Context
On May 15th, 2026, an Ebola virus disease outbreak caused by Bundibugyo ebolavirus (BDBV) in Ituri province was declared (1). An alert from local health authorities was raised on May 5th, 2026 about an unidentified disease cluster with high mortality in Mongbwalu, Ituri province, including deaths among health care workers (2, 3). The reports of these clusters of deaths date back to March and April, consistent with preliminary estimates of the start of the outbreak (4, 5). Rapid deployment of resources for response and investigation followed, with initial confirmed cases in four health zones across Ituri (Mongbwalu, Rwampara, Nyankunde, and Bunia) and in the neighbouring province North Kivu (Butembo, Katwa and Goma) (6), before multiple more were added as surveillance increased.
We used laboratory line list data to characterise the early epidemiological patterns of the outbreak, including key delay distributions, spatial dynamics and transmission.
Description of the spatial expansion of the outbreak
The early outbreak data indicate that the cases were concentrated in Mongbwalu, Rwampara and Bunia, with cases along the major transit route and into a border town (Aru), as well as Butembo and Katwa, Goma, and Miti-Muthesa in Nord and South Kivu, respectively (Fig. 1a). Confirmed cases show the highest reported burden in Mongbwalu and Rwampara, with rapidly rising cases in Bunia throughout May. The earliest positive case from the outbreak had a symptom onset date of April 24, 2026. Earlier samples from suspected cases were available but showed invalid results as early as April 9th (Fig. 1b). Invalid results might be due to insufficient blood taken and/or difficulties transporting samples to the laboratory during the early phase of the outbreak, when local laboratories were just being established. However, most samples up until May 23rd have test results (Fig. 1b). Since May 12th, we observe a rapid increase in the number of health areas and health zones reporting positive cases (by 6 and 30, respectively, between May 12th and June 1st; Fig. 1c), and increasingly, health zones outside the main affected ones are reporting cases (Fig. 1d).
Figure 1. The epidemiology of the 17th Ebola virus disease outbreak in the Democratic Republic of the Congo (DRC). A) Map of health zones in eastern DRC; red represents where suspected or confirmed cases have been reported, with radius of black circles representing the number of positive cases by RT-qPCR. B) Number of cases by case status in Ituri province (positive = red, negative = blue, dark grey = invalid, grey = unknown) from the outbreak line list. The purple line is the 5-day rolling positivity and the orange line is the proportion of samples without known outcome. The shaded areas represent 95% CIs based on a binomial sampling model. C) Cumulative number of health areas (dotted) and health zones (solid) with non-negative cases in Ituri province. D) Time series of reported positive cases by health zone and symptom onset date in Ituri province. Circle size represents the number of reported positive cases on a given date.
Early estimates of the epidemiological parameters
Testing laboratory-based epidemiological records from the outbreak that include case demographics, sample collection and testing dates, results, and symptom onset dates (for a subset of cases) indicate high test positivity rates between 5-11 May (68%, 95%CI: 48, 83) based on symptom onset dates. Positivity remains high with 30-40% between 12 May and 26 May (95%CI: 0.27, 0.54) following the scale-up of testing and formal detection of the outbreak on May 15th (Fig. 1b). The fraction of untested samples however increased over the last week of the data (26 May - 2 June), likely due to a testing backlog (roughly half of the samples remain untested, Fig. 1b). Samples were predominantly from Bunia and Rwampara, with fewer samples available from Mongbwalu despite early reports of an ongoing outbreak there. Over the entire period, roughly half the patients (56.8%) had a test taken within 3 days of symptom onset, and 34% of test results were available within seven days of samples being taken. The delays between symptom onset and testing across positives, negatives and invalid tests have been declining from around 20 days in late April 2026 to 5-10 days from May onwards (Fig. 2b). Across data available for onset dates May 5th to May 25th, for delays from symptom onset to sample (n = 249) and from symptom onset to first test result (n = 313), the best-fitting models from maximum likelihood estimates (with censoring corrections) were Weibull (shape = 1.157, scale = 4.859) and Gamma (shape = 5.335, rate = 0.620) distributions, respectively, implying mean delays of 4.616 and 8.611 days, respectively, and will likely change as the laboratory infrastructure is expanded locally. We excluded more recent days (May 26th to June 1st) because of delays between symptom onset and when cases are recorded in the line list. Testing time intervals are similar across different laboratories (Bunia and Kinshasa).
