One of the primary objectives of the Big Corona Study was to gain insight into the frequency and nature of social interactions of people. Since the coronavirus is mainly transmitted through close contacts, data on our social behavior and contact patterns are essential for understanding the dynamics of this infectious disease. In other words, the answer to the question "who sees whom?" can also help clarify the question of "who infects whom?".

However, during a pandemic, contact patterns naturally change a lot, for example, due to lockdowns. It's not just the measures themselves, but how well they are followed and by whom. Therefore, it is crucial to continuously monitor how our contact behavior evolves in practice. This can provide more insight into the actual impact of certain measures - which measures influence the social behavior of which groups in society? Which measures are the most efficient in actually reducing virus circulation? The answers to these questions help to manage the crisis in real-time.

COVID-19 was certainly not the first respiratory infectious disease to raise these types of questions. Colds, the flu, pneumonia... We and other scientists across the globe have been researching for years how and when infected individuals make contact with others and potentially infect them, as well as how disease spread could be prevented.

Surveys to map contact patterns

In a reference study published in 2008, we, for the first time, surveyed people on a large scale about their social interactions in order to model the spread of infectious diseases. Participants were asked to keep track of how many physical and non-physical social contacts they had each day. Physical contacts included actions like shaking hands or giving a kiss, while, for instance, a conversation with someone that didn't involve physical touch was categorized as non-physical contact. For each contact, participants also noted place (e.g. inside or outside?), duration, as well as relevant characteristics of their contactperson (male or female, age, etc.). Over 7,000 people from eight different European countries participated in the study, including Belgium, Germany, Finland, the United Kingdom, Italy, Luxembourg, the Netherlands, and Poland. On average, they were found to have between 13 and 14 contacts per day. Although there was little difference across countries in which groups had relatively more contacts, the number of contacts itself varied widely. For example, children had more social contacts than adults, people from larger families also had more contacts, and the number of contacts was lower on Sundays.

When we used this much richer data set to model the spread of a fictitious respiratory infection, we could perform much more detailed analyses and draw conclusions tailored to specific age groups or population segments.

Mathematical models have always taken into account the number of contacts people maintained to predict the course of epidemics, but that type of information was initially often based on assumptions and other indirect factors, never on concrete observations. Meanwhile, it has become common practice to use survey data in this field of research, but at the time, this was very innovative. Epidemiologists refer to the "social contact hypothesis," which states that reported contact patterns provide a good approximation of the spread of airborne infectious diseases.

More studies followed. Other researchers also began using various methods to measure social behavior: from surveys or diary studies to direct observations with electronic sensors or mobile phone data (similar to the Corona Alert app). Not long ago—just a few months before the SARS-CoV-2 virus would turn the world upside down—we conducted a comprehensive literature search of all scientific studies on contact surveys for modeling infectious diseases. Such a systematic overview is useful not only for global scientific evidence (are the results of a particular approach comparable in studies conducted by different researchers, in different contexts, and in various forms?) but also to bring together best practices (which specific approach or variation in the method yields the best results?).

We compared a total of 64 different studies from 24 countries that used surveys to map contacts. Some surveys targeted the general population, while others focused on very specific groups such as pupils or students. In some cases, participants were questioned about contacts in a period in the past (retrospective), while other studies asked them to keep track of contacts as they happened (prospective). The precise definition of a 'contact' varied from study to study: not all surveys distinguished between physical and non-physical contacts, and the criterion for what exactly constituted a non-physical contact sometimes depended on a certain proximity (e.g., closer than 2 meters) or whether there was a conversation (e.g., with at least 3 words, more than just a simple 'hello'). The overview of all these studies reveals that existing contact research typically includes around a thousand participants, and the number of contacts per day generally ranges from 8 to 27.

Most studies collect data from participants through self-reporting in a daily or online log. While face-to-face interviews result in more complete participant responses, they are far more labor-intensive for researchers. In terms of accuracy, there appears to be little difference between paper and online surveys. Where data was collected via proximity sensors, they were found to be more accurate in gathering short contacts, but interaction patterns picked up were very similar, regardless of the measurement method.

What proved to be of fundamental importance is the definition of what precisely constitutes 'a contact.' Given that this parameter is essentially used as a surrogate for exposure to an infectious disease, it's crucial to adapt your survey to fit your research question. Even for infectious diseases transmitted through droplets, a personal conversation as a definition or criterion for a non-physical contact can lead to an underreporting of the number of risk contacts, as susceptible individuals could also be infected by someone who sneezes or coughs in close proximity. Even more challenging to account for is transmission via objects such as doorknobs or water taps through which you can indirectly infect others.

A second crucial element is when participants are surveyed. The data shows that about 6% fewer contacts are reported on the second day of the survey, a sign of what we call 'reporting fatigue.' Moreover, it holds that the more contacts reported on the first day, the greater the drop in the number of contacts reported on the second day. For retrospective studies (which look back), the longer the period, the more 'recall bias,' meaning vaguer recollection and poorer reporting of contacts.

Finally, the comparison of many international studies highlights the importance of the precise characteristics queried for each social contact. Collecting the age of both participants and their contacts is useful for mapping age-specific interaction patterns that have been effectively observed for example for pertussis, varicella, or parvovirus. Also, the location, duration, and frequency of contacts can help provide a nuanced view of effective disease prevention or control. An example is a study on the spread of varicella zoster among non-vaccinated children. Here, close contacts with a duration of at least 15 minutes and skin-to-skin contact best described who was effectively infected.

