Rapid Geographic Expansion, Decelerating Growth, and Six-Month Scenario Projections of Bundibugyo Virus Disease in the Democratic Republic of the Congo: Analysis of Situation Reports 1–61

1. Introduction

Ebola virus disease (EVD) is a severe viral haemorrhagic fever caused by viruses belonging to the genus Orthoebolavirus, family Filoviridae, and order Mononegavirales. Among the recognised orthoebolaviruses associated with human disease, Bundibugyo ebolavirus represents one of the least extensively characterised species because relatively few outbreaks and clinical datasets have been documented [1]. Bundibugyo virus disease was first recognised during an outbreak in western Uganda in 2007 and was subsequently reported in the Democratic Republic of the Congo (DRC) in 2012 [2]. Compared with Zaire ebolavirus, the principal cause of previous large DRC outbreaks, Bundibugyo virus remains associated with major gaps in available medical countermeasures. Vaccines and therapeutics developed primarily against Zaire ebolavirus, including rVSV-ZEBOV and Zaire-specific monoclonal antibody products, cannot be assumed to provide equivalent protection against Bundibugyo virus because of substantial antigenic differences between species [3].

Taxonomically, Bundibugyo virus belongs to the species Bundibugyo orthoebolavirus. It is an enveloped, non-segmented, negative-sense single-stranded RNA virus with a genome of approximately 19 kb encoding seven major structural and functional proteins: nucleoprotein, VP35, VP40, glycoprotein, VP30, VP24, and the RNA-dependent RNA polymerase [4]. Comparative genomic studies have shown that Bundibugyo virus is genetically distinct from Zaire, Sudan, Taï Forest, and Reston orthoebolaviruses, particularly within the glycoprotein gene and other regions relevant to antigenicity, host-cell entry, immune evasion, and molecular diagnostics [5]. Whole-genome sequencing during an outbreak is therefore essential for confirming species identity, evaluating genetic relatedness to historical strains, distinguishing a new zoonotic introduction from sustained human-to-human transmission, and monitoring mutations that could affect diagnostic assay performance. Detailed phylogenetic data from the 2026 Bunia outbreak should be incorporated when officially released by INRB, WHO, or other authorised genomic surveillance partners [6,7].

On 15 May 2026, the DRC Ministry of Health officially declared an EVD outbreak following laboratory confirmation of Bundibugyo virus in Ituri Province. The outbreak subsequently expanded across eastern and northeastern DRC, affecting Ituri, North Kivu, South Kivu, Haut-Uélé, and Tshopo. By 14 July 2026, national surveillance reports documented 2,073 confirmed cases, 796 confirmed deaths, and transmission across 45 health zones. The rapid geographic expansion occurred in a setting characterized by recurrent insecurity, population displacement, mining-related mobility, cross-provincial travel, restricted access to healthcare, and variable availability of decentralized molecular diagnostics and treatment capacity [8,9].

Retrospective epidemiological investigations suggested that transmission may have preceded formal recognition by several weeks. Delayed detection may have been influenced by limited access to Bundibugyo-specific molecular assays in peripheral health zones, reliance on diagnostic platforms primarily configured for Zaire ebolavirus, incomplete community-based surveillance, and logistical barriers to specimen referral. Such delays are particularly consequential for EVD because late isolation, delayed contact identification, community deaths, and undetected transmission chains can accelerate geographic dissemination and increase mortality [10,11].

This study provides an updated epidemiological synthesis of the 2026 Bundibugyo virus disease outbreak using official Situation Reports issued between 14 May and 14 July 2026. The objectives were to: (1) characterise the overall epidemic trajectory; (2) estimate phase-specific growth rates and doubling times; (3) assess changes in the geographic distribution of cases across provinces and health zones; (4) compare provincial burden and reported case fatality ratios; (5) describe available demographic and operational indicators; and (6) evaluate evidence of temporal deceleration or continuing transmission. Based on the observed epidemiological trajectory, we also aimed to project the possible six-month progression of the outbreak under alternative rapid-containment, continued-deceleration, and sustained-incidence scenarios. Additional objectives were to: (7) compare candidate growth models and quantify the reduction in growth rate between phases; (8) characterize expected-trajectory versus posterior-predictive uncertainty using a complementary Bayesian framework; (9) extend the scenario comparison with deterministic sensitivity bounds; and (10) assess forecast calibration through rolling-origin validation.

2. Methods

2.1. Data Sources

Primary data were derived from 53 available official SitReps numbered 1–61 and issued by the INSP Centre des Opérations d'Urgence de Santé Publique (COUSP-RDC), available at https://insp.cd/ebola-17eme-epidemie/, covering 14 May to 14 July 2026 [12]. Not all sequentially numbered SitReps were available; analyses therefore used the reporting dates and observations contained in the available reports without interpolation. Supplementary contextual information was drawn, where necessary, from official communications issued by national authorities, WHO, and United Nations agencies. The INSP/COUSP series remained the primary source for all quantitative case, death, CFR, provincial, and health-zone analyses.

2.2. Case Definitions

Confirmed cases: Any person with a positive BDBV nucleic acid amplification test (NAAT) performed by the INRB reference laboratory, specifically its authorized mobile diagnostic units deployed in Ituri and all affected provinces

Confirmed deaths: deaths in laboratory-confirmed BDBV cases.

Case fatality rate (CFR) = confirmed deaths / confirmed cases × 100%.

2.3. Statistical Analysis

Each SitRep reporting date was treated as a discrete epidemiological observation point. Descriptive analyses were conducted across all available SitReps from 14 May to 14 July 2026 and included cumulative confirmed cases, confirmed deaths, recoveries, suspected cases, patients in isolation, contact follow-up rates, and the numbers of affected provinces and health zones. Absolute and percentage changes between consecutive available SitReps were calculated where data were reported. Because several SitReps documented retrospective cleaning, harmonisation, and reclassification of DHIS2 data, changes in cumulative totals were not automatically interpreted as incident events unless explicitly reported as new cases.

