AESOP Early-Alert System of Outbreaks with Pandemic Potential

A data-driven system for early-warning of respiratory viral disease outbreaks contributing to preparedness against epidemics.

Manoel Barral-Netto M.D.,Ph.D.

Instituto Gonçalo Moniz - Fundação Oswaldo Cruz (Fiocruz) - Salvador, Bahia

This proposal addresses the initial phases of an outbreak, trying to understand the complex forces operating in its initiation and spreading.
Worldwide dissemination of an infectious disease requires a high transmission rate and an immunologically naïve population. These conditions are most likely for new infectious agents, especially zoonotic respiratory viruses. Insidious agents with a high ratio of infected symptomless individuals over clinically identified cases have a higher probability of cruising stealthily of traditional surveillance systems, allowing for unimpeded progression. We assume that a respiratory viral infection has the highest likelihood of achieving pandemic proportions in a presently highly connected World. This assumption stems from: a) there is a myriad of viruses capable of animal-to-man transmission, and b) many viral respiratory diseases have a large proportion of infected vs. diseased individuals able to seed and disseminate the pathogen. 
Recently, several viral epidemics caused suffering and death on a large scale. During 2019,  more than 100 viral outbreaks and only a handful of non-viral outbreaks as listed by WHO. 
To better understand the factors that determine pandemic viruses' emergency, we propose integrating and analyzing data potentializing the early signals emitted by an emergent outbreak.

The target population is local sanitary authorities at distinct administrative levels and policymakers.
The health authorities' needs in forecasting new infection outbreaks are known and originate from current surveillance systems' incapacity to timely pinpoint areas of concern.
Fiocruz, linked to the Brazilian Ministry of Health (MoH), includes in its mission the scientific support for the Unified Health System (SUS), including health surveillance. We have already obtained Primary Health Care data across the country to explore time-series analysis data on respiratory symptoms and confirmed cases. Additionally, one team member is mainly responsible for analyzing and interpreting the flu surveillance component in the whole country. 
ÆSOP will benefit from routinely collected data from distinct sources to increase the alert's predictive capability. The intended output is anticipating approximately one month of the currently available early alert approaches for outbreaks. Gaining one month ahead of current alert systems may be crucial to streamlining logistical issues for mitigating the impact and adopting containment measures for avoiding its spreading.

The ÆSOP platform's main product is an integrated/holistic early warning system of potential new epidemics or pandemics. It will send high-precision early warnings to health surveillance authorities of local, regional, and national levels free of charge. 

Given that the model will provide predictions of respiratory viral disease outbreaks with quantified uncertainty, the report to surveillance authorities will propose strategies and actions to minimize risk with suggestions for preparedness, response, and situation awareness.

The set of environmental, social, and context data spatially integrated into their territories' health units is a valuable resource for other purposes. We will make this data accessible via API for consumption or integration with distinct signals by the scientific community.

ÆSOP's reports will allow authorities to focus on surveillance and case investigation, enabling precision public health strategies as sample collection for genomic investigation for rapid and reliable pathogen diagnostics and enhancing predictive capabilities. 

The solution will be pilot tested in Brazilian municipalities in different regions, including the Amazon, to calibrate it for different climates and socioeconomic and cultural contexts. We estimate this pilot to cover half a million km2 and help the health surveillance of two million people. A successful pilot test is easily scalable for virtually the whole country.

We anticipate providing reliable information one month earlier than the usual recognition of a SARS outbreak, above the levels observed in previous years in the same area. Alerts will allow for isolation of the affected area, identify vulnerable population segments, and provide up-to-date information regarding diagnosis and clinical orientation for health professionals, structuring sample collection and their analyses.

Modeling results will provide possible scenarios for the risk of the infection spreading. Health authorities will benefit from the ÆSOP visualization platform. Additionally, the data access to the scientific community will advance knowledge on pathogen’s spread and their determinants.

First-year

We will test the EHR-based app for early outbreak signaling in municipalities, defeating a total of 500,000 individuals. Early identification of the increased risk of outbreaks will intensify health authorities' actions and justify improving registrations at the PHC level. Health professionals will notice a concrete use of their annotations for the public good besides an administrative requirement. 

Another expected benefit comes from adding metagenomic analyses for infectious disease surveillance in Brazil, a likely nest for zoonotic viral spillover to men due to the Amazon forest and other diversified ecosystems. Early identification of new viruses infecting men has the potential of preventing large epidemics. However, such protection is difficult to quantify due to the self-defeating prophecy (a prediction that prevents what it predicts from happening).

