The initial COVID-19 wave (February-June 2020) prompted widespread NPI implementation across Europe, including business closures, school suspensions, and gathering bans. While these measures achieved partial control, a second wave followed societal reopening (August 2020-January 2021). With vaccines initially limited and population immunity waning, alongside new variants of concern, effective interventions remained critical. Existing studies extensively analyzed first-wave NPI effectiveness, comparing transmission trajectories with and without NPIs. However, these first-wave estimates, based on baseline contact patterns lacking organizational safety measures and individual protective behaviors, are insufficient for assessing NPI impact during an ongoing pandemic. Post-first wave, contact patterns didn't revert to pre-pandemic levels; individuals and organizations adopted protective measures like distancing and improved ventilation. These changes likely mitigated the impact of strict NPIs. Furthermore, governments needed effectiveness estimates for the more granular NPIs used in the subsequent waves (e.g., sector-specific business closures, small group gathering bans). This necessitated a multinational, subnational dataset to account for varying NPI implementation timings and regional heterogeneity in transmission dynamics.

Numerous studies investigated the effectiveness of NPIs during the first wave of COVID-19. These studies primarily compared transmission rates before and after the implementation of NPIs, often using aggregate national-level data. While these studies provided valuable insights into the broad impact of NPIs, their limitations included a lack of granularity in both the NPI definitions and the geographical scale of analysis, and an overreliance on aggregate national data. Several studies (e.g., Flaxman et al., 2020; Brauner et al., 2021; Hsiang et al., 2020; Salje et al., 2020) used various methodologies but tended to focus on the early phase of the pandemic when safety measures and individual protective behaviors were minimal. These first-wave analyses served as proxies for the impact of various activities on transmission in the absence of safety measures. However, these first-wave estimates are potentially inadequate for assessing NPIs during a pandemic's later stages.

To address the challenges of the second wave, the researchers developed a semi-mechanistic hierarchical Bayesian model. This model improves upon previous models by incorporating a latent random walk to account for unmodeled changes in transmission and by allowing for stochasticity to prevent artifacts from low case counts. This approach enabled the estimation of the effects of 17 individual NPIs using case and death data from 114 regions across seven European countries (Austria, Czech Republic, England, Germany, Italy, Netherlands, Switzerland). The dataset, manually compiled, involved a systematic categorization of interventions and rigorous validation procedures to ensure high data quality. The model accounted for the concurrent implementation of multiple NPIs by leveraging the variation in NPI implementation timing across regions. Subnational data were crucial to avoid ecological fallacies and biased effect estimates. The model's robustness was assessed through sensitivity analyses that explored variations in data, model parameters, epidemiological assumptions, and potential unobserved confounding factors. The model was implemented in NumPyro using a No-U-Turn Sampler (NUTS) for posterior inference. Key epidemiological parameters (generation interval, incubation period, onset-to-death delay, onset-to-case confirmation delay) were derived from a combination of meta-analyses and linelist data, carefully accounting for potential biases and censoring. To mitigate bias from the emergence of new variants, data points after the emergence of variant B.1.1.7 were excluded. Data preprocessing included excluding observations from the early phase of the study period to avoid confounding due to infections originating before the start of the study.

The overall effect of all NPIs was smaller in the second wave (66% reduction in Rₜ) compared to the first wave (77-82% reduction). This difference is attributed to the widespread adoption of organizational safety measures and individual protective behaviors, making various public life areas safer and diminishing the added impact of NPIs. Specifically: * **Business closures:** remained highly effective (35% Rₜ reduction), with significant effects observed for gastronomy (12%), nightclubs (12%), and retail/close-contact services (12%). Leisure and entertainment venue closures showed a smaller effect (3%). * **Gathering bans:** banning all gatherings, including one-on-one meetings, had a large effect (26% reduction), but only the strictest thresholds (2 people) were particularly effective. Less stringent limits (10 or more people) had smaller effects, possibly due to voluntary protective behaviours and limited adherence to rules on private mixing. * **Educational institution closures:** had a small effect (7%) during the second wave, compared to a substantial effect in the first wave. This difference is attributed to safety measures implemented in schools, preventing large undetected clusters. * **Mask mandates:** a stricter mask-wearing policy reduced transmission by 12%. * **Nighttime curfews:** reduced transmission by 13%. Sensitivity analyses across 86 experimental conditions demonstrated that the results were robust to model structure, parameter distributions, and data variations, indicating consistent findings across different conditions. Furthermore, second-wave estimates outperformed first-wave estimates in predicting transmission during the third wave, indicating their greater relevance to ongoing pandemic scenarios.

This study's findings highlight the dynamic nature of NPI effectiveness and underscore the influence of pre-intervention safety measures and behavioral changes on NPI impact. First-wave estimates overestimate NPI effectiveness in an ongoing pandemic because they lack the context of protective measures and behaviors adopted during the later stages. The study's robust findings offer valuable insights for policymakers. While NPIs like closures and bans continue to reduce transmission, their effectiveness is context-dependent. The results suggest that implementing adequate safety protocols can significantly reduce the need for school closures and that only the strictest gathering limits remain impactful. The study also emphasizes the importance of factors such as mask mandates and curfews in curbing transmission.

This research provides crucial evidence on the effectiveness of individual NPIs during the second COVID-19 wave in Europe. The findings highlight the importance of considering the dynamic nature of NPI effectiveness and the influence of both organizational safety measures and individual protective behaviors. The study’s results can inform policy decisions by offering data-driven insights into the effectiveness of various interventions in an ongoing pandemic. Future research should focus on exploring the interplay between NPIs, emerging variants, and evolving societal behaviors.

The study relied on observational data and is subject to potential unobserved confounding factors, including variations in adherence to NPIs. The findings are based on a specific set of European countries during a particular period, and generalizability to other contexts should be approached cautiously. The model's assumptions, while carefully considered, may introduce uncertainty into the estimations, particularly regarding the assumptions made on delays between infections and reported cases/deaths.