Carnivals, air pollution and Covid-19 in the proximal regions of Germany, Belgium and Netherlands
The grotesque failure of the most widely cited epidemiological models to correctly predict the dynamics of Covid-19 epidemics (that this author has documented quite early in the Lombardian case) raises the question whether many areas of science today are not too much focused on modeling at the expense of actual understanding of the processes that are being modeled. Epidemiology is certainly not an exception in this regard, my own field of economics is replete with the same kind of modeling, where the only reason for the extremely distortionary assumptions is that practitioners desire to do models. But in epidemiology, the damage caused by those models may have been particularly huge because the Covid-19 pandemic has been one of those rare cases where expert predictions have been at the core of the debate (even though cynical people among us could say that politicians only embraced the alarmist models because it suited them).
It is now clear from the data from multiple countries and regions that Covid-19 epidemics do not at all manifest the exponential growth pattern postulated for them by prevailing models. Instead of uniform exponential growth (the same R0) until intervention, there seems to be a brief period of explosive growth followed by constantly slowing growth and then, decrease. There is clearly something (or potentially, several patterns) that takes place at the initial explosive phase that urgently requires understanding. Especially given that some countries and regions (Japan, Denmark, Israel, New Zealand) have for some reason avoided that stage or had it in a milder form.
To see this clearly one could look at the case growth data plotted on the logarithmic scale, for instance, for Netherlands and New Zealand. It is obvious that the initial growth was dramatically faster in the Netherlands.
So why do some countries have initial explosive growth, while others do not? I think that perhaps the best way to try to understand this is through going at least one level lower on the territorial scale.
This approach can often provide insights that are impossible without it. For instance, once one realizes that the worst hit Swiss canton is Ticino that borders the Italian region of Lombardy, one can understand a lot more what has come to pass in Switzerland. And once one learns that there is the phenomenon of many Lombardians regularly going to Ticino for work and back to Lombardy, the puzzle gets even clearer.
But perhaps, even more insights can be obtained from a context where proximate regions from not two but three countries are all badly hit by Covid-19. I am talking about the area combining areas in Eastern Belgium, Western Germany and Southern Netherlands. Amazingly, in all three countries, it is exactly those areas that are among, if not the, worst hit. The notorious commune of Gangelt is the Wuhan of Germany, and it is right next to the border with the Dutch province of Limburg. The Dutch provinces of North Brabant and Limburg (bordering both Germany and Belgium) have been the worst-hit ones in Netherlands, especially early on. The Belgian province of Limburg bordering the Dutch provinces of North Brabant and Limburg is the epicenter of Covid-19 in Belgium.
Initially, I had the hypothesis that the carnival superspread in Gangelt was behind this regional clustering pattern of Covid-19. However, the scale of the event in Gangelt makes this rather unlikely, and it probably happened too late to explain so much spread in three countries.
However, could there be something about all the regions in question in Germany, Netherlands and Belgium that could explain their misfortunes? The first part of the answer is provided by carnivals just like the one in Gangelt.
Carnivals are a Catholic tradition that survived all the way from the Middle Ages, and in Western Europe, they tend to take place in mid-February. Wait a minute, you could say, but Netherlands is a predominantly protestant country. You would be right to make this objection but there are two Dutch provinces that are predominantly Catholic, and you may have already guessed what provinces those are: Limburg and North Brabant. The region of North Rhine Westphalia where Gangelt is located is also Catholic, as is the Limburg province of Belgium.
The role of carnivals in the spread in Gangelt seems to be well-accepted. There seems to even be genetic evidence. Is there more direct evidence for the relevant provinces of Netherlands and Belgium than the fact that carnivals are held there in February?
The best evidence from the Netherlands comes from a preprint summarizing the results of testing of healthcare workers (HCWs) in three hospitals in exactly the regions that interest us. In the study, 96 out of the 1,796 participating HCWs tested positive for SARS-CoV-2. Sixty (sic!) out of those 96 had participated in carnivals 14 days before onset of symptoms.
As for Belgium, the worst-hit cities are Hasselt and Sint Truiden. They both had carnivals in the end of February — early March.
But interestingly, carnivals are only part of the story of what may have gone wrong in those regions. At least one more major factor has probably been in play: air pollution. There is a growing body of research suggesting that high air pollution in certain regions may have significantly contributed to the spread of Covid-19 either through predisposing people in them to infection through long-term exposure to harmful particles, or through the facilitation of aerosol spread, or, perhaps, both.
Most interestingly, in a recent preprint, Andrée finds that PM2.5 concentrations are a significant predictor of Covid-19 incidence in the 355 Dutch municipalities that he analyses. He writes:
It is intriguing, however, that the highest case density in the Netherlands is in Brabant, the southeastern part of the country, while major cities like Amsterdam and Rotterdam are in the west part of the country where the case density is lower. While Brabant is not the most populous province, it accounts for the highest contribution to nation-wide industrial GDP. Within the province, the sub-region Zuidoost-Noord-Brabant produces the highest contribution to industrial GDP. This area approximately spans the COVID-19 case cluster that can be seen on the map.
He further notes that the German region of North-Rhine Westphalia where Gangelt is located is highly industrialized. Finally, one can also easily discover that the Belgian province of Limburg is home to areas with significant industrial concentrations, too. Genk is the industrial centre of the province and it is right next to Hasselt on the map.
The most interesting questions about the Covid-19 pandemic will not be answered with the help of sterile modeling, or even mere statistical hypothesis testing. Instead, there is a wealth of opportunity waiting for enterprising researchers who are willing to do what is essentially detective work. One has to hope that science fields like epidemiology will re-embrace qualitative research of this kind, which is badly needed. I suspect that the more such research will be done, the more starkly evident the role of superspreader events like carnivals will become, coupled with other factors like air pollution.