solution-oriented collaborative computational planning for smoother decision-making

COVID-19: which simulation?

Declared by WHO as Public Health Emergency of International Concern on January 2020, coronavirus disease 2019 (COVID-2019) is challenging health systems and governments everywhere, revealing their helplessness and lack of preparation towards new forms of pandemic.


The absence of a vaccine, the uniqueness of disease features (long asymptomatic incubation period), the unpredictability regarding the efficiency of quarantine policies and in the epidemiological and economic outlooks, all those factors of uncertainty have advocated for the need of more epidemiological simulation. I am not sure if (actually I hope that) government advisers were able to find geniuses to create high-tech simulations to properly guide political decisions, but if you would surf the internet in March and April 2020, you could come across a mixed bag of epidemiological simulations. Here is a brief overview of those initiatives. 

Limitations of analytic approach

One of the most acclaimed "simulation" on Twitter is Gabriel Goh's Epidemic Calculator. Rather than a proper simulation, it is more like a parametric model (numerous variables can be set up with sliders) helping to forecast the general likely evolution of the pandemic. 

The problem with such analytic approach is that it does not reflect the dynamic and the complexity of the reality: health system carrying capacity is not taken into account, second wave of infection is not simulated etc. That is why agent-based models are favoured in epidemiological simulations.

Mainstreaming agent-based simulation

If agent-based epidemiological models are often included in simulation software as examples or tutorials, it was a surprise to find them in mainstream media. In the first place, this article of March 14 in the Washington Post (translated in 13 languages!) has caught the attention of a large number of web users. As confessed in the text, the simulation proposed here is spatially simplistic ("a network of bouncing balls on your screen"), and vastly oversimplify the complexity of real life (human contacts in public spaces, workplaces).

Indeed, bouncing balls flying freely inside a large blank area is not really what we can expect from a simulation that is supposed to deal with extremely complex social and behavioural factors. Still, the article does its job of education / popularization of science and the agent-based simulation gets clear-cut results in terms of different shapes of evolution of the disease according to the implemented policy (and the winner is of course extensive social distancing #stayathome).

To add or not to add spatial complexity

Following Washington Post's article, the maths YouTube channel 3Blue1Brown has made another attempt of simulating COVID-19 evolution through beautiful agent-based models. The presentation of pandemic dynamics and of the simulation methodology is truly amazing but again, the tentative at the end of the video of adding some spatial complexity with the "shared central locations" is not mastered and does not give exploitable results (at some point ALL agents gather in ONE central location...). Adding complexity is a commendable effort, but might be misleading if not done properly.


Solution-oriented simulations

The modelling of central hubs in the city where people gather (and where they most probably get infected by the virus), has also been performed, but with a higher level of complexity and realism in the agent-based simulation EpiSim by our friend and colleague Ira Winder (with F. Catalang and D. Goldman). Indeed, here the land use of a city is recreated more faithfully with the modelling of dwellings, workplaces, schools, retails, hospitals, etc. all having their own "gathering densities" (relating to infection risk), changing based on the number of agents gathering there at that specific time. We're already very far from the "bouncing balls wandering around randomly" in other simulations. But the main difference of this simulation is that it integrates a "hospitalization capacity" indicator that could make the simulation truly useful for decision-making in real life deployment. 

This simulation is still a WIP. We cannot wait to see more complexity, more realism added to it. A big next step would be to recreate different transport modes (with their own contagion rate) which share could be modified on the influence of certain policies. Also, rather than a random city, implementing the simulation proactively in a real city plan would be something never seen and pioneering to get our cities ready for the next pandemic outbreaks (let's hope not). Perhaps only then can we pride ourselves on having effectively smart cities.

Supporter of agent-based simulation and fervent defender of SDG 3, Pragmetrics intends to support epidemiological simulation effort parallel to the search for solutions in the field of urban planning.

April 16, 2020