Potential for broad-scale transmission of Ebola virus disease during the West Africa crisis: lessons for the Global Health security agenda

Background The 2014–2016 Ebola crisis in West Africa had approximately eight times as many reported deaths as the sum of all previous Ebola outbreaks. The outbreak magnitude and occurrence of multiple Ebola cases in at least seven countries beyond Liberia, Sierra Leone, and Guinea, hinted at the possibility of broad-scale transmission of Ebola. Main text Using a modeling tool developed by the US Centers for Disease Control and Prevention during the Ebola outbreak, we estimated the number of Ebola cases that might have occurred had the disease spread beyond the three countries in West Africa to cities in other countries at high risk for disease transmission (based on late 2014 air travel patterns). We estimated Ebola cases in three scenarios: a delayed response, a Liberia-like response, and a fast response scenario. Based on our estimates of the number of Ebola cases that could have occurred had Ebola spread to other countries beyond the West African foci, we emphasize the need for improved levels of preparedness and response to public health threats, which is the goal of the Global Health Security Agenda. Our estimates suggest that Ebola could have potentially spread widely beyond the West Africa foci, had local and international health workers and organizations not committed to a major response effort. Our results underscore the importance of rapid detection and initiation of an effective, organized response, and the challenges faced by countries with limited public health systems. Actionable lessons for strengthening local public health systems in countries at high risk of disease transmission include increasing health personnel, bolstering primary and critical healthcare facilities, developing public health infrastructure (e.g. laboratory capacity), and improving disease surveillance. With stronger local public health systems infectious disease outbreaks would still occur, but their rapid escalation would be considerably less likely, minimizing the impact of public health threats such as Ebola. Conclusions The Ebola outbreak could have potentially spread to other countries, where limited public health surveillance and response capabilities may have resulted in additional foci. Health security requires robust local health systems that can rapidly detect and effectively respond to an infectious disease outbreak. Electronic supplementary material The online version of this article (10.1186/s40249-017-0373-4) contains supplementary material, which is available to authorized users.

. Distribution of Ebola virus incubation period (from exposure to symptoms), by days of incubation Notes: Adapted from Legrand et al. [3] and Eichner et al. [4]. *Frequency related to the number of patients out of a total of 5,000. Source: Meltzer et al. [1]