To estimate growth rates in the context of changing health care seeking behavior, rapid changes in surveillance, testing availability, and reporting lags, we implemented multiple different approaches with varying assumptions regarding the observation process. Our first approach adjusted for unknown test results by fitting Negative Binomial (NB) regression models to daily positive test counts by onset date under the assumption that the proportion of positives will be the same in cases without/with test results. Analysis of cases with symptom onsets in the three-week range May 5th, 2026 to May 25th, 2026 (to minimise day-of-week effects) gave a growth rate estimate of 0.087 (95% CI: 0.050, 0.119), corresponding to a doubling time of \~8 days (95%CI: 5.8, 13.9 days, Fig. 2a). Taking advantage of regular updates to the line list, we performed out-of-sample (OOS) evaluation of our growth model and find that the OOS root-mean-square-error (RMSE) drops when compared to the in-sample (1.902 in sample vs 1.883 OOS RMSE). Although typically unexpected, this increase in model performance OOS is because unresolved tests become resolved in subsequent reports.
To account for the expected number of samples that might be added to the data, we used the fitted distribution of the onset-to-sample delay to impute the likely number of additional samples that would be added (Fig. 2a yellow). As expected, the closer we get to the present, the higher the uncertainty in these estimates (Fig. 2a, yellow bars). Using the same model as above, we re-fit the growth rate model and find that growth rates slightly increase (0.094 and 95% CI: 0.057, 0.127, doubling time of 5.5 - 12.2 days), which is expected due to a larger number of untested cases that might become positive. When estimating growth rates for positive cases reported from Bunia and Rwampara, doubling times were 10.5 days (95% CI: 7.0, 34.7) and 9.4 days (95% CI: 6.5, 24.7) for the model with and without the sampling adjustment (Fig. 3).
We note that for models that extended the period to the 30th of May, doubling times were lower (8.9 - 11.9 days). It is therefore possible that the growth of the epidemic may have slowed following the declaration of the outbreak and scale-up of public health responses, but data at the time of writing do not allow a definitive conclusion or the magnitude of any reduction to be reliably estimated.
Figure 2. Growth, demographics, and surveillance. A) Daily number of positive, invalid, unknown, cases Ituri province based on symptom onset date. Yellow bars represent imputed number of samples (unknown status) based on the delay distribution between onset and sample estimated between 5th and 25th May. Red whiskers show 95% CIs. The purple line represents the 5-day rolling positivity rate, and shaded bars represent 95% CIs based on a binomial sampling model. The black line shows the estimated growth rate of positive cases when adjusting for the number of unseen samples, and the green line without that adjustment. The shaded area represents 95%CIs. B) Daily time between symptom onset and date of sampling. The line shows the median and the grey bars uncertainty. The inset shows the histogram of delay values and Weibull (orange) and Gamma (blue) distribution fits. The colour of the dots represent their status. C) Age-sex distribution for positive and negative cases (top: female; bottom: male; blue: negative, orange: positive).
Figuer 3 Growth in Bunia and Rawampara Health Zones: Daily number of confirmed cases in Bunia and Rwampara provinces based on symptom onset date. Lines correspond to different model fits to the data (with and without adjusting for the number of potential samples based on delay distributions estimated between 5 and 25 May). Shading represents 95% CIs. Grey bars show the number of untested cases, yellow the number of predicted samples to be received, and red the number of positives. The purple line represents the 5-day rolling positivity and uncertainty based on sample size.
Spatial risk mapping
Next, we investigated the spatio–temporal dynamics of the outbreak using a stochastic compartmental spatial modelling framework informed by empirical population, travel, and March 2026 mobile phone-derived mobility data and EVD natural history parameters from previous outbreaks (7). Using daily new confirmed cases at the national level and cumulative positive cases at the health zone level, we calibrated transmission rates and spatial spread. We find that models with seeding of infections in Mongbwalu, Rwampara, and Bunia in mid-April best reconstruct observed dynamics (Fig. 4). The 95% credible intervals (CrIs) for the origin span mid-March to late-April, which is consistent with reports of clusters of deaths in the Mongbwalu/Rwampara/Bunia region and preliminary evidence from viral genomic data suggesting transmission started between early April and early May 2026 (4).
