practical identifiability analysis (see Supporting Information). We are using data from a specific setting, which may limit the external validity of our results; however, the qualitative dynamic results obtained should be consistent for other refugee camp settings and have been seen in similar mathematical modeling analyses [
43,
80].
The strengths of this model include the novelty of considering vaccination scenarios while explicitly accounting for environmental transmission of cholera in a refugee camp. Additionally, our identifiability and sensitivity analyses methodically considered parameter uncertainty. Another strength is the fact that we used real-world data to inform our model which in turn, provided insight for the public health response. Specifically, we liaised with the CDC as well as local nongovernmental organizations in real-time and used data from the OCV campaign to directly inform the model. Further, the model was then used to explore counterfactual forecasting scenarios to help answer outstanding questions among trial investigators about whether or not a booster campaign was necessary. Finally, the consistency of our results with other analyses indicates the robustness of our findings.
In general, our results indicate that vaccination should be considered in conjunction with WaSH with the caveat that immunity may wane over time. Furthermore, the trade-off between vaccine effectiveness (i.e., one-dose compared with two-dose) and timing of reactive vaccination should be carefully considered. The WHO currently holds a stockpile of over 3 million doses of oral cholera vaccine to allow countries or institutions to request doses of vaccine during cholera outbreaks. The average time from when requests were approved to receipt in country has been 14.4 days with an additional 9.5 days until vaccination actually started [
81,
82]. As seen in our vaccination scenario results, timing is crucial to the impact of vaccination. Recent mathematical modeling work has examined how best to allocate global stockpile reserves [
83].
Ideally, pre-vaccination should be considered as a short-term transmission reduction strategy, compared to the potentially longer-lasting effects of improved WaSH. A recent study used a static model fit to data from Malawi to estimate cases averted in Haiti by implementation of oral cholera vaccine and/or WaSH and found that a combined implementation of WaSH and vaccine resulted in the greatest reductions in cases [
84]. Furthermore, WaSH reduces transmission for a wide range of infectious diseases and if maintained is more permanent, while Shanchol targets cholera, and its effects do not last as long—although vaccine campaigns may be easier to implement, as they do not require sustained maintenance. Thus, both WaSH and vaccination may have their roles to play in an effective intervention strategy.
Our analyses suggest that vaccination provides an effective strategy for preventing cholera outbreaks in refugee camps and that cholera vaccination should be considered, even in the absence of an ongoing outbreaks. Given the dramatic increases in displaced populations and refugee settlements across the world it is critical that vaccination be considered with water sanitation and hygiene improvements. If a camp is facing an outbreak, delayed distribution of vaccines can substantially alter the effectiveness of a reactive vaccine campaign, suggesting that quick distribution of vaccines (e.g., using a first come, first served approach) may be more important than ensuring that every individual gets both vaccine doses.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.