Context-Aware Realistic Epidemic Simulator: a data-driven approach to mimic real-world interactions and guide containment measures
The COVID-19 pandemic has highlighted the importance of timely interventions in the presence of an epidemics and raised the attention towards models able to illustrate and possibly predict the impact of mitigation strategies at different levels. In this context, we designed and developed CARES, a Context-Aware Realistic Epidemic Simulator that, based on the accurate representation of the interactions between individuals of the population under study, permits to replicate and compare multiple scenarios that would be prohibitive to study analytically.
CARES is a data-driven simulator of epidemic outbreaks within a synthetic population of interacting agents, designed so as to capture the peculiarities of specific places and amenities in order to understand their role in the epidemics and guide precise mitigation strategies. To this end, CARES incorporates the characteristics of locations, services and infrastructures (for example parks, hospitals, school, etc.) and allows to isolate the impact of simulated context-specific interactions with respect to a baseline of aggregated interactions modeled upon geographic and social constraints.
CARES' underlying model is based on the representation of social interactions by means of a graph whose vertices are the individuals and whose edges are decomposed into several layers:
- the Social Base (SB), an underlying geo-referenced social network mostly built upon census data;
- the Social Aggregate (SA), a baseline network of interactions that depend on the structure of the social network;
- the Social Contexts (SC), a multitude of networks of location-specific interactions drawn based on both the social network and data available in public geographic information systems (GIS).
The SB layer encodes information on the social fabric and allows to define realistic patterns of fruition of services (e.g., friends and relatives may go together to a place and are more likely to interact if they meet by chance), thus increasing the verisimilitude of the interactions simulated by the other layers. By being able to study the evolution of the epidemic for any combination of the defined layers, it will be possible to isolate the effect of each individual social context or of their combinations with respect to the baseline composed of all those interactions that are not directly measurable and/or controllable.
For the construction of the graph we integrate heterogeneous data from multiple sources. In addition to geo-referenced socio-demographic statistics, we use data extracted from regional and municipal geographic information systems, in order to model services and infrastructures with great precision. CARES is released as open-source software under the GPLv3.