Model characteristics
Progression only: EbolaResponse uses specific features to derive the number of Ebola virus disease (EVD) cases. A patient can only progress forward through the model (e.g., can never go from incubating (I) back to susceptible (S)), and no patient can skip a state (e.g., go from incubating to recovered, skipping infectious).
Community size: the model uses a community size equivalent to the total population in the city, and assumes that all the population is susceptible to the disease.
Incubation period: we adapted the probability distribution data from Legrand et al. [3] and Eichner et al. [4] to generate a lognormal probability distribution of Ebola incubation ( Figure S1). The model uses a mean incubation period of 6.3 days, a median of 5.5 days, and a standard deviation (SD) of 3.31 days.
Chowell et al. [2] estimated mean incubation periods of 5.30 (SD 0.23) and 3.35 (SD 0.49) days based on data from EVD outbreaks in the Democratic Republic of the Congo (formerly Zaire) in 1995 4 and in Uganda in 2000. These estimated incubation periods are lower than other estimates, such as Legrand et al. [3] and Eichner et al. [4]. These differences may be partly attributed to different virus subtypes [4]. WHO estimated a mean incubation period for the first 9 months of the West African outbreak of 11.4 days, with an upper limit of 21 days [5] and, more recently, Chowell estimated mean incubation period of 12.7 days and mean infectious period of 6.5 days [6]. EbolaResponse allows adjusting the probability distribution to almost any structure desired, with an upper limit of 25 days incubation. While the specific probability distribution used for incubation periods affected the point estimates of the results, they do not affect the overall conclusions of the report.
Infectious period: Several studies of EVD outbreaks have estimated the infectious period to range between 6 to days approximately. The WHO Ebola Response Team estimated the interval from symptoms onset to death was 7.5 days by September 2014 [5] and 8.2 days by the end of November 2014 [7]. A detailed study of chains of transmission between February and August 2014, estimated an infectious period of 8.9 days [8]. A recent parameter review of all EVD epidemics found that the mean time from symptom onset to death in all previous epidemics ranged from 6 to 10.1 days [9], and a modeling study for previous outbreaks estimated 6.5 days as the infectious period [10]. Because the purpose of our estimate is to illustrate what could have happened had the Ebola outbreak of West Africa expanded beyond the three mostly affected countries, we used the best estimate from WHO for the West Africa outbreak, i.e., an infectious period of 8 days [7]. We found no data reporting measurement of changes in the risk of onward transmission over the duration of fulminate illness. It would be conceivable, therefore, that such risk does change as a patient becomes sicker and requires more and more care. We assumed that the risk of onward transmission (infection) from patient to susceptible was equal throughout the 8-day period. Last, burial practices in the region show potential risk of EVD transmission during a traditional burial due to possible contact with a victim's body fluids [11]. Safe 5 burial practices were considered as a component of the safe community isolation intervention in our model (see section on distribution of patient by category over time).
Population governor: EbolaResponse includes a population "governor" that prevents the model from calculating more cases than the inputted population. This overestimate could happen if one assumes that most of the patients remain in the "not effective home isolation," which has the highest risk of onward disease transmission (Table S1).
The population governor was programmed by simply reducing the daily estimate of the persons newly infected proportionate to the cumulative reduction in the susceptible population, as follows: Factor to reduce estimate of newly infected at day t = [Model populationcumulative total of newly infected up to day (t-1)] / model population.
This governor reduces the effective number of persons infected daily (i.e., effectively lowers the risk of transmission inputs shown in Table S1). With "large populations," this governor is unlikely to affect the calculations. The "governor" only begins to appreciably affect estimates (i.e., reduce them) when approximately 40% -50% of the population have become infected.
Distribution of patient by category over time: The model splits the patients who have become symptomatic [12,13] into three categories of isolation: (1) hospitalized, (2) effective home isolation, and (3) no effective home isolation. These three categories reflect the ability, or risk, to transmit Ebola onward. The distribution of patients into these categories affects the overall progress of the epidemic.
The more patients in the categories "hospitalized," and "effective home isolation", the slower the 6 progress of the epidemic (because these two categories have transmission rates less than 1 person infected per infectious person). It is possible that a proportion of patients in the "effective home isolation" and the "no effective home isolation" scenarios would end up in the hospital; however, we have assumed that they go so late in the progression of disease as to make no notable change in the risk of onward transmission.  [3]. ‡When these values remain at less than one person infected per infectious person, the epidemic eventually ends. The EbolaResponse modelling tool uses the shown values to fit the model to the data, assuming 6 days of infectiousness. §This patient category refers to patients at home or in a community setting such that there is a reduced risk for disease transmission (including safe burial when needed). Source: Meltzer et al. [1].

Growth scenarios based on transmission patterns
To estimate the potential broad-scale transmission of Ebola, we modeled three growth scenarios based on transmission patterns observed in Liberia during the 2014-15 EVD epidemic ( Figure S2) [14].
For all scenarios we assumed that within the first week of case detection, 10% of the EVD cases would be hospitalized or effectively isolated, as was estimated for the EbolaResponse tool based on reported epidemiological data [1,14]. We used the EbolaResponse tool to estimate the total number of EVD cases in each of the three scenarios using the parameters summarized in Table S2. 1) Liberia-like scenario. Was based on the compartmentalization of Ebola cases that was fitted to data collected in Liberia during the 2014-2015 Ebola epidemic. We assumed a 5-6 percentage points increase per week in the number of cases hospitalized or effectively isolated during weeks one through 11, and a two percentage increase per week during weeks 12 through 16. This resulted in a total of 66% of cases being effectively isolated by week 15.
2) Delayed-response scenario. Based on the assumption that the implementation of control measures would proceed slower than observed in Liberia and ultimately reach a smaller proportion of the population, we assumed a 1.5 percentage point increase per week in the number of cases hospitalized or effectively isolated during weeks one through three and between 2-4 percentage point increases per week during weeks four through 16. The final proportion of Ebola cases in effective isolation was 50% at the end of week 16.
3) Fast-response scenario. Based on the assumption that control measures would be implemented more quickly compared to Liberia and ultimately reached a larger proportion of the population. 8 We assumed a 10-percentage point increase per week in the number of cases hospitalized or effectively isolated during weeks one through four, seven percentage point increase per week during weeks five through seven, and a four percentage point increase per week in weeks eight through 12. The final proportion of cases in effective isolation leveled off at 81% at week 13.  Notes: * The estimated implementation of control measures to prevent disease in the baseline Liberia-like scenario was fitted to reported Ebola cases beginning on day 91 of the outbreak, because the data suggests that a large, coordinated response began around this day.

