1.1 Problem and Background (200 words max)
As societies around the world react to the COVID-19 pandemic, vulnerable communities including displaced people living in crowded camps and slum-dwellers are at enormous risk of contagion if the Coronavirus reaches their communities. Overcrowding and poor sanitation make social distancing and careful hygiene maintenance virtually impossible, which threatens to increase Coronavirus transmission. Insufficient healthcare infrastructure, meanwhile, leaves already vulnerable people at risk of not receiving needed care, which threatens to increase the mortality rate of COVID-19.
In addition to the risks they create for their residents, conditions in refugee camps or urban slums make it difficult to trace the arrival and spread of the Coronavirus, or to monitor its evolution among the people it infects. This paucity of data complicates both prevention and management planning: it is impossible to flatten a curve if the curve cannot be identified.
To help mitigate and manage this risk, the AIforGood Simulator team is developing a model that would predict the spread of COVID-19 in an unfavorable setting, such as a refugee camp or urban slum. This model would predict not just the spread of the disease in this unfavorable environment, but also simulate the impact of different intervention methods.
1.2 Solution summary in simple terms (150 words max)
The AIforGood Simulator models the Coronavirus’ behavior in an unfavorable setting--such as a refugee camp or urban slum. Using data inputs obtained from residents and other stakeholders, the simulator maps out the human geography and interpersonal behavior of the setting and its residents, identifying likely transmission hotspots and predicting the rate of transmission as a function of not only of the novel Coronavirus’ typical behavior, but also of the environmental conditions it will encounter in that setting.
Beyond just mapping how the novel Coronavirus would spread through this setting, the simulator would also predict how different interventions would affect its spread. The simulator could thus be used as a tool for residents of the setting, or service providers (such as NGOs, local health authorities, or neighborhood committees) to plan and coordinate collective responses.
1.3 Solution summary in technical terms (200 words max)
The AIforGood Simulator consists of two parallel models: a compartmental model and an agent-based model.
Compartmental models are widely used epidemiological models that predict the life cycle of an infectious disease across a target population’s ‘compartments’: susceptible, exposed, infected, and removed. They are useful for predicting a disease’s spread through a community, but they assume that every individual in the target population has the same characteristics and behavior. Human societies, however, are much less simple: differences in age, ethnicity, or sex; medical histories, and living conditions can affect transmission in complex ways.
To balance the generalities of the compartmental model, the Simulator also includes an agent-based model, built off of 2014 cholera research conducted at George Mason University. Agent-based models trace individual behavior with greater fidelity than compartmental models, incorporating the specificities of individual lives and of social and societal behavior in different communities.
Using each model, the Simulator encodes how different intervention methods alter a disease’s transmission through a community. It supports interactive dashboards where stakeholders can input characteristics about their community and their intervention of choice, and generates predictive transmission curves based on that data.
1.4 State of advancement of the project (100 words max)
As of late April 2020, the AIforGoodSimulator is still being developed and prepared for rollout in Greece. The Simulator team is obtaining information from aid workers in Moria and elsewhere in Greece to refine the model’s accuracy, and a team member is preparing to deploy to Lesvos to help advise strategic decision making by NGOs, add nuance to advocacy efforts, and further inform the modeling with field observations.
1.5 Project Timeline
- Proof of concept developed in HackfromHome Hackathon (Apr.5-Apr.6)
- Further developments of models gather data from the ground, outreach to NGO partners (Apr.7 - Apr.27)
- Start to provide custom reporting for NGOs working in Greece and user research with NGOs working in other camps (Apr.28-May.12)
- Development of a platform-level solution and working in a broader range of activities in partnership with stakeholders (May.13-June.30)
2.0 Project Implementation
2.1 Solution, research, or intervention? (choose accordingly) (1000 words max)
- Solution: describe the solution imagined (app, data lake, video game), its beneficiaries, locality of implementation, etc
- Research: describe hypothesis and research objectives
- Intervention: describes objectives (long-term, mid-term, short term, or goal and specific objectives)
The AIforGood Simulator will provide a solution to aid workers, local authorities, and community members trying to develop strategies to contain COVID-19 transmission in their environments. It will consist of an application built on two dimensions: a simulator on the back end, and an interface on the front end allowing the user to input demographic data representative of the target community, and to select different intervention models. The application would then run a simulation of the disease’s transmission through that community, modeling the effect of the intervention. This tool would allow stakeholders to select the specific intervention -- or combination of interventions -- best suited given their environment and available resources.
