FGV EMAp's Dengue Forecasting Model Makes Waves in PNAS for Brazil's Public Health

FGV EMAp's Breakthrough in Dengue Forecasting Earns PNAS Spotlight

A groundbreaking study led by researchers from the Escola de Matemática Aplicada (EMAp) at Fundação Getulio Vargas (FGV) has been published in the prestigious Proceedings of the National Academy of Sciences (PNAS), marking a significant achievement for Brazilian higher education in applied mathematics and public health. The paper, titled "Leveraging probabilistic forecasts for dengue preparedness and control: The 2024 Dengue Forecasting Sprint in Brazil," details an innovative ensemble model designed to predict dengue outbreaks with unprecedented accuracy across Brazil's diverse regions. This collaboration underscores EMAp's leadership in data-driven solutions to national health challenges.

The research, spearheaded by Flávio Codeço Coelho, a professor at FGV EMAp and creator of the InfoDengue system, integrates advanced machine learning and statistical methods to generate probabilistic forecasts. These predictions help public health officials anticipate outbreak timing, intensity, and locations, enabling proactive resource allocation and prevention efforts.

The Escalating Dengue Threat in Brazil

Dengue fever, transmitted by the Aedes aegypti mosquito, has long been endemic in Brazil, but recent years have seen explosive growth. In 2024, the country recorded over 6.5 million probable cases—surpassing the cumulative total from the previous decade—and thousands of deaths, marking the worst epidemic on record. Factors like climate change, urbanization, and the Aedes mosquito's adaptation to new environments exacerbated the crisis, with cases spreading to previously unaffected areas.

By early 2026, while cases dropped significantly in 2025 (over 70% reduction to about 1.6 million), vigilance remains critical. The Ministry of Health reports ongoing transmission, with states like Mato Grosso do Sul noting hundreds of cases in the first weeks of the year. Forecasts for 2026 suggest a moderate season similar to 2023 or 2025, but tools like FGV EMAp's model are essential to avert repeats of 2024's devastation.

FGV EMAp: Pioneering Applied Mathematics in Public Health

Established as FGV's hub for applied mathematics, EMAp bridges theoretical rigor with real-world applications, particularly in epidemiology and data science. Flávio Codeço Coelho, with his background in mathematical modeling, has been instrumental through projects like InfoDengue—a decade-old platform monitoring arboviruses nationwide—and Mosqlimate, which provides open climate data for health modeling.

Luiz Max Fagundes de Carvalho, another key contributor, specialized in ensemble techniques, ensuring the model's robustness. Their work exemplifies how Brazilian universities are leveraging interdisciplinary expertise to address societal issues, positioning FGV EMAp as a leader in computational epidemiology.

The Infodengue-Mosqlimate Dengue Challenge: Fostering Global Collaboration

In response to the 2024 crisis, the Infodengue-Mosqlimate consortium, involving FGV EMAp and Fiocruz, launched the 2024 Dengue Forecasting Sprint (IMDC24). Six international teams from Brazil, Saudi Arabia, Spain, and the US competed to forecast cases in five diverse states: Amazonas, Paraná, Minas Gerais, Rio de Janeiro, and São Paulo.

Teams received comprehensive datasets: historical dengue incidence, sociodemographics, and high-resolution climate variables from ERA5-Land (temperature, humidity, rainfall, El Niño indices). This open challenge not only accelerated innovation but highlighted higher education's role in international scientific diplomacy.

Step-by-Step: How the Ensemble Model Works

The model's power lies in its ensemble approach, combining diverse predictions to mitigate individual weaknesses. Here's the process:

• Data Integration: Weekly dengue cases from SINAN (Brazil's notifiable diseases system), climate covariates via Mosqlimate API, and mobility patterns.
• Individual Models: Teams used Bayesian hierarchical models, LSTM neural networks, generalized additive models (GAMs), and SARIMA for time-series forecasting.
• Scoring and Weighting: Performance evaluated with probabilistic metrics (e.g., logarithmic score) across regions, epidemic phases, and seasons. Weights assigned dynamically—e.g., one model excels in Amazon's low-transmission dynamics, another in Southeast peaks.
• Ensemble Aggregation: Logarithmic pooling merges forecasts into probabilistic bands (e.g., 50% chance of high incidence), using open-source code on GitHub.
• Output Generation: Annual and weekly scenarios for exceptional vs. typical years, now vaccine-adjusted.

This step-by-step fusion ensures reliability, with the ensemble outperforming singles in validation.Read the full PNAS paper for technical details.

Proven Performance: Regional Insights and Validation

No single model dominated; variability was key in 2024's extremes. Ensembles captured complementary strengths—e.g., machine learning for nonlinear patterns, statistics for uncertainty quantification. In Paraná, models underestimated peaks, but ensembles balanced via regional scoring.

Post-challenge, expanded to 15 teams for 2025, with forecasts adopted nationally. Early 2026 data aligns with moderate predictions, validating the approach amid vaccination rollout.

Seamless Integration with Brazil's Ministry of Health

The model's crowning achievement: official adoption into the Ministry's dengue response agenda. It informs hospital bed allocation, campaign timing, and mosquito control, preventing care delays that caused 2024 fatalities. Annual sprints now provide pre-season projections, evolving with new data like vaccination coverage from the National Immunization Program.

Fiocruz webinars share updates, fostering transparency.See Fiocruz's 2026 forecast webinar summary.

Boosting Brazilian Higher Education and Research Ecosystem

FGV EMAp's PNAS publication elevates Brazil's global research profile, demonstrating applied math's societal impact. Collaborations with Fiocruz, LNCC, and international partners exemplify interdisciplinary higher ed models. It attracts funding, talent, and positions Brazil in global health innovation.

In a landscape where universities face funding pressures, such successes highlight research's role in policy and economy. EMAp trains next-gen modelers via theses like Transformer-based forecasting.

Future Horizons: Evolving Challenges and Arbovirus Expansion

2025's sprint grew to 15+ teams, incorporating vaccines. 2026 forecasts predict no 2024-like surge, aiding planning. Plans include Zika/chikungunya integration and climate-resilient extensions amid El Niño variability.

Challenges persist: data quality, extreme events, equity in underserved areas. Yet, open platforms like Mosqlimate democratize access, empowering universities nationwide.Explore Mosqlimate forecasts.

Photo by Lucas Vasques on Unsplash

Stakeholder Perspectives: From Researchers to Policymakers

Coelho emphasizes: "Our ensemble transforms uncertainty into action, learning yearly." Health officials praise timely insights for saving lives. Internationally, it inspires similar sprints.

For higher ed, it showcases Brazil's math prowess, drawing PhD students to EMAp's programs.