Gomez-Barroso, Velasco, Varela, Leon, and Cano: Spread of Ebola virus disease based on the density of roads in West Africa

Spread of Ebola virus disease based on the density of roads in West Africa


On March 23rd 2014 the World Health Organization announced that a new Ebola outbreak had appeared in West Africa involving three countries. The objective of this study was to show how a road density index (RDI) could be constructed and a study of its association with Ebola cases during the outbreak. The study was carried out at the district level across the affected countries. RDI was calculated by km2 of territory as a proxy for the mobility of the population. To calculate this index, the number of km of road constructed in each district was estimated and subsequently divided by the area of each district expressed in km2. The accumulated incidence of Ebola was calculated per district. A generalised linear model with a Poisson distribution was used. The RDI varied from 0.12 to 0.84 between the districts. An RDI increase of 0.01 indicates a 3% increase in Ebola infection risk (RR is 1.03; CI 1.03-1.04). The density of the road network can influence the increased incidence of Ebola cases in the affected zone. An exhaustive mapping of the area could help the relevant organisations to manage another outbreak in the future and it could help the distribution of resources in an emergency situation.


Disease caused by the Ebola virus was first reported in 1976 (Bowen et al., 1977), virtually simultaneously in Sudan and what is now the Democratic Republic of Congo. The outbreaks that have occurred since have mostly appeared in remote rural areas of central Africa, resulting in around 2,400 cases according the Centers for Disease Control (CDC), Atlanta, GA, USA (CDC, 2016a, 2016b). On March 23rd 2014, the World Health Organization (WHO) announced that a new Ebola outbreak had occurred in the Republic of Guinea and four days afterwards it was announced that cases had been reported in the capital, Conakry, making it the first large city with Ebola transmission. On 30th March, Liberia announced its first cases and on 25th May Sierra Leone did the same. This was the origin of an unprecedented Ebola epidemic as regards both geographic scope and number of cases (WHO Ebola Response Team, 2014).

Ebola virus disease (EVD), formerly known as Ebola haemorrhagic fever, is a zoonotic infection with a reservoir in bats that can transmit the virus to other animals including humans. The transmission cycle is animal-to-animal, animal to human (spillover event) and human-to-human. Transmission of the virus from one human to another can occur through contact with blood and body fluids of infected people or from the bodies of those having died of Ebola (CDC, 2016a, 2016b). The current outbreak occurred in West Africa, a region characterised by high population densities and elevated population movement levels as much within as across national borders (WHO, 2015a, 2015b). The epidemic spread across almost the whole territory of the three affected countries, including rural and urban areas and also the capital cities (WHO, 2016).

The objectives of this study were to show how a road density index (RDI) could be constructed and used as a proxy of population movements in the region, and which could also be replicated in other regions or countries if the need arises. This Index was based on information about the national road networks and we studied its association with occurrences of cases during the recent EVD outbreak in the Republic of Guinea, Sierra Leone and Liberia.

Materials and Methods

Study area

The study was carried out at the district level across the three affected countries: the Republic of Guinea, Liberia and Sierra Leone. The maps for each country were obtained from the global administrative areas (GADM, 2015) website (http://www.gadm. org/), which is a spatial database of the location of the world’s administrative areas (or administrative boundaries) for use with geographical information systems (GIS) and similar software. The administrative areas in this database are countries and lower level subdivisions, in this case specific districts of the three countries. The administrative areas of Guinea include 33 prefectures and the capital, Conakry. Liberia is divided into 15 regions, while Sierra Leone is divided into 12 districts in addition to the capital, Freetown, itself divided into two zones: the urban west and the rural part. The distribution of administrative areas in the region of study can be found in Figure 1.

Road network and road density index calculation

The spatial information regarding the road network of the three countries was downloaded from Socioeconomic Data and Applications Center (SEDAC, 2013a, 2013b), where road information on a global level may be obtained through an open-source system. This road network includes primary, secondary, tertiary and local roads, as well as urban thoroughfares and tracks. From this information, the RDI was calculated by km2 of territory as a proxy for the mobility of the people resident in the three affected countries. To calculate this index the number of km of road in each district was first estimated and then divided by the area of each district expressed in km2.

Population data Ebola virus disease case information

Population data for 2010 was obtained from the Gridded Population of the World, V.3, from Socioeconomic Data and Applications Center (SEDAC, 2013a, 2013b). This information is available in raster format with a resolution of 0.00833 degrees (30 arc seconds); i.e. this information allowed the extraction of the number of inhabitants for every 0.5 km2 of territory studied. In order to calculate the population in each district, the population of each cell in a district was added up using a geographical calculator.

