Marking five years since the emergence of COVID-19, we revisit the early days of the pandemic to explore how in silico models helped detect outbreaks, understand airborne transmission, and guide public health decisions, shedding light on the digital tools that shaped our response.
“There is currently an outbreak of pneumonia of unknown aetiology in Wuhan, China”.
This observation was a chilling reality, the beginning of a chapter that would reshape our world: The COVID-19 pandemic. For most of us, pandemics belonged to history books or sci-fi thrillers. Even eerily prophetic works like the movie Contagion (2011) or David Quammen’s best-selling book Spillover (2012) were seen as hypotheticals, not as blueprints of an imminent reality. Yet, as the pandemic unfolded, we quickly learnt how wrong we were.
The first alert came on December 30, 2019, as an email from the Program for Monitoring Emerging Diseases (ProMED), a network run by the International Society for Infectious Diseases. That same day, hospitals in the Wuhan region convened emergency meetings, and health agencies launched investigations. All the initial cases seem to point to a crowded seafood market that also traded live wild animals in cramped and unsanitary conditions. All necessary ingredients for a spillover.
On January 5, 2020, the World Health Organisation (WHO) issued its first formal notice to member states, advising precautionary measures for a mysterious cluster of pneumonia. Just four days later, the possible suspect was identified: a novel coronavirus, later named SARS-CoV-2.
March 11 came around and the WHO officially declared COVID-19 a global pandemic. Soon after, flights were grounded, borders closed, cities fell silent, and the world stood still.
In the face of this invisible threat, scientists turned to every tool available. Among these, computer models and simulations played a key role, though they often went unnoticed.
In this article– and in two upcoming pieces marking five years since the pandemic’s beginning– we explore how computer models helped the world respond to the pandemic, from early recognition to vaccine development, and beyond. This first chapter is dedicated to the early days: from outbreak detection to how it spread.
Early Warnings: the AI that spoke before the world listened
Dr. Kamran Khan, an infectious disease specialist in Toronto, had seen enough in his career to know the world wasn’t ready for the next big outbreak. He was in New York during the 2001 West Nile virus outbreak and anthrax attacks. Then, he moved back to Toronto in 2003, just in time for the SARS crisis. These experiences shaped his mission to predict and contain future epidemics before they spread.
In 2013, he founded BlueDot, a startup that uses computer modelling to detect emerging infectious diseases. BlueDot’s platform analyses epidemiological and travel data, and scans over 25.000 news sources in 100 languages. This huge amount of data is then processed by natural language processing tools, AI, and human experts, resulting in a powerful vigilance and intelligence tool that can sniff out emerging outbreaks.
In July 2019, BlueDots’report in the Journal of the American Medical Informatics Association validated the model’s ability to detect outbreaks. The authors reviewed the data from July 2017 to June 2018, comparing the model-generated alerts with the real world disease outbreaks identified by the WHO, including HIV, Avian Influenza, Dengue, Ebola, and many others. Impressively, BlueDot's model accurately identified 35 out of 37 outbreaks, and even predicting them an average of 43 days before the WHO. The authors of the paper also underlined the usefulness of the proposed approach alongside the traditional indicator-based surveillance, as it helped timely inform authorities and the population if, and where, a new epidemic is surfacing up.
Within five months of BlueDot's publication on the validation of its event-based outbreak surveillance, an email alert on a new suspicious pneumonia outbreak in Wuhan was sent out by the ProMED. Within a day, BlueDot confidently flagged that cluster, a week before the WHO made its first official statement on the COVID-19 outbreak.
This wasn’t science fiction anymore. It was a wake-up call for global health systems and a glimpse into how computer modelling and simulations, paired with human insight, could effectively support global and national health agencies.
References:
What’s in the Air? How Models Helped Us Understand COVID Indoors
As hospitals were overflowing and vaccines were still under development, scientists raced to answer one pressing question: how does COVID-19 spread?