Second-year

The early-warning app will be running in the whole country, numerically extending the potential benefit to 100 million individuals. 

Third-year

We will increase the previously obtained benefit by improving alerts' spatial-time precision. AESOP's adaptation to Mozambique's health system, planned for the 3rd year, has the potential of reaching other Portuguese-speaking countries in a short time.

In the test phase: 

• Percentage of predicted outbreaks by syndromic analyses confirmed by ARI cases notifications;
• Modeling success will be measured by comparing predicted data (and its corresponding quantified uncertainty) to field measurements, number of cases, number of deaths, etc.;
• Validated consistency and representativeness of the historical series used;
• Percentage of correct identification of previous outbreaks;
• Time difference between regular detection versus ÆSOP’s detection of previous outbreaks;

During operation:

• Percentage of ÆSOP’s predicted outbreaks confirmed by data registered outbreaks;
• Percentage of false-positive alerts (AESOP’s warnings unconfirmed by in site investigations);

Environmental and socioeconomic data in the regions pinpointed by first-level alerts may have a low spatial and temporal coverage. We will combine data from different sources (numerical models and satellites) to improve data analysis. 

Besides, the analysis demands the use of a large volume of data from different sources and formats. There will be a need to validate the consistency and representativeness of the historical series used. 

The large and multimodal databases to be integrated will require robust computational infrastructure with a considerable cost. We plan to use part of the Challenge award for infrastructure and seek additional support.

Financially, the costs of metagenomics may be a barrier. Costs are high due to reagents and equipment maintenance and the safe transportation of biological samples to the labs responsible for processing and characterization. We contacted the Fiocruz Genomic Network to use the network's data and limit the number of samples processed by us. 

Depending on the project's development, we may not be able to set up a pilot study in Mozambique within three years.

Fundação Oswaldo Cruz (Fiocruz) a Foundation linked to the Brazilian Ministry of Health. Fiocruz has its main campus in Rio de Janeiro and units in 12 Brazilian States. My affiliation is with Instituto Gonçalo Moniz, the Fiocruz branch in the State of Bahia.

The possibility of finding partners at TTC proved correct as we could connect with Prof. Cordula Robinson group (Northeastern University) for evaluation and possible adaptation of the University's social media approaches to increasing AESOP's precision. Similarly, we got a commitment from Dr. Tamer Farag (Facebook Strategic Partnerships) to explore their symptomatological survey and mobility data. As our proposal includes monitoring electronic health records, area-matched analyses with Facebook's symptomatologic survey may increase AESOP's precision and provide valuable information on the relatedness of symptoms' auto-evaluation and EHR.

We will try TTC partnerships in high spatial and temporal coverage of environmental data from several sources. Additionally, we would like to explore the validation of the historical series's consistency and representativeness in a large volume of data from different sources and formats.

Presently, Brazil faces a combination of health, economic, and political crises, making it very hard to secure funds for research and development of technological approaches. TTC's support will be necessary for: a) allowing investment in locally hard-to-finance areas, such as computational infrastructure and advanced molecular approaches for surveillance; b) being awarded by TTC significantly increases the likelihood of obtaining local support from both public and private sources.

Besides the mentioned collaboration with Northeastern University and Facebook under the section Partnership & Growth Opportunities, we have collaborators from several universities in different countries.

With KU Leuven, we will collaborate in the field of viral molecular characterization. Our group has long-standing links with Prof. Vandamme’s group in a long and fruitful collaboration. Anne-Mieke Vandamme (https://orcid.org/0000-0002-6594-2766); Philippe Lemey (https://orcid.org/0000-0003-2826-5353); Guy Baele (https://orcid.org/0000-0002-1915-7732).

Janelle Thompson (https://orcid.org/0000-0003-0445-3720), from Nanyang Technological Technological University, collaborates with us on environmental and wastewater surveillance.

With the London School of Hygiene & Tropical Medicine, we also have an existing successful partnership in epidemiology and the use of large-scale data to study infectious diseases with Associate Prof. Elizabeth Brickley (https://orcid.org/0000-0003-0280-2288), who has expertise in the training of rapid response teams for outbreak investigations and will contribute to the development AESOP. 

Prof. Alessandro Reali (https://orcid.org/0000-0002-0639-7067) from the University of Pavia collaborates with colleagues from UFRJ in the field of spatio-temporal numerical simulation of COVID-19 outbreaks. He is engaged in our proposal.