The EVD cases used for this study were those reported during the outbreak in western Africa, from the start of the outbreak in the first week of 2014 through week 21 (18-24 May) of 2015 (73 weeks in total) in the three affected countries. The case details were obtained from the patient database of WHO’s Global Health Observatory (GHO) (http://www.who.int/gho/en/). The data were available on this web page with weekly updates throughout the outbreak. The data published on 27 May 2015 was used for this study (WHO, 2015a, 2015b) with confirmed and probable cases analysed jointly. The weekly EVD incidence by district and the accumulated incidence were calculated per district during the period of study using the population references. A video with the spatio-temporal evolution of Ebola rates during the outbreak was done.


A spatio-temporal mixed zero inflated negative binomial model was used in order to estimate the association between Ebola risk and RDI. We used this model because the dependent variable, number of Ebola cases, had many zero values. The offset was population and the fixed parameter RDI. The random effects were the district and the weekly number of the cases.

logtijk=β0 + β1RDI11ijk + β2weekjk + β3districtk

Where tijk=number of Ebola cases in the week j and the district k; RDIijk=RDI in the week j and the district k; Weekjk=week j in the district k; Districtk=district k. We used a ArcGIS 10.1 (ESRI, Redlands, CA, USA) software to perform the spatial information and R and INLA packages for statistical analysis.


A total of 20,667 cases were studied (16,723 confirmed and 3,944 probable). By country, the distribution was 3,627 in Guinea, 5,553 in Liberia and 11,487 in Sierra Leone. The video of the Ebola rates (Appendix 1) shows the evolution of rates in time and space during 73 weeks (week 1/2014 to week 21/2015). The rates start to go up in week 28/2014 and started to come down in week 1/2015.

Figure 2 shows the global incidence rates for each district. The highest incidence rates were in Sierra Leone’s Western Urban district (3,989.59 cases/100,000 inhabitants), followed by Monserado in Liberia (563.21 cases/100,000 inhabitants), Port Loko in Sierra Leone (551.25 cases/100,000 inhabitants). After that came the Liberian districts of Margibi (504.47 cases/100,000), Lofa (391.93 cases/100,000), and Bomi (320.04 cases/100,000). In the Republic of Guinea, the highest incidence rate was in the capital Conakry with a rate of 259.52 cases per 100,000 inhabitants.

Figure 3 shows the population distribution in the study area. The population tends to be concentrated in large urban areas with much lower densities in the remaining territory. Table 1 shows the RDI for the districts in each of the three countries studied. The highest RDI was found in Labé with 0.84, a district of the Republic of Guinea. The district with the lowest RDI was Sinoe in Liberia with 0.12. The RDI mean was 0.44 and the median 0.4. Table 1 also shows the number of accumulated cases, population and area for each district. The district with the highest number of accumulated cases was Western Urban in Sierra Leone followed by Monserrado in Liberia.

The results of the spatio-temporal model showed an association between RDI and the rates of Ebola infection. An increase of 0.01 in the density of the road network per km2 resulted in a 3% increase in Ebola infection risk (RR is 1.04; CI 1.01-1.06). Table 2 shows the result of the model, while Figure 4 shows the effect by week of RR of EVD cases adjusted by RDI. The risk increased in the middle of the outbreak when the transmission was higher.


In this study we have investigated the effects of RDI as a proxy for movement between districts and related that to the spread and occurrence of Ebola cases. The results support the hypothesis that the magnitude and area affected by the outbreak is due to a combination of factors, one of which being the magnitude of the number of people living and moving around the affected area. Ebola risk appears to increase 3.7% with each increase of the RDI of 0.01 km/km2.

The region of Africa studied is characterised by high population density, which has increased across the three countries by 200% in the last 50 years with a strong focus on the cities (Alexander et al., 2015). Another distinctive feature of the region is the high mobility levels of the population (Maconachie et al., 2007), which some studies estimate to be seven times higher than any other region globally (Wesolowski et al. 2015). This is potentially linked to the search for improved socio-economic conditions and flight from the armed conflicts that have afflicted the region for a number of decades. Furthermore, our study concerns a region which is highly interconnected with relatively easy movement between villages and cities, including the capitals, and which has substantial cross-border traffic in an area where borders barely exist and where people identify more with the region rather than particular nation states (Médecins Sans Frontières,, 2015). For this reason, this study has considered the three countries jointly as a common area.