Estimating the number of Ebola cases
We selected 21 cities with both high volumes of air traffic from West Africa and a high percentage of the city's population living in slums for modeling (Table S3 shows the main characteristics of the cities chosen; air traffic is shown in Appendix 3). To calculate the number of projected cases in each city, we assigned a proportion of patients to either "hospitalized", "effective community isolation" or "no effective isolation" for each of the growth scenarios outlined above. Using the EbolaResponse model [1], we input each city's population and assumed either 10 or 100 cases occur before detection and initiation of an effective response (hereafter seeding). We estimated low and high case count estimates based on number of cases that occur before detection and initiation of an effective response (hereafter seeding; low case count: 10 cases; high case count: 100 cases). The difference in speed of initiation of an effective response could be due to multiple causes, including the speed of outbreak detection and the speed with which resources can be gathered and deployed. The total number of new cases was then added over 120 days to produce the total cumulative case counts. Bed capacity and diagnostic capabilities are an important limitation to the effectiveness of a response to Ebola [15,16].
Because it was calibrated with data from the field, the Liberia curve considers these variables as limiting factors in our isolation proportions. The delayed and fast response scenarios thus assume a capacity to expand bed capacity that is slower and faster than Liberia. Because our estimates are only meant to illustrate what could have happened, we are assuming that countries in each category of response could achieve similar capabilities in short time. Notes: * Population density was estimated for each city [17]. † Lower-middle-inc. denotes lower-middle-income country, and upper-middle-inc. denotes upper-middle-income country, as classified by the World Bank [18]. ‡Estimates in US$ at average exchange rate in nominal terms (current prices). National Health Accounts are not maintained or updated by all countries; estimates of health expenditures may have been estimated through technical contacts from the country or public documents and reports [19]. §The population living in urban slums is reported at the national level [20]; to derive an estimate for the city we assumed that population living in slums is proportional to population size.

13
There is substantial uncertainty in our estimates for a hypothetical broad-scale transmission of Ebola. We have no data to accurately predict the capacity of countries that were at high risk of Ebola transmission, as determined by air travel volume, of detecting and initiating an effective response to prevent or slow down an Ebola outbreak. Country's response capabilities are complex, and affected by a range of factors, as we discuss in section 2.3.2. The plausibility and uncertainty of our estimates may be well illustrated by the case of the Ebola outbreak in Nigeria.
Nigeria was quick to control an outbreak of Ebola that originated from a single infected traveler who flew from Liberia to Lagos in July 2014, which provides a good example for discussion [21][22][23]. In response to the rapid detection of Ebola, the Nigerian government, in collaboration with CDC and other partners, created an incident management system largely using staff from the Nigerian Polio Eradication Program and support from the Bill and Melinda Gates Foundation. Nigeria rapidly initiated a series of effective outbreak response measures, including training of healthcare workers, contact tracing, household visits, effective isolation of infectious patients, airport screening, and the creation of an Emergency Treatment Unit in two weeks. All these efforts by a well-staffed and prepared health workforce, which included support from the Polio Eradication Program, resulted in quickly controlling the Ebola outbreak. In contrast, the outbreak in Guinea took several weeks to be detected, and several more weeks before an effective response was put in place [24,25]. The index case in Nigeria collapsed upon arrival to the airport in Lagos, which resulted in a very early detection of Ebola infection. That, plus the efforts from the local workforce of diversion of resources from the polio program resulted in a total of 19 cases.
For Nigeria (a lower-middle income country), if we considered a "Liberia-like" Ebola transmission and response and that only 10 cases of Ebola had occurred before outbreak detection and initiation of an effective response, our model results in an expected total of 627 cases of Ebola.
14 However, partially because Nigeria had public health capabilities from the Nigerian Polio Eradication Program and support from the Bill and Melinda Gates Foundation and thus quickly available resources and a trained health workforce, they put together a fast response. If we put these data into our model, assuming that only one Ebola case occurred before detection and effective response (as happened in reality) and that Nigeria had a "fast" response, we would expect a total of 28 Ebola cases, which is comparable with the 19 Ebola cases reported in Nigeria in a somewhat unique scenario. Our estimate is also comparable to a model based on the days before the intervention, assuming 12 exposed individuals from an index case [21].

