In 2022, cholera tore through Uganda’s Adjumani refugee settlements with devastating force. Nearly six out of every hundred infected people died, a mortality rate that shook even experienced epidemiologists. It was six times higher than what the World Health Organization considers acceptable even in emergencies. While most observers saw an inevitable tragedy, Peter Mukiibi and Dr. Prisca Kizito saw something different: a solvable problem hidden in the data.
The statistics painted a stark picture. Refugee-hosting districts were generating roughly 60 percent of Uganda’s cholera burden, yet these areas represented only a small fraction of the country’s population. Meanwhile, health surveillance systems were detecting outbreaks only after the disease had already been spreading quietly for weeks. By the time health workers knew what was happening, hundreds of people were already sick, and the critical window for prevention had slammed shut.
Today, Mukiibi and Kizito have engineered what public health experts consider a genuine breakthrough. As the technical architect behind the Multimodal AI for Early Detection of Cholera Outbreaks in Uganda’s Refugee Settlements, Mukiibi has created the first predictive surveillance system specifically designed for the messy, complicated reality of displaced populations in sub-Saharan Africa. Working closely with Kizito, whose medical expertise ensured the system would work in real clinics with real patients, the team has fundamentally changed how health authorities can prepare for epidemics before they start.
The system now forecasts district-level outbreaks eight to twelve weeks before the first patient shows symptoms. That lead time is transformative. Health ministries in Kenya, Tanzania, and Mauritius are rushing to implement the framework, recognizing it addresses vulnerabilities that extend far beyond Uganda’s borders.
A Track Record Built on Solving Real Problems
This breakthrough didn’t come out of nowhere. Mukiibi has spent years applying engineering thinking to concrete health challenges in East Africa. In previous work, he developed integrated surveillance platforms that connected veterinary and human health data to track diseases that jump between animals and people.
He built cold-chain monitoring systems that help ensure vaccines stay effective as they move through districts where electricity is unreliable and roads wash out during rainy season.
These earlier systems demonstrated his ability to build technology that functions in difficult environments. They shared common principles: start by understanding what health workers face every day, bring together data from multiple sources, and design for places where the power might cut out and internet might crawl at speeds that would frustrate anyone used to fiber optics. But the cholera system represents something ambitious, not just monitoring what’s happening right now but predicting the future before anyone gets sick.
From American Education to African Implementation
The foundation for this innovation took shape in 2024, when Mukiibi and Kizito were graduate students at Brandeis University in Waltham, MA, United States, pursuing their master’s degree in Global Health Policy and Management. Their professor, Moaven Razavi, encouraged them to dig into a question that had frustrated epidemiologists for years: why does disease surveillance consistently fail the people who need it most?
The idea was crystallized during their time at Brandeis, where they had access to advanced computational resources, machine learning expertise, and the interdisciplinary environment needed to tackle such a complex problem. They started asking hard questions about refugee health that most surveillance systems simply ignored.
The challenge they were tackling was enormous. Uganda hosts over a million refugees, making it one of the world’s largest refugee host nations. Most of these people fled countries where cholera is a constant threat: South Sudan, the Democratic Republic of Congo, Burundi. They settled in districts that were already struggling with crumbling infrastructure and insufficient resources.
In some refugee settlements, nearly half the residents depend on untreated water sources. In others, dozens of people share a single latrine, far exceeding what humanitarian guidelines recommend. Traditional surveillance systems, built for stable populations with predictable patterns, couldn’t begin to capture what was really happening in these communities.
“The models we had were treating every district the same,” Mukiibi explained. “They couldn’t see the difference between a settlement where half the people are drinking untreated water and a district with modern infrastructure. They didn’t understand how people moving between South Sudan and Uganda during the dry season could create pathways for disease, or how new arrivals from an area where cholera is endemic could spark an outbreak.”
“What we were seeing in the field was heartbreaking,” Kizito added. “Health workers would get overwhelmed by outbreaks they never saw coming. We needed to give them real intelligence with enough time to actually do something about it, not just confirmation of what they could already see walking through their clinics.”