In addition to the normally high levels of population movement in the region, the epidemic provided additional reasons for people to move across the region. Patients have moved in search of medical assistance or in order to follow the ancient custom of returning to their birth villages to die and be interred alongside their ancestors. People also moved to care for or visit sick friends or family and to assist with their funerals when needed. On occasion, entire villages have been abandoned by people fleeing the epidemic. Studies carried out during previous Ebola outbreaks showed that less than 1/5 of those who lived with a confirmed or probable case developed the disease, and that secondary cases only occurred when there was close contact with the patient (Mylne et al., 2014). This requirement means that secondary transmission is primarily limited to three scenarios: in hospitals with insufficient protection measures, during traditional funeral ceremonies and in communities where close relatives care for the sick. What distinguishes this outbreak from previous strains of the Ebola virus is that the levels of transmission in the community remained high, while the measures taken to control transmission in hospitals and at funerals limited the transmission of the disease (Faye et al., 2015).

In Africa, the most plentiful source of data about population movements are national censuses (Tatem and Smith, 2010; Garcia et al., 2014) which quantify patterns of movement according to changes of residence over the course of a year and over large spatial areas. Another method of gathering more up-to-date data is the use of mobile phone call data (González et al., 2008; Wesolowski et al., 2014). However, when interpreting the data, it is necessary to take into account the particularities of the region, such as socioeconomic heterogeneity and the effects of gender on ownership and use (Wesolowski et al., 2012). This evidently makes the use of such data a difficult task for the area under study.

The highest infection rates, more than 500 cases for every 100,000 inhabitants, were found in the Western Urban Area of Sierra Leone, where the capital Freetown is located; in Montserrado County, Liberia, which includes the city of Monrovia; and in the Margibi district of Liberia. In the Republic of Guinea, however, the incidence of the disease in the Conakry District – which has the highest population density and includes the capital of the same name – was broadly similar to the rest of the country, but still the highest in the country.

A number of studies attempting to predict the geographical spread of the Ebola virus in West Africa based on the movement of the population have been conducted. One of these was developed by the Spatial Ecology and Epidemiology Group of Oxford University, who constructed a model to identify which areas were at risk of importing Ebola cases and published their results in the form of maps on the Web (http://seeg-oxford.github.io/ebolaspread/) (SEEG-Oxford, 2015). Weslowski et al. (2014) have produced spatial model of movement patterns in 15 countries in West Africa by analysing data obtained from national micro-censuses, telephone calls and spatial data about the population. These models estimate the most important travel routes between different settlements in the region and the relative volume of traffic. This study, however, used historical data without taking into account the changes in behaviour induced by the epidemic.

Valdez et al. (2015) developed a stochastic compartmental model using the movement patterns from Weslowski et al. (2014) with the aim of understanding how the movement of the population affected the propagation of the Ebola virus across the different regions (counties) of Liberia. They concluded that although a reduction in movements slows the propagation of the disease it does not contain it. In previous Ebola outbreaks, control was based on following and monitoring contacts with rapid isolation of suspected cases. This task is potentially much more difficult when the cases occur in multiple locations, especially in densely populated ones. High levels of population movement produce significant problems for controlling the disease: on the one hand, the teams tracking cases are frequently detained at the borders (Médecins Sans Frontières, 2015), while on the other, the improvement of the situation in a given country attracts patients from neighbouring countries looking for treatment centres and thus restarts the chain of transmission.

One of the limitations of this study is that, although information about the road network is fairly complete, there could still be unmapped neighbourhood byways that could be used for movements in between villages. Another limitation is derived from the data itself, which only provide disaggregation at the district level with the consequence that the data can only be studied in aggregate form. To remedy this, complete information about road networks is needed, which is not always available in the rural areas. New technologies, e.g., satellite-generated remote sensing images or application of the global positioning system (GPS) offer, however, a new opportunity to study the roads and the movements of people in close to real time. This approach can contribute to better public health decisions during future Ebola outbreaks.


The density of the road network influences the incidence of Ebola cases in the affected zone. This information should be considered for risk evaluation in a way that would aid the distribution of resources in emergency situations. Finally, an exhaustive mapping of the area could help the relevant organisations to better manage another outbreak in the future.



KA Alexande, CE Sanderson, M Marathe, BL Lewis, CM Rivers, J Shaman, JM Drake, E Lofgren, VM Dato, MC Eisenberg, S Eubank, 2015. What factors might have led to the emergence of Ebola in West Africa? PLoS Negl Trop Dis 9:e0003652.


ET Bowen, G Lloyd, WJ Harris, GS Platt, A Baskerville, EE Vella, 1977. Viral haemorrhagic fever in southern Sudan and northern Zaire. Preliminary studies on the aetiological agent. Lancet 1:571-3.


