Spreading of infectious diseases can be prevented or substantially slowed down by measures like social distancing or curfews. However, a long-term lock down of the whole country will not be sustainable for society. At some point, measures will need to be relaxed and replaced by highly dynamical and targeted measures to prevent spreading.
How could measures of social distancing be adapted selectively to local diffusion dynamics of COVID-19? How can local high-risk areas be identified? And on which occasions is the probability of transmission particularly high? Based on this knowledge highly targeted measures could be selected.
This is our proposal:
We build a graph from information of people including their close contacts and their relations to locations they regularly visit. Based on this, we identify high-risk areas. In order to ensure anonymity and, at the same time, act sufficiently locally, we use gridded locations, e.g. at the level of postcodes. We also evaluate the type of relationships (for example, work, kindergarten, study, relatives…) to understand in which field of social life distancing needs to be addressed primarily. From that, high-risk areas and facilities can be determined statistically.
From which data is the graph built?
Patients would be asked to specify anonymous information about locations they have visited or are going to visit. Examples of questions could be: Where do you live? Which places do you regularly visit? Where do your parents live? Where is the school of your kids? If possible, this information would be collected for contact persons of the patients as well, at least for the people in the same household. Data of a large number of volunteers can be used to give better statistics of people’s movements between different locations. Under the assumption that a larger number of discovered infections also hints at a larger number of unknown infections, this can be used to reason about the potential spreading of hotspots. Potentially, these data could be enriched by anonymized mobility data from the mobile phone providers to give a complete picture about people movements during the day, however, without the distinction of different purposes.
When is a postal code zone considered critical?
The criticality of an area increases with the number of connections of infected people and close-contact persons to this area. Additionally, aggregation on the type of connection can be retrieved. Are the connections mainly due to students, or due to sports activities?
Who shall use the tool?
We provide a web interface that shall be used by patients and volunteers to enter their information to populate the graph. Health authorities could encourage people to enter their data, which would also be beneficial for reducing the high workload of authorities in case of a crisis. The map indicating high-risk areas shall be used by local authorities and decision makers.
How do established social distancing measures interact with the graph?
Whenever public authorities decide on social distancing measures, for example, they close a particular university, an edge or relation in the graph is cut. Once a high-risk area is identified the authorities can evaluate which measures to adopt to cut out more and more edges to isolate that area, so that the network gets more and more disconnected. This would limit the spread of the virus. Public authorities enter this information in the web application.
What happens if patients are no longer infected, or when close contacts have passed the quarantine time?
When people are officially declared healthy by a doctor this information is entered. The corresponding edges are no longer contributing to the criticality of regions. In particular, if people living in an area with a high infection rate recover, the cluster can at some point be deemed safe, because most people are immune to new infections.
Can public mobility data improve the accuracy of the model?
Our first, basic approach asks for the number of principal connections between locations. It does not account for frequency how often these places are visited. Weighted graphs could improve the situation and public mobility data could be taken into account for this purpose. On the other hand, our concept foresees to distinguish the different purposes of mobility, which anonymized mobility data does not tell.
The informative value of the graph can be further increased by using weighted relations. This means not only asking is there in principle a connection of this number of patients to this area, but also how often are people traveling from this zone to the other. This would mean to include anonymous public mobility data.
Challenges we ran into
Building a back-end and front-end prototype from scratch in two days, that is a challenge!
Accomplishments we are proud of
This was a tremendous effort. We learnt about WirVsVirus on Thursday, we delivered the idea Thursday evening, and Sunday evening a prototype exists!
What we have learnt:
We experienced great team work under pressure. We learned a lot about the dynamics of a pandemic which is new for everyone in our team and probably everyone involved in WirVsVirus. We also got a glimpse on the huge challenges authorities have to face during a crisis like this.
Build the form for entering patient’s data from a simple prototype to a more elaborate version (e.g. using point-and-click location selection)
Build up infrastructure and database (Neo4j)
Study integration of public mobility data
The technologies that we have chosen for the prototype are perfectly suited for building a robust, reactive, and highly scalable web application.