Fine-Scale Risk Mapping to Identify and Disrupt Viral Spillover

Using artificial intelligence to map fine-scale zoonotic disease risk to predict and prevent future pandemics.

Peter Daszak, Ph.D., President of EcoHealth Alliance

The spillover of animal microbes into people causes most pandemics, and is increasing in frequency, with over three new zoonotic viruses emerging each year (1). Emerging infectious diseases (EIDs) cause millions of deaths and cost billions of dollars annually, with COVID-19 alone costing tens of trillions of US dollars in economic losses. Pandemic prevention efforts depend on identifying underlying causes, targeting the places where pathogens are most likely to spillover, and predicting the ability of new diseases to spread. Despite challenges, prevention programs are likely to provide significant return on investment. Initiatives to reduce emerging disease risk via conservation and increased surveillance would cost US$17.7-26.9 billion, two orders of magnitude less than the US$1 trillion in annual damages pandemics produce(2).

Predicting disease emergence is a complex multidisciplinary problem requiring analysis of interactions among human activities, livestock production, and biodiversity (3,4). To address this challenge, we will use artificial intelligence (AI) to re-envision global predictive models that pinpoint areas and conditions with the highest future spillover risk. We will also develop a finer-scale predictive model for Southeast Asia and China, incorporating human mobility data, and ground-truth our predictions with data from multiple outbreaks of influenza and other diseases.

Our solution supports researchers, policy-makers, and public health practitioners by making EID risk maps widely available and up-to-date with the latest data, which will help optimize surveillance and other programs to target locations at the highest risk of disease spillover. Further, by identifying high-risk locations and risk factors for emergence, our work will support intervention programs that are not only spatially targeted but also targeted toward the greatest drivers of spillover (e.g., deforestation).

More broadly, our project will serve communities where the risk of epidemics is highest (most often vulnerable, impoverished populations), and ultimately will protect global health as we aim to prevent the spread of infectious diseases across national and continental borders.

EcoHealth Alliance (EHA) prides itself on our strong partnerships with local organizations, researchers, and governments across the world. Throughout this project, as in all of our work, we will collaborate with our partners to understand the local realities surrounding EID hotspots. Context-specific data (e.g., participatory research methods) will be incorporated in the model, and we will engage our partners to ensure we utilize the best available data and produce model products that are informative and useful to the local context.

Our solution provides a public good by creating an open-source, science-based tool to identify geographic areas with the highest probability of pathogen spillover. It will improve disease surveillance, influence policy, and ultimately protect global human health. By delineating highest-risk hotspots and risk factors for emergence, we will support researchers, public officials, and medical practitioners to allocate resources for surveillance, informed policy interventions, and other preemptive measures. At a regional level, our fine-scale Southeast Asia and China EID risk model will allow governments to allocate disease surveillance and mitigation resources to the areas of their country with the greatest need.

Additionally, our solution provides a public good through improved information sharing. The global model will be widely accessible and interpretable. With deployment infrastructure, the model results can be available via APIs into web applications and other project uses. We will prototype a public-facing risk assessment tool that will allow users to adjust model data and assumptions in order to explore hypothetical or future risk scenarios (e.g., how would risk change under a given reforestation effort?). Finally, because code will be open-source and well-documented in a public repository, this work can serve as an adaptable blueprint for future research efforts.

Our solution will help to inform surveillance and interventions in areas with high EID spillover risk. Predicting the spatial distribution of EID risk is a cost-effective strategy to optimize surveillance programs that can anticipate, target, and intervene in potential outbreaks. Preventing an outbreak can save millions of human lives and is far more economically sustainable than trying to control a disease once it has emerged and spread. 

Our target populations will be well-served by the development and deployment of our tool. This type of approach assists researchers in understanding EID risk; demonstrably, previous incarnations of our hotspot mapping tool have been cited over 3,000 times. We know through our extensive One Health work across the globe that policy makers and governments are more inclined to invest in preemptive pandemic prevention and One Health-based policy when the approach is reinforced by a data-based tool such as this. Similarly, data-based tools allow intergovernmental agencies and governments to direct resources for aid and intervention in a more streamlined approach to target particularly vulnerable populations, while communities at highest risk will be able to use the interactive combinations of novel datasets to direct their own individualized protection of local public health. 

We will scale our impact in Southeast Asia and China--a region at high risk for EID spillover-- by developing a finer-resolution risk map and analysis for this area. Our overall approach to developing region-specific risk maps is scalable and could be further applied to other regions of the world with unique risk factors and disease landscapes, and it will allow us to predict risk on a city-level and village-level rather than on a broad country-level. With this predictive capacity, surveillance and intervention efforts can be most efficiently applied, with the potential to impact millions of lives.