CDC, 2016a. Ebola virus disease. Communication resources. Virus ecology graphic. Available from: https://www.cdc.gov/ vhf/ebola/resources/virus-ecology.html


CDC, 2016b. Ebola virus disease distribution map. Ebola hemorrhagic fever. Available from: https://www.cdc.gov/vhf/ebola/ outbreaks/history/distribution-map.html


O Faye, PY Boëlle, E Heleze, O Faye, C Loucoubar, N Magassouba, B Soropogui, S Keita, T Gakou, EHI Bah, L Koivogui, AA Sall, S Cauchemez, 2015. Chains of transmission and control of Ebola virus disease in Conakry, Guinea, in 2014: an observational study. Lancet Infect Dis 15:320-6.


GADM, 2015. Database of global administrative areas. Available from: http://www.gadm.org/


AJ Garcia, DK Pindolia, KK Lopiano, AJ Tatem, 2014. Modeling internal migration flows in sub-Saharan Africa using census microdata. Migrat Stud 3:89-110.


MC González, CA Hidalgo, AL Barabási, 2008. Understanding individual human mobility patterns. Nature 453:779-82.


R Maconachie, T Binns, P Tengbe, R Johnson, 2007. Temporary labour migration and sustainable post-conflict return in Sierra Leone. GeoJournal 67:223-40.


Sans Frontières Médecins, 2015. Ebola: Tackling the epidemic across country borders. Available from: http://www.msf.org/en/article/ebola-tackling-epidemic-across-country-borders


A Mylne, OJ Brady, Z Huang, DM Pigott, N Golding, MU Kraemer, SI Hay, 2014. A comprehensive database of the geographic spread of past human Ebola outbreaks. Sci Data 1:140042.


SEDAC, 2013a. Global Roads Open Access Data Set, Version 1 (gROADSv1). Available from: http://sedac.ciesin.columbia.edu/data/set/groads-global-roads-open-access-v1/data-download


SEDAC, 2013b. Gridded population of the world (GPW). Available from: http://sedac.ciesin.columbia.edu/data/collection/gpw-v3


SEEG-Oxford, 2015. Ebola-spread. Available from: http://seegoxford.github.io/ebola-spread/


AJ Tatem, DL Smith, 2010. International population movements and regional Plasmodium falciparum malaria elimination strategies. PNAS 1071:2222–7.


LD Valdez, HHA Rêgo, Braunstein LA StanleyHE, 2015. Predicting the extinction of Ebola spreading in Liberia due to mitigation strategies. Sci Rep 5:12172.


A Wesolowski, N Eagle, AM Noor, RW Snow, CO Buckee, 2012. Heterogeneous mobile phone ownership and usage patterns in Kenya. PLos One 7:e35319.


A Wesolowski, WP O’Meara, N Eagle, AJ Tatem, CO Buckee, 2015. Evaluating spatial interaction models for regional mobility in Sub-Saharan Africa. PLos Comput Biol 11:e1004267.


A Wesolowski, G Stresman, N Eagle, J Stevenson, C Owaga, E Marube, T Bousema, C Drakeley, J Cox, CO Buckee, 2014. Quantifying travel behavior for infectious disease research: a comparison of data from surveys and mobile phones. Sci Rep 4:5678.


A Wesolowski, O Caroline, L Bengtsson, E Wetter, X Lu, AJ Tatem, 2014. Commentary: containing the Ebola outbreak – the potential and challenge of mobile network data. PLos Curr 1:6.


WHO, 2015a. Ebola situation report. Available from: http://apps.who.int/ebola/current-situation/ebola-situationreport-28-october-2015


WHO, 2015b. Factors that contributed to undetected spread of the Ebola virus and impeded rapid containment. Available from: http://www.who.int/csr/disease/ebola/one-year-report/factors/en/


WHO Ebola Response Team, 2014. Ebola virus disease in West Africa. The first 9 months of the epidemic and forward projections. New England J Med 371:1481-95.

Figure 1.

Administrative areas of the study region and road system.

Figure 2.

Global Ebola virus incidence rates during the outbreak.

Figure 3.

Distribution of the population in the studied region.

Figure 4.

Weekly relative risk evolution.

Table 1.

Road density index, km of roads, populations, cases and area by district in the study region.

