Managing epidemics better thanks to simulation

Cutting-edge geographic modelling to predict the evolution of an epidemic

“Macroscopic models can be used to predict the evolution of epidemics,” notes Maxime Colomb, doctor in Geomatics and research engineer in the ASCII project team (shared by Inria, CNRS and the Institut Polytechnique de Paris), at the Inria Saclay Centre. But it's impossible for them to predict, for example, how the epidemic will react if lockdown is enforced, because they are running simulations on too large a scale. ” 

In 2020, just as the first period of lockdown had been introduced in response to the Covid pandemic, Denis Talay, head of the ASCII team, took up the challenge of creating a simulator capable of making such predictions, thanks to cutting-edge geographic modelling. It was only natural that he would approach the IGN (National Institute for Geographic and Forest Information), where Maxime Colomb was doing his thesis at the time, and set up the ICI (which stands for “Inria - Collaboration – IGN”) project. 

Reproducing the mechanisms of everyday life

“To achieve this, we divided the project into two parts: firstly, the generation of geographical data, and secondly, the epidemiological simulation.” explains Maxime Colomb.

The young researcher was in charge of the first part, and turned his focus firstly to modelling the buildings and their interiors. To do so, he developed an innovative method for merging heterogeneous data. This is a means of exploiting numerous sources, such as OpenStreetMap or the Paris Urbanism Agency, to create models of buildings and their occupancy (what activities take place there, at what times, how many people work there, what sort of public is catered for, etc.). 

Cross-referencing multiple sources to create a simulation population

Then, the researcher developed a synthetic population generator based on the same principle: cross-referencing multiple sources to create a simulation population that best represents the actual population of a given area, using INSEE statistical categories to define different types of households (single, couple, house-share, family, etc.) and assign socio-demographic characteristics to them. “We know, for example, that in 83% of cases, the adult in a single-parent family is a woman”, illustrates Maxime Colomb. 

All that was left to do, then, was to bring this little world and this environment to life. The engineer assigned a Markov model of the geolocalised timetable to each individual, once again based on their sociodemographic characteristics: where they work, at what times, when and where they go shopping, play sports and so on. 

The outcome was an authentic digital twin of the town, its occupants, their activities and their movements. These results have enabled the ICI project to participate in both the Mobidec PEPR and the “Epidemic Propagation” section of the “Digital Twin of France and its Territories” call for projects launched in 2024 by IGN, Cerema (Centre for Studies and Expertise on Risks, the Environment, Mobility and Urban Planning) and Inria. 

A fast and accurate simulator

“It's an individual-centric simulation model that's much more accurate than those we have seen up until now,” enthuses Maxime Colomb. What's more, as we've developed it using a highly optimised language, it's very fast. ”

If we take the 11th arrondissement of Paris as an example, we can simulate, in just a hundred seconds, the spread of a pathogen among the 145,000 individuals who live there and the 1.3 million people who pass through the area over a period of 100 days! 

This speed is far from insignificant, as it enables us to use a special algorithmic method known as the “Monte Carlo method”. 

The aim of this method is to compensate for the probabilistic nature of simulations by running a very large number of them for each set of parameters. The goal is to obtain a statistically robust result. 

Visualising the evolution of an epidemic from all angles

The second part of the project involves using this digital twin for epidemiological simulation. In concrete terms, the modelled geographical data is used to simulate the spread of a pathogen within the population, according to various criteria such as incubation time, distance and duration of contamination, and health policies. Hospital epidemiologists from the CRESS (Centre for Research in Epidemiology and Statistics, affiliated with Inserm) contribute their expertise to the project. 

“Users of the simulator can explore results by sub-population, for example by age group or socioeconomic category, or using specific indicators such as the number of daily hospital admissions, the number of deaths, etc.”, explains Maxime Colomb. 

To ensure the effectiveness of their simulator, the researchers are currently comparing the results provided by ICI with Santé Publique France's actual data on the Covid epidemic. Further steps will follow, such as widening the scope of the simulator to the whole of mainland France and to French overseas departments and territories. 

Other planned developments include extending the simulation to the spread of diseases outdoors (currently only indoors) and passed on by vectors such as mosquitoes, and not just from person to person. Another goal is to further improve the accuracy of simulations, notably by integrating large-scale events such as concerts, during which an individual's probability of contamination changes. 

Not only limited to healthcare applications

Once these adaptations have been made, the possible applications will seem almost limitless. For example, the simulator could be used to assess retrospectively whether the measures introduced during the Covid epidemic were the most relevant and effective, whether they were introduced at the right time or not, and so on. Of course, it will also be a valuable decision-support tool for policy-makers. 

Every year, the human respiratory syncytial virus spreads in waves and overwhelms paediatric emergency services. The simulator could be used to test health measures and determine which ones would reduce the pressure on hospital services.

The researchers are currently working with the ARS (Regional Health Agency) in the Île-de-France region to better define their needs in terms of simulation and adapt their research accordingly. 

Today, an online interface already allows healthcare stakeholders to display pre-calculated simulation results on the Genci (Grand Equipement National de Calcul Intensif, devoted to academic and industrial research) platform for certain parameters. But in a few months' time, the simulator code itself will be published as open source. This will give anyone the possibility to adapt it by integrating new data... and will open the door to applications beyond healthcare. “As the simulations are very precise, they could also be used, for example, to design transport plans, or for the management of air pollution based on people's movements,” concludes Maxime Colomb.