Estimated number of Ebola cases using city-specific weights
To account for differences in living conditions between Monrovia, Liberia, and the cities used in the analyses, we performed two additional sets of analyses. We weighted the estimated number of Ebola cases that would occur using the ratio of: 1) each city's population density (pop/sq. mile) to Monrovia's population density, and 2) the ratio of the national proportion of the population living in slums in each country to the proportion of the population living in slums in Liberia (see main text for results).
We estimated the correlation coefficient between these two variables to check for co-linearity. Table S5 shows the Pearson's R correlation coefficients for the following variables related to sociodemographic conditions within the cities selected for analysis: country population [19], total urban population living in slums [20], gross domestic product (GDP) per capita [26], World Bank economic category [18], population density [17], and proportion of the country's population living in urban slums  Notes: *Country population corresponds to estimates for year 2013 [19]. † Population in slums shows the estimated total urban population living in slums; the estimates were obtained at the country level from modelling [20]. ‡ GDP per capita denotes the gross domestic product per person, values in 2015 US dollars [26].
§ Each year on July 1, the World Bank revises its analytical classification of the world's economies based on estimates of gross national income (GNI) per capita for the previous year [18]. ¶ Pop. in slums denotes the proportion of the country's population living in urban slums (compared to total population in the country) in percentage [19,20].
Last, for ease of comparison, Table S6 Figure 2 of the main text. * Adjustments based on population density and population in slums are included to account for the differences in living conditions between Monrovia, Liberia, and the cities used in the analyses, to account for sociodemographic factors in the risk of Ebola transmission [11,[27][28][29][30][31][32].
† Each year on July 1, the World Bank revises analytical classification of the world's economies based on estimates of gross national income (GNI) per capita for the previous year [18]. ‡ The number of cases for each estimate are based in the economic category of the country. We assumed that low-income countries would have a delayed implementation of control measures to prevent Ebola spread, lower-middle income countries would have a Liberia-like implementation of control measures, and upper-middle and high-income would have a relatively fast implementation of control measures. § Low and high case count estimates based on number of cases that occur before detection and initiation of an effective response (low: 10 cases; high: 100 cases).

Potential broad-scale transmission of Ebola cases to major cities in within five selected countries
To illustrate the potential spread of Ebola within countries once a case had been imported into the main city, we selected five countries: Nigeria, Ethiopia, Kenya, South Africa and India. For each country we projected the number of EVD cases in major urban centers (populations of 100,000 or more).
We chose these urban centers based on population size and travel access to the country's major urban area, either through air travel or that were located along major highways. The number of cases represents possible scenarios of intervention, and we used the same criteria as before: delayed transmission for lower income countries, Liberia-like transmission for lower-middle income countries, and fast transmission for upper income countries. Table S7 shows the estimated number of Ebola cases at four months (120 days) based on national resources and response scenarios, by city in selected countries (column A shows the baseline estimates, column B shows estimates weighted by population density). By design, the projected number of cases depends on the speed with which we assume that effective control measures are implemented. With our current assumptions, early detection of cases and rapid initiation of control measures is particularly important in the major cities of low-income countries (Ethiopia).  Figure 2 of the main text. The highlighted areas in the table represent most likely response scenario based on the country's economic category. Lower-middle-inc. denotes lower-middle-income country, and upper-middle-inc. denotes upper-middle-income country, as classified by the World Bank. * Each year on July 1, the World Bank revises analytical classification of the world's economies based on estimates of gross national income (GNI) per capita for the previous year [18]. † Sources of data: CIA Factbook [33], Guangzhou population [34], Kenya 2009 Census [35], India 2011 census [36], South Africa 2011 Census [37], National Population Commission, Nigeria [38], Demographia [17]. ‡ Ratio of city population density to population density of Monrovia, Liberia [17]. Population densities from Eldoret, Kisumu, and Nakuru were not available; the numbers reflect the average population density in Nairobi and Mombasa, Kenya. § Low and high case count estimates based on number of cases that occur before detection and initiation of an effective response (low: 10 cases; high: 100 cases). Table S8 and Figure S3 show the air-traveler volumes from Sierra Leone, Guinea, and Liberia to the top 35 destination cities in the world, from July 2014 through December 2014. The table shows traveler volumes from each country to the main destination cities and the aggregate volume. We did not include travelers between Guinea, Liberia, and Sierra Leone. Data was obtained from BlueDot (Dr.

Appendix S3. Travel data
Kamran Khan's database) which stratifies the arrivals by metro areas/cities [39]. Similar data was used to model the potential risk for international dissemination of EVD during the 2014-2015 outbreak in West Africa, and expected travelers infected with Ebola departing from Sierra Leone, Guinea, and Liberia [40]. We also included Addis Ababa (Ethiopia), Ouagadougou (Burkina Faso), and Wuhan