Any stakeholder in a target community could be a user of this application: an aid worker managing a refugee camp, a regional politician or public health leader trying to limit transmission, or a member of a vulnerable community trying to protect neighbors and kin. The simulator will be sufficiently adaptable to incorporate varying levels of complexity. This would make it possible for a stakeholder working nationally to predict transmission rates across multiple communities, or for a local leader to predict transmission rates within their own neighborhood.
On the back end, the simulator will run two parallel models: a compartmental model and an agent-based model (which are described in greater detail in the following section). The compartmental model will incorporate our best knowledge of the transmission rate of the novel Coronavirus, along with our best understanding of how conditions in particular environments, such as overcrowding and poor hygiene, can accelerate its transmission, to predict the spread of the disease through a given target community.
The agent-based model, for its part, will incorporate not only the broad, environmental factors employed by the compartment model, but also simulate the daily behaviors of individuals within the target community. This will enable users to test how altering collective behaviors (for example, moving community meetings from a crowded room to an outdoor area, or replacing in-person meetings with video-calls) will alter the disease’s transmission rate.
On the front end, the simulator will present the end-user an interface allowing them to input characteristics of the target population: its total numbers, its demographic bell curve, its population density, prevalence of pre-existing medical conditions, etc… By inputting these characteristics, the end-user will be able to see the evolution of a disease through the target community if no intervention is carried out.
The interface will also allow the end-user to simulate the effects of different interventions: distributing PPE to the whole population, or perhaps only to health workers; isolating the infected and conducting contact— tracing; or experimenting with different degrees of social distancing.
Our intention is for this tool to help inform strategic decision-making by stakeholders at any level of an intervention. Humanitarian and development operations tend to be fragmented between multiple actors and stakeholders and to require extensive coordination. Properly used, the AIforGood Simulator will allow a fragmented field of actors to make decisions based on sound evidence and predictive data.
The first pilot of the Simulator will take place in Greece. We will deploy a team member to the island of Lesvos, which hosts the Moria Reception and Identification Center - the largest refugee camp in Europe. We will use the simulator to help inform and coordinate local actors’ collective response, weaving between small NGOs, local and regional authorities, and multilateral actors such as UNHCR and IOM. As the tool consolidates and gains in accuracy, the project team may expand its use elsewhere in Greece, and eventually beyond.
2.3 Methodology (500 words max)
As described above, the Simulator will be built on two parallel and complementing models: a compartment model and an agent-based model. For both models, we have created a baseline interactive tool using
A compartment model is used to trace a disease’s evolution through a target population. As the disease progresses, individuals in the target population move through specific compartments: susceptible, exposed, infected, and removed (SEIR). The compartment model is also flexible enough to allow for additional compartments, that are more specific to the COVID-19 situation, or even the COVID-19 situation in Moria. For example, isolated, removed to local hospitals, etc, thus allowing for future growth in the project.
For the baseline SEIR model, at the beginning of the simulation, every subject is susceptible to the disease — anyone might get ill if there is an outbreak. When an outbreak occurs, those affected by it move from susceptible to exposed. Not every person exposed will get sick: some people may be naturally resistant to the disease, or have immunity. However, some will get sick, and move from exposed to infected. While some of the infected will quickly recover, and remain a part of the target population, others may need to be hospitalized, quarantined, or may succumb to the disease. These people will move from infected to removed. Compartment models can help us understand transmission at a broad, environmental level, and to identify how certain universal factors — such as population density or hygiene conditions — can accelerate or slow down transmission.
