The Price of Late Detection: What a CDC Model Reveals About the 2026 Ebola Bundibugyo Outbreak

The Ebola outbreak spreading across the Democratic Republic of the Congo (DRC) and into Uganda has become one of the most rapidly escalating in the virus’s recorded history. To understand the sheer scale of this outbreak, we only need to look at the approximately 35,000 contacts that the Africa Centers for Disease Control and Prevention (Africa CDC) is attempting to follow.¹ The explosive growth is the focal point of a new transmission model from the US. CDC, published earlier this month.²

How many people might already have had Ebola before public health even knew there were cases? On June 5, 2026, the CDC published a modeling study in the Morbidity and Mortality Weekly Report (MMWR) that provides some insight.² Using a transmission model built for this outbreak, analysts projected that without a large and sustained response, this outbreak could rival the 2014 to 2016 West Africa epidemic.

The weeks since have made the warning ever more real. As of June 23, 2026, the CDC reported more than 1,000 confirmed cases in the DRC, with the case count climbing faster and higher than the 2 prior Ebola Bundibugyo outbreaks on record (Figure 1).³

What Does the Transmission Model Predict?

The CDC used a branching process model: a method that starts with a single infection and simulates the tree of transmission that grows outward from it.² Analysts ran that simulation under 4 levels of a single intervention, the share of symptomatic patients who are found, isolated, and treated before they can infect anyone else: 20% (poor), 50% (moderate), 70% (high), and 95% (extremely high). Because the true number of deaths early in the outbreak was uncertain, they calibrated the model 3 ways, to 50, 100, and 200 cumulative deaths as of May 24.²

The results line up along a single gradient (Figure 2). The more symptomatic patients are found and isolated before they can infect others, the steeper the odds of a catastrophic outbreak fall, in projected cases and projected deaths alike. At poor isolation, very large outbreaks dominate the simulations; at high isolation, they become the exception rather than the rule. The worst outcomes never fully vanish. They simply become rare once enough patients are isolated in time.

The lever, in other words, is isolation. And isolation is the daily work of infection prevention.

Why Is It Already This Big?

Here, the model says something counterintuitive and worth sitting with. It did not find an unusually contagious virus. The estimated basic reproductive number was approximately 2.51, squarely within the normal range for Ebola.² The high odds of a very large outbreak come almost entirely from one fact: how big the outbreak already was on the day it was first confirmed.

The model also worked backward, inferring when the outbreak most likely began (Figure 3). Calibrated to the lowest death toll, it placed the spillover from an animal reservoir to a human months before the first laboratory confirmation in May.² Calibrated to a higher death toll, the inferred start moved earlier still, and the projected outbreak grew larger, because the virus had simply been spreading in the community for longer before public health intervened. Even a strong isolation effort cannot fully undo that head start.

At its core, the data shows a direct link: delay detection, and you all but ensure a larger, longer epidemic.^2,4 A late-seen outbreak is a large outbreak, almost by definition.

This is a principle we already know in practice. The model puts a human cost on it.

What a Projection Can and Cannot Foretell

A projection is not a prediction, and the honest reading of this one depends on understanding its limitations.² The true number of Bundibugyo virus disease (BVD) deaths in late May was unknown, so the model was calibrated to a range rather than a fact. Estimates of the reproductive number vary widely across Ebola outbreaks, and the real value here may sit above or below the figure used. The model did not account for people changing their own behavior to avoid infection, for rising immunity in affected communities, or for the rare phenomenon of relapse in survivors.² Nor did it factor in the on-the-ground realities now shaping the response: the strain on contact tracing, community mistrust of public health, and ongoing conflict in the region.

Each of these limitations cuts a different way. Some would make the actual outbreak smaller than the simulations suggest; others, even larger. The purpose of a scenario projection is not to forecast a precise case count. It is to show how sharply the outcomes diverge depending on a variable we can actually move.

That variable is how fast we find cases and either confirm or rule them out.

The Illusion of a Less Virulent Strain

Early reports suggest some patients are experiencing illness on the milder end of the spectrum, consistent with the lower fatality rates seen in this strain’s 2 prior outbreaks: 32% in Uganda in 2007 and 34% in the DRC in 2012, against the 80% to 90% that the Zaire species has reached in past epidemics.⁵ These observations are preliminary, and the true death toll is almost certainly higher than the confirmed count reflects.

A milder course, however, is not the reassurance it appears to be. Less severe illness can keep infected people moving through their communities longer, increasing the number of exposures and delaying their seeking care and getting tested. The model has already told us what happens when cases are found late.