Country District names Density roads index (km/km2) Km roads Infected cases (n) Population (n) Area surface (km2)
Liberia Bomi 0.51 1093.68 197 61553 2145.85
  Bong 0.36 3020.01 180 177886 8491.07
  Gbapolu 0.15 1442.96 23 73440 9388.39
  Grand Cape Mount 0.28 1398.54 141 214762 5039.07
  GrandBassa 0.33 2531.41 118 248602 7580.91
  GrandGedeh 0.13 1364.97 4 603750 10330.03
  GrandKru 0.17 633.68 22 90666 3717.26
  Lofa 0.27 2758.65 472 120427 10393.15
  Margibi 0.48 1358.5 832 164923 2851.71
  Maryland 0.48 1060.07 6 50512 2205.75
  Montserrado 0.68 1247.57 2685 476731 1839.52
  Nimba 0.30 3609.01 243 277011 12008.77
  River Cess 0.25 1312.16 36 68538 5346.59
  River Gee 0.17 1030.53 12 88356 6236
  Sinoe 0.12 1114.06 36 324498 9457.34
Republic of Guinea Boffa 0.62 3330.11 52 190280 5403.38
  Boké 0.33 3666.40 7 353886 11243.15
  Fria 0.70 1138.21 9 62957 1634.6
  Gaoual 0.24 2810.28 0 164602 11497.44
  Koundara 0.28 1519.42 0 101851 5392.83
  Conarky 0.39 169.57 572 220400 433.94
  Dabola 0.45 2418.09 14 106156 5328.41
  Dinguiraye 0.40 4665.31 1 145548 11668.89
  Faranah 0.40 5230.60 71 169553 13084.78
  Kissidougou 0.58 3618.54 43 182057 6283.78
  Kankan 0.36 6378.30 34 299502 17490.81
  Kérouané 0.48 4520.65 163 274921 9480.91
  Kouroussa 0.34 5461.39 20 177038 16028.3
  Mandiana 0.32 3739.48 0 10 11677.27
  Siguiri 0.35 6242.07 33 310690 17878.05
  Coyah 0.39 520.60 236 847217 1329.46
  Dubréka 0.52 2030.75 145 184279 3921.91
  Forécariah 0.38 1667.69 422 231005 4348.32
  Kindia 0.49 4450.56 115 355263 9051.52
  Télimélé 0.48 3708.75 43 243365 7778.86
  Koubia 0.48 1751.16 0 95928 3623.54
  Labé 0.84 1814.43 0 250915 2159.36
  Lélouma 0.46 1274.49 0 172500 2795.62
  Mali 0.27 2288.24 5 176675 8563.8
  Tougué 0.32 1991.65 2 171709 6129.4
  Dalaba 0.64 2019.82 9 137402 3178.26
  Mamou 0.35 3606.30 0 334786 10441.84
  Pita 0.55 2536.04 8 282633 4630.33
  Beyla 0.49 6165.30 46 213168 12562.89
  Guéckédou 0.58 2479.27 379 368837 4301.38
  Lola 0.38 1711.66 110 197385 4511.67
  Macenta 0.40 3239.01 743 317012 8147.29
  Nzérékoré 0.39 1615.24 255 317439 4092.44
  Yamou 0.27 878.26 12 130993 3259.45
Sierra Leone Kailahun 0.44 1847.68 737 291416 4222.36
  Kenema 0.40 2529.75 535 429137 6339.38
  Kono 0.45 2463.35 545 324811 5475.36
  Bombali 0.35 2966.81 1066 360760 8418.39
  Kambia 0.41 1302.42 248 243651 3169.5
  Koinadugu 0.44 5532.24 162 216104 12649.59
  Port Loko 0.45 2782.60 2144 388932 6116.69
  Tonkolili 0.42 2750.74 610 298065 6510.5
  Bo 0.55 3118.48 423 397460 5707.32
  Bonthe 0.42 1586.81 2 135711 3789.6
  Moyamba 0.45 3191.62 313 253921 7142.72
  Pujehun 0.46 1846.25 35 181658 3979.5
  Western rural 0.44 284.98 1568 646580 654
  Western urban 0.60 50.82 2928 73392 85.29
Table 2.

Spatio-temporal zero inflated multilevel results.

  RR CI low CI upper
RDI 1.037 1.010 1.066
Random effects variance      
    District 0.162    
    Week 1.14    

[i] RR, relative risk; CI, confidence interval; RDI, road density index.

Abstract views:


Article Metrics

Metrics Loading ...

Metrics powered by PLOS ALM

Copyright (c) 2017 Diana Gomez-Barroso, Emilia Velasco, Carmen Varela, Rosa Cano

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
© PAGEPress 2008-2018     -     PAGEPress is a registered trademark property of PAGEPress srl, Italy.     -     VAT: IT02125780185