Model-based evaluation of school- and non-school-related measures to control the COVID-19 pandemic

The role of schools in SARS-CoV-2 transmission remains unclear. While school closures were implemented globally during the first wave of the COVID-19 pandemic, their effectiveness relative to other non-school-based measures is debated. This study aims to quantify the impact of school-based interventions in the Netherlands, specifically examining whether school closures would have altered the trajectory of the second wave in autumn 2020. The context of the study is the Netherlands' experience with COVID-19, which involved a phased approach to control measures, including school closures and relaxations. The importance of the study lies in providing data-driven evidence to inform policy decisions regarding school-based interventions during pandemics. Existing studies have yielded conflicting results, some suggesting that school closures are crucial, while others minimize their impact. This study addresses this uncertainty by using a sophisticated model calibrated to real-world data, allowing for a more precise assessment of the interplay between school and non-school measures.

Several studies have modeled the impact of school closures on COVID-19 transmission, yielding varying results. Some predict significant impacts, emphasizing the potential for schools to act as super-spreading events. Others downplay the role of school closures, highlighting the importance of broader societal measures. The discrepancy may stem from differences in model structures, data used for calibration, or the specific contexts examined. Some studies found that school closures could lead to a postponement of the peak, although it might not significantly alter the overall case numbers. The effectiveness of school closures also seems to be context-dependent. Studies on influenza have shown that school closures are an effective mitigation measure, but whether this effect translates to COVID-19 is not fully understood. This study contributes to the literature by using an age-structured model fitted to actual hospitalization and seroprevalence data, providing a robust and data-driven assessment of the role of school closures.

The researchers developed an age-structured deterministic compartmental model of SARS-CoV-2 transmission in the Netherlands. The model divides the population into age groups and infection stages (susceptible, exposed, infectious (multiple stages), hospitalized, recovered). The model incorporated age-specific contact rates from pre-pandemic and post-lockdown surveys, as well as school-specific contact rates. The model parameters were estimated using a Bayesian framework, fitting the model to age-specific hospitalization data (February 27 – April 30, 2020, n=10,961) and seroprevalence data (April/May 2020, n=3207). The Hamiltonian Monte Carlo method, implemented in Stan, was employed for parameter estimation. The model allowed the researchers to simulate different scenarios by manipulating the contact rates associated with schools and other settings. They calculated the effective reproduction number (R_e) under each scenario using the next-generation matrix method. The model’s ability to reproduce observed patterns of hospitalization and seroprevalence data increases the confidence in the accuracy of its predictions. In addition, the model used a frequency-dependent transmission, which is more accurate for densely populated areas like the Netherlands. The model was calibrated by setting its parameters to match the effective reproduction numbers estimated by the Dutch RIVM at specific points during the pandemic. This calibration step, coupled with the use of real-world data, improved the model’s predictive power. Separate analyses were performed for August 2020 (when schools reopened) and November 2020 (during the second wave) to assess the impact of measures at different pandemic phases. The age-specific impact of school closures was also investigated.

The model accurately replicated the observed age-specific hospitalization and seroprevalence patterns during the first wave. The estimated probability of hospitalization increased exponentially with age, except for individuals under 30. The estimated basic reproduction number (R₀) before interventions was 2.71 (95% CrI 2.15–5.18), dropping to 0.62 (95% CrI 0.29–0.74) after the full lockdown. In August 2020 (R_e ≈ 1.31), reducing non-school contacts by 60% would have reduced R_e to 1, while closing schools only decreased R_e by 10%. In November 2020 (R_e ≈ 1.00), reducing non-school contacts or closing schools resulted in similar R_e reductions (around 16%). Analyses investigating the age-specific impact of school closures in November 2020 showed that closing schools for 10–20-year-olds had the greatest impact on R_e (8% reduction), followed by 5–10-year-olds (5%), with negligible effects for 0–5-year-olds. The study also notes a strong positive correlation between age-specific hospitalization rates, highlighting the importance of using both hospitalization and seroprevalence data for parameter estimation.

The findings demonstrate that the effectiveness of school-based interventions is highly context-dependent. When non-school measures are sufficient to bring R_e below 1, the added value of school closures is limited. However, if non-school measures fail to control the pandemic, targeted school closures, especially for older children, can significantly contribute to epidemic control. These findings highlight the importance of a comprehensive approach that balances the benefits of different interventions and accounts for the specific conditions of the pandemic. The age-specific effect aligns with the observed age-dependent susceptibility to SARS-CoV-2 infection, with older children showing higher susceptibility compared to younger children. The study's implications are significant for policymakers, suggesting that prioritizing comprehensive non-school measures should be the primary approach. School closures should be considered a supplementary strategy when other measures prove insufficient. The model's limitations underscore the need for continued research and data collection to refine our understanding of school-based transmission dynamics.

This study provides crucial insights into the relative importance of school- versus non-school-related measures for COVID-19 control. The effectiveness of school closures is highly dependent on the level of non-school control measures already in place. While comprehensive non-school interventions should remain the priority, targeted closures in secondary schools can offer additional benefits when other measures are insufficient. Further research is needed to investigate the impact of specific school-based mitigation strategies, and to refine our understanding of the age-specific dynamics of SARS-CoV-2 transmission.

The study's limitations include the age grouping used for susceptibility estimation (0–20 years), potentially underestimating the effect of school closures for older children. The assumption that school contact patterns in August-November 2020 mirrored pre-pandemic levels might also overestimate the potential impact of school closures. The model also did not differentiate between various types of non-school contacts, and the lack of hospitalization data from the second wave limited the analysis. Additionally, the model’s conclusions are sensitive to the accuracy of serological data in identifying past infections, which is influenced by asymptomatic cases.