Ironically, a less virulent strain can be the harder one to contain.

What Does an Ebola Model Have to Do With My Facility?

In terms of Ebola preparedness, very little. The CDC continues to assess the risk to the general US population as low.⁶ The relevance of this model to your day-to-day work is the principle. The first clusters of this outbreak were identified among health care workers (HCWs), and the World Health Organization (WHO) has been explicit that BVD transmission is amplified in health care settings when infection prevention and control (IPC) measures fall short.^7,8 The same basic inputs that drive this outbreak’s projection are universal. The cost of a positive patient identified too late is paid by the HCWs, family, visitors, and other patients who were exposed in the meantime. Every surveillance trigger we maintain, every isolation protocol we teach, every hour shaved off the gap between a patient walking in and a clinician asking the right clinical question is, in miniature, the lever this model measures.

When IPC processes are in effect, a potentially infectious patient is identified and placed in appropriate isolation while diagnostic testing is conducted. No exposures. No outbreak.

When they fail, that same patient is not. And every hour that passes without detection multiplies the risk of exposure. The cost of a delayed diagnosis is always paid in the cases that follow.

Which Trajectory Is This Outbreak On?

The model did not predict a single outcome. It offered three paths, contingent on how well and how quickly the outbreak can be contained. Based on what we know so far, which one are we on?

Start with the dead. The model spread its scenarios across 3 plausible late-May death tolls: 50, 100, and 200, and the true figure was never certain.² But the confirmed toll has since climbed past 269 and is still rising,³ already beyond the highest of those calibrations, the one built on the earliest spillover and the longest stretch of unseen spread. The isolation picture points in the same direction. Even if every one of the roughly 35,000 contacts were reached and agreed to isolate, the model still leaves a large outbreak as a substantial risk (Figure 3). And the case count confirms it: more than 1,000 infections, climbing faster than in either prior outbreak of this strain (Figure 1), while responders are still chasing tens of thousands of contacts and the peak, by Africa CDC's assessment, has not yet arrived.¹

None of this is set in stone. We are roughly one month into the model's 3-month horizon. The official counts almost certainly trail the real ones, and the single variable that decides the outcome is the ability to identify and isolate positive patients. If case finding and isolation reach the high end of the model’s description, the worst projections are avoided. Thankfully, there is time.

These projections do not tell us what will happen, but they do inform us what can still be changed, and the consequences of delayed action.

Find it early or count it later.

References

1. Getachew S, Musambi E. Africa CDC chief says the continent needs to invest its own funds in Ebola response, vaccine. AP News. June 19, 2026. Accessed June 23, 2026. https://apnews.com/article/congo-ebola-africa-cdc-ituri-a5bfda53dbef567146cc1b39cce6f3f3
2. Mooring EQ, Koval WT, Routledge I, et al. Modeled scenario projections for the Ebola disease outbreak caused by Bundibugyo virus, 2026. MMWR Morb Mortal Wkly Rep. 2026;75(22):285-289. doi:10.15585/mmwr.mm7522e1
3. Ebola outbreak: current situation. CDC. Accessed June 23, 2026. https://www.cdc.gov/ebola/situation-summary/index.html
4. Matson MJ, Chertow DS, Munster VJ. Delayed recognition of Ebola virus disease is associated with longer and larger outbreaks. Emerg Microbes Infect. 2020;9:291-301. doi:10.1080/22221751.2020.1722036
5. Ebola outbreak history. CDC. Accessed June 23, 2026. https://www.cdc.gov/ebola/outbreaks/index.html
6. Richard DM, Routledge I, Koeller S, et al. Assessment of risk to the US population from the Ebola disease outbreak caused by Bundibugyo virus, 2026. MMWR Morb Mortal Wkly Rep. 2026;75(22):290-292. doi:10.15585/mmwr.mm7522e2
7. Ebola disease caused by Bundibugyo virus, Democratic Republic of the Congo and Uganda. Disease Outbreak News. World Health Organization. Accessed June 23, 2026. https://www.who.int/emergencies/disease-outbreak-news
8. Zomahoun DL, Boyd MA, Honein MA, et al. Notes from the field: outbreak of Ebola disease caused by Bundibugyo virus, Democratic Republic of the Congo and Uganda, May 2026. MMWR Morb Mortal Wkly Rep. 2026;75(22):293-294. doi:10.15585/mmwr.mm7522e3