Corona predictions TU/e data scientists attract huge interest

At the beginning of March, when the seriousness of the corona crisis slowly began to dawn upon the world, Edwin van den Heuvel decided to take action. “I saw the growing number of coronavirus cases and realized that this could go dramatically wrong.” he says. “I spoke with a colleague: what can we do to help?” They took a closer look at the data for China, performed a few analyses, and quickly realized that they were able to successfully predict corona cases for all provinces in China, for Iran, Italy and South Korea.

His experience with epidemiology, when he conducted research into what causes people to get sick or how risk factors develop, proved helpful. Nor was he unfamiliar with the curves that describe the spread of the virus. “I worked in the pharmaceutical industry before, where the same kind of curves are used for drug concentration measurements.”

What started out as a small exercise, grew into a large project. Over the next six weeks, he predicted the number of confirmed infections and deaths several days ahead for no fewer than 24 countries and 4 Dutch provinces. For some of these, he was also able to estimate the maximum number of confirmed infections. Van den Heuvel was among the first ones to predict the definitive flattening of the curve in the Netherlands, and he made the first public estimation for the maximum.

Model with three parameters

Van den Heuvel worked closely together with two regular colleagues: Marta Regis and Zhuozhao Zhan. For six weeks, they started their days with the same morning routine. “Marta or Zhan collected the most recent data from the websites, ran the simulations, after which we discussed it with each other via Skype,” Van den Heuvel says. “We ran through all countries and discussed the most notable trends.”

The scientists based their predictions on a so-called logistic population growth model,  developed by Belgian mathematician Pierre-François Verhulst around 1845. The researchers used this model with its three parameters to analyze the officially reported mortality rates, starting at the moment of the first reported death. Based on this ‘fit,’ his team made a forecast of the number of new infections and deaths for the next days. They added new data every day to make the estimated number more reliable. Later they made another adjustment, based on a generalization of the Von Bertalanffy model, to improve the predictions.

Swamped mailbox

Once his predictions started to appear on the TU/e website, his mailbox got swamped with reactions. “That first week was very intense, with lots of reactions and media requests. I didn’t expect it to cause such a stir,” he says. The reactions come from all corners of society: hospitals and companies that want to use his data, as well as numerous interested people with suggestions and questions, which Van den Heuvel tries to answer as much he could.

The page with his predictions broke all records: it has been visited almost 358,000 times now. “I think many people started to follow our daily updates to have some kind of indicator for how the coronavirus pandemic evolves globally.”

The many requests led to a concrete collaboration with several hospitals, at first in the region, but also with a hospital in Rotterdam, for which he, together with other researchers, translated the expected numbers into hospital beds. “That is what you do it for of course, to have an actual impact in society.”

Reliable

He also got into contact with the RIVM (the Dutch Institute for Public Health and the Environment). “There were some uncertainties at first about what we did exactly and how we worked, because we use the accepted diffusion models in a slightly different way, but eventually we developed a good working relationship. Our page is also mentioned on their website with reliable data sources.”

However, there were negative reactions as well. “We were told for example that we couldn’t be more wrong and that we are a kind of amateurs. We take everything seriously because we want to do the best work possible with the data we have. But I have to say I was surprised by how fast people came to conclusions sometimes, without seriously examining first what it is we do. That can make me angry.”

Much of the criticism was aimed at the limited value of the predictions, for example, since the available data is incomplete. “We acknowledge the problems with the data and we were careful not to make strong statements,” Van den Heuvel says. “We didn’t take the traditional route according to which you need good data first to build a model. Instead, we focussed on the data and tried to retrieve knowledge from that data by testing several different models and making comparisons to other countries. We did much more behind the scenes than people see on the website.”

Next steps

Meanwhile, Van den Heuvel has stopped the daily updates. Now that the pandemic is beginning to level off in many countries, the daily charts no longer provide many new insights. But that does not mean that the research has completely stopped. Behind the scenes, the team continues to monitor the situation to see whether the pandemic is flaring up again, now that in many countries the lockdown is slowly being lifted. As soon as this is the case, Van den Heuvel will inform the public.

Van den Heuvel is also thinking about the next steps. “A PhD candidate is investigating what effects the measures have; I expect to see the first results from that soon.  We will also teach a course for math students in the coming quartile year about models for infectious diseases, and I want to try and set up a (research) project in collaboration with the RIVM.”

In conclusion, is there anything he would have done differently in retrospect? “If I had known what this would have led to, I might have wanted a larger team,” Van den Heuvel says laughing. “Then someone could focus exclusively on the search for better data, and someone else could run more simulations.” But at the same time, he is proud of his team. “We are a small, clos-knit team. It’s special how we came up with this together and then put it into practice. We will keep trying to make a contribution.”

The daily forecasts by Edwin van den Heuvel and his team can be retrieved here.

This research is part of TU/e against COVID-19, a TU/e wide initiative to help solve the corona crisis.