International changes in respiratory syncytial virus (RSV) epidemiology during the COVID‐19 pandemic: Association with school closures

Abstract Background Little RSV activity was observed during the first expected RSV season since the COVID‐19 pandemic. Multiple countries later experienced out‐of‐season RSV resurgences, yet their association with non‐pharmaceutical interventions (NPIs) is unclear. This study aimed to describe the changes in RSV epidemiology during the COVID‐19 pandemic and to estimate the association between individual NPIs and the RSV resurgences. Methods RSV activity from Week (W)12‐2020 to W44‐2021 was compared with three pre‐pandemic seasons using RSV surveillance data from Brazil, Canada, Chile, France, Israel, Japan, South Africa, South Korea, Taiwan, the Netherlands and the United States. Changes in nine NPIs within 10 weeks before RSV resurgences were described. Associations between NPIs and RSV activity were assessed with linear mixed models. Adherence to NPIs was not taken into account. Results Average delay of the first RSV season during the COVID‐19 pandemic was 39 weeks (range: 13–88 weeks). Although the delay was <40 weeks in six countries, a missed RSV season was observed in Brazil, Chile, Japan, Canada and South Korea. School closures, workplace closures, and stay‐at‐home requirements were most commonly downgraded before an RSV resurgence. Reopening schools and lifting stay‐at‐home requirements were associated with increases of 1.31% (p = 0.04) and 2.27% (p = 0.06) in the deviation from expected RSV activity. Conclusion The first RSV season during the COVID‐19 pandemic was delayed in the 11 countries included. Reopening of schools was consistently associated with increased RSV activity. As NPIs were often changed concomitantly, the association between RSV activity and school closures may be partly attributed to other NPIs.


| INTRODUCTION
Changes in the seasonality of respiratory viruses including respiratory syncytial virus (RSV) have been a collateral effect of the COVID-19 pandemic. RSV is highly seasonal and circulates during winter in temperate climates and rainy season in tropical climates. 1 RSV is a major cause of lower respiratory tract infections in young children and the elderly. 2,3 In 2015, 3.2 million hospital admissions and 59 000 in-hospital deaths in children younger than 5 years were attributed to RSV worldwide. 2 Since the WHO declared COVID-19 a pandemic in March 2020 (Week (W)11), 4 unusual RSV circulation patterns have been reported.
In both hemispheres, little to no RSV activity was observed during the first expected RSV season, with reductions in RSV detections or acute bronchiolitis hospitalizations up to 90% compared with previous years. [5][6][7][8] Multiple countries subsequently experienced out-of-season increases in RSV cases. [9][10][11][12] However, the characteristics of RSV resurgences varied greatly. Changes in testing practices, evolution of surveillance systems, or viral interference between RSV and SARS-COV-2 may have contributed to these observations. 13 Also, nonpharmaceutical interventions (NPIs) that were implemented to slow the spread of COVID-19 likely played a role. 14,15 Although several countries have reported their own conclusions regarding the effect of the COVID-19 pandemic on RSV circulation, few assessed the impact of NPI policies and their timing on RSV activity. 9,10 In addition, most studies considered lockdown periods, rather than individual NPIs. It is therefore still unclear which NPIs may have triggered RSV resurgences. Understanding the drivers of RSV resurgences may improve forecasting of future epidemics and potentially reveal NPIs that could be implemented annually to reduce RSV circulation.
The two objectives of this study were to describe the international changes in RSV epidemiology during the COVID-19 pandemic and to improve the understanding of the association between NPIs and RSV resurgences. We hypothesized that including different countries would allow to identify the role of individual measures. First, we described the changes observed in the circulation of RSV using RSV surveillance data from 2017 to 2021 in 11 countries. Then, we identified the NPIs that were downgraded in the 10 weeks before the start of the RSV resurgence in each country. Finally, we investigated the associations between NPIs and the difference between observed and expected RSV activity.

