Dataset: 11.1K articles from the COVID-19 Open Research Dataset (PMC Open Access subset)
All articles are made available under a Creative Commons or similar license. Specific licensing information for individual articles can be found in the PMC source and CORD-19 metadata
More datasets: Wikipedia | CORD-19

Logo Beuth University of Applied Sciences Berlin

Made by DATEXIS (Data Science and Text-based Information Systems) at Beuth University of Applied Sciences Berlin

Deep Learning Technology: Sebastian Arnold, Betty van Aken, Paul Grundmann, Felix A. Gers and Alexander Löser. Learning Contextualized Document Representations for Healthcare Answer Retrieval. The Web Conference 2020 (WWW'20)

Funded by The Federal Ministry for Economic Affairs and Energy; Grant: 01MD19013D, Smart-MD Project, Digital Technologies

Imprint / Contact

Highlight for Query ‹Coronavirus disease in birds medication

Ebola virus disease in the Democratic Republic of the Congo, 1976-2014


Ebola virus disease (EVD) outbreaks are rare and knowledge of the transmission and clinical features of this disease is sparse. As of May 2015, the devastating outbreak in West Africa has resulted in more than ten times the number of cases reported in all previous outbreaks and will ultimately provide improved insights into EVD. Here, for the first time, all the databases from EVD outbreaks in the Democratic Republic of the Congo (DRC) have been cleaned and compiled into one anonymised individual-level dataset (See Supplementary file 1). The data provided are an invaluable addition to the West Africa data and will allow a more complete picture of the disease. The DRC is the country that has experienced the most outbreaks of EVD. Since the virus' discovery in 1976, there have been six major outbreaks (Yambuku 1976, Kikwit 1995, Mweka 2007, Mweka 2008/2009, Isiro 2012, and Boende 2014) and one minor outbreak (Tandala 1977) reported in the DRC, four in the northern Equateur and Orientale provinces and three in the southern provinces of Bandundu and Kasai-Occidental (Figure 1). Some of these have been described in the literature (World Health Organization, 1978; Heymann et al., 1980; Khan et al., 1999; Muyembe-Tamfum et al., 1999, 2012; Maganga et al., 2014). However, the individual-level data and corresponding lessons from these outbreaks have not been collated or made publicly available; by doing so, we aim to permit a more powerful statistical analysis and a fuller understanding of the disease. The end of the most recent outbreak in the DRC was declared on the 21st of November 2014. This provides an unparalleled opportunity to assemble all the information gathered about EVD in the DRC through almost four decades, learn from the Congolese experience with this disease, and compare the features of EVD in DRC with the epidemic that has had such a devastating effect in West Africa.

Case demographics

The number of cases and case-fatality ratios (CFRs) varied greatly between outbreaks (Table 2). It can also be observed that laboratory confirmation became more readily available over time. Across all outbreaks, 57% of cases were female (95% CI = 53.9–60.1). In the second Mweka outbreak and in the Isiro outbreak, more than 70% of cases were females. However, in the other outbreaks, the percentage of females was lower (53–59%). When comparing the probable and confirmed cases by age with the overall DRC population (Figure 2), we observed a high concentration of cases in the 25–64 age category compared to the baseline population. This might be because at this age individuals are more likely to be carers. The occupation was only recorded during three outbreaks: Kikwit, Boende and Isiro. During Kikwit, 23% (73/317) of cases were known healthcare workers (HCWs) and 0.6% (2/317) were possible HCWs. During Boende, the occupation was recorded for 85% (58/68) of cases. 14% (8) were known HCWs and 3% (2) were possible HCWs. During Isiro, occupation was reported for 94% (49/52) of cases. 27% (13) were HCW. Although occupation was not recorded on an individual level, during Yambuku, 13 of the 17 Yambuku Hospital workers contracted EVD (World Health Organization, 1978).

