To support data-driven decision making in response to the COVID-19 pandemic, a team made up of Cox’s Bazar public health and information management professionals, scientists and domain experts from UN Global Pulse, UNHCR Innovation, OCHA, and academics from Durham University and IBM/MIT are working to model the impact of possible public health operational interventions in the Cox’s Bazar settlement. In this blog they describe the what, the how and the impact of their work. In the coming weeks, the methodology and results will be detailed in a scientific report.
Working to create data-informed response scenarios to COVID-19
As COVID-19 continues to spread across continents, the world’s most vulnerable populations will be significantly affected and must not be left behind. Camps housing refugees and internally displaced persons (IDPs) often have to deal with overcrowding, insufficient sanitation facilities, and are constantly threatened by natural disasters like flooding.
The Cox’s Bazar refugee operation in Bangladesh raced to respond to the COVID-19 pandemic, with numerous measures being taken in the camps as early as two months before the first case was confirmed in May 2020. However, certain measures that have previously proven effective in slowing down the spread of the virus, including social distancing and household quarantine, are not always possible or easy to implement in such settings given the high population density and the sharing of shelters between multiple households.
Through modeling the spread of the virus in different scenarios, we can better determine what the most effective public health interventions and mitigation measures against COVID-19 are. As a team, we put our collective and multi-disciplinary expertise and skills at work to better understand, from a scientific perspective, how different types of interventions (e.g. wearing masks or new care delivery mechanisms) might best mitigate the spread of COVID-19 in these challenging contexts.
‘SIR’ and ‘ABM’ models
Two of the most common mathematical approaches to modeling disease spread are compartmental models, such as Susceptible-Infected-Recovered (SIR) models, and agent-based models (ABMs). SIR models are commonly used and are based on mathematical equations describing the high-level dynamics of the epidemics based on national or sub-regional level statistics. They can also be relatively easily built and manipulated. Although SIR models can be helpful in informing and providing rapid insights to decision-makers, they are generally not designed to produce information based on geographically disaggregated, individual and group level dynamics. In contrast, an agent-based approach simulates the movement of people at the individual or group level, where interactions in specific physical locations, and between different groups of people, can be modeled precisely based on available data – imagine creating a virtual world. This approach can result in a better understanding of the geographical spread of the virus, and allows for the modeling of specific interventions, like closing a community center or enforcing mask-wearing at distribution centers
Innovation through co-creation
We’ve been simulating the effects of various interventions that could help the Cox’s Bazar refugee operation stay ahead of the pandemic using expertise from the ground. Our simulation takes an agent based approach (see Box 1) to simulate the movement of each of the almost 600,000 individuals in the Kutupalong-Batukhali Expansion Site region of the settlement (see Box 2) using publicly available data from the Humanitarian Data Exchange (HDX) programme, UNHCR Microdata, and WHO EWARS.
Over the past several months, there has been an increase in the spread of rumours and misinformation surrounding the virus. For example, one such rumour was that many individuals might be harmed or imprisoned if they were symptomatic and presented themselves for testing. While this information is false, it has serious implications for public health and the data being collected. Acknowledging that testing data might be incomplete, and the difficulty of getting precise estimates of the number of people in the settlement who might have contracted the virus, our team is simulating the potential efficacy of interventions, rather than attempting to predict highly accurate hospitalization or fatality rates, based on the idea of supporting data-driven decision making.
How did we build this computer simulation?
We designed a two-step procedure: (i) create a ‘digital twin’ of the settlement and (ii) simulate the movement of individuals to understand how their interactions change under certain public health interventions. To create the digital settlement, we used open source census datasets that contain information on the number of people living in different sections of the camps and their demographics. From there, we included the locations where people might interact, like gathering places, water pumps and food distribution centers. After we created the digital twin to provide a bird’s eye view of where people live and where they might go during the day, we proceeded to simulate their daily routines and how they interact. For instance, we know that men are more likely than women to collect aid rations from distribution centers, so this is something we accounted for when building the model. In collaboration with the Community Based Protection team based in the Cox’s Bazar settlement, we are also in the process of finalizing the design and distribution of surveys to residents to better understand their current routines.
Once we can model where and how people interact, we can simulate what would happen if no additional measures were taken against COVID-19 and analyse the possible effectiveness of different interventions.
Simulating the effects of wearing masks in the Kutupalong-Batukhali Expansion Site
The Kutupalong-Batukhali Expansion Site, which comprises a collection of camps, is the most densely populated and interconnected site in the Cox’s Bazar settlement. This area hosts almost 600,000 Rohingya refugees, with more than 44,000 people estimated to be living per square kilometer. This is a lot higher than the UNHCR set standards, and one and a half times higher than the number of people living in New York City.
We have worked to understand the possible effects of a variety of public health interventions in the camps. One example is how effective mask wearing in certain locations is. The simulation was based on various data inputs and assumptions, validated by our teams working in the camps, about what people do on a given day, how they interact, and how often they go to certain common facilities. While there is much evidence that masks are effective in reducing the probability of an infected person transmitting SARS-CoV-2, general estimates of the efficacy vary between experimental setup and the type of masks worn, e.g. cotton mask versus surgical masks.
To account for this, we ran multiple simulations in an attempt to capture these differences, as well as differences in compliance with mask wearing. Preliminary results from these simulations suggest that mask wearing in the settlement setting can be an effective measure to mitigate the spread of COVID-19.
Throughout the course of this project we co-created and reimagined what might be possible, adapting scenarios based on the rapidly changing realities which people on the ground are confronted with.
This continues to be a learning experience for all of us, and proof of how much we can accomplish when we put our collective knowledge to work. Currently, we are also designing a web-based tool to allow relief organizations, decision makers and modelers to visualize multiple scenarios at the same time in a dynamic way (see figure above).
We hope that this will aid us, and others, in presenting and interpreting complex epidemiological simulation results.
Meet the Team:
Sandra Harlass, Senior Public Health Officer (UNHCR); Egmond Evers, Health Operations Lead (WHO); Charles Cambell, Senior Field/Technical Coordinator (UNHCR); and David Kennedy, UK Public Health Rapid Support Team
Leonardo Milano, Lead, Predictive Analytics Team
UN Global Pulse
Joseph Aylett-Bullock, Data Scientist and Researcher; Anjali Katta, Scaling and Development Fellow; Katherine Hoffman Pham, Data Scientist and Researcher; and Miguel Luengo-Oroz, Chief Data Scientist
Rebeca Moreno-Jimenez, Innovation Officer and Data Scientist
Joseph Aylett-Bullock, Carolina Cuesta-Lazaro and Arnau Quera-Bofarull who are part of the Institute for Data Science
IBM/MIT Watson AI Lab
Hendrik Strobelt and Benjamin Hoover who are experts in data visualization
For general inquiries contact: email@example.com
For media inquiries contact: firstname.lastname@example.org.