With our fitted model, we can visualise the spatial invasion dynamic over time. Initial spatial spread shows early seeding around Mongbwalu, Rwampara, and Bunia and then outward into other health zones (Fig. 4b). As new locations experienced established transmission, they acted as local hubs for further spread (e.g., Butembo and Katwa, Fig. 4b). Using Approximate Bayesian Computation to explore posterior distributions over fitted parameters, we estimated R0 = 3.5 (95% CrIs 2.2, 4.0) and overdispersion of kappa = 0.53 (95% CrIs: 0.04, 0.88, which characterises superspreading, with lower values of kappa indicating more superspreading, Fig. 4c). Our results are consistent with estimates from prior outbreaks, though they are primarily from outbreaks of EVD (7). Forcing a single seeding event lowers estimates of R0 and increases the estimated extent of superspreading, but fails to capture the spatial spread of cases (18x under-estimation in Rwampara and 5x under-estimation in Buni).
Simulating outbreaks by sampling from the posterior distribution of parameters, we observe overall strong agreement in the trajectory of case accumulation at the national level and cumulative positive cases at the health zone level (latest date, May 31st, log RMSE 0.563 for daily new positive cases nationally and 0.770 for cumulative positive totals across health zones, Figure 4b,d). The model was calibrated against data through May 21st and continued to show strong agreement with reported data through May 31st for most health zones. We note a sizable under-prediction for Nyankunde (<1 model prediction vs >20 observed cases). That zone has a small population size and, despite its close proximity to Mongbwalu, Rwampara, and Bunia, has low empirically observed mobility from that region. We note that Nyankunde does contain a large, 150-bed regional referral hospital that is treating patients from around the impacted region.
Using our calibrated model, we assessed the spatiotemporal risk of BVD in every health zone in the DRC (Fig. 4a). Risk is calculated by dividing the log predicted number of infections from the metapopulation model by the log of the highest estimated number in any health zone.
Figure 4. Spatial risk assessment and model fit. A) Spatial relative risk of BVD transmission across health zones in the DRC. Darker colour indicates higher relative risk. Logged values were normalised by the logged maximum value from the model. B) Predicted spatial spread of BVD cases from the epicenter zones. C) Predicted total cases per zone estimated by simulating from the posterior distribution over model parameters vs. observed cases*. D)** Predicted mean trajectories for the three most impacted zones determined via the same simulations as in panel C. (blue dashed line mean, 95% CrIs shown in blue shading) vs observed (red line). All other health zones reporting cases are shown in the supplementary materials.*
Notes on methods and assumptions
Epidemiological data
We used multiple complementary epidemiological case datasets in this work. First, we extracted the daily number of suspected and confirmed cases, and deaths at the health zone level from INSP situation reports. These were reported in situation reports 001 - 017 and are archived here: GitHub - INRB-UMIE/BDBV2026-Data: Data and scripts for epidemiological analysis of the 2026 Bundibugyo Ebola outbreak · GitHub . Secondly, laboratory metadata from Ituri and North Kivu containing information on spatial locations (health area, health zone), symptom onset and sample collection dates were used for estimating epidemic growth rates, delay distributions, and the spatiotemporal dynamics of the outbreak. These data were made available via a data sharing agreement and at this stage cannot be made public.
Epidemiological delay distributions
We analysed a line-list dataset compiled from 942 laboratory records across 27 health zones in Ituri District, DRC. Where relevant/appropriate and complete, the dataset contained the following five key types of dates for each individual: symptom onset date, sample collection date, laboratory receipt date, result analysis date, and date of death. Where complete, we also had access to information about the patients’ age, sex, health zone, viral CT value and test result. We classified a record as positive if (i) the Kinshasa (INRB) result is positive or (ii) the Bunia (LPSP) result is positive and the Kinshasa result is not yet available. Discordant records where the Bunia result is positive but the Kinshasa result is negative are classified as negative — Kinshasa’s definitive negative overrides the Bunia positive.