By mid-2020, consensus had formed around two primary non-contact transmission routes: contaminated surfaces (fomites), and airborne particles– tiny aerosols and droplets released through breathing, talking, coughing, and sneezing.
Airborne transmission, especially indoors, became a major concern. In enclosed spaces like homes, restaurants, offices, and gyms, where people spend hours in close proximity, the virus could linger in the air. Poor ventilation, loud speech, or physical exertion made things worse. But measuring virus-laden aerosols is notoriously difficult. The particles are small and often at concentrations below detectable thresholds. That’s where studies based on mathematical modelling helped.
One such study by Dr. Tareq Hussein and colleagues, from the University of Amman, published in the International Journal of Environmental Research and Public Health, addressed how aerosols deposit in different parts of the respiratory tract, as well as how changes in breathing rate for different genders and activity levels, impact the transmission of the virus.
Hussein’s team combined an indoor aerosol behaviour computer model with a respiratory tract deposition model. Subsequently, they simulated scenarios where an infected person emits 10 virus particles per second into an enclosed room. They compared outcomes in well- and poorly-ventilated environments, as well as with individuals at rest or physically exercising.
As a result, they found that the extent of room ventilation alone had a significant impact: better ventilation lowered the inhaled dose of virus particles by around threefold. Likewise, physical activity could substantially increase the risk, which helped identify why gyms were among the first venues that needed to be shut down.
The insight from this work, and many more similar computer modelling and simulation studies, informed public health guidelines and helped business owners make safer choices when opening up, while protecting their individual and their customers’ health.
Reference:
Modelling the Storm: Predicting the path of a pandemic
During the pandemic, governments and leaders faced tough decisions: lockdown or stay open? Prioritise health or economy? Every decision carried immense weight– and needed data and science to provide informed support.
A great help came from Covasim, an agent-based simulation tool developed by Dr. Cliff C. Kerr, from the Bill & Melinda Gates Foundation. Covasim modelled real-world communities– households, schools, workplaces– and forecasted COVID-19 spread under various scenarios and circumstances. The tool could simulate interventions like physical distancing, virus detection, tracing of contacts, quarantine, or vaccination.
To ensure its benefit and impact, the computer model was created to be open and accessible. It didn’t require supercomputers and was also released as an easy-to-use webapp. In the related paper, published on PLOS Computational Biology, Covasim is described to handle the simulation of tens of thousands of infections among a susceptible population of hundreds of thousands of people, spanning 12 months, within a minute, on a personal laptop.
The Covasim model was also tested in real-life applications. In King County, where the city of Seattle is located, a simulation was run to estimate the impact of necessary restrictions on the mobility of its inhabitants. The model predicted that even modest restrictions could flatten the curve, which turned out to be true, once applied in practice. The projections also showed how ramping up contact tracing could halve infections, while doing nothing would triple them.
Later on, the same Covasim model was also tested in the state of Victoria, Australia, where Melbourne is located. It helped officials to evaluate how restrictions could be lifted, without overwhelming hospitals. Also, the computer simulations showed that public health measures had to be slowly minimised in order to mitigate vaccine hesitancy and the threat of new variants. Thus, driving home the message that herd immunity is unlikely without broad vaccination coverage.
Covasim was also employed in India, South Africa, the United Kingdom, and beyond. It helped political and healthcare leadership to make informed decisions in real-time.
References:
The COVID-19 pandemic wasn’t just a public health crisis, it was a stress test for how humanity responds to uncertainty. Alongside doctors and nurses, virologists and vaccine makers, computer models became instrumental not only in helping fight the pandemic but also in upporting quick decision-making, demonstrating their potential to shape future responses to health crises.
This is the first in a three-part reflection on the role of in silico medicine during a global crisis. In the next chapter, we’ll dive deeper into how computer models based on digital twins helped in treatment development, and how they could serve for future outbreaks.