To characterise epidemic growth over time, log-linear regression models were fitted to cumulative confirmed case counts for three predefined periods: an early growth phase, an intermediate phase, and a late phase. The early phase covered SitReps 15–27, the intermediate phase covered SitReps 27–47, and the late phase covered SitReps 47–61. For each phase, the epidemic growth rate was estimated from the slope of the regression of the natural logarithm of cumulative confirmed cases on calendar time. Doubling time was calculated as T2 = ln(2)/r, where r represents the estimated daily growth rate. Model fit was assessed using the coefficient of determination, and estimates were interpreted descriptively because cumulative observations were serially correlated and subject to retrospective revision.

Six-month scenario projections were generated from 14 July 2026 to 14 January 2027 under three conditional assumptions. A logistic model represented rapid containment and progressive saturation; a Gompertz model represented slower but sustained deceleration; and a constant-incidence scenario assumed continuation of the mean absolute increase observed from 30 June to 14 July (47.6 confirmed cases per day). Logistic and Gompertz parameters were estimated by nonlinear least squares using all available cumulative-case observations. Residual bootstrap resampling with 1,000 iterations was used to derive empirical 95% uncertainty intervals. An unbounded exponential model was examined only as a stress test and was not retained as a plausible six-month forecast because it generated biologically and operationally unrealistic totals.

Province- and health-zone-level burden was summarized using cumulative confirmed cases, confirmed deaths, reported case fatality ratios, and proportional contribution to the national total. Geographic expansion was assessed by tracking the number of affected provinces and health zones over time. Reported case fatality ratios were calculated as cumulative confirmed deaths divided by cumulative confirmed cases multiplied by 100. These ratios were interpreted cautiously because they may have been influenced by delayed outcome reporting, community deaths, small denominators, incomplete case ascertainment, and retrospective reclassification. To quantify the statistical precision of province-level CFRs directly rather than relying solely on qualitative caution, exact Wilson score 95% confidence intervals were calculated for each province's CFR [13].

Demographic and exposure-related variables were analysed only when sufficiently disaggregated data were available. Because age, sex, occupation, household linkage, exposure category, and healthcare-worker status were not consistently reported across SitReps, no formal demographic modelling was undertaken.

2.3.1. National-Level Growth-Model Comparison

Four candidate nonlinear growth models exponential, logistic, Gompertz, and generalized-logistic (Richards) were fitted to the complete national aggregated cumulative-case series (21 unique data points spanning SitReps 1–61, combining the values reported in Table 1 and the values annotated in Figure 2, which agreed exactly on all overlapping SitReps) [14,15]. Models were fitted by nonlinear least squares and compared using the root-mean-square error (RMSE), the small-sample-corrected Akaike Information Criterion (AICc), and the Bayesian Information Criterion (BIC) [16,17]. This comparison was necessarily conducted at the national level: disaggregating growth-model comparison by province or health zone, would require case time series indexed by health zone and date, which were not available in the aggregated SitRep summaries used for this analysis.

Interval-level growth-rate and doubling-time trajectory. To complement the three-phase summary above, instantaneous growth rates and doubling times were additionally estimated between each pair of consecutive available data points (ri = [ln(yi+1) − ln(yi)] / Δti; Td,i = ln(2)/ri), producing a finer-grained trajectory of acceleration, deceleration, and plateauing than the three discrete phases alone (Figure 3).

Growth-rate reduction between phases. The proportional reduction in growth rate between successive phases was calculated as Δr = rearlier − rlater and as the percentage reduction (Δr / rearlier × 100), to quantify the magnitude of epidemic deceleration between phases between phases rather than describing deceleration only qualitatively.

Bayesian sensitivity analysis. As a complementary Bayesian sensitivity analysis, a Bayesian nonlinear Gompertz growth model with a negative-binomial observation model (to accommodate overdispersion in count data) was fitted to the national aggregated case series using Markov chain Monte Carlo sampling, implemented with the affine-invariant ensemble sampler in the Python package emcee [18,19]. Sampling used 40 walkers for 25,000 iterations, with an 8,000-iteration burn-in and a thinning interval of 25. Posterior draws of the expected epidemic trajectory were used to quantify parameter and trajectory uncertainty, while posterior-predictive draws additionally incorporated observation-level dispersion. This Bayesian analysis was conducted as a complementary sensitivity analysis and was reported separately from the residual-bootstrap intervals presented in Table 5.

Extended deterministic sensitivity scenarios. Two further scenarios were generated by proportionally scaling both the growth rate (r) and carrying capacity (K) of the Gompertz curve underlying the primary six-month projection (Section 3.6) by ±30%, representing an illustrative improved-control scenario (reduced transmission and lower ultimate epidemic size) and an illustrative insecurity-driven resurgence scenario (increased transmission and higher ultimate epidemic size). The Gompertz parameters used for this perturbation were those calibrated to reproduce the originally primary Gompertz projection (Section 3.6), rather than the independently refitted Gompertz model used for the model-comparison exercise above (Section 3.2.2), which was fitted to a coarser aggregated series for the purpose of comparing functional forms on equal terms; the two Gompertz fits are therefore not numerically identical and are not intended to be combined. These sensitivity scenarios are deterministic bounds rather than probabilistic forecasts.

Forecast calibration. Forecast calibration was assessed through rolling-origin temporal validation: the Gompertz model was re-fitted using only data available up to each of four earlier truncation points (SitReps 39, 44, 50, and 55), and the resulting projection for SitRep 61 was compared with the actually reported value.

All statistical analyses and visualisations were performed using R version 4.3.x and Python, including the pandas, NumPy, SciPy, stats models, emcee, and matplotlib libraries. A two-sided significance level of 0.05 was applied where inferential statistics were reported.

2.4. Ethical Considerations

All data are derived from publicly available, aggregated official surveillance reports. No individual-level patient data were accessed. Ethical approval was not required per applicable guidelines for secondary analysis of aggregated public health surveillance data.

