The researchers have already made an estimate of the final number of confirmed infections and deaths for China, South Korea, Italy, Spain and Iran. It’s still too early to make such an estimation for the Netherlands, France, Belgium, Germany, Switzerland, Canada and the United States, but Van den Heuvel expects to present a conservative estimate for the Netherlands this week.

In order to make a more accurate comparison between the analyzed countries, the team decided since this weekend to display all numbers in the charts per million inhabitants. It then becomes immediately apparent that the virus seems to have caused relatively little damage in China; the model shows that the number of deaths in that country is three per one million inhabitants. This seems mainly due to the stringentquarantine measures in the province of Hubei, as a result of which the virus could spread across the rest of the country only marginally. It should be noted though that Dutch health organisation RIVM, among others, warns that the virus might resurface at a later stage under such circumstances.

Italy

Of all analyzed countries, the situation is most serious in Italy at this point: that country seems to be heading for about 180 deaths per one million inhabitants. With a population of sixty million, the virus could cause approximately eleven thousand deaths per 113 thousand confirmed infections in Italy, according to current predictions. That doesn’t mean, incidentally, that the mortality rate is ten percent: like in the Netherlands, many people will never be diagnosed with the coronavirus because they only show mild symptoms.

Random sampling would be required for more reliable data, Van den Heuvel emphasizes. “The current tests for the coronavirus are targeted, which leads to a selection that can cause both an overestimation and an underestimation.” He expects to be able to present longer term predictions for the Netherlands this week already. These should be dealt with cautiously, he emphasizes. “Because it’s still based on a testing policy that can’t be translated to the population that easily.”

Medical statistics

Edwin van den Heuvel is no stranger to medical data analysis: before he came to TU/e in 2014, he worked at the University Medical Center Groningen as professor in medical statistics for a few years, and before that he was head of the statistical department at pharmaceutical manufacturer MSD. As a data scientist, he simply felt compelled to analyze the current pandemic, he said earlier.

The predictions from Van den Heuvel and his colleagues Marta Regis and Zhuozhao Zhan are based on a so-called logisticpopulation growth model developed by Belgian mathematician Pierre-François Verhulst around 1845. The researchers use 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,’ they make a forecast of the number of new infections and deaths for the next day(s) and – when they have enough data – an estimate of the final total number of infections and deaths. They add new data every day to make the estimated number more reliable.

Reliable prediction

“The predictions for the short term, a maximum of three days into the future, deviate less than five percent from the reported figures in most cases,” Van den Heuvel says. However, it turns out to be more difficult than expected to provide a reliable forecast of the maximum number of infections and deaths in countries. “We still have to adjust that daily, despite the fact that we believed that some countries were stable. We still haven’t found the cause of this.”

The researchers are now trying to analyze how the various measures taken by governments influenced the growth rate. Van den Heuvel: “Our first findings seem to show that measures actually have a positive effect, but we are still in a validation phase in which we need to prove that this isn’t a problem of the data.”

Gratifying

Van den Heuvel’s work elicits many responses, he says. “We are closely followed by a group of people who want to know how the situation unfolds, and who also thank us for what we do. That is very gratifying. They send us articles, which we look into of course, share ideas on how we could improve our work, sometimes with regard to presentation, sometimes contents.” But there are negative responses as well. “We are sometimes told 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. I want to thank everyone who takes the trouble to visit our website and tries to help us.”