Introduction
The COVID-19 pandemic and its huge negative impact on health, national economies and social cohesion has created an enormous collective challenge for countries around the globe. As of 20 May 2021, 163.7 million cases and more than 3.3 million deaths have been reported worldwide [
1]. To reduce the spread of the disease in the absence of widespread vaccination, governments have been relying on social distancing non-pharmaceutical interventions (NPIs) such as restrictions on gatherings, school closure requirements, or stay at home requirements, as well as on more conventional NPIs such as testing and contact tracing.
There is evidence that the implementation of social distancing NPIs have helped to control the spread of the COVID-19 epidemic and its consequences [
2‐
4]. However, stringent applications of these policies have adverse consequences. They place a huge burden on the economy [
5], they have equity implications [
6,
7] and they can have strong psychological effects on the population [
8,
9]. In addition, the public’s willingness to adhere to restrictions (which has been high throughout the first wave of the pandemic [
10]) is likely to wane over time, particularly where end dates to these restrictions are uncertain [
11]. Considerations like these outline the fact that the choice of type, mix and intensity of NPIs is, while crucial, a difficult policy decision.
The aim of this study was two-fold. First, to explore the association between the intensity and time delay in the implementation of a wide range of NPIs and the time-varying rate of growth of the COVID-19 epidemic during its initial phase in the 37 member states of the Organisation for Economic Co-operation and Development (OECD). Second, to assess whether any effects found in the initial phase were similar at a later stage of the pandemic. The results of this study can help policy makers make decisions about which NPIs to prioritize in their fight against COVID-19.
Discussion
This cross-country longitudinal study investigated the effect of a comprehensive set of NPIs on the growth of the COVID-19 epidemic during its initial phase in the 37 OECD member states and explored if the results from this initial phase were similar for the period October-December 2020. In the initial phase of the epidemic, (1) restrictions on gatherings, (2) workplace closing requirements, (3) school closing requirements, (4) mask wearing requirements, and (5) the total number of tests performed per thousand population were significant predictors of the average daily growth in cumulative weekly COVID-19 cases, with restrictions on gatherings having the highest effect. For the first four NPIs, higher levels of intensity in the application of the measures tended to be associated with a higher impact on epidemic control, although only marginally for restrictions on gatherings. These results were robust to changes in the model-fitting procedure. The results from the initial phase of the epidemic were not similar in the period October–December 2020. During this period, workplace closing requirements and the testing policy were significant predictors of, respectively, a decrease and an increase in the epidemic growth.
Several data-driven multi-country studies support our findings that restrictions on gatherings were associated with decreases in the COVID-19 epidemic growth in the initial phase of the epidemic [
4,
22‐
25]. Haug et al. [
4] found a higher effect of restrictions on small gatherings compared with mass gatherings. Brauner et al. [
22] found that limitations on gatherings to 1000 people or less were associated with a 23% reduction in the effective reproduction number R
t, limitations on gatherings to 100 people or less with a 34% reduction and limitations on gatherings to 10 people or less with a 42% reduction. Liu et al. [
23] found that restrictions on gatherings of 1000 people or more were not effective while restrictions on gatherings of 10 people or less were. In contrast, we found only a marginal “dose–response” relationship in the impact of this NPI. Work closing requirements were also shown to be effective in a number of multi-country studies [
3,
22,
23,
26,
27] which used data from the first phase of the epidemic. Like us, Brauner et al. [
22] and Hunter et al. (preprint) found evidence of a differential effect when different levels of intensity were implemented for this NPI. There is evidence from several studies that school closure requirements have also been effective in the early stages of the epidemic [
4,
22,
23,
25,
28‐
30]. In their analysis with data from 130 countries and territories, Liu et al. [
23] found that the effect of school closing on reducing R
t was present both when this NPI was implemented at any level of intensity and when it was implemented only at the highest intensity (which is evidence of an effect at different levels of intensity). This is consistent with our finding of a “dose–response” relationship in the effect of school closing requirements. Our findings that mask wearing requirements were a significant predictor of the reduction in epidemic growth in the early stages of the epidemic are supported by other data-driven studies, such as Bo et al. [
31] and Chernozhukov [
32]. Leffler et al. [
15] found that in countries with cultural norms or policies supporting mask wearing, per capita mortality increased less than in other countries. However, there is some debate in the literature regarding the impact of the use of face masks by the public. In a recent meta-analysis of RCTs exploring the use of masks to prevent the transmission of respiratory infections in community settings, Gomez-Ochoa et al. [
33] found no effect. More recently, in a multi-disciplinary review of the literature, Howard et al. [
34] concluded that the preponderance of the evidence supports the widespread use of face masks by the public. Further research is required to understand the effect of wearing face masks on disease transmission in community settings, including research on the impact of different types of masks and on the impact of adherence to face mask use. With regards to our findings that the number of tests per thousand population were associated with the decrease in the rate of growth of the epidemic in its initial stage, Chaudry et al. [
35] found an opposite effect: in their study, testing volume was a significant predictor of the increase in the number of COVID-19 cases per million population in 50 countries. Koh et al. [
36] and Islam et al. [
24] found that early implementation of lock-down
type NPIs have been effective at containing the epidemic. In contrast, we did not find that the delay in the initial NPI response to the epidemic was a predictor of the flattening of the epidemic growth in OECD countries.
