Assessing the forecasting power of mean-field approaches for disease spreading using active systems

Abstract

The forecasting of the temporal evolution of real-world diseases is systematically performed via mean-field, compartmental models, such as the SIR model and its various variants. Here, we investigate the spreading of SIR dynamics over a system of interacting, active agents – whose dynamics generate random contacts among the agents, as well as temporal and spatial correlation and fluctuations – to assess whether mean-field-like SIR models are able to accurately describe the temporal evolution of an epidemics. We find that such models display temporal dynamics that are intrinsically and fundamentally different from the ones obtained in active agent simulations. Specifically, we numerically prove that no effective mean-field SIR model is consistent with the dynamics that emerge in agent-based simulations. Our results call into question the use of such models to forecast the evolution of real-world epidemics, independently of the method used to estimate the basic reproduction number.