Epidemic curves

The epidemic curves were plotted for the six major outbreaks (Figure 3). The date of infection was based on symptom onset when available (701/995). When it was not, hospitalisation dates were used (5/995). In cases where these were also absent (281/995), the notification dates were used as proxy. For Mweka 2007, the date of infection was mostly based on the notification date (98%), whereas in the other outbreaks, infection dates refer to onset of symptoms almost exclusively (>90%). In time, case definitions became more specific. With the exception of Kikwit, in which notification and the closure of healthcare facilities coincided closely in time, outbreaks seemed to peak before major interventions were initiated.


The proportion of probable and confirmed cases reporting EVD symptoms is shown in Figure 4. Overall, the most commonly reported symptom was fever, which was reported by 95% of cases (95% CI = 92.6–97.3%) and at least 90% of cases in every outbreak. Reports of vomiting were also similarly common across all major outbreaks, reported by 75% of cases (95% CI = 69.3–79.2) and between 57% and 76% of cases for all major outbreaks. There was considerable variation in how frequently the remaining symptoms were reported for different outbreaks. In particular, hemorrhagic symptoms were present in 61% (95% CI = 51–71) of cases during Kikwit but only 10% (95% CI = 5–18) during Mweka 2007. The Bundibugyo ebolavirus (Isiro outbreak) did not present a symptom profile that was particularly different from that seen for the Zaire ebolavirus (all other outbreaks). However, this was difficult to conclude given the large variation between outbreaks.


The mean CFR overall was 79% (95% CI = 76.4–81.6), but there were significant differences between epidemics and within epidemics over time (Figure 5 and Figure 5—figure supplement 1). The highest average CFR was seen during the first outbreak in Yambuku (mean = 96%, 95% CI = 92.6–97.9 in our subset of 262/318 cases). Kikwit, Mweka 2007, and Boende had high average CFRs ranging from 74% to 78%. During the Isiro and Mweka 2008 outbreaks, the CFR was lower, at 54 and 44% (95% CI = 39.5–67.8 and 26.4–62.3), respectively.

All EVD patients under 2 years of age died (N = 29, Figure 5—figure supplement 1). CFRs generally decreased during childhood and then increased again to plateau at around 70–80% in adulthood (Figure 5—figure supplement 2). This pattern was less readily observed for the CFRs in the Yambuku outbreak, which remained high and similar for all ages.

In the regression model that included the delay between symptom onset and hospitalisation as a factor but excluded three outbreaks for missing data (Table 3), the baseline CFR in individuals over 15 years of age during the first month of an EVD outbreak who were admitted to hospital after 0.3 days (the average time from symptom onset to admission to hospital) during the Boende outbreak was 74% (95% CI = 17.8–99.3). The CFR was similar during the Isiro outbreak but was significantly higher during the Kikwit outbreak (94%). The CFR in 0–5 year olds was 76%, and in 5–15 year olds, it was significantly lower at 36%. The odds of dying declined on average by 31% (95% CI = 3.1–52.0%) each month after the start of an outbreak and increased by 11% (95% CI = 1.8–20.7%) per day that a symptomatic person is not hospitalised (Table 3).

In the regression model that included all major outbreaks, the CFR for individuals over 15 years of age during the first month of the outbreak during the Boende outbreak was estimated at 79% (95% CI = 25.8–99.5). The Yambuku, Kikwit, and Mweka 2007 outbreaks had significantly higher CFRs (96%, 94% and 93%) and the Mweka 2008 outbreak had a significantly lower CFR (48%). 0–5 year olds had significantly higher CFRs (90%) than those over 15 years of age. For the 5–15 year olds, the CFR was significantly lower (57%). The odds of dying declined on average by 35% (95% CI = 22.6–45.9) each month after the start of each outbreak (Table 4).

Reproduction numbers through time

Changes in the effective reproduction number, R, over the course of the outbreaks were plotted in Figure 6. In Yambuku, Mweka 2008, and Boende 2014, R dropped below one within 3–5 weeks after the initial case and the outbreak was rapidly brought under control. In these settings, the spread of EVD during the first 2 weeks had been high (R > 3). By contrast, in Kikwit 1995, Mweka 2007, and Isiro 2012, where the initial transmission rate was lower, spread of EVD was sustained for more than 13 weeks. Overall, we can see that R declines before the major interventions occurred, which could point to behavioural changes that occurred spontaneously in the populations.

