Our results show that non-pharmaceutical interventions were effective in decreasing transmission, with an overall estimated decrease in the 7-day cumulative incidence of 22% in one week, starting 5 days after an increase in restrictions of 1 standard deviation. Our models assigned the highest and most consistent effects to interventions in social distance and bars and restaurants, particularly indoors, which each decreased incidence by 9–13%, depending on the model. Inconsistent associations with decreased COVID-19 transmission were found for culture and leisure venues, ceremonies and religious celebrations, and indoor sports, possibly explained by a certain collinearity remaining in the model. For any model, no effect of outdoor sports, commerce or mobility restrictions was found in decreasing transmission.
Our study has some limitations. First, the definition of a quantitative index entails discretionary decisions. For example, restrictions in bars and restaurants included different measures, some of them qualitative, such as the prohibition of using the bar or stay standing-up, limitations to number of persons per table, to the capacity or in opening hours. Further, the resulting scale is being compared to NPI applied, for example, to sport activities. Decisions were taken by a panel of expert collaborators and are documented and freely available. Moreover, NPIs were graded from a theoretical maximum that in some cases was not achieved. Indexes were normalised to improve comparability, but still, an increase of one standard deviation may not have an equivalent meaning in all fields. Second, there was important correlation between different fields, meaning that they tended to increase or decrease together. This could difficult identifiability of individual effects (8, 26). The attribution of effect to NPIs in specific fields should be done with caution, as it could be sensitive to analytical choices and model specification. Third, we analysed official restrictions, but not adherence to them, nor precautions decided by individuals on top of existing recommendations. Some studies point to difficulties of the population in understanding complex and changing norms(27, 28), while others show how people may increase precautions and decrease mobility by their own decision(29). Pandemic fatigue may have further decreased compliance throughout the study period. Finally, we left our some NPIs to avoid noise in the index. For example, measures in betting and gaming venues, or in swimming pools were overlooked, considered to represent a small fraction of interactions in the field. Any effect of these measures would be spuriously attributed to restrictions implemented simultaneously, for example, in bars and restaurants or in sports facilities in general.
As a strength, the study was conducted over a long period of 8 months and three epidemic waves, with on and off measures at different points in time across 50 territories, providing a rich natural experiment with sufficient variability. Of note, there were no significant changes in the testing and surveillance recommendations in this period, making the time series reliable within each territory. Moreover, many cultural, socioeconomic factors and other contextual variables are shared by the territories, being more homogeneous than international comparisons, facilitating the attribution of differences to different levels of NPIs(13, 30).
Some previous studies have estimated the effectiveness of NPIs in the second and successive COVID-19 epidemic waves. A study in 114 regions of 7 European countries using subnational data and analysing 17 different NPIs found a combined effect of all NPIs of 66% reduction in the (instantaneous) reproduction number Rt(10). However, it evaluated strict measures, such as closure of leisure and entertainment venues, gastronomy, retail and close-contact services, night clubs or educational institutions. Therefore it is expected that the impact on SARS-CoV-2 transmission is larger than the one estimated in our study, where softer restrictions are considered. More in line with our results, in Italy, implementation of the less stringent “yellow” tier was associated to decreases in Rt after 3 weeks of 13–19%, the “orange” tier, including closure of restaurants and restrictions to intra-municipality mobility to 27–38% and the strictest “red” tier to 36–45%(15). However, in countries such as Italy or the UK, measures were implemented in tiered levels with fixed combinations of NPIs, making it more difficult to evaluate the relative contribution of the different types of NPIs(13–15).
Regarding the effectiveness of NPIs in particular fields of activity, in Switzerland, business closures, recommendations to work from home and restrictions on gatherings were particularly effective (31). In 7 European countries(10), the highest effect was found for closing non-essential business, which decreased COVID-19 transmission by 35%, including night clubs, restaurants, retail and close-contact services and, to a smaller extent, closing leisure and entertainment venues (theatres, museums and zoos) which only contributed 3% of the total effect). Limiting gatherings to 2 people also had a very important effect, reducing COVID-19 transmission by 26%. Curfew alone decreased transmission by 13%. As indicated before, these were more stringent interventions than in our study, therefore a higher impact in transmission is expected. For example, our estimated effect for bars and restaurants (combined indoors and outdoors) was 12% reduction in 7-day cumulative incidence in 1 week, and restrictions in culture and leisure venues were only significant in one of the two models, reducing incidence by 14%. We did not evaluate measures in nightclubs, since they were closed in the entire study period. Contrary to the findings in the mentioned study, we did not find any effect for restrictions in commerce, although in Spain total closure was seldom applied, and restrictions in capacity or opening hours were more common. We also did not find any effect of mobility restrictions, including curfew, possibly because it was in place in the vast majority of the study period, with only variations in hours affected. A study in Spain in 7 provinces during a shorter period, also found no effect of regional mobility restrictions but found that more strict curfews were associated with increased transmission(32). However, the range of NPIs included was limited, making spurious associations a greater threat, as argued in the limitations section. This same study found an effect of limiting gatherings, the intervention that is found associated with decreased transmission more consistently (in our study, social distance decreased transmission between 9% and 13%).
In conclusion, our results estimate that increasing restrictions had a considerable effect in decreasing COVID-19 transmission, with interventions in social distance, bars and restaurants having the higher and more consistent effect. Our results can contribute to the corpus of evidence that will help inform future decisions in response to COVID-19 resurgence or to future pandemics. This must include, not only studies in effectiveness of NPI, but also in their cost-effectiveness and potential negative effects. Continued partnership and collaboration between epidemiologists in the public administration and scientists, particularly mathematicians and data scientists, is crucial to ensure adequate and timely analysis of data that can be used for evidence-based recommendations.