Predicting dengue fever in a changing climate
Every year, Vietnam prepares for dengue fever season. With up to 200,000 cases reported annually, the circumstances shaping these outbreaks are becoming increasingly difficult to read.
Rising temperatures and fluctuations in rainfall are pressing in Aedes aegyptithe mosquito that transmits dengue fever to northern and mountainous areas. Climate change, rapid urbanization, population growth, and complex human movements are making it increasingly difficult to predict dengue fever using traditional surveillance alone.
The Dengue Advanced Response Tool (DART) was designed to help address this challenge. The platform, supported by Wellcome, integrates case notification, climate, and population data to predict weekly dengue incidence at the district level one to 12 weeks in advance, giving public health teams more time to prepare, allocate resources, and respond before outbreaks take hold.
DART’s scope and outcomes were shaped through direct consultation with local partners and were highly customized to Vietnam’s needs and realities.
New prediction method
Making effective dengue predictions is more difficult than you might think. Many predictive tools have traditionally relied on statistical models. Statistical models can show different possible outcomes and help decision makers better understand uncertainty.
However, researchers need to decide in advance which factors should shape predictions and uncertainties such as temperature, precipitation, and population density. If you miss something important, your predictions can be inaccurate.
Machine learning models behave differently. Analyze large amounts of data to identify patterns, rather than being told which elements are important. The problem is that most machine learning models generate only one predicted number of cases, without a range, and with no indication of how confident they are. The missing context is important for public health managers deciding whether to take action.
“DART was designed to integrate these strengths because it was previously an either-or dilemma,” said Huỳnh Ngọc Tuyên, a DPhil candidate at the University of Oxford and a predictive modeler for OUCRU Vietnam’s DART project.
DART combines machine learning and advanced techniques to add a confidence range to each prediction. In fact, rather than relying on a priori human judgment to determine whether humidity, rainfall, or other factors are most important, DART identifies relationships between them from the data itself. Instead of a single predicted number, users can see a range of, say, 400 to 650, giving them a clear basis for deciding whether and how to act.
“This gives us the flexibility of machine learning while producing meaningful output for public health decision-making,” Tuen said. “That combination is what sets DART apart.”
The platform integrates three components: data pipelines, predictive models, and visualization interfaces. Disease surveillance records, weather data, and population density estimates are input into the system and automatically processed and aggregated.
It takes about 8 hours to process data for all of Vietnam from 2002 to 2025. Once you start your daily operations, you can automate your pipelines and run them weekly with little manual input.
The current system already generates weekly forecasts of dengue incidence in Ho Chi Minh City up to 12 weeks ahead at city or district level. In 2025, district-level government will be abolished in Vietnam, and DART aims to adapt the model to the commune level to reflect this change.
Predictions are only useful if they are used.
Over three years, DART has grown in a way that reflects the complexity of the problem itself. The project began with a focused technical team and ended with a broader range of partners including local governments, public health agencies and academic institutions. All this shaped the shape of the platform.
“It’s good to see more people taking part,” said project leader Sarah Sparrow, an associate professor at the University of Oxford. Its expansion was not just a matter of scale.
Involving the National Institute of Health and Epidemiology (NIHE), Hanoi Center for Disease Control (Hanoi CDC), Ho Chi Minh City Center for Disease Control (HCDC), Ho Chi Minh City Department of Health (HCMC DoH), University of Science and Technology Hanoi (USTH), and other local stakeholders meant that the platform was tested not only against research assumptions but also against operational realities.
The initial goal was to compare the dynamics of dengue fever in multiple cities in Vietnam. Over time, the team moved to a more down-to-earth approach. We started by properly building and validating the framework in Ho Chi Minh City before considering moving it to other locations.
“This is not an academic exercise,” Associate Professor Sparrow said. “We want to develop tools that public health teams can understand, trust, and use.”
By 2026, it will no longer be a question of whether DART can produce reliable predictions. It was whether the institutions in which it was built had enough confidence to act on it.
On May 21st and 22nd, this question was taken to the closing workshop. Participants observed data moving through the pipeline and reviewed what the data produced. Participating institutions, including Wellcome, NIHE, HCDC, Hanoi CDC, Ho Chi Minh City Pasteur Institute, USTH, Save the Children, and academic partners in Vietnam and the UK, reflect the project’s translational ambitions.
Master Truong Thanh Tan Lan, Head of Epidemic Surveillance, Early Warning, Preparedness and Emergency Response Department at HCDC, described the platform as “an effective tool to support dengue forecasting in Ho Chi Minh City”, but noted that due to recent changes to the city’s administrative structure, further improvements are needed to ensure that it remains relevant and easy to use in practice.
The conference also highlighted challenges that the field is grappling with more broadly. Associate Professor Phạm Quang Thái, deputy director of infectious disease control at the NIHE, observed that initial enthusiasm for modeling tools among public health managers has given way to caution.
“Models get caught up in numbers, visuals, and technique,” he says. “For public health managers, the questions are simpler: What does this mean, where should we look, and what should we do next?”
He argued that gaining broader trust requires closer involvement of surveillance experts from the outset, clearer evidence of practical benefit, and formal validation by recognized scientific institutions to give tools like DART the legitimacy they need for government adoption.
what happens next
The workshop also focused on the more difficult question of what happens after the grant ends. Forecasting systems require ongoing technical support, up-to-date data, and most importantly, institutional ownership. The research team will continue to work closely with key stakeholders, including HCDC and local partners, to integrate DART into routine dengue prevention and control activities.
The findings of another study conducted at OUCRU on the barriers and enablers of early warning systems, Exemplary Advance Warning and Response, emphasized that stakeholder engagement should be treated as a continuous process throughout the project lifecycle.
The conversation also looked further ahead. Interest has been expressed in adapting this platform to Brazil and Nepal, stress testing the framework in different epidemiological settings, and making it more scalable.
Wellcome’s Dr. Felipe J. Colon-Gonzalez made his ambitions clear. “This is not purely philanthropic. This is a strategic investment to create tools to guide policy and decision-making on the ground.”
Therefore, the long-term value of DART is likely to depend more on Vietnam’s public health authorities’ ability to own, test, and use the system over time than on any single technological advance. In dengue control, predictions, like weather forecasts, only matter if someone is prepared to act on them.
Read the full story on the OUCRU website