Delays in case detection

The delay distributions from onset of symptoms to notification, from onset of symptoms to hospitalisation, from onset of symptoms to death, length of hospital stay, and from hospitalisation to death were plotted for each outbreak (when available) in Figure 7. The largest delays between symptom onset to notification and to hospitalisation were seen during the Kikwit outbreak (12.9 days and 5.0 days, respectively). The largest delay between symptom onset and death and the longest duration of hospitalisation were seen during the Isiro outbreak (11.4 and 8.0 days, respectively). However, this was only recorded for the Kikwit, Mweka 2008, and Isiro outbreaks. The longest delay between hospitalisation and death was observed during the Mweka 2008 outbreak (11.0 days) (Table 5).


This article provides for the first time a description and a line list for all outbreaks that have occurred in the DRC. This represents almost 40 years of surveillance data, seven outbreaks, and 996 suspected, probable, or confirmed cases. It is an invaluable resource for studying the epidemiology and clinical features of EVD. We highlight the importance of reducing the delay between symptom onset and hospitalisation, as the odds of dying increase by 11% per day that a patient is not hospitalised. We also observe higher incidence in those between 25 and 64 years of age and a higher CFR in patients under 5 or over 15 years of age than in those between 5 and 15 years old. These trends mirror those observed during the West African outbreak, where cumulative incidence was highest in those between 16 and 44 years of age and CFR progressively dropped from 89.5% in those under 1 year of age to 52.1% in those between 10 and 15 years, to rise again to 78.7% in those over 45 years old (WHO Ebola Response Team et al., 2015a). These distinctions could inform the choice of target age groups for interventions such as vaccination.

Another important finding is that during outbreaks with an initially lower reproduction number, R, (≤3) national and international response was slower, outbreaks took longer to control, and (with the exception of Yambuku, where the virus was first discovered) were larger outbreaks than those with initially high R. This occurred during the current outbreak in West Africa, where the basic reproduction numbers for Guinea, Sierra Leone, and Liberia have been estimated at 1.51, 2.53, and 1.59, respectively, and indicates the need for any future EVD to be met with rapid national and international response (Althaus, 2014).

Our estimates largely coincide with those recently reviewed in the literature (Van Kerkhove et al., 2015). The basic reproduction numbers reported for the Kikwit outbreak (3.00) is comprised in the range found by other studies (1.36–3.65) (Chowell et al., 2004; Ferrari et al., 2005; Lekone and Finkenstädt, 2006; Legrand et al., 2007; Forsberg White and Pagano, 2008; Ndanguza et al., 2011), and our estimate for the Yambuku outbreak (5.00) is similar to that reported by Camacho et al. (4.71, range = 3.92–5.66) (Camacho et al., 2014). The mean delay of onset of symptoms to hospitalisation and to death estimated here for Kikwit (5.0 and 9.5, respectively) was also similar to that found by other authors (4–5 [Khan et al., 1999; Rowe et al., 1999] and 9.6–10.1 [Bwaka et al., 1999; Khan et al., 1999]). Our estimated mean delay of onset of symptoms to death during the Boende outbreak (9.4) was slightly lower than that found by other authors (11.3) but included in their reported range (1–30) (Maganga et al., 2014). The delay between hospitalisation and death during the Kikwit outbreak found in the literature (4.6) coincided with our estimate (4.5) (Khan et al., 1999). In addition, our estimates of the overall CFR for Kikwit and Boende (78% and 74%, respectively) coincided with other estimates reported in the literature (74–81% [Muyembe and Kipasa, 1995; Khan et al., 1999; Ndambi et al., 1999; Sadek et al., 1999; Chowell et al., 2004] and 74% [Maganga et al., 2014], respectively). The remaining outbreak estimates have not been studied by other authors and are reported here for the first time.