| Country selection
Countries with publicly available RSV surveillance data were selected between November 2020 and February 2021 ( Figure 1) Table S1). RSV surveillance data were obtained from community clinic networks in South Africa and Israel and from laboratory networks in other countries. One reference laboratory (in Sao Paulo) of the three in Brazil was included. In South Africa, the surveillance had sites in four of nine regions. For the other countries, the entire territory was represented. Data were reported year-round except in France and Israel that published reports only during winter (W40 to W14). In Israel, reports were published year-round in 2020.
In 2021, another Israeli surveillance system was used between W18 and W40. In France, no data were retrieved from W20-2020 to W29-2020 and from W25-2021 to W29-2021. We assumed that no RSV was detected in the other weeks. For each week, we extracted the number of RSV detections, number of RSV tests done, and the RSV positivity rate (number of detections divided by the number of tests). In the Netherlands, Japan and Taiwan, the total number of tests was not available and was approximated by the sum of the number of detections for all respiratory viruses.
The OxCGRT reports publicly available information on NPIs collected from news articles, government press releases, and briefings. We selected nine NPIs in the OxCGRT: the eight containment and closures policies available, plus facial coverings policies. We did not include COVID-19 specific or economic measures that were outside the scope of this study. The nine NPIs were school closures, workplace closures, facial covering policies, gathering restrictions, cancellation of public events, stay-at-home restrictions, public transport closures, restriction on internal movement in the country (e.g., within regions or states), and international travel controls (Supplementary Table S2). The NPIs stringency levels (0 to 2, 3, or 4, with 0 indicating no measures) were grouped when necessary to avoid empty categories. From the OxCGRT, we also extracted the stringency index, which is a composite score of the eight containment and closure policies and the public information campaign variable (scale ranging from 0 to 100). The OxCGRT reports information daily. For this study, the highest stringency level of the week for each country was used.

| Outcomes
To describe the changes in RSV epidemiology, we reported the delay of the onset and the peak of the first RSV epidemic observed in each country during the COVID-19 pandemic, relative to the expected onset and peak of the RSV season. We qualitatively assessed the meteorological season in which onset of that first RSV epidemic was expected and observed. We described changes in the shape of the RSV peaks during the COVID-19 pandemic relative to previous RSV seasons. Then, we examined which NPIs were downgraded before the RSV resurgences. To identify NPIs that could explain the deviation in RSV cases from pre-pandemic RSV epidemics, we assessed associations between the weekly difference between observed and expected RSV activity in the pandemic period W18--2020 to W44-2021 and the concurrent stringency index and level of NPIs.

| Statistical analysis
We determined the onset and offset week of RSV seasons for each country using the average annual percentage (AAP) method. 1 The delay in onset was calculated as the total number of weeks between the expected and the observed onset of the first RSV epidemic during the COVID-19 pandemic. Delays in onset <40 weeks (i.e., 9 months) were classified as "delayed season," whereas longer delays were classified as "missed season." To describe the NPIs that were downgraded before the RSV resurgence, the start of the RSV increase was defined as the first of the consecutive weeks with >0.5% of RSV detections that included the first RSV epidemic during the COVID-19 pandemic, with 2-week gaps accepted. The RSV detection rate of 0.5% was selected based on the observation that RSV detection rates were lower while RSV was not circulating. We arbitrarily selected a period of 10 weeks before the start of the increase in RSV to allow enough time for detection by the selected surveillance system despite differences in surveillance population coverage, country size, and population density. We reported the number of weeks between downgrading the NPIs and the peak of the RSV epidemic.
Association between NPIs and deviations in RSV activity compared with pre-pandemic seasons were tested using linear mixed models (LMM). As the dependent variable, we considered the weekly deviation in RSV detection rate, defined as the difference between the observed weekly RSV detection rates and the average prepandemic detection rates for the same week number. Random effects for hemisphere and country were used with countries nested within hemisphere. The stringency index or the individual NPIs were included as fixed effects. NPIs were entered as continuous or categorical according to best fit, defined as the model with smallest Akaike information criterion (AIC). First, each NPI was included in a separate model (referred to as "single-NPI models") with hemisphere and country included as random effect. All NPIs that showed an association with p < 0.5 were included in the full model from which the final model was selected with a backward elimination procedure. At each step, we removed the single NPI whose omission yielded the best fit model (smallest AIC) among all candidate models with a single NPI removed. This was repeated until the model contained only NPIs that were associated with average deviance in RSV activity (p < 0.1) and whose contribution to the model was significant (ANOVA for nested models, p < 0.1) or whose removal worsened model fit when assessed using AIC. As some levels of NPIs were only rarely observed, we performed a sensitivity analysis where NPIs were dichotomized. A cutoff value was chosen independently for each country and based on the frequency distribution of weeks over the categories and univariate analyses where level of the NPI was entered as a categorical variable (Supplementary Table S3).
LMM were fitted in RStudio Version 1.3 with the maximum likelihood method and included a first-order autoregressive error ("lme" function from R package "nlme").