3. Results

3.1. Overall Epidemic Trajectory

The epidemic expanded rapidly during the observation period. Confirmed cases increased from 8 in SitRep 1 on 14 May 2026 to 2,073 in SitRep 61 on 14 July 2026, representing a 259-fold increase over 61 days. During the same period, cumulative confirmed deaths increased from 4 to 796, while cumulative recoveries reached 377. The reported case fatality ratio increased from 16.0% in SitRep 15 on 29 May to 26.7% in SitRep 42 on 25 June, 31.2% in SitRep 47 on 30 June, and 38.4% in SitRep 61 on 14 July 2026. The number of affected health zones rose from 3 on 14 May to 22 by 29 May, 34 by 25 June, 37 by 6 July, and 45 by 14 July. The number of affected provinces increased from one to five following the recognition of Haut-Uélé and Tshopo in July 2026.

Overall, the updated trajectory indicates that transmission remained active through mid-July 2026. Although the estimated epidemic growth rate slowed over time, the outbreak had not entered a sustained decline by 14 July. Continued increases in confirmed cases, deaths, affected health zones, and provinces indicate ongoing community transmission and persistent gaps in early detection, referral, contact follow-up, and access to supportive care.

Figure 1. Confirmed Cases and Deaths Across SitReps 1–61, DRC Ebola (Bundibugyo) Outbreak, May 14th–July 14th 2026.

Figure 1. Confirmed Cases and Deaths Across SitReps 1–61, DRC Ebola (Bundibugyo) Outbreak, May 14th–July 14th 2026.

3.2. Epidemic Growth Rate and Doubling Time

3.2.1. Three-Phase Log-Linear Analysis

Log-linear regression applied to cumulative confirmed case counts showed progressive deceleration in epidemic growth. During the early phase (SitReps 15–27), the estimated daily growth rate was r = 0.078/day, corresponding to a doubling time of 8.9 days. During the intermediate phase (SitReps 27–47), the growth rate declined to r = 0.036/day, with a doubling time of 19.5 days. In the late phase (SitReps 47–61; 30 June–14 July 2026), the growth rate decreased further to r = 0.028/day, corresponding to a doubling time of 25.1 days.

These findings indicate a clear slowing of epidemic expansion between late May and mid-July 2026. However, the outbreak had not entered a sustained decline by 14 July, as cumulative confirmed cases continued to increase from 1,406 at SitRep 47 to 2,073 at SitRep 61. The progressive lengthening of the doubling time may reflect the combined effects of expanded Ebola Treatment Centre capacity, decentralized diagnostic services, intensified contact tracing, improved cross-border coordination, and broader implementation of response measures. Nevertheless, persistent community deaths, interruptions in response activities, variable contact follow-up, and continued geographic expansion to five provinces and 45 health zones indicate that containment remained incomplete.

These estimates should be interpreted cautiously because they are based on cumulative SitRep data that were repeatedly revised through DHIS2 cleaning, harmonisation, and reclassification. In addition, cumulative observations are serially correlated, and changes between consecutive SitReps do not always correspond exactly to incident cases. Therefore, the estimated growth rates and doubling times are best understood as descriptive indicators of epidemic trajectory rather than precise measures of transmission intensity.

Figure 2. Log-linear growth of confirmed cases across SitReps 15–61.

Figure 2. Log-linear growth of confirmed cases across SitReps 15–61.

The estimated doubling time increased from approximately 8.9 days during the early phase (r = 0.078/day) to 19.5 days during the intermediate phase (r = 0.036/day) and 25.1 days during the late phase (r = 0.028/day), indicating progressive deceleration of epidemic growth.

3.2.2. National-Level Growth Model Comparison

Four candidate nonlinear growth models exponential, logistic, Gompertz, and generalized-logistic (Richards) were fitted to the reconciled 21-point national aggregated case series and compared by RMSE, AICc, and BIC (Section 2.3.1). The Gompertz model provided the best fit by both AICc and BIC, followed by the Richards model, the logistic model, and, considerably worse, the unbounded exponential model, which is not biologically plausible over a six-month horizon (Table 2). The Gompertz model's asymmetric shape implying a comparatively rapid deceleration after the inflection point followed by a long, slowly saturating tail is consistent with the progressive but incomplete deceleration described in Section 3.2.1, and provides model-based support for the Gompertz curve chosen a priori for the six-month projection reported in Section 3.6.

3.2.3. Interval-Level Growth-Rate and Doubling-Time Trajectory

Instantaneous growth rates and doubling times, estimated between each consecutive pair of the 21 reconciled data points, provide a finer-grained view of epidemic deceleration than the three discrete phases in Section 3.2.1 (Figure 3). Growth was fastest immediately after the outbreak's recognition (r = 0.463/day between SitRep 1 and SitRep 5, doubling time 1.5 days), a value that reflects the very small denominator at SitRep 1 (8 cases) rather than sustained exponential transmission. The growth rate fell rapidly over the following intervals, from 0.184/day (doubling time 3.8 days) around SitRep 10–15 to below 0.05/day by SitRep 40, after which it declined more gradually, reaching 0.024/day (doubling time 28.6 days) in the final interval (SitRep 55–61). This trajectory shows an early deceleration phase, an intermediate approach toward a slower, comparatively stable growth rate, and no evidence of a plateau or reversal by SitRep 61.

Figure 3. Interval-level instantaneous growth rate (top) and doubling time (bottom) across SitReps 1–61, estimated between consecutive reconciled data points.

Figure 3. Interval-level instantaneous growth rate (top) and doubling time (bottom) across SitReps 1–61, estimated between consecutive reconciled data points.

3.2.4. Growth-Rate Reduction Between Phases

Using the three phase-specific growth rates reported in Section 3.2.1, the absolute and proportional reduction in growth rate between phases was calculated (Δr = rearlier − rlater; Table 3). The growth rate fell by 53.8% between the early and intermediate phases and by a further 22.2% between the intermediate and late phases, for an overall reduction of 64.1% between the early and late phases. The comparatively smaller proportional reduction between the intermediate and late phases, relative to the early-to-intermediate transition, indicates that epidemic deceleration became less pronounced by mid-July 2026, although the aggregated data do not permit attribution of this pattern to specific control measures.

3.3. Geographic Spread

The geographic footprint expanded substantially, increasing from 3 affected health zones in one province at SitRep 1 on 14 May 2026 to 45 health zones across five provinces by SitRep 61 on 14 July 2026, representing a 15-fold increase in the number of affected health zones.

The most rapid early expansion occurred during the second half of May, when the number of affected health zones rose from 7 on 19 May to 22 by 29 May. Geographic spread continued through June, reaching 35 health zones by 27 June and 36 by 30 June. During July, the outbreak expanded further with the recognition of additional affected zones and the formal inclusion of Haut-Uélé and Tshopo. By 14 July, affected areas included 26 health zones in Ituri, 11 in North Kivu, 1 in South Kivu, 4 in Haut-Uélé, and 3 in Tshopo.

Although the rate of newly affected health zones slowed relative to the initial expansion phase, geographic spread had not stabilised by mid-July.

Figure 4. Geographic Spread: Number of Affected Health Zones Over Time, DRC Ebola (Bundibugyo) Outbreak.

Figure 4. Geographic Spread: Number of Affected Health Zones Over Time, DRC Ebola (Bundibugyo) Outbreak.

3.4. Province-Level Comparisons

By SitRep 61 on 14 July 2026, Ituri remained the principal epicentre, accounting for 1,863 of 2,073 confirmed cases (89.9%) and 668 of 796 confirmed deaths (83.9%). North Kivu contributed a substantially smaller proportion of the national caseload, with 189 confirmed cases (9.1%), but accounted for 111 deaths (13.9%), corresponding to a markedly higher province-level CFR of 58.7% compared with 35.9% in Ituri. This disparity may reflect delayed case detection, higher proportions of community deaths, limited or delayed access to treatment, differential case ascertainment, or outcome-reporting biases rather than an intrinsic difference in disease severity.

The remaining provinces contributed relatively few cases. Haut-Uélé reported 14 cases and 13 deaths (CFR 92.9%), Tshopo 4 cases and 3 deaths (CFR 75.0%), and South Kivu 3 cases and 1 death (CFR 33.3%). These province-specific CFRs, particularly in Haut-Uélé and Tshopo, should be interpreted with considerable caution because they are based on very small denominators and are highly sensitive to the addition or reclassification of a single case or death. Exact Wilson score 95% confidence intervals were calculated for each province's CFR to make this denominator-driven imprecision explicit (Table 4): the intervals for Haut-Uélé (68.5–98.7%) and, especially, South Kivu (6.1–79.2%) are extremely wide, confirming that these point estimates alone should not be used to rank province-level severity. Overall, the provincial distribution at SitRep 61 indicates that the outbreak remained heavily concentrated in Ituri, while mortality was disproportionately high in North Kivu and in newly affected provinces with limited case counts.

Figure 5. Confirmed Deaths and CFR (%) by Province, SitReps 17-24-61.

Figure 5. Confirmed Deaths and CFR (%) by Province, SitReps 17-24-61.

3.5. Bubble Analysis: Cases, Deaths, and CFR

The bubble plot (Figure 6), in which bubble size is proportional to the reported case fatality ratio, illustrates marked heterogeneity in epidemic burden and mortality across the five affected provinces at SitRep 61. Ituri accounted for the overwhelming majority of confirmed cases and deaths, with 1,863 cases and 668 deaths, but had a comparatively moderate CFR of 35.9%. North Kivu reported a much smaller caseload 189 cases yet contributed 111 deaths, corresponding to a substantially higher CFR of 58.7%.

The newly affected provinces showed even higher reported CFRs despite very small case counts: 92.9% in Haut-Uélé (13 deaths among 14 cases) and 75.0% in Tshopo (3 deaths among 4 cases). South Kivu reported 3 cases and 1 death, yielding a CFR of 33.3%. These patterns suggest that high-burden settings and high-fatality settings were not necessarily the same. The elevated CFRs in North Kivu, Haut-Uélé, and Tshopo may reflect delayed case detection, community deaths, restricted access to treatment, incomplete ascertainment of mild cases, or outcome-reporting delays. However, estimates for provinces with small denominators should be interpreted cautiously as quantified directly by the Wilson confidence intervals in Table 4 because a single additional case or death can substantially alter the CFR.

3.6. Six-Month Scenario Projections

3.6.1. Primary Scenario Projections

The three prediction scenarios diverged substantially after the forecast origin, illustrating the sensitivity of long-range forecasts to assumptions about future control. The logistic model projected approximately 2,624 cumulative cases by 14 January 2027 (bootstrap 95% interval: 2,446–2,980). The Gompertz model projected approximately 3,942 cases (95% interval: 3,519–4,668). Continuation of the recent absolute incidence of 47.6 cases per day yielded approximately 10,839 cumulative cases. These scenarios represent conditional planning bounds rather than predictions of what will necessarily occur.

Table 5. Conditional Cumulative-Case Projections from 14 July 2026, Including Two Additional Deterministic Sensitivity Scenarios.

Scenario	Model	Basis	1 month	3 months	6 months	6-month uncertainty
Rapid containment	Logistic	Residual bootstrap 95% CI	2,535	2,623	2,624	2,446–2,980
Moderate deceleration	Gompertz (published)	Residual bootstrap 95% CI	3,031	3,787	3,942	3,519–4,668
Persistent recent incidence	Constant 47.6 cases/day	Not model-based	3,502	6,361	10,839	Not model-based
Improved control	Gompertz, r & K −30%	Deterministic sensitivity	1,525	2,339	2,701	Deterministic bound (not probabilistic)
Insecurity-driven resurgence	Gompertz, r & K +30%	Deterministic sensitivity	4,567	5,082	5,139	Deterministic bound (not probabilistic)

Note: The Gompertz curve underlying the two new sensitivity rows was calibrated to reproduce the originally primary Gompertz projection (row 2) exactly at the observed SitRep 61 value and at the three published projection horizons (K=3,953.96, b=3.9367, r=0.02963/day), and is distinct from the independently refitted Gompertz model used for functional-form comparison in Table 2 (Section 2.3.1). The two sensitivity rows should be read as illustrative deterministic bounds (±30% on r and K), not as probabilistic forecasts with a defined confidence level.

Table 5. Conditional Cumulative-Case Projections from 14 July 2026, Including Two Additional Deterministic Sensitivity Scenarios.

Scenario	Model	Basis	1 month	3 months	6 months	6-month uncertainty
Rapid containment	Logistic	Residual bootstrap 95% CI	2,535	2,623	2,624	2,446–2,980
Moderate deceleration	Gompertz (published)	Residual bootstrap 95% CI	3,031	3,787	3,942	3,519–4,668
Persistent recent incidence	Constant 47.6 cases/day	Not model-based	3,502	6,361	10,839	Not model-based
Improved control	Gompertz, r & K −30%	Deterministic sensitivity	1,525	2,339	2,701	Deterministic bound (not probabilistic)
Insecurity-driven resurgence	Gompertz, r & K +30%	Deterministic sensitivity	4,567	5,082	5,139	Deterministic bound (not probabilistic)

Note: The Gompertz curve underlying the two new sensitivity rows was calibrated to reproduce the originally primary Gompertz projection (row 2) exactly at the observed SitRep 61 value and at the three published projection horizons (K=3,953.96, b=3.9367, r=0.02963/day), and is distinct from the independently refitted Gompertz model used for functional-form comparison in Table 2 (Section 2.3.1). The two sensitivity rows should be read as illustrative deterministic bounds (±30% on r and K), not as probabilistic forecasts with a defined confidence level.

Figure 7. Six-month conditional projections. The divergence between scenarios reflects structural uncertainty and the unknown future effect of interventions.

Figure 7. Six-month conditional projections. The divergence between scenarios reflects structural uncertainty and the unknown future effect of interventions.

3.6.2. Bayesian Sensitivity Analysis

As a complementary Bayesian sensitivity analysis (Section 2.3.1), a negative-binomial Gompertz model fitted by MCMC (emcee) to the 21-point reconciled national series yielded posterior parameter estimates of K = 2,396.95 (95% credible interval [CrI] 2,082.95–2,727.82), b = 4.996 (95% CrI 4.784–5.255), r = 0.0497/day (95% CrI 0.0445–0.0565), and a negative-binomial dispersion parameter φ = 327.3 (95% CrI 102.0–1,216.4). The posterior median fitted value at SitRep 61 (1,883) under-predicted the actual observed value (2,073) by approximately 9%, a discrepancy discussed further below and in Section 4.

Table 6 compares, at each forecast horizon, the point estimate and bootstrap interval from the primary Gompertz projection (Section 3.6.1) with the Bayesian posterior expected-trajectory and posterior-predictive intervals. The Bayesian analysis produced a materially lower central estimate at every horizon than the originally primary bootstrap-based projection, together with a wider posterior-predictive interval at six months (1,948–2,839) than the bootstrap interval (3,519–4,668) the two intervals do not overlap. This divergence most plausibly reflects the different likelihoods and data weighting used by the two methods (least-squares fitting of a single point per SitRep-group versus a negative-binomial likelihood fitted to the full reconciled series, which gives relatively more influence to the earliest, smaller counts) rather than a single method being definitively correct; both are reported transparently as sensitivity analyses rather than reconciled into one number. Because the Bayesian median itself under-predicted the most recent observed count (Section 3.6.2, first paragraph), the true six-month burden may plausibly exceed even the wider Bayesian predictive interval, reinforcing rather than resolving the uncertainty already flagged for the moderate-deceleration scenario.

Figure 8. Bayesian negative-binomial Gompertz fit: posterior median expected trajectory (solid line) with 95% expected-trajectory credible band (darker ribbon) and 95% posterior-predictive band (lighter ribbon), against observed confirmed cases (points).

Figure 8. Bayesian negative-binomial Gompertz fit: posterior median expected trajectory (solid line) with 95% expected-trajectory credible band (darker ribbon) and 95% posterior-predictive band (lighter ribbon), against observed confirmed cases (points).

3.6.3. Extended Deterministic Sensitivity Scenarios

The two additional deterministic scenarios added in Table 5 (improved control and insecurity-driven resurgence, ±30% on r and K of the calibrated Gompertz curve) substantially widen the plausible six-month range: from approximately 2,701 cases under improved control to approximately 5,139 cases under an insecurity-driven resurgence, compared with the 2,624–10,839 range spanned by the three original scenarios. Figure 9 places all five scenarios on a common axis, showing a monotonic ordering from rapid containment through improved control, moderate deceleration (published), insecurity-driven resurgence, and persistent recent incidence.

3.6.4. Forecast Calibration: Rolling-Origin Validation

To assess how well the Gompertz projection would have performed if issued earlier in the outbreak, the Gompertz model was re-fitted using only data available up to four earlier truncation points and the resulting projection for SitRep 61 was compared with the value actually reported (Table 7). Projections issued earlier in the outbreak under-predicted the eventual SitRep 61 case count substantially by 27.0% when only data through SitRep 39 were available — and this under-prediction narrowed progressively as more data accrued, to 5.7% using data through SitRep 55. At every truncation point tested, the Gompertz model under-predicted rather than over-predicted the subsequent case count, a directionally consistent pattern that is discussed further in Section 4.

3.7. Demographic Profile

Demographic characteristics were not consistently disaggregated across the SitReps reviewed through 14 July 2026. Limited age- and sex-specific information was available in selected early reports, but the data were insufficient for robust longitudinal comparison or formal demographic modelling. Early summaries suggested that working-age adults constituted a substantial proportion of cases and that healthcare workers were affected; however, age-specific denominators, outcomes, occupational exposures, and healthcare-worker case counts were not reported consistently. These observations should therefore be regarded as descriptive signals rather than definitive demographic patterns. Individual-level line-list data would be required to evaluate age, sex, occupation, exposure history, healthcare-worker status, and clinical outcome reliably.

3.8. Key Epidemiological Findings Summary

Table 8. Summary of Key Epidemiological Findings, DRC Bundibugyo Virus Disease Outbreak, SitReps 1–61, 14 May–14 July 2026.

Finding	Updated detail	Priority
Rapid epidemic expansion	Confirmed cases increased from 8 on 14 May to 2,073 on 14 July 2026, representing a 259-fold increase over 61 days. Confirmed deaths reached 796.	CRITICAL
Persistent active transmission	Although proportional growth slowed over time, cumulative cases continued to rise through SitRep 61, including an increase from 1,406 cases on 30 June to 2,073 on 14 July. The epidemic had not entered a sustained decline.	CRITICAL
Ituri dominance	Ituri accounted for 1,863 of 2,073 confirmed cases (89.9%) and 668 of 796 deaths (83.9%) at SitRep 61, remaining the principal epidemic epicentre.	CRITICAL
Progressive growth deceleration	Estimated doubling time increased from 8.9 days in the early phase (r = 0.078/day) to 19.5 days in the intermediate phase (r = 0.036/day) and 25.1 days in the late phase (r = 0.028/day).	HIGH
Continued geographic spread	The number of affected health zones increased from 3 to 45, a 15-fold increase, while the number of affected provinces rose from one to five: Ituri, North Kivu, South Kivu, Haut-Uélé, and Tshopo.	CRITICAL
Rising national CFR	The reported national CFR increased from approximately 16.0% on 29 May to 38.4% on 14 July 2026. This may reflect true mortality, delayed detection, community deaths, incomplete ascertainment, and retrospective data reconciliation.	CRITICAL
Disproportionate mortality in North Kivu	North Kivu reported 189 cases and 111 deaths, corresponding to a CFR of 58.7% (Wilson 95% CI 51.6–65.5%), substantially higher than the 35.9% CFR in Ituri (95% CI 33.7–38.1%).	CRITICAL
Very high CFRs in low-count provinces	Haut-Uélé reported a CFR of 92.9% (95% CI 68.5–98.7%) and Tshopo 75.0% (95% CI 30.1–95.4%), but these estimates were based on very small denominators, as reflected in their wide Wilson confidence intervals, and should be interpreted cautiously.	HIGH
Small but persistent South Kivu burden	South Kivu reported 3 confirmed cases and 1 death, corresponding to a CFR of 33.3% (95% CI 6.1–79.2%).	MODERATE
Surveillance and response constraints	Persistent community deaths, variable contact follow-up, treatment-access limitations, insecurity, population mobility, and repeated DHIS2 data cleaning complicated interpretation and containment.	CRITICAL
Absence of consistently reported demographic data	Age, sex, occupation, healthcare-worker status, and exposure history were not consistently disaggregated across SitReps, preventing robust demographic or risk-factor analyses.	HIGH
Model-supported deceleration, quantified reduction	Among four candidate national growth models compared by AICc/BIC, the Gompertz curve was best supported; the national growth rate fell by 64.1% overall between the early and late phases (Δr = 0.050/day), with a smaller marginal reduction (22.2%) between the intermediate and late phases than between the early and intermediate phases (53.8%).	HIGH
Persistent under-prediction of future burden	Rolling-origin validation showed that Gompertz-based projections issued earlier in the outbreak under-predicted the subsequently observed SitRep 61 case count by 5.7–27.0%, narrowing but not eliminating the gap as more data accrued; a complementary Bayesian sensitivity analysis likewise produced a lower central six-month estimate (2,397) with a wide posterior-predictive interval (1,948–2,839) than the original bootstrap-based projection, suggesting the moderate-deceleration scenario in Table 5 may understate future burden.	CRITICAL
Limited disease-specific countermeasures	The response remained dependent primarily on early detection, isolation, contact tracing, infection prevention and control, and supportive clinical management. Any statement regarding the regulatory status of specific vaccines or therapeutics should be independently verified and referenced.	CRITICAL

Table 8. Summary of Key Epidemiological Findings, DRC Bundibugyo Virus Disease Outbreak, SitReps 1–61, 14 May–14 July 2026.

Finding	Updated detail	Priority
Rapid epidemic expansion	Confirmed cases increased from 8 on 14 May to 2,073 on 14 July 2026, representing a 259-fold increase over 61 days. Confirmed deaths reached 796.	CRITICAL
Persistent active transmission	Although proportional growth slowed over time, cumulative cases continued to rise through SitRep 61, including an increase from 1,406 cases on 30 June to 2,073 on 14 July. The epidemic had not entered a sustained decline.	CRITICAL
Ituri dominance	Ituri accounted for 1,863 of 2,073 confirmed cases (89.9%) and 668 of 796 deaths (83.9%) at SitRep 61, remaining the principal epidemic epicentre.	CRITICAL
Progressive growth deceleration	Estimated doubling time increased from 8.9 days in the early phase (r = 0.078/day) to 19.5 days in the intermediate phase (r = 0.036/day) and 25.1 days in the late phase (r = 0.028/day).	HIGH
Continued geographic spread	The number of affected health zones increased from 3 to 45, a 15-fold increase, while the number of affected provinces rose from one to five: Ituri, North Kivu, South Kivu, Haut-Uélé, and Tshopo.	CRITICAL
Rising national CFR	The reported national CFR increased from approximately 16.0% on 29 May to 38.4% on 14 July 2026. This may reflect true mortality, delayed detection, community deaths, incomplete ascertainment, and retrospective data reconciliation.	CRITICAL
Disproportionate mortality in North Kivu	North Kivu reported 189 cases and 111 deaths, corresponding to a CFR of 58.7% (Wilson 95% CI 51.6–65.5%), substantially higher than the 35.9% CFR in Ituri (95% CI 33.7–38.1%).	CRITICAL
Very high CFRs in low-count provinces	Haut-Uélé reported a CFR of 92.9% (95% CI 68.5–98.7%) and Tshopo 75.0% (95% CI 30.1–95.4%), but these estimates were based on very small denominators, as reflected in their wide Wilson confidence intervals, and should be interpreted cautiously.	HIGH
Small but persistent South Kivu burden	South Kivu reported 3 confirmed cases and 1 death, corresponding to a CFR of 33.3% (95% CI 6.1–79.2%).	MODERATE
Surveillance and response constraints	Persistent community deaths, variable contact follow-up, treatment-access limitations, insecurity, population mobility, and repeated DHIS2 data cleaning complicated interpretation and containment.	CRITICAL
Absence of consistently reported demographic data	Age, sex, occupation, healthcare-worker status, and exposure history were not consistently disaggregated across SitReps, preventing robust demographic or risk-factor analyses.	HIGH
Model-supported deceleration, quantified reduction	Among four candidate national growth models compared by AICc/BIC, the Gompertz curve was best supported; the national growth rate fell by 64.1% overall between the early and late phases (Δr = 0.050/day), with a smaller marginal reduction (22.2%) between the intermediate and late phases than between the early and intermediate phases (53.8%).	HIGH
Persistent under-prediction of future burden	Rolling-origin validation showed that Gompertz-based projections issued earlier in the outbreak under-predicted the subsequently observed SitRep 61 case count by 5.7–27.0%, narrowing but not eliminating the gap as more data accrued; a complementary Bayesian sensitivity analysis likewise produced a lower central six-month estimate (2,397) with a wide posterior-predictive interval (1,948–2,839) than the original bootstrap-based projection, suggesting the moderate-deceleration scenario in Table 5 may understate future burden.	CRITICAL
Limited disease-specific countermeasures	The response remained dependent primarily on early detection, isolation, contact tracing, infection prevention and control, and supportive clinical management. Any statement regarding the regulatory status of specific vaccines or therapeutics should be independently verified and referenced.	CRITICAL

4. Discussion

Although the 2026 Bundibugyo virus disease outbreak in the Democratic Republic of the Congo showed a progressive deceleration in proportional growth, the epidemiological situation remained critical. The estimated doubling time increased from 8.9 days during the early phase to 19.5 days during the intermediate phase and 25.1 days during the late phase, an overall 64.1% reduction in the national growth rate between the early and late phases (Section 3.2.4). This slowing should not be interpreted as evidence of outbreak control. By 14 July 2026, more than 2,000 confirmed cases had been reported, alongside increasing mortality and continued geographic expansion into additional health zones and provinces. Transmission therefore remained intense, even though the epidemic curve had become less steep than during the initial phase. Two structural factors may have contributed substantially to the rapid early expansion: delayed recognition and confirmation of the outbreak, and the absence of licensed BDBV-specific vaccines and therapeutics, which limited the applicability of ring-vaccination strategies used during the 2018–2020 North Kivu Zaire ebolavirus outbreak [20,21].

The reported WHO declaration of a Public Health Emergency of International Concern shortly after the national outbreak declaration reflected concern regarding rapid geographic spread, the humanitarian context of eastern DRC, and the limited availability of species-specific medical countermeasures [22]. In this setting, clinical management depended largely on supportive care, while investigational products such as MBP134/MBP431, obeldesivir, and remdesivir required evaluation within appropriately regulated emergency-use or adaptive clinical-trial frameworks. These constraints highlight the comparatively limited clinical and operational evidence base available for BDBV relative to Zaire ebolavirus.

Marked heterogeneity in reported case fatality ratios between provinces was among the most operationally important findings. By SitRep 61, North Kivu had a CFR of 58.7% (Wilson 95% CI 51.6–65.5%), compared with 35.9% in Ituri (95% CI 33.7–38.1%); the non-overlapping confidence intervals for these two high-case-count provinces indicate that this difference is unlikely to be attributable to chance alone, even though it may still reflect reporting and ascertainment differences rather than a true difference in disease severity. This difference is unlikely to be explained solely by the natural history of BDBV. During the 2007 Uganda outbreak, the reported CFR was approximately 32% [23,24]. Several mechanisms may account for the higher CFR in North Kivu: delayed detection, confirmation of cases only at or near death, limited access to Ebola treatment centres, a high proportion of community deaths, and delayed care-seeking related to insecurity or mistrust of health authorities. These mechanisms are likely to be interrelated and may also have influenced the very high CFRs observed in newly affected provinces with small case counts; the correspondingly very wide Wilson intervals for Haut-Uélé (68.5–98.7%), Tshopo (30.1–95.4%), and South Kivu (6.1–79.2%) confirm that these point estimates should not be used, on their own, to rank province-level disease severity (Section 3.4).

The estimated growth rates and doubling times should nevertheless be interpreted cautiously. They were derived from cumulative SitRep observations that were serially correlated and subject to reporting delays, retrospective revisions, reclassification, and model uncertainty [25,26]. Operational constraints affecting surveillance, referral, laboratory confirmation, contact tracing, and outbreak control may also have influenced the observed trajectory [20]. The progressive lengthening of the doubling time is consistent with epidemic deceleration, but it does not establish that transmission was declining in absolute terms. Growth indicators should therefore be interpreted alongside incident case counts, community deaths, contact follow-up performance, geographic expansion, and treatment-centre occupancy. The reported international dimension of the outbreak further highlighted gaps in preparedness for non-Zaire orthoebolaviruses. Reports of cases in Uganda and an imported case in France, if confirmed by authoritative sources, would underline the importance of cross-border surveillance and international coordination [22,27,28]. Point-of-entry screening for Ebola disease remains challenging because early BDBV symptoms are non-specific and overlap with common febrile illnesses. Screening strategies should therefore emphasise epidemiological linkage, exposure history, longitudinal symptom monitoring, and rapid access to molecular confirmation rather than reliance on haemorrhagic manifestations alone. The DRC–Uganda cross-border response plan, coordinated through regional mechanisms, included rapid response teams, strengthened surveillance, laboratory confirmation, and infection prevention and control activities. Its effectiveness depends on two complementary components: real-time analysis of contact-tracing performance, case-detection delays, and epidemiological indicators; and sustained community engagement based on culturally appropriate risk communication, accepted safe-burial practices, and meaningful community participation. In settings affected by displacement, insecurity, and mistrust, community engagement should be treated as a core response function rather than a supporting activity. Reporting of community participation, refusal rates, and risk-perception indicators alongside epidemiological metrics would improve adaptive outbreak management [29,30].

By 14 July 2026, the provincial distribution remained highly asymmetric. Ituri accounted for 1,863 confirmed cases across 26 health zones, North Kivu reported 189 cases across 11 health zones, and South Kivu recorded 3 cases in one health zone. Haut-Uélé and Tshopo contributed relatively few cases but reported very high CFRs. These contrasts may reflect differences in transmission intensity, population mobility, surveillance sensitivity, diagnostic access, health-seeking behaviour, and retrospective geographic reassignment. Because provincial contact-follow-up indicators were not consistently reported across all SitReps, comparisons between provinces should be made cautiously and should not be interpreted as direct measures of response effectiveness [31,32].

4.1. Model Comparison, Growth-Rate Reduction, and Forecast Reliability

The multi-model comparison provides model-based support for the Gompertz curve used in the primary six-month projection: among four candidate national-level growth models, Gompertz was best supported by both AICc and BIC, and the Richards (generalised-logistic) model which nests both logistic and Gompertz-like behaviour performed similarly well, consistent with genuine asymmetric deceleration rather than symmetric logistic saturation [14,15]. However, three findings from the additional analyses temper confidence in the specific numerical projections reported in Table 5. First, rolling-origin validation showed that Gompertz-based projections issued earlier in the outbreak systematically under-predicted the subsequently observed case count, by as much as 27.0% when only data through SitRep 39 were available and still by 5.7% using data through SitRep 55 the most recent truncation point tested. Second, the complementary Bayesian sensitivity analysis (Section 3.6.2), fitted independently using a negative-binomial likelihood, produced a central six-month estimate (2,397) well below the originally primary bootstrap-based projection (3,942), with a posterior-predictive interval (1,948–2,839) that does not overlap the original bootstrap interval (3,519–4,668); the Bayesian model's own median fitted value at the most recent observed SitRep (1,883) itself under-predicted the actual count (2,073). Third, the growth-rate reduction between the intermediate and late phases (22.2%) was proportionally smaller than the reduction between the early and intermediate phases (53.8%). This pattern may be consistent with diminishing gains in epidemic control, although the aggregated data do not permit attribution of changes in growth rate to specific interventions (Table 3). Taken together, these findings highlight substantial structural uncertainty in the six-month projections and indicate that neither the moderate-deceleration nor rapid-containment scenario should be interpreted as an upper bound on future burden. All projections should therefore be considered conditional planning scenarios rather than deterministic forecasts [16,17,18,19].

4.2. Limitations and Future Research

The analyses reported here were conducted at the national and province levels because the aggregated SitRep summaries used in this study (Section 2.3.1) do not resolve case counts by health zone and date. This bears directly on interpretation. The three-phase deceleration we document nationally, and the widening forecast-calibration error at earlier truncation points (Section 3), are consistent with either a broadly uniform slowdown or a national curve dominated by localised flare-ups offset by decline elsewhere. The second reading is plausible: accounts of the 2018–2020 Nord-Kivu/Ituri epidemic describe transmission as spatially heterogeneous, with resurgences in specific health zones (e.g. Katwa, Butembo, Mabalako) tied to insecurity and interrupted response access [33,34]. A single national growth-model fit averages over exactly this structure.

The same gap limits covariate analysis. Forest cover has been linked to Ebola spillover location [35]; road infrastructure and mobility have shaped this outbreak's spatial trajectory [36]; and healthcare accessibility in Nord-Kivu is markedly uneven [37]. Testing these associations against growth rate, doubling time or final size requires health-zone-indexed case series and matched geospatial covariates, which the aggregated data cannot provide; we report this as open rather than approximate it. Trajectory-clustering of epidemic types [38] and daily-resolution early-warning indicators acceleration, autocorrelation, spatial synchrony [39,40,41] face the same constraint.

We regard health-zone-level line-list acquisition as the highest-priority extension, since disaggregated growth-model comparison, covariate analysis and clustering all follow directly from it; intervention-timing counterfactuals [42,43] are a natural subsequent step once that baseline exists.

Several limitations constrain this analysis. The study relied on aggregated surveillance data, which are vulnerable to reporting delays, incomplete ascertainment, retrospective reconciliation, and changing case classification. CFRs calculated from cumulative counts during an active outbreak may overestimate or underestimate the true risk of death because outcomes are not yet known for all cases and mild infections may be under detected. Demographic and exposure data were inconsistently disaggregated, limiting risk-factor analysis. The regression models were based on cumulative rather than incident case counts and did not explicitly account for serial correlation, changes in surveillance intensity, or intervention timing. Despite these limitations, the consistency of the overall epidemic pattern across INSP, WHO, and UN sources supports the qualitative conclusion that the outbreak remained active, geographically expanding, and operationally demanding through mid-July 2026.

5. Conclusion

The 2026 Bundibugyo virus disease outbreak remained active through mid-July despite progressive deceleration in proportional growth. Continued mortality, geographic expansion, and sustained incidence show that transmission had not yet been brought under control. The six-month projections demonstrate that future burden will depend strongly on the timeliness and effectiveness of containment and should be interpreted as conditional scenarios rather than deterministic forecasts. A multi-model comparison, growth-rate-reduction analysis, complementary Bayesian sensitivity analysis, extended deterministic sensitivity scenarios, and rolling-origin forecast validation provided additional evidence regarding epidemic dynamics and forecast uncertainty. The divergence between modelling approaches and the systematic under-prediction observed in rolling-origin validation indicate substantial structural uncertainty and suggest that the moderate-deceleration scenario should not be interpreted as an upper bound on future burden. The significance of this analysis lies in showing how routinely reported SitRep data can provide timely operational intelligence when growth rates and doubling times are interpreted alongside incident cases, community deaths, contact follow-up, treatment access, geographic spread, and an explicit, multi-method characterization of forecast uncertainty. These findings support intensified surveillance, earlier case detection, improved access to supportive care, accelerated development and evaluation of BDBV-specific medical countermeasures, the acquisition of health-zone-level time series and spatial covariate data needed to extend growth-model comparison, clustering, and intervention counterfactual analysis below the province level.