The prototype they developed at Brandeis University immediately caught attention when they presented it at the Heller Cybersecurity Conference. The technical sophistication, integrating artificial intelligence with real-time disease surveillance in ways that hadn’t been done before was impressive. But what really resonated was that you could tell they understood the actual problems health workers face, not just the theoretical ones.
When they presented again at the Brandeis Africa Forum later that year, something shifted. International health organizations and African health ministries realized this wasn’t just an interesting academic exercise but a working solution developed by people who understood both the technology and the communities it would serve.
Mukiibi started building partnerships with implementation organizations while still finishing his degree. Kizito’s credentials as an emergence physician proved crucial in convincing skeptical health officials that this system wouldn’t just create more work for people who were already stretched impossibly thin. Within months of graduating, the system was live in Uganda’s refugee-hosting districts, processing real-time data and generating predictions.
At its heart, the system uses a specialized form of an AI algorithm called an LSTM neural network, which is particularly good at finding patterns in data that unfolds over time. But what makes version different is the engineering and how its customized to work in places where resources are scarce, and conditions are unpredictable.
The system pulls together four different streams of real-time information:
Health surveillance data from Uganda’s district health system, which tracks disease patterns across thousands of health facilities nationwide. The algorithm looks for subtle signals that most people would miss: a modest uptick in diarrheal illness, a shift in how much oral rehydration solution clinics are using, cases of acute stomach illness clustering in a specific area. By watching weather patterns weeks and months ahead, the system can anticipate when conditions are ripening for transmission.
Population movement data from Uganda’s statistics bureau, combined with refugee registration information from UNHCR. This helps the algorithm understand how new arrivals from areas where cholera is common might change the risk profile in a particular settlement.
Detailed settlement mapping that tracks everything from how many water sources are working to how far people must walk to find clean water. The system monitors hundreds of variables for each settlement, building a granular picture of vulnerability.
Mukiibi calls the result “refugee-aware modeling.” The system explicitly accounts for the unique circumstances of displaced populations. It understands that a newly arrived refugee family settling in a camp near the border carries different risks than people who’ve lived in Kampala for generations with access to treated municipal water. It recognizes that a settlement where infrastructure is crumbling creates fundamentally different outbreak dynamics than a well-resourced district.
What makes it practically viable is that after the initial training phase, the system runs on ordinary computers drawing minimal power, which matters as most settings have no electricity or its intermittent and expensive. Computational efficiency reflects years of experience building systems that must keep working when everything else is failing.
In testing against historical outbreak data, the system correctly predicted roughly nine out of ten cholera outbreaks weeks before they happened, with relatively few false alarms. That’s substantially better than previous early warning systems, which typically got it right maybe six times out of ten while crying wolf far more often.
Proof in the Field: Kyangwali Settlement
The real test came during Uganda’s 2025 rainy season. In March, the algorithm flagged Kyangwali settlement in western Uganda as extreme risk, projecting an outbreak would hit in late May.
On paper, the prediction made no sense. Kyangwali hadn’t seen cholera in years. Local health officials were understandably skeptical. But the algorithm had spotted something humans easily miss: several hundred new refugee arrivals from an area of Congo where cholera was active, settlement water infrastructure that was falling apart, and weather forecasts showing unusually heavy rains ahead.
Acting on the prediction, district authorities pre-positioned thousands of oral cholera vaccine doses, sent mobile health teams into the settlement for preventive outreach, and implemented emergency water treatment protocols. The interventions weren’t cheap for a resource-constrained district. When the rain came in April, harder and earlier than anyone expected, flooding contaminated the settlement’s main water sources exactly as the system had predicted. Cholera cases appeared in the third week of May, right on schedule.
But because of the advance warning, the outbreak contained fewer than fifty confirmed cases with no deaths. Researchers who analyzed what happened afterward estimated that without the predictive intervention, Kyangwali likely would have seen 550-700 cases and potentially about 40 deaths, based on how similar outbreaks had played out historically. The upfront investment in prevention ended up averting emergency response costs estimated at nearly a million dollars.