Overall, CFRs and delays between symptom onset and hospitalisation, symptom onset and death, and hospitalisation and death reported in our study do not differ substantially with those reported for the current outbreak (WHO Ebola Response Team et al., 2015b). The data presented were originally collected for the containment of the outbreaks rather than for providing the basis of an epidemiological study of the disease. As such, variables are not recorded consistently across all outbreaks and there are missing data. This dataset does not take into consideration undetected cases. A surveillance study carried out in northwestern DRC between 1981 and 1985, through clinical records and serological testing, detected 21 cases likely to be EVD, suggesting that sporadic cases do occur (Jezek et al., 1999). Another serosurvey carried out in Yambuku after the outbreak suggested that that 17% of the population in the village was infected asymptomatically (Breman et al., 1978). Under-reporting may differ between and during outbreaks and may impact the calculated estimates such as CFRs, which limits the validity of direct comparisons of values between outbreaks. Other limitations include the different case definitions employed in different outbreaks and that the method used to calculate the effective reproduction numbers is susceptible to changes in reporting during the outbreak (as most methods are). However, it is robust if the extent of underreporting remains constant during each outbreak. Moreover, it is robust to different reporting sensitivity between outbreaks.

The regular re-emergence of EVD in human hosts is likely to be connected to the presence of the virus in animal reservoirs, such as bats and monkeys (Leroy et al., 2009; Muyembe-Tamfum et al., 2012). The presence of vast tropical rainforests covering entire regions of the DRC and the strong link existing between local economies and the forest makes a re-emergence of the virus in the country in the near future very likely (Pigott et al., 2014). Although the Mweka 2007 outbreak has been linked to the consumption of fruit bats that migrate to the area (Leroy et al., 2000), the epidemiological link between index cases (when known) and animal reservoirs has not been found for any of these outbreaks.

All outbreaks except for the 2007 Mweka outbreak have involved hospital transmission during the early part of the outbreak, illustrating the amplifying effect that poor infection control can have on EVD epidemics. A study of the 1976 outbreak has highlighted the importance of community infection to transmission (Camacho et al., 2014). Traditional burials are an important mechanism of transmission of EVD. Funeral data can help inform mathematical models that explore the importance of this route of transmission and can help guide resource allocation. This will be explored in subsequent analysis.

Mweka 2008 was the shortest and smallest outbreak with the lowest CFR. This could be due to the short delay between the first notification and the opening of the isolation centre (10 days). The low CFR during Isiro could be due to infection by a less virulent type of virus (B. ebolavirus) and is in line with what has been reported for this virus in other outbreaks (Van Kerkhove et al., 2015).

In most outbreaks, major interventions arrived when the reproduction number, R, was less than one and the epidemic was already under control. This suggests an important role of other factors, such as changes in contact behaviour, in shaping the changes of R. For example, there is evidence that an increase in the proportion of patients admitted to hospital was associated with a reduction in the size of EVD transmission chains in Guinea in 2014 (Faye et al., 2015) and the community acceptance of EVD control measures in West Africa improved dramatically over the course of the epidemic, which led to better infection control (Dhillon and Kelly, 2015).

The Boende outbreak began whilst the West African outbreak was gaining international importance. This much smaller outbreak, with an initial R of five, which consisted of 68 cases, lasted only 10 weeks. The more remote setting, a background antibody presence in the area and a greater preparedness to EVD (that led to its notification 3 weeks after the first case and the opening of the first isolation centre a month later) could have contributed to the avoidance of a larger outbreak (Heymann et al., 1980; Busico et al., 1999; Maganga et al., 2014).

The high number of EVD cases between 25 and 64 years of age compared to the background demographics, the high CFR in children under five, the decrease in CFRs in those 5 to 15, and the subsequent increase in CFR during adulthood are phenomena that warrant further investigation. The variation in symptoms reported during different outbreaks is also a matter for further research.