An agent-based model, for its part, allows for a more granular understanding of how a disease spreads. Agent-based models assume that individuals (agents) have certain goals, and will try to move towards those goals given certain circumstances (for example, visiting shops in order to sustain themselves). Agent-based models allow us to simulate the individual behaviors within a target population--and thus allow us to simulate behaviors that might increase or decrease the rate of disease transmission. Agent-based models also allow us to break that population down demographically: toddlers remaining at home, children going to school, adults going to work or to markets, the elderly remaining at home. They also allow us to model alterations in their behavior: staggering visits to markets to reduce density within, or replacing a single distribution of items with several, smaller distributions where fewer people gather.
2.4 Results/Expected results (500 words max)
We intend to use the AIforGood Simulator to guide decision-making by community leaders, NGOs, local authorities, and other stakeholders in Lesvos within a month of deployment. We will achieve this by using it in tabletop exercises, presentations to relevant stakeholders, and in humanitarian coordination forums.
We hope that the Simulator will help NGOs in the field reach decisions on the base of sound data analysis, rather than relying solely on common sense. We intend to use it to reinforce the decisions of medical actors in Lesvos, and eventually in other camp and slum settings.
The AIforGood Simulator will allow the NGOs to simulate different cases based on different interventions and compare the relative efficacy of these interventions. We understand the uniquely precarious situation that exists in refugee camps. Our hope is that our tool can use information about infectiousness and the characteristics of the social network within a refugee camp to then model what could happen if certain intervention methods are implemented. Given these input parameters and intervention methods, the tool would then return statistics about the potential number of hospitalisations, critical cases, and deaths that the camp would endure.
3.0 Safety, quality assurance and regulation
3.1 What steps have you taken to ensure your solution’s safety? How advanced are you in this process (if applicable)? Please check the Biosafety and Biosecurity guideline of OpenCovid19
3.2 Have you planned the conduct of your manufacturing process that ensures quality, what are the steps you have taken? How advanced are you in this (if applicable)?
3.3 Will you need assistance with the regulation system? If not, which regulatory system do you plan on using to distribute the product? Please elaborate (please see: Regulatory-Strategies) (if applicable)
3.4 Have you talked to medical staff about the feasibility of your project? What did they say?
We have contacted medical organisations within the refugee camps and they do have needs with modelling support.
3.5 Have you planned the testing, verification and validation of your solution? How advanced are you? (if applicable)
4.0 Impact, issues and risks
4.1 What impact do you feel your project could have? (100 words max)
On a small scale, we believe our tool could help NGOs and community leaders working on the ground in refugee camps to better understand the potential risks and benefits of feasible intervention methods. On a larger scale, our efforts are part of an international, interdisciplinary push to mitigate potential outbreaks of COVID-19, particularly in at-risk populations. Our team is constantly learning and in conversation with all of the other research and science currently being conducted globally. We hope that our tool can not only inform efforts on the ground, but also the pursuit of optimal models and interventions for situations that may be similar to refugee camps.
4.2 What do you think would make your project a success?(100 words max)
4.3 Please list the known issues, potential risks, grey-areas, etc in your project
5.1 What other projects on JOGL are like yours? Search for them and Link them!
5.2 Is this an innovative project? What makes this project different if it’s unique on JOGL?
5.3 Is there already an open source version of this project?
6.0 Team experience
6.1 Please cite your team members and their roles in the project.
(if applicable) If the project involves several locations or labs, list them too.
7.0 Funding and Costs
7.1 Please provide a costing of your project be as detailed as you can, all funding requests must be transparent and be for specific needs. The maximum grant is 3000 euros. Smaller grants are more likely to be funded.
7.2 How is your project being funded so far?
7.3 How much funding do you need and how do you plan to use that funding?
- Projects (new, funded and refused ones) are able to ask for funding at each round.
- If your project involves several locations and entities, please indicate how you plan to distribute the grant money.
- If your project is a consortium of existing projects, the upper limit of the grant only applies to sub-project entities.