SDG 3: Good Health and Well-being
SDG 10: Reduced Inequality
2. Project Details
Company or Institution
European Space Agency, Φ-Lab team
Quantifying Dengue Outbreaks from Space using AI
General description of the AI solution
Dengue fever is one of the most common and rapidly spreading arboviral-diseases in the world, with major public health and economic consequences in tropical and sub-tropical regions. When forecasting an infectious disease outbreak, it is crucial to model the spatio-temporal behaviour of the transmission. However, this is challenging because the complex dynamics of the dengue spread across different countries and weather conditions. When dengue records are highly heterogeneous, a single AI solution might not be capable of capturing all the patterns. Applying a second AI solution to interpret the same dataset will potentially describe unseen patterns from the first AI approach.
In this context, the European Space Agency’s Φ-Lab (ESA Φ-Lab), in collaboration with UNICEF, developed an AI-ensemble solution using multiple machine learning (ML) architectures. The dengue transmission patterns are decomposed into multiple domains by combining different ML-algorithms; therefore, it increases the understanding of the association between dengue transmission and meteorological conditions. This project selected Long-Short-Term Memory (recurrent neural network) and gradient boosting as the AI-ensemble solution to predict dengue cases one-month ahead in Peru. The AI solution was supported by the latest advances from Earth Observation satellites and satellite-based products to describe the geographical and meteorological characteristics. The results showed that the AI-ensemble model was able to forecast dengue cases at different transmission behaviours (e.g., low, seasonal, and endemic) and demonstrated that temperature is a significant variable. The findings also identified that departments located at higher altitudes kept the weather less favorable to dengue transmission, while proximity to the Amazon rainforest had a strong effect in keeping some departments in an endemic scenario.
This ESA-UNICEF cooperation provided knowledge about the impact of climate factors on dengue transmission that could be the basis for prevention and mitigation policies. These results have been already published at International Conference on ML (ICML2021).
European Space Agency
Excellence and Scientific Quality: Please detail the improvements made by the nominee or the nominees’ team or yourself if your applying for the award, and why they have been a success.
Dengue is a mosquito-borne viral infection mostly found in urban and peri-urban areas located in warm and tropical climate regions. The World Health Organization (WHO) reported a significant increase in dengue cases over the last two decades and it has been partly attributed to globalisation, climate change and urbanisation. The main limitation to stop the spread is that there is no specific antiviral treatment for dengue, or mass dengue testing campaign to limit dengue transmission. Therefore, the most important policy measure to control an outbreak is still by promoting an early warning system. This strategy is supported by forecast models that trigger an alert when the probability of dengue cases is higher than a threshold. Several published projects have tried standard statistical regression methods and single machine learning (ML) architectures to forecast outbreaks worldwide. However, a suitable methodology has not been yet found due to the high spatial variability of dengue cases across locations and detrended time series, making it a difficult task for a parametric or a single ML method to fully describe the transmission behaviour. In this context, the complex heterogeneity in dengue cases across Peru (pilot study) motivated ESA Φ-Lab team to innovate and explore an AI-ensemble forecast model. The success of this approach is the connection of multiple learning algorithms working together to capture different patterns from the same dataset. This innovation improved the identification of unseen information compared to when performed by a single method. For example, we used the Long-Short-Term Memory, a deep learning method well-known for its ability to process the input data as a sequence of values with a long-short-term memory of past inputs. Similarly, CatBoost, a gradient boosting model was also explored since it can deal with learning problems based on heterogeneous features, noisy data, and complex dependencies.
Scaling of impact to SDGs: Please detail how many citizens/communities and/or researchers/businesses this has had or can have a positive impact on, including particular groups where applicable and to what extent.
The recent outbreaks of life-threatening infections and re-emerging infectious diseases highlight the need for prevention, preparedness, as well as effective emergency outbreak response. UNICEF launched its Health Emergencies Preparedness Initiative (HEPI) to strengthen its health emergency response capacity, in coordination with the World Health Organization (WHO) and the US Centers for Disease Control and Prevention (CDC).
UNICEF identified diseases for which it is making organizational preparations and support packages. They include disease-specific technical guidance documents, pre-positioned stock and supply requirements, and includes dengue.