“What happened in Kyangwali fundamentally changed our thinking,” says Dr. Jane Aceng, Uganda’s Minister of Health. “We went from hoping we could respond fast enough to knowing we can actually prevent outbreaks before anyone dies. That’s transformational.”
Expanding Across East Africa
The success in Uganda has sparked rapid adoption across the region. Kenya is implementing the system in its massive refugee complexes in the north and Tanzania is adapting it for camps hosting hundreds of refugees from Burundi and Congo. Even Mauritius, which doesn’t host large refugee populations but faces cholera risks from maritime trade, is customizing the framework to monitor its port cities.
“What makes this so powerful is how adaptable it is,” explains Dr. Matshidiso Moeti, WHO Regional Director for Africa. “This isn’t some rigid system that only works in one place. It’s a framework that genuinely understands vulnerability and disease transmission in ways that can be adjusted for completely different contexts.”
The World Health Organization has expressed formal interest in incorporating the framework into its global cholera control strategy, which aims to dramatically reduce cholera deaths by 2030.
Mukiibi’s work has gained recognition beyond East Africa majorly at public health conferences in Africa and beyond. In uganda, the system currently protects over a million people across refugee-hosting districts. As implementations roll out over the coming months, that coverage will expand to protect more people across East Africa.
Health economists who’ve analyzed the numbers suggest the cost-effectiveness is compelling. If the system achieves even a modest reduction in cholera deaths across implementing countries, it could prevent thousands of deaths annually while saving health systems tens of millions in emergency response costs. The cost per death prevented compares favorably to vaccination campaigns and is dramatically better than reactive outbreak response.
What This Means for the Future
This innovation arrives at a moment when the old ways of responding to health crises are clearly breaking down. The COVID-19 pandemic showed us how catastrophically expensive reactive response can be, both in money and in human suffering. Climate change is steadily expanding the geographic range of waterborne diseases. Conflict keeps displacing people at scales we’ve never seen before.
The traditional approach of waiting for crises to explode, then scrambling to contain the damage, simply isn’t sustainable anymore.
What Mukiibi and Kizito have demonstrated is a different path: predictive prevention that sees crises coming and stops them before they start. The same principles could apply to other epidemic-prone diseases. Research is already underway to adapt the framework for measles, meningitis, typhoid fever, and other threats.
Beyond preventing disease, the system generates valuable information about infrastructure gaps and climate vulnerabilities that help with long-term planning. Humanitarian organizations now use risk assessments to decide where to invest in water systems. NGOs use maps to target health education. Government planners use the predictions to shift budgets from emergency response toward prevention.
“What’s been demonstrated here is that we can actually move from perpetual crisis mode to strategic prevention,” says Sir David Nabarro, Global health leader, WHO Special Envoy on COVID-19. “All the pieces existed. The technology, the data, the need. What was missing was someone with the vision and skill to put it all together in a way that works for the people who need it most.”
A Different Model for Innovation
Perhaps what matters most about this story is what it shows us about where real innovation comes from in global health. This system didn’t emerge from a gleaming tech campus in California or a well-funded research institute in Geneva. It came from someone who spent years working inside the health systems he was trying to improve, who understood the daily realities facing health workers in places most people will never visit, combined with advanced training and resources available at a major American university where the original idea took root.
“This is what genuine global health leadership looks like,” says Dr. Agnes Binagwaho, who served as Rwanda’s Minister of Health and now leads a university focused on health equity. “Not outsiders imposing solutions they think should work, but innovators who bring together deep knowledge of local context with world-class technical training to develop interventions precisely designed for specific challenges.”
For the million-plus refugees in Uganda and the millions more living in precarious conditions across Africa, this combination of technical sophistication and real-world understanding means something profound. It means that disease outbreaks might become something we can see coming and prevent, rather than disasters we simply endure.
In refugee settlements where every week of warning can mean the difference between a handful of cases and a deadly epidemic that claims dozens of lives, those eight to twelve weeks of prediction time represent more than a technical achievement. They represent hope that we can protect the world’s most vulnerable people before catastrophe strikes, not just count the casualties afterward.
