Within the framework of the COMET project DEXHELPP, a team of scientists working at the Institute for Information Systems Engineering at the Vienna University of Technology and at the TU spin-off company dwh is working on complex methods to be able to provide sound forecasts on the spread of the corona virus.
We and our digital twin
This does not result in projections, but in complex simulations: »There would of course be mathematical equations with which the spread of an epidemic in the population could be described - at least to a certain extent«, says Niki Popper, CSO and co-founder of dwh. »But our approach is much more flexible. We work with an agent-based approach. This means that we simulate the behaviour of many individual people and can observe on the computer how these virtual agents pass on the virus to each other.«
Real persons are thus represented by digital twins on the computer and tracked over the entire time course of the epidemic. The virtual person travels certain distances every day - for example, to work and back home. Day after day, it is simulated which person has which contacts with which other people.
This results in dynamic networks: there are people with whom a person has regular contact, for example in the household or at work. In addition, there are changing contacts with random people - for example, with customers in a shop. Each individual virtual contact has a certain probability of infection. Thus, under normal conditions, there is an exponential spread of the infection at the beginning - not because exponential functions were used to describe the epidemic, but as a natural consequence of the model, as a result of the simulated random contacts.
Data, data, data
A wide range of population data is taken into account. For example, the age distribution is of crucial importance because it has an important influence on the number of contacts. Gender and the exact spatial distribution of places of residence is also important. In rural areas, the likelihood of one person infecting another depends strongly on the spatial distance of their homes. In a large city such as Vienna, this correlation is weaker because there is more chance of being infected through purely random contacts, for example on public transport.
»Much of our data is provided by Statistik Austria - for example, on the size of households, regional population distribution or the distribution and size of jobs«, explains Martin Bicher, research assistant at dwh. In addition, there is a lot of scientific literature that can be taken into account in the models - from the typical contact probability per location to estimations of infection probabilities.
If all this is taken into account, it is possible to simulate the effects of quarantine measures, event bans or school closures. »Such measures suddenly change the structure of the contact networks - and we see very clearly in our models that this also has an effect on the spread of the disease«, explains Martin Bicher.
Recalculate whether the resources are sufficient
The most important thing now is that health care can be maintained. The agent-based computer models can also be used to assess which measures are necessary for this: The model distinguishes between mild, severe and critical cases, each of which requires different care. The age-dependent distribution of severity was taken from a case number study from China and converted to the Austrian population structure. The number of hospital beds and intensive care beds is also taken into account in the modelling in order to be able to examine whether the resources are sufficient on the basis of different scenarios.
»We are constantly improving and refining our models«, says Niki Popper. »New findings are constantly being added that we can take into account. So we hope to be able to tell you step by step in the near future how Covid-19 will develop. In any case, the fact that most people in Austria now seem to be adhering to the quarantine recommendations makes us optimistic.«