In addition to the concrete recommendations we distilled from our meta-analysis, we compiled those best practices and existing databases into a standard format that we made freely available online, in combination with an open-source software package so that other researchers can use it to analyze new data on social contacts. With this, we primarily aim to facilitate data exchange in the future. If different research groups collect the same type of data using the same survey and methodology, the results can easily be compared between different countries, settings, or over time.

When a lockdown in our country seemed inevitable in early 2020, we knew what to do to reliably and efficiently map changing contact patterns through the Big Corona Study. This example of contact analysis shows that research, even on a completely new virus in an unprecedented societal context, always builds on decades of accumulated knowledge and expertise.

A look across the border

By the end of May 2020, data was already available on our platform from several countries, including France, China, Hong Kong, Peru, the UK, Russia, Zimbabwe, Vietnam, South Africa, and Zambia. Lander Willem and his colleagues could immediately use these international data to investigate the effect of social distancing measures in the workplace and the closure of schools on the spread of the coronavirus.

For example, a reduction of 60% in workplace contacts would reduce the reproduction number by 10% in most countries, even more in Poland and Hong Kong. An exception is Peru, and to a lesser extent, Zimbabwe, where workplace measures have much less impact because they constitute a much smaller proportion of total social contacts compared to elsewhere. The effects of school closures show even more regional differences. In our country and Vietnam, a school closure is estimated to be associated with a 10% reduction in the reproduction number, but in Italy, Luxembourg, or France, the effect is twice as large. Much still depends on the context, about which we knew very little at that time. If we assume that the elderly are much more vulnerable to the virus than children, which was already suspected for COVID-19 at that time, then the impact of closing schools is much smaller.

We collected additional data through the European CoMix study. This study follows the impact of the coronavirus crisis on awareness and behavior for hundreds of participants in different European countries over an extended period. CoMix is a survey, but unlike the Big Corona Study, it is a longitudinal study, allowing changes over time to be tracked for each participant. Moreover, it is a panel study, meaning that the participants were selected in such a way that they form a representative group in terms of age, gender, and region—for our case, the Belgian population.

CoMix started in March 2020 in Belgium, the Netherlands, and the UK but was later expanded to 17 other European countries, thanks to collaborations with local universities and health institutions and support from the European Centre for Disease Prevention and Control and funding through the European Commission. We brought all the data from these different countries together within our online tool.

The results from CoMix showed that the average number of contacts was significantly lower than in non-pandemic times. We compared the data with Flemish research from 2010 as a reference for social contact patterns. During the lockdown, the number of contacts decreased by a whopping 80%, compared to 2010. This figure was similar in other European countries. In Shanghai and Wuhan, there was an even stronger decrease in social contacts, while in the United States, the decrease was more limited at that time. Even after reopening businesses and non-essential shops, the number of contacts remained low. Then, we began to make more and more contacts again, especially among younger age groups. During the summer, the number of contacts decreased again. The median values, however, remained relatively stable, meaning that if you lined up all participants according to their number of social contacts, the person in the middle had approximately the same number of contacts. In other words, the increase and decrease in the average are mainly due to a smaller group of people who significantly change their behavior.

For the entire period from March to August 2020, fewer people kissed, hugged or shaked hands (skin-to-skin contacts) than in the reference period in 2010. Belgian data revealed that residents of Flanders had more social contacts on average than residents of Brussels or Wallonia.

A large proportion of the selected panel participated, especially in the first rounds (75.7%), but this percentage decreased to about 50% by the last round after 16 weeks, in the middle of the summer. Importantly, there was no link between the number of contacts and a person's decision to stop participating. In the survey, participants were asked to record their number of contacts in the past period. For this specific study, the effect of potential underestimation or forgetfulness is considered small, given that we are dealing with very recent contacts each time and the total number is very low due to strict measures and an exceptional lockdown situation. Of course, the spread of the coronavirus depends not only on social contact patterns. How easily the virus can be transmitted also depends on the use of face masks, possible ventilation, maintained distance, and even larger-scale seasonal effects.

And what about children? For practical reasons, minors were excluded from participating in the CoMix study. The ethical evaluation of study protocols targeting minors is naturally more extensive and therefore time-consuming, which would have slowed down the start of the study. Making predictions based solely on the contact behavior of adults has significant limitations, which is why we asked CoMix participants with children to also report how many contacts their child had each day.

We also collaborated with international colleagues from outside Europe, including for a similar study in Quebec, Canada. There, too, the average number of contacts dropped significantly, from eight per day before the lockdown to three per day during the spring 2020 lockdown. The average also gradually increased to five contacts per day in September 2020 when schools reopened but dropped again to four per day during the second wave.

In Japan, the most recent data on contact patterns dated back to 2011. Especially with the Tokyo Olympics, which were ultimately postponed from 2020 to 2021, it was necessary to have a better understanding of the pandemic's impact and measures on social behavior. In collaboration with researchers from various Japanese universities, we conducted a local study to assess changes in contact patterns before and during the Olympics. We found that, typically, people had about three contacts per day (compared to twelve in the 2011 study). The number of contacts varied depending on the age group of the participants, but there were no significant differences during the Olympics. The strict measures during the Games themselves and in and around Tokyo seemed to have had an effect.

With all these studies, we were able to provide not only our own models but also those of fellow researchers worldwide with up-to-date data on contact patterns to map the spread of the virus as accurately as possible. In addition, the results provide insight into the actual impact of the imposed measures. Policymakers can use this information to adjust decisions if needed and strive to balance the positive and negative effects of COVID-19 policies.