Of the 942 individuals, there were 284 individuals classified as positive and 119 of those individuals had a recorded symptom onset date (41.9%). The dataset allows for calculating delays from symptom onset date to laboratory analysis, from sample to laboratory test, from laboratory test to analysis, from symptom onset to analysis, and from symptom onset to death. These each reflect hospital care-seeking delays, transport/logistical delays, and durations of clinical illness. Restricting the analysis to individuals where both contributing dates were non-missing and the resulting delay was non-negative, we obtained different eligible sample sizes per delay: symptom onset to first-analysis (n = 325), symptom onset to sample (n = 425), symptom onset to death (53), and laboratory receipt to first-analysis (n =542).
To visualise age trends, we binned the continuous age values into eight groups with left-closed intervals: 0–4, 5–9, 10–19, 20–29, 30–39, 40–49, 50–59, and 60+ years. To estimate delay distributions, we used both frequentist and Bayesian methods, focusing mainly on 5th-25th May to align with windows for growth rate estimation. We obtained our censoring-corrected maximum likelihood estimates (MLEs) using the fitdistrplus package (8) in R with Akaike Information Criterion (AIC) used for model selection and compared these results to estimates from Bayesian censoring-corrected distributions using the epidist package (9) in R (which importantly, corrects for double-interval censoring and right truncation).
Epidemiological parameter estimation
Case classification and daily counts
Laboratory records were classified by combining the initial result and the final confirmatory result (final result), after correcting for some clear data entry errors. Each record was assigned to one of four mutually exclusive categories (positive, negative, invalid, or unresolved) the final result was used whenever it gave a definitive value (positive/negative/invalid), falling back to the initial result otherwise; a record was ‘not yet tested.’
For each symptom-onset day t we count the number of positive samples cₚ(t), the number with a valid (positive or negative) result cₛ(t) = cₚ(t) + cₙ(t), and the total number of possible cases (including untested and invalid) samples c_T(t) = cₛ(t) + (invalid + not-yet-tested).
Negative-binomial growth model with a testing-completeness offset
We modelled the number of positive samples by symptom-onset day using negative-binomial (NB) regression with a log link, to accommodate over-dispersion relative to the Poisson (although we note similar results were obtained using a Poisson regression):
cₚ(t) ~ NB(μ_t, θ), Var(cₚ) = μ_t + μ_t²⁄θ,
log μ_t = α + r·t + log[cₛ(t) ⁄ c_T(t)].
The exponential growth rate r (per day) is the coefficient of onset day t, with doubling time log(2)⁄r. The term log[cₛ(t)⁄c_T(t)] is an offset (coefficient fixed at 1) equal to the log proportion of suspect samples tested on day t; it accounts for incomplete testing, so that the expected observed positive count is exp(α + r·t) scaled by the proportion tested. This assumes that, conditional on onset day, untested samples have the same probability of being positive as tested ones (missing-at-random).
Adjustment for the onset-to-sample reporting delay
Because samples are collected with a delay after symptom onset, recent onset days are under-represented at any data cut-off (right truncation). We estimated the onset-to-sample delay distribution from records with both dates by fitting a gamma distribution by maximum likelihood (with a half-day continuity adjustment); the gamma distribution was preferred over Weibull and log-normal by AIC. With G the fitted CDF and T the cut-off date, the expected proportion of an onset day’s eventual samples already collected by time T is p(t) = G(T − t + 1/2). A sample-adjusted model added log p(t) as a second offset, log μ_t = α + r·t + log[ cₛ(t) ⁄ c_T(t) ] + log p(t), up-weighting recent, still-incomplete onset days so that r additionally accounts for onset-to-sample right truncation. We report both base and sample-adjusted growth rates. When estimates for delays are only estimated for symptom onset dates between the 5th and 25th of May, the inferred growth rates are higher (Fig. 3).
As an alternative to entering testing completeness as a fixed offset, we fitted a model in which the per-day proportion of samples without a definitive result was included as a covariate with a freely estimated coefficient. Define the unresolved proportion on onset day t as u(t) = [c_T(t) − cₛ(t)] ⁄ c_T(t) = (invalid + notadj-yet-tested) ⁄ c_T(t), equivalently one minus the proportion tested. The negative-binomial growth model was then log μ_t = α + r·t + b·u(t), with the same NB(μ_t, θ) variance structure as the previously described model. This relaxes the constraint that the testing term enter with a coefficient fixed at 1: because log[cₛ(t)⁄c_T(t)] = log[1 − u(t)], the offset specification of Model 1 is approximately the special case b ≈ −1, whereas here b is estimated from the data. The coefficient b therefore quantifies the association between observed positive counts and the unresolved fraction, conditional on onset day. A sample-adjusted version retained the onset-to-sample completeness as an offset while keeping testing as a covariate, log μ_t = α + r·t + b·u(t) + log p(t). Confidence intervals for r and b were obtained by the same negative-binomial parametric bootstrap described.
Uncertainty
Confidence intervals were obtained by a parametric bootstrap with 500 replicates. In each replicate, the positive count for every onset day was redrawn from a negative-binomial distribution with mean equal to the observed count and dispersion equal to the fitted θ of the corresponding model, so that the resampling reflects the same over-dispersion as the fitted model rather than Poisson variability. The testing-completeness term log[cₛ(t)/c_T(t)] was treated as a known offset and held fixed at its observed value (the proportion tested is a design quantity and the resampled response is a component of it). For the sample-adjusted model, each replicate additionally drew gamma delay parameters from a parametric bootstrap of the fitted shape and rate, entering through log p(t). Reported 95% confidence intervals are the 2.5th and 97.5th percentiles of the bootstrap distribution of r (and of the doubling time log(2)/r); all windows converged in all 500 replicates.
Out-of-sample validation (temporal back-fill RMSE)
To assess predictive performance out of sample, we exploited two sequential data sets of the same line-list: an earlier data set (31 May 2026; 942 records) used for fitting and a later data set (2 June 2026; 995 records) used as the evaluation target. Each model was fitted to daily positive counts by onset day from the 31 May data set, over the 5–25 May and 5–30 May onset windows separately. The onset-to-sample delay distribution used by the sample-adjusted models was re-estimated from the 31 May records, and the completeness term used the training cut-off, p(t) = G(31 May − t + ½). For each onset day we formed a delay-completed prediction — the fitted linear predictor with the onset-to-sample completeness term log p(t) removed, so that recent days are projected to their eventual (filled-in) count, evaluated at the 31 May covariates:
μ̂*(t) = exp(α̂ + r̂·t) · [cₛ(t)⁄c_T(t)] (offset models),
μ̂*(t) = exp(α̂ + r̂·t + b̂·u(t)) (covariate models),
with cₛ/c_T and u taken from the 31 May data set. These predictions were compared with the observed positive counts by onset day in the 2 June data set, which have partially filled in (and been reclassified) relative to the training data set. We summarised accuracy by the root-mean-square error (RMSE) and mean absolute error across onset days in each window. To separate the contribution of generalisation from the effect of the de-censoring projection, we computed the same delay-completed predictor scored both in-sample (against the 31 May counts it was fitted to) and out-of-sample (against the 2 June counts); the fitted mean including all offsets was also recorded as the conventional in-sample residual. A naive persistence baseline, carrying forward the 31 May observed counts unchanged, was included for reference.
Estimation windows
Growth rates were estimated over several symptom-onset windows (5–25 May, 5–30 May, 1–25 May, 5–18 May) to assess sensitivity to the analysis period. We also ran sensitivity analyses for Bunia and Rwampara (see supplementary figures).
Software
Analyses used R (≥ 4.4), with MASS::glm.nb for negative-binomial regression and fitdistrplus for delay-distribution fitting and parametric bootstrap.
Spatial modelling framework
We implemented a discrete-time tau-leap Susceptible Exposed Infectious Removed (SEIR) metapopulation model that allows for overdispersed transmission. We partitioned DRC into 519 health zones. Each zone i is treated as its own well-mixed sub-population with an initial total population N_i (from GRID3 v4.4) and its own per-zone S_i / E_i / I_i / R_i compartments. Zones interact through a mobility kernel M_ij. The state evolves in discrete daily time steps. At each step, every susceptible individual in zone i has a probability 1 − exp(−λ_i Δt) of becoming exposed, with the per-zone time-varying force of infection
λ_i = (1 − ε) · B_i / N_i + ε · Σ_j M_ij · B_j / N_j,
where B_i is the summed per-individual transmission rate over all infectious cohorts currently in I_i, and a mixing parameter ε ∈ [0, 1] sets the fraction of contacts drawn from outside the zone. Individuals exit the latent period (E→I) following a gamma distribution with parameters taken from past outbreaks for BDV (mean 6.3 days and coefficient of variation ~0.4) (10). We implement the gamma distribution by splitting E into n_E sequential exponential sub-stages each with rate n_E·σ, where σ = 1/(6.3 days), such that the total time governing the E→I transition is Erlang(n_E, n_E·σ) ~ Gamma(shape=n_E, mean=6.3 d). The infectious period (I→R) is assumed to be exponential with a mean of 7 days.
Overdispersion is implemented at the individual level such that rate β_j is drawn from Gamma(k, R0/(k·D_inf)) when a transition happens from E→I. The result is that the per-individual distribution of secondary infections is roughly NegBinom(R0, k). Individuals entering I in the same time step are tracked as a cohort, and recoveries split the cohort’s transmission rate exactly via the Beta thinning identity Gamma(a+b) → Beta(a, b)-fraction.
Between-zone coupling M_ij is a row-normalized convex combination of three independently constructed components. Each component is a normalized origin→destination matrix. The three components are a long-range gravity model, a short-range gravity model, and an empirical flow model. We use driving times to calibrate the gravity models and Flowminder data for empirical flows and combine them such that
M_gravity_ij = (1 − w_short) · M_long_ij + w_short · M_short_ij
if row i has empirical outflow:
M_ij = (1 − w_emp) · M_gravity_ij + w_emp · M_emp_ij
else:
M_ij = M_gravity_ij;
where M_long_ij = N_j^α · exp(−D_ij / scale_long), M_short_ij = N_j · exp(−D_ij / scale_short), α is the gravity model exponent and was fixed at 1 during simulations, scale_long and scale_short are normalization factors, M_emp_ij are the empirical movements, w_emp a weighting factor, and D_ij is the pairwise distance, and N_j the destination population.
Authors
Key contributors to data collection, molecular testing, data interpretation, analysis, and writing:
Institut National de Recherche Biomédicale (INRB) and Institut National de Santé Publique (INSP), Kinshasa, Democratic Republic of the Congo, and partners
Dav M. Ebengo (INRB)*
Pierre Akilimali (INSP)*
Tania Bishola Tshitenge (INRB)
Ciara Judge (Oxford)
Bernardo Gutierrez (Oxford)
Ellie Bourgikos (Oxford)
Olivier le Polain de Waroux (WHO)
Joel Kosianza (WHO)
Abdil Mahamud (WHO)
Benjamin Kanku (INSP)
Etien Koua (WHO)
Olga Ntumba (WHO)
Lorenzo Subissi (WHO)
Nicksy Gumede (WHO)
Marie Roseline Darnycka Belizaire (WHO)
Otim Patrick Cossy Ramadan (WHO)
Pablo N. Perez-Guzman (Imperial College)
Christian Morgenstern (Imperial College)
Victoria M. Cox (Imperial College)
Ruth McCabe (Imperial College)
Anne Cori (Imperial College)
Neil M. Ferguson (Imperial College)
Linus Bengtsson (Flowminder)
Benjamin Reddy (Oxford)
Simon Cauchemez (Pasteur)
Joseph L.-H. Tsui (Oxford)
Justus Nsio (AfricaCDC)
Yap Boum (AfricaCDC)
Jeanine Nkakulu (AfricaCDC)
Yenew Kebede (AfricaCDC)
Collins Tanui (AfricaCDC)
Didier Bompangue Nkoko (INOHA)
Francine Ntoumi (Fondation Congolaise pour la Recherche Médicale)
Jean-Jacques Muyembe Tamfum (INRB)
Cathal Mills (Oxford)
Samuel V. Scarpino (Northeastern U)
Moritz U.G. Kraemer (Oxford)
Christian Ngandu (INSP)
Dieudonne Mwamba Kazadi (INSP)
Placide Mbala-Kingebeni (INRB)*
Acknowledgements
We are grateful to all institutions and partners for their support to surveillance efforts in the DRC.
Statement on continuing work and analyses before publication
Please note that this data is based on work in progress and should be considered preliminary. Our analyses are ongoing, and a publication communicating our findings is in preparation. If you intend to use data for similar analyses and before our publication, please contact Prof. Placide Mbala-Kingebeni (INRB, DRC), Prof. Dav Ebengo (INRB, INOHA) or Pierre Akilimali (INSP).
References
1. DRC - Ministry of Public Health, Hygiene and Social Welfare, official declaration Ebola outbreak, 15 May 2026. https://logcluster.org/en/documents/drc-ministry-public-health-hygiene-and-social-welfare-official-declaration-ebola-outbreak[.](http://paperpile.com/b/tEVyge/34h9M)
2. A. L. Phelan, J. B. Nuzzo, L. O. Gostin, The PHEIC for Ebola disease caused by Bundibugyo virus: an inflection point for solidarity and health equity. Lancet, doi: 10.1016/s0140-6736(26)01037-8 (2026).
3. D. Mwamba, P. Akilimali, F. Mboussou, F. Kabasubabo, K. Angendu, C. Ngandu, R. Fotsing, A. Ancia, R. S. Kamba, Bundibugyo virus disease outbreak in Ituri, Democratic Republic of the Congo. Lancet 0 (2026).
4. Initial genomes from May 2026 Bundibugyo Virus Disease Outbreak in the Democratic Republic of the Congo and Uganda, Virological (2026). https://virological.org/t/initial-genomes-from-may-2026-bundibugyo-virus-disease-outbreak-in-the-democratic-republic-of-the-congo-and-uganda/1032[.](http://paperpile.com/b/tEVyge/whqR5)
5. SitRep MVE N° 001/2026. https://insp.cd/sitrep-mve-n-001-2026/[.](http://paperpile.com/b/tEVyge/OynOp)
6. Health Cluster, World Health Organization, “Cluster Sante de la RDC Rapport situationnel - Humanitaire MVE #3 : Épidémie de la maladie à virus Ébola (Souche Bundibugyo) dans le contexte humanitaire (19 mai 2026)” (2026); https://reliefweb.int/report/democratic-republic-congo/cluster-sante-de-la-rdc-rapport-situationnel-humanitaire-mve-3-epidemie-de-la-maladie-virus-ebola-souche-bundibugyo-dans-le-contexte-humanitaire-19-mai-2026[.](http://paperpile.com/b/tEVyge/62Li5)
7. R. K. Nash, S. Bhatia, C. Morgenstern, P. Doohan, D. Jorgensen, K. McCain, R. McCabe, D. Nikitin, A. Forna, G. Cuomo-Dannenburg, J. T. Hicks, R. J. Sheppard, T. Naidoo, S. van Elsland, C. Geismar, T. Rawson, S. I. Leuba, J. Wardle, I. Routledge, K. Fraser, Pathogen Epidemiology Review Group, N. Imai-Eaton, A. Cori, H. J. T. Unwin, Ebola virus disease mathematical models and epidemiological parameters: a systematic review. Lancet Infect. Dis. 24, e762–e773 (2024).
8. Help to Fit of a Parametric Distribution to Non-Censored or Censored Data [R package fitdistrplus version 1.2-6], Comprehensive R Archive Network (CRAN) (2026). https://cran.r-project.org/web/packages/fitdistrplus/index.html[.](http://paperpile.com/b/tEVyge/gbUAX)
9. Estimate Epidemiological Delay Distributions With brms. https://epidist.epinowcast.org/[.](http://paperpile.com/b/tEVyge/AKjJV)
10. M. D. Van Kerkhove, A. I. Bento, H. L. Mills, N. M. Ferguson, C. A. Donnelly, A review of epidemiological parameters from Ebola outbreaks to inform early public health decision-making. Sci. Data 2, 150019 (2015).