The Impact of School and Family Networks on COVID-19 Infections Among Dutch Students: A Study Using Population-Level Registry Data (preprint)

Understanding the impact of different social interactions is key to improving epidemic models. Here, we use extensive registry data -- including PCR test results and population-level networks -- to investigate the impact of school, family, and other social contacts on SARS-CoV-2 transmission in the Netherlands (June 2020--October 2021). We isolate and compare different contexts of potential SARS-CoV-2 transmission by matching pairs of students based on their attendance at the same or different primary school (in 2020) and secondary school (in 2021) and their geographic proximity. We then calculated the probability of temporally associated infections -- i.e. the probability of both students testing positive within a 14-day period. Our results highlight the relative importance of household and family transmission in the spread of SARS-CoV-2 compared to school settings. The probability of temporally associated infections for siblings and parent-child pairs living in the same household was 22.6--23.2\%, and 4.7--7.9\% for family members living in different household. In contrast, the probability of temporally associated infections was 0.52\% for pairs of students living nearby but not attending the same primary or secondary school, 0.66\% for pairs attending different secondary schools but having attended the same primary school, and 1.65\% for pairs attending the same secondary school. Finally, we used multilevel regression analyses to examine how individual, school, and geographic factors contribute to transmission risk. We found that the largest differences in transmission probabilities were due to unobserved individual (60\%) and school-level (34\%) factors. Only a small proportion (3\%) could be attributed to geographic proximity of students or to school size, denomination, or the median income of the school area.

Predicting COVID-19 Infections Using Multi-layer Centrality Measures in Population-scale Networks (preprint)

Understanding the spread of SARS-CoV-2 has been one of the most pressing problems of the recent past. Network models present a potent approach to studying such spreading phenomena because of their ability to represent complex social interactions. While previous studies have shown that network centrality measures are generally able to identify influential spreaders in a susceptible population, it is not yet known if they can also be used to predict infection risks. However, information about infection risks at the individual level is vital for the design of targeted interventions. Here, we use large-scale administrative data from the Netherlands to study whether centrality measures can predict the risk and timing of infections with COVID-19-like diseases. We investigate this issue leveraging the framework of multi-layer networks, which accounts for interactions taking place in different contexts, such as workplaces, households and schools. In epidemic models simulated on real-world network data from over one million individuals, we find that existing centrality measures offer good predictions of relative infection risks, and are correlated with the timing of individual infections. We however find no association between centrality measures and real SARS-CoV-2 test data, which indicates that population-scale network data alone cannot aid predictions of virus transmission.