As of spring 2020, there are about 70 million displaced people worldwide, more than 6.6 million of whom live in camps. While some of these camps offer dignified conditions, all too many are overcrowded and lacking in sanitation. Inhabitants might share a water tap with thousands of other people. The air might be foul, there might not be enough food, and security may be dire, leaving people malnourished and vulnerable to a range of chronic, respiratory, and mental health illnesses.
In normal circumstances, life in a refugee camp is challenging — at best. The novel Coronavirus threatens to amplify these challenges dramatically. Displaced people — the vast majority of whom live in developing countries — may not be able to access healthcare facilities, let alone ICUs equipped with enough ventilators and staffed by enough providers to survive complex cases of COVID-19.
Many countries have imposed social distancing to combat COVID-19. How do you isolate when you leave in a tent — set on a pallet and sheltered by a tarpaulin, if you’re lucky — less than a meter away from your neighbour? An abundance of news stories and research papers suggests it would be disastrous if the Coronavirus seeps into a cramped refugee camp like Moria in Greece. Yet, there is no clear strategy for camp authorities to avert an outbreak.
To tackle this issue, a team of 8 enthusiasts from the AI for Good community decided to participate in HackFromHome, the 48 hours global hackathon. Through this process we have had the pleasure of expert advice from Shikta Das who is an epidemiologist and from NGO workers providing support in Moria Camp.
Our idea was inspired by the article in the Washington Post and the 3blue1brown approach to modelling the COVID-19 spread and how different measures like lockdowns and isolation of confirmed cases can slow down the spread. We realised that to assess whether these measures could offer a solution for crowded and unsanitary environments, such as refugee camps or urban slums, we would need to simulate the camp environment and assess the outcome of potential interventions.
Our research informed us that an SEIR model (Susceptible-Exposed-Infected-Recovered) would be most applicable given the nature of the camp inhabitants’ movements, which is mostly within camp boundaries. We explored two approaches to SEIR models — an agent-based and a compartment model.
The agent-based model (ABM) captures activities performed by agents, like visiting the health clinic, markets, and washing facilities. The agents are located on a digital prototype of the Moria camp with different functions. Agents have different probabilities of infection as advised by the latest scientific research according to age. In this simulation, we programmed three different interventions: wearing PPE equipment, isolating vulnerable individuals and total lockdown.
The compartment model assumes the population is homogenous, as opposed to treating the population as heterogeneous individuals like the agent-based model does. By zooming out to the whole population all parameters of the SEIR model can be set and the simulation can proceed via respective differential equations. In particular, R0, incubation time, and infectious period were set as the same as the princess diamond cruise ship based on the advice from the epidemiologist.
Agent-based model results
The plot above shows the result of the agent-based model and the heatmap produced below shows the results from the compartment model. Each of the models and their outputs are hosted on our website for you to explore. We have a special section called model card that outlines the detailed assumptions, explanations and limitations of the model in order to ‘open the black box’.
Compartment model results
Our ABM model concluded that providing Personal protective equipment (PPE) to the full population, in addition to isolating the vulnerable population(elderly, pre-existing conditions) were the most effective interventions. The compartment model tells us that intervening strictly within 5 days can reduce fatalities by 75%.
We hope our work can be a tool for NGOs working on the ground to mobilise resources and expect the models to become more accurate with data from doctors/organisers on the ground. We hope to use this to help camp authorities to prepare mitigation/suppression strategies. We believe the tool rightly placed will have the power to save many lives.
- We are using different models (compartmental, individual- and interaction-based) to simulate a refugee camp environment as close to reality as possible.
- We have conducted user research with over a dozen NGOs around the world to make our tool more useful for them.
- We work closely with two camps in order to model the potential outbreak of the virus and share insights with the local NGOs.
- We are trying to use satellite imagery to get more information about the camps to quickly scale and replicate our models across multiple regions of the world.
- We are looking for ways to conduct better research into simulation approaches and parameters to make the models more robust and accurate.
- We need front end developers for our web app.
- We need epidemiologists to help us validate our modelling assumptions
- We are looking for contacts of humanitarian professionals, mainly from the NGOs responding to COVID on the ground.
- We are looking for partners in NGOs and local grassroots organisations who work with the displaced people.
- We are looking for contacts of relevant decision-makers on a country level and in global humanitarian organisations (WHO, WFP, Unicef, MSF, etc).
- We are looking for collaborations with humanitarian and medical research groups to validate our model assumptions.