Line list data and reports for each outbreak were retrieved from the Direction de Lutte contre la Maladie (DLM) (Ministère de la Santé Publique (Direction de la Lutte contre la Maladie), 2007, 2009, 2012; Ministère de la Santé Publique (Comité National de Coordination), 2014). The DLM is the public body in charge of containing EVD outbreaks in the DRC. These data were designed for outbreak containment rather than for epidemiological analysis; therefore, appropriate cleaning was undertaken. The fields selected were age, sex, date of symptom onset, date of hospitalisation, date of hospital discharge, outcome, case definition, date of notification (when the case was first reported to the DLM), date of death, occupation, fever, diarrhoea, abdominal pain, headache, vomiting, hiccups, and hemorrhagic symptoms. Where this information was not available, it was left blank. A unique ID was assigned to each patient in the dataset. The Tandala outbreak (1977) included only one reported case; therefore, only the context and history of this outbreak was analysed. We included 262 of the 318 cases reported in Yambuku (those for which these data were available). The aggregated line lists can be found in Supplementary file 1.

Case definitions

According to the WHO EVD case definitions for outbreak settings; suspected cases are all individuals (alive or dead) who had a fever and had contact with a suspected, probable, or confirmed EVD case or a sick or dead animal; any individual with a fever and more than three additional EVD symptoms; or any person with unexplained bleeding or whose death is unexplained (World Health Organization, 2014). Probable cases are suspected cases that have a clear epidemiological link with a confirmed case. Confirmed cases are individuals who were tested positive via PCR. In the DRC setting, the case definitions employed varied somewhat between outbreaks (Appendix 1, Section A). Unless stated otherwise, where the case definitions distinguished susceptible cases from probable and confirmed cases, all estimates presented (CFRs, symptom delays, and reproduction numbers) were computed omitting suspected cases.

Patient demographics, epidemic curves, and symptoms

DRC national demographics between 1975 and 2010 were used as reported by the UN Department of Economic and Social Affairs (United Nations (Department of Economic and Social Affairs), 2013). For temporal comparison of patient reports, we used the date of infection. When available, we used the date of symptom onset. When these were unavailable, hospitalisation dates were used instead. If these were also absent, the notification dates were used as proxy.

When calculating the proportion of confirmed and probable cases that presented with EVD symptoms, we assumed that patients for whom the presence or absence of at least one symptom was reported did not display any additional symptoms unless those were also reported.

CFRs, reproduction numbers, and delay distributions

The odds of dying from EVD were estimated through binomial regression with age group and year of outbreak as factorial covariates and the number of months since the start of the outbreak and the delay from symptom onset to hospitalisation as continuous covariates. The age groups used were 0–5 years, 5–15 years, and >15 years. The delay from symptom onset to hospitalisation was present for 63% of probable or confirmed cases. These dates were not recorded for Yambuku and Mweka 2007 and only for four cases for Mweka 2008. For this reason, these outbreaks were excluded from this first analysis. A second regression model was conducted that excluded the delays from symptom onset to hospitalisation as an explanatory variable, enabling the use of data from all major outbreaks and increasing statistical power. The start of an outbreak was defined by the earliest onset of symptoms of any detected case. The CIs were calculated using profiled log-likelihood.

We calculated the weekly effective R, the average number of individuals that were infected by a typical EVD case during the period of infectiousness, by reconstructing the transmission tree of each outbreak on the basis of date of infection for each case (Wallinga and Teunis, 2004). To link a case to its most likely source, we assumed a serial interval of 15.3 days with a standard deviation of 9.3 days as reported during the current outbreak in West Africa (Maganga et al., 2014). Delays in care were only calculated for those outbreaks for which the necessary dates were recorded.


R-3.1.2 was used for the cleaning, analysis, and plotting of figures (R Development Core Team, 2011).

Ethical approval

This study was approved by the LSHTM Research Ethics Committee (approval number PR/1541/1541).

Role of the funding source

The funders had no role in the design, collection, analysis, and interpretation of data, or in the writing of the manuscript. The corresponding authors had full access to all the data and were responsible for the final decision to submit for publication.