Given that most OECD countries implemented strong social distancing measures during the initial phase of the COVID-19 epidemic, it is surprising that stay at home requirements and restrictions on internal movement did not have an impact on epidemic growth during that time. One possible explanation is that, during the 11 weeks studied, these two policies were implemented by OECD countries as recommendations (rather than actual restrictions or bans) much more frequently than was the case for other NPIs such as restrictions on gatherings, work closing and school closing requirements. Similarly, mask wearing requirements had a relatively small absolute effect on the wADGR. This may be partially explained by the fact that the intensity of this NPI increased relatively slowly across the OECD member states over the 11 weeks compared to other NPIs. Two NPIs standard to many health systems that did not impact on the epidemic growth rate in the initial phase of the epidemic were the contact tracing policy and the testing policy. In line with these results, Liu et al. [
23] found inconclusive evidence of the effectiveness of these two policies. The lack of effect of contact tracing in the early phase of the epidemic may be explained at least in part by the fact that for the most part of the 11 weeks studied, most OECD countries did not trace the contacts of all confirmed cases. Similarly, a possible reason for the lack of effect of testing policies was that comprehensive testing strategies (i.e. testing of anyone with COVID-19 symptoms or open public testing) were not widespread in the OECD member states until the last few weeks. Our model did, however, identify the volume of testing per unit of population (a proxy for the testing policy) as a significant predictor of epidemic growth.
Our study detected significant variability across countries regarding the impact of the NPI response on the epidemic growth in the early stage of the epidemic, as evidenced by the percentage of the total outcome variance (34.9%) that is unexplained by the fixed effects in the mLMM. Besides differences in the implementation time delay, type and intensity of the measures applied in each country, other factors may have been at play to explain this variability. For example, differences in adherence to the NPIs, variations in the quality of epidemiologic data, or unobserved sociodemographic and health system effects across countries. These and other factors can be investigated further with data from subsequent waves of the epidemic.
In the period October-December 2020, the results were not similar to those of the initial phase. Two NPIs were found to have an effect on the wADGR: work closing requirements (with a very small negative effect) and testing policy (with a positive effect). This effect of the testing policy does not have a straightforward interpretation. On the one hand, higher volume of testing will lead to an increase in the number of COVID-19 cases diagnosed. On the other hand, cases diagnosed will typically be isolated and their contacts traced, tested and quarantined if testing positive, which should lead in the mid-term to fewer cases overall. Perhaps the time lag to that mid-term effect is long enough not to be picked up by the model.
The evidence from data-driven multi-country studies analysing the impact of workplace closing requirements on epidemic growth with data from the latter part of 2020 is not consistent. Sharma et al. (preprint), using data between August 2020 and January 2021, found that business closures were more effective than other NPIs at reducing R
t. They found that the effect was similar (about a 12% reduction on R
t) for closure of restaurants, night clubs, retail and close contact services (such as hairdressers) and lower for entertainment venues. Wibbens et al. [
28] using data between March and November 2020 (a period which also included the initial phase of the pandemic) found that workplace closing had the strongest impact on the epidemic growth rate out of all NPIs evaluated. In contrast, Ge et al. (preprint) found that in the second wave of the epidemic workplace closures were not effective at mitigating COVID-19 transmission. Out of the three studies above including data from late 2020, only Wibbens et al. [
28] analysed the impact of the testing policy and found that its impact on the growth rate, albeit negative, was lower than that of most other policies. We found a positive effect of the testing policy on the epidemic growth rate, possibly due to the increase in cases confirmed through testing.
In the initial phase of the epidemic, shifting from the lowest to the highest levels of intensity of the NPIs had the following effects: shifting from no restrictions on gatherings to restrictions of gatherings of 10 people or less was associated with an average reduction of the wADGR of 2.81%, 57% higher than shifting from no workplace closing requirements to requiring closing of all but essential workplaces, 70% higher than closing all school levels, and about three times as much as requiring country-wide mask wearing in all public places or in all public places where social distancing is not possible. Shifting from no mask requirements to requiring country-wide mask wearing in all public places was associated with an average reduction in the wADGR of 0.96%, twice as much as shifting from no mask requirements to recommending mask wearing. These results should be interpreted with care. This is an ecological study, and as such causality in the associations between the NPIs and epidemic growth cannot be unequivocally inferred. Although the results from the initial phase of the pandemic were not similar for the period between October and December 2020, this does not limit their validity. The dissimilarity in results may be partially explained by the differences in the impact of NPIs between initial and further stages of the epidemic. For example, Sharma et al. (preprint), comparing the first and second waves for seven European countries, found that the effect of NPIs on the reproductive number Rt was considerably smaller in the second wave compared with the first. The authors argued that a number of factors played a role in lowering the effect of NPIs, including persistent behavioral change in the population (e.g. avoiding close contact) and the generalized adoption of safety measures (e.g. distancing rules). They added that the effects of NPIs on epidemic control in the early stages of the epidemic were measured relative to the population behavior and safety protocols which were prevalent before the epidemic started and hence may not adequately inform policy at later stages once these factors have changed.
Our study has a number of limitations. The analysis reflects national level outcomes and policies based on available data. It does not explore the differential impact of NPIs implemented regionally or locally within countries. The characterization of the intensity of the NPIs using a limited number of ordinal levels (as is done in the OxCCGRT and as we have done with the mask wearing requirements) might mask smaller variations in effect. To detect such variations, databases would have to use interval scales to distinguish NPI intensity. In addition, the compilation and inclusion of data on NPI enforcement and adherence could strengthen the results.
Our study adds to the existing literature by exploring, using a data-driven, longitudinal approach, the impact of NPIs on the COVID-19 epidemic growth in the OECD member states during the initial stage of the epidemic, as well as exploring whether any effects found in the initial phase were similar at a later stage of the pandemic. An important methodological feature of our study is the use of mixed effects longitudinal models to explore the impact of NPIs on the average daily growth rate in weekly confirmed cases. These models use repeated measures to simultaneously explain changes in the growth rate within and across countries. The model presented here was robust, as similar results were obtained using different estimation procedures (maximum likelihood and Bayesian estimation).
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.