Our improvements to the global EID hotspot model will likewise be scaled for impact through greater accessibility via a public-facing interface and prototype risk assessment tool. This tool will allow users to adjust model assumptions and run current or future scenarios to receive results in real time. This spatial risk user-interface will be comparable to a viral ranking tool, Spillover, that EHA helped develop and released this month (12). Our model and data infrastructure improvements mean that our model predictions will always be current with the best-available information, and thus more usable by policy makers around the world.

We will measure the success of our project in several key ways: 

1) We will track usage of our updated disease hotspot model (e.g., database downloads, usage of the risk assessment application, citations in scientific papers). 

2) We will track dissemination of our risk model and relevant outputs to project partners and policy makers and document direct changes to surveillance strategies and intervention efforts.

3) We will host roundtable meetings with government partners and scientists in hotspot countries/regions to solicit feedback and understand if and how they have operationalized pandemic prevention measures based on our solution.

Example specific measurable indicators:

Indicator 1: Number of annual peer-reviewed citations of our improved model through 2025

Indicator 2: Number of individual users who download our data and open-source code

Indicator 3: Number of intergovernmental and other institutional organizations using our maps, data, and/or risk assessment application

Indicator 4: Number of in-country government partners that have implemented our risk mapping tools to prioritizing disease surveillance or pandemic prevention efforts each year.

Our primary barrier is access to data, including disease outcome data and human mobility data. By working with our Trinity Challenge partners (Hong Kong University, Tencent, and Facebook) we will be able to access the best-available data in each relevant field to support the development of models that are representative, informative, and accurate on both the global- and country-level. 

As a non-profit organization, EcoHealth Alliance requires financial support to implement the far-reaching scale and impact of this innovative, multi-year project. With funding from the Trinity Challenge, we will be able to harness both our internal resources and partner networks at EcoHealth Alliance that uniquely qualify us to excel in this project. The technical expertise of our multidisciplinary team, consisting of computational scientists, wildlife veterinarians, epidemiologists, ecologists, virologists, economists, technologists, anthropologists, and conservation biologists, allows us to implement cutting-edge research. EcoHealth Alliance heads a global network of partners, including those in China and Southeast Asia, that provides exceptional support for our core scientists in developing real-time, on-the-ground insights into disease emergence with all technical, political, and cultural barriers addressed.

EcoHealth Alliance maintains a network of over 100 local collaborators at the community, nonprofit, academic, private sector, governmental and intergovernmental levels around the world that often serve as implementing partners. A list of EHA’s organizational partners may be found here: https://www.ecohealthalliance....

The Trinity Challenge network will enable EHA to initiate and cultivate relationships with key collaborators who are crucial to our project’s success, while also playing an important role in the dissemination of our finished products to various stakeholders in both the public and private sector.

Support from the Trinity Challenge will allow us to dedicate the necessary amount of staff time and energy to complete this project, and vital partnership with other organizations through the Challenge will provide access to crucial data. From these datasets, we will characterize human density and connectedness along rural to urban gradients, and we will assess how these networks intersect with spillover risk. For example, by adding a layer of human connectivity in our regional model with our collaborators’ assistance, we will rapidly identify the spread potential for each high risk site for new zoonotic disease emergence that we identify (even without specific transmission factors being known for individual novel pathogens). Consulting with our expert collaborators will allow us to understand data provenance, limitations, and potential biases to address. Their input and feedback may inform potential use cases and impacts of our work, which will help us to refine our models and create useful end-products.

Our solution matches EHA’s expertise in predictive modeling and risk mapping with the key strengths of our collaborators. Hong Kong University’s expert scientists will help us determine how well model assumptions track in on-the-ground conditions. Surveillance and outbreak data from HKU and the China CDC (Guangdong) will help address ascertainment biases, particularly with epidemiological data around avian influenza and other EIDs in our regional model. We will include caseload, individual outbreak, driver, background surveillance (poultry, pigs, people) and global sequencing data from the CEIRS network program at HKU.

Experts from Facebook and Tencent will help us better understand human movement and demography as EID risk factors. Mobility data from China (Tencent) and Southeast Asia (Facebook) will be critical to translate spillover risk into a network of disease spread risk. We will maintain open dialogue with these collaborators on the design and implementation of our work, and hope to build lasting partnerships that also benefit their public health efforts.

We will also work closely with our partners at CSIRO to incorporate unique biodiversity datasets including degree of habitat loss, changes in composition of wildlife communities, and new estimates of croplands and livestock systems to use in our updated global hotspot model.