During the first wave COVID-19 wave in the spring of 2020, our testing capacity was limited to about 2,000 to 3,000 PCR tests per day. To track the evolution of the epidemic and the effect of lockdown measures through the presence of antibodies among the population over time, we analyzed between 3,000 and 4,000 "surplus" random blood samples from different diagnostic laboratories from across the country at fixed intervals. In total, more than 22,000 samples were collected between late March and mid-October 2020. We checked each of them for the presence of antibodies to the S1 spike protein of SARS-CoV-2.

On April 1st, 2020, it was estimated that about 3% of the Belgian population had antibodies to the coronavirus. Three weeks later, that proportion had doubled to about 5 to 6%. At that time, only 0.4% of the population had had a confirmed COVID-19 infection, effectively confirming that a multitude of infections occurred that were not recorded in the official statistics.

There are good reasons to suspect that even the estimate based on serological testing was still an underestimate of the true figures given that people with suspected symptoms of COVID-19 were asked to stay home as much as possible and defer non-urgent care, which means they will also be underrepresented in our pool of routine blood samples.

Moreover, there appeared to be large differences between age groups: at the very beginning, the number of people in their twenties with antibodies was the lowest of all age groups at only 0.6%, while during the same period, it was estimated that almost 6% of children under 10 years of age had antibodies against the coronavirus. During the month of May, nearly 7% of all Belgians were reported to have antibodies, but by mid-June that number had already dropped to 5.5%, and by July even further to 4.5%. These figures suggest that at least in part of the cases, the accumulated immunity was short-lived, and therefore, that the results do not provide a conclusive answer as to who had ever been infected with the virus.

In any case, the sensitivity of the serological test depends on how long after infection it is taken. Very recent infections with insufficient build-up of antibodies may be missed. Some studies suggest that antibody levels begin to decline after just two to three months, while other studies suggest that protection remains much longer. The figures and trends for Belgium are in line with those from Switzerland, Spain and the UK, although comparisons remain difficult because the nature and timing of the measures differ across studies.

In the early stages of an epidemic, there is also very little information about the reliability of diagnostic tests, simply because testing has been limited and because clear negative controls are not always available (samples from someone for whom we are 100% sure they did not have COVID-19).

How good a particular test method is is usually expressed in terms of two parameters: sensitivity and specificity. Sensitivity expresses how well the test can detect a positive case, in other words: the probability that the test will detect the presence of, in this case the coronavirus, in someone who is actually infected. Specificity on the other hand expresses how good the test is at correctly identifying someone who is not infected as negative.

PCR tests are quite reliable, even though their sensitivity and specificity are not 100%. In other words, there is a non-negligible number of false-negative results: people who are still infected but where the test cannot detect the virus. Reliability is also affected by when (at the first symptoms? Too early or too late after infection?) and how well the sample is taken, etc. Therefore, it is important to also look at those parameters in context: which test is most useful for screening how many asymptomatic cases there are? Which one to follow up contacts? Which one to identify a cluster? Which one for prevention or for diagnosis? In hospitals, an accurate and sensitive test is important to administer appropriate care for each individual patient, but a strategy with rapid, weekly testing may be better as a public health strategy, even though that rapid test may be a little less sensitive. In practice, factors such as price, speed and ease of use are certainly as important as sensitivity and specificity.

In addition to developing a convenient and reliable test, its rollout also depends on lack of reagents, lab capacity and manpower. Especially in the case of a global pandemic, a shortage of raw materials and equipment was a real concern. To test as efficiently as possible, samples can be pooled. There is a lot of expertise and research data on how to test most efficiently, even under time pressure or when resources are scarce. Again, an example where existing knowledge for other infectious diseases can help to develop a decisive approach even in a pandemic caused by an entirely new virus.

When testing capacity and policies were up to speed, we could finally use these data to determine Rt. The big advantage of using new cases instead of hospitalizations, is speed: you pick up a possible increase (or decrease) in the reproduction number much faster and can thus adjust immediately when the epidemic gains strength. After all, hospitalizations can remain stable for an entire period, even when the number of cases is already increasing (rapidly). Especially when there is an increase in infections among young people, who only rarely end up hospitalized.

By asking all participants of the Great Corona Study whether they are experiencing the typical corona symptoms, we wanted to immediately attempt to gauge the severity of the epidemic, and where exactly it is occurring. Of course, this is only a survey, not a diagnosis, and the sample is not immediately representative of the general population for various reasons. We can account for this by, for example, taking into consideration the participation rate per municipality when interpreting the number of people with symptoms for that municipality. But even with careful analysis that corrects for various selection biases, the question remains: is the survey sensitive and comprehensive enough to really say something about virus circulation? In other words, can COVID-19 symptoms as reported by participants in the Great Corona Study be used to make predictions about COVID-19 figures in different regions in Belgium? To find an answer to that question, we compared the self-reporting of symptoms by participants in the third round of the Great Corona Study on March 31, 2020 (about 3.5% of the Belgian population), with the actual infection figures a week later, as reported by Sciensano between April 7 and 9, 2020.

It seems that the estimations could indeed be more accurate and comprehensive if we were to gather more precise data than just the postcode, and if clusters, in other words, could also be mapped on a much smaller scale. Especially during the period at the end of March and April 2020, when the country was in lockdown, the typical transmission routes were limited (large gatherings and mass events were prohibited), and virus transmission therefore occurred on a much more local and localized scale. A second important way to improve the predictive value of this kind of crowd-sourcing data is if we knew exactly when symptoms began. Currently, the questions in the Great Corona Study only inquired about which symptoms someone had "in the seven days prior to the day of the survey." However, the symptoms could have started earlier.