| Ethics
As this study used only publicly accessible surveillance data, it did not require ethics approval.

| Pre-pandemic RSV seasonality
During the pre-pandemic years, one peak of RSV was observed annually, except in Taiwan where two peaks occurred in 2017 and 2018.
As expected, the RSV season onset and offset weeks varied between countries but stayed consistent over the years in most countries (Supplementary Table S3

| Pandemic impact on onset of RSV season
COVID-19 was declared a pandemic in March 2020 (W11-2020). The 2019-2020 RSV season ended in the eight preceding weeks in Canada, France, Israel, the Netherlands, South Korea, and the United States (Table 1). However, only in Israel the number of detections had returned to baseline by W11-2020 ( Figure 2). In the southern hemisphere, the number of RSV detections was increasing in South Africa and Brazil. The season was expected to start later in Chile. Japan and Taiwan were also between RSV seasons.
During the pandemic, a delay in onset of the first RSV epidemic during the COVID-19 pandemic was observed in all 11 countries (

| Impact on shape of RSV epidemic
Whereas in South Africa pre-pandemic RSV seasons showed one peak, two consecutive periods of high RSV activity were observed in 2020-2021 (Figure 2). After a first peak when RSV activity was  On the contrary, pre-pandemic RSV seasons in Taiwan could have two annual peaks, but only one was observed during the COVID-19 pandemic (Figure 2). The RSV positivity rate at the peak was higher than in pre-pandemic seasons. However, the positivity rate may have been overestimated as the total number of RSV tests was approximated using respiratory viruses' detections.

| NPIs before RSV resurgence
In the 10 weeks preceding the RSV increase, the most commonly downgraded NPIs were school closures, workplace closures, and stayat-home requirements, respectively, in six, four, and four countries (

| Association between NPIs and RSV resurgences
We hypothesized that lower NPIs levels were associated with higher difference between observed and expected RSV activity. As expected, the stringency index was negatively and significantly associated with the difference in RSV activity (Table 3)

| DISCUSSION
This study aimed to describe and compare changes in RSV seasonality observed during the COVID-19 pandemic in 11 countries and to explore their association with NPI policies using publicly accessible surveillance data. In all countries included in the analysis, the first RSV epidemic during the COVID pandemic was delayed (13-88 weeks, mean: 39 weeks). Whereas one RSV season was missed in Brazil, Chile, Japan, Canada, and South Korea (delay ≥40 weeks), the onset of the first RSV season during the COVID pandemic was <40 weeks late in the other six countries. RSV activity was consistently associated with school closures in the descriptive analysis and the models.
School closures were the NPI most often downgraded before the RSV resurgences and lower levels of school closures were systematically associated with increased RSV activity in univariate and multivariate analyses. This association was expected as children are recognized as the main transmission group for RSV. 17 However, school closures were not downgraded in all countries before the RSV resurgence. Accordingly, whereas some previous publications also underlined the role of school closures, others found no temporal link with RSV resurgences. 9,11,18,19 Lower levels of school closures are probably not sufficient to trigger RSV resurgences.
Other NPIs may have contributed to RSV resurgences. Similarly to the effect of school closures, lower levels of stay-at-home requirements would likely increase between-households social interactions in all age groups, and its association with RSV activity was expected.
Surprisingly, public transport closure was found to be associated with RSV activity when dichotomized. Although not often implemented, the periods when public transport were closed corresponded with high levels of NPIs stringency in general. