Washington D.C., 15 April 2019 (PAHO) - In 2018, PAHO’s Health Emergencies Department and UNICEF’s Office of Innovation partnered to test machine learning as a tool to predict yellow fever risk areas in the Americas. Leveraging PAHO’s datasets, the project demonstrated the value of digital epidemiology—using digital and environmental data to better understand disease dynamics.
By applying the AdaBoost algorithm and factors such as precipitation, temperature, primate presence, and forest canopy loss, the model achieved 100% recall and 95% precision for cases reported between 2000 and 2018. It correctly identified all 640 municipalities with past cases and flagged 31 additional municipalities—mostly adjacent to infected areas—as potential sites for future outbreaks. Key predictors included latitude, canopy loss, ecoregion type, nonhuman primate populations, and temperature.
The findings, presented at a PAHO workshop in Brasilia with UNICEF, the Brazilian Ministry of Health, Fiocruz, and international universities, highlighted how machine learning can inform preparedness and outbreak response strategies for yellow fever.
Yellow fever remains endemic in Africa and the Americas, but between 2016 and 2018 Brazil experienced an unusually large epidemic that spread to coastal regions not affected for decades. Over 2,100 cases and 700 deaths were reported, prompting emergency vaccination drives and straining vaccine supplies.
In the Americas, transmission is mainly sylvatic—occurring in forested areas where mosquitoes (Haemagogus and Sabethes) spread the virus between monkeys and humans. Environmental conditions such as rainfall, temperature, altitude, forest cover, and the presence of monkeys and mosquito vectors strongly influence transmission risk. For PAHO, mapping and analyzing these factors is crucial for risk assessment, guiding vaccination strategies, and reducing the disease’s impact.
By analyzing environmental drivers of yellow fever, machine learning allows for faster, more affordable risk mapping and the use of large data sources like satellite imagery and social media. These approaches enhance understanding of yellow fever epidemiology and strengthen preparedness beyond Brazil, aligning closely with existing vaccination recommendations across South America.
In 2019, Brazil, Bolivia, and Peru continued to report cases, underscoring that 13 countries in the Americas remain at risk. PAHO’s ongoing work aims to identify both recognized and emerging high-risk areas to enable timely, cost-effective interventions that can reduce the spread and impact of future outbreaks.