Any rapid response to an acute public health emergency requires a risk assessment based on reliable monitoring and prediction to ensure defensible decision-making, including the implementation of appropriate control measures. A systematic approach to a public health risk assessment can provide the basis to prioritize actions to alleviate the consequences on affected populations. In most areas, children have the highest incidence of infection, with severe dengue being a leading cause of serious illness and death in some Asian and Latin American countries. Please, check: https://www.unicef.org/supply/documents/dengue-fever-health-emergency-supply-note
UNICEF and ESA Φ-Lab team are working on developing an AI-ensemble solution to accurately monitor and predict the spread of dengue cases under a rapidly changing environment (e.g. climate change). We are initially developing a locally fitted model for Peru. However, with the ability of AI to be scaled and additionally available health data from the global UNICEF network, we are aiming at developing a globally applicable model which can reduce the gaps in the monitoring and prediction capacity in different regions. We strongly believe such AI-based monitoring and prediction systems will greatly enhance our ability in promoting SDG Goal 3 by enhancing the health of children in vulnerable situations and SDG Goal 10 by reducing inequalities in addressing health threats.
This project summarises how important it is to perform collaborative work with complementary expertise from intergovernmental organizations and agencies to advance knowledge and address societal challenges.
Scaling of AI solution: Please detail what proof of concept or implementations can you show now in terms of its efficacy and how the solution can be scaled to provide a global impact ad how realistic that scaling is.
This project developed a dengue forecast application based on an AI solution, supported by the latest advances from Earth Observation (EO) satellites and satellite-based products. Peru was the Pilot study; however, this AI solution implemented by ESA Φ-Lab was planned to be a fully-reproducible framework, where UNICEF can construct, validate and implement a working forecast model for other countries. Specific EO products were strategically selected because they are freely available to the public and offer worldwide coverage. As well as the use of free census data and dengue records, both data collection processes complied with GDPR.
ESA Φ-Lab strategically designed the outcome package to be replicable by UNICEF in other Latin American countries. The unsuitable approach here would be implementing the same AI-fitted model for Peru to other locations, which is unrealistic since dengue transmission and weather conditions patterns vary from one country to another. For this reason, this project does not provide one-fit-all AI model, but it offers one-fit-all AI strategy through scalability and sustainability solutions, named AI4EO4Dengue-cookbook. This document is a ‘do-it-yourself’ guided book that explains step-by-step all processes, going from data collection/management to the test/train/validation of the AI-ensemble solution. All the algorithms will be open and freely available on the ESA Φ-Lab's GitHub account.
UNICEF's goal is to generate evidence on how climate change and dengue impact children's health. Therefore, the outcomes of this project are likely to support the emergence of an enthusiastic AI4Good and AI4SDGs ecosystem, inspired by the vision and foundation of a truly public health purpose. This project started with only two UNICEF Offices (New York Innovation Office, Latin America and the Caribbean) and now it counts with five UNICEF offices engaged with the ESA Φ-Lab on the mission of providing knowledge and awareness about the impact of climate factors on dengue transmission.
Ethical aspect: Please detail the way the solution addresses any of the main ethical aspects, including trustworthiness, bias, gender issues, etc.
The equity-based approach in UNICEF’s programmes and policies seeks to understand and address the root causes of inequity so that all children, particularly those who suffer the worst deprivations in society, have access to education, health care, sanitation, clean water, protection and other services necessary for their survival, growth and development. The same principles are applied when technologies such as AI-algorithms are used to support operations of UNICEF and they are summarized in a recent UNICEF report https://www.unicef.org/globalinsight/media/1171/file/UNICEF-Global-Insight-policy-guidance-AI-children-draft-1.0-2020.pdf
The application of AI technologies to estimate dengue outbreaks in a specific age-group (children) across Peru has a powerful potential to promote equal environmental health information and reduce inequalities across different administrative areas.
The ethical processes of collecting, handling, and analysing the data follow the same principles of equality, diversity, and respect to the intellectual property of other parties involved. The project is aligned with modern and robust considerations of ethics for scientific production, building upon the team’s experience in publishing at world-class scientific journals, such as Nature Climate Change. The careful consideration of data allowed the project to critically evaluate the data sources and ensure that the information cannot, in any case, be traced back to one individual. The analytical processes were designed to align with rigorous ethical procedures aiming at publications in high-impact journals in sectors notoriously compliant with ethical regulations, like healthcare. Therefore, the project protects the confidentiality of all involved since the data collected was already anonymised and aggregated at the district level. The procedures adopted to ensure that there are no risks to participants, as they will not be directly identified in any of the methodological processes, and the information cannot lead to their identification. The strong ethical principles followed provide additional layers of transparency and build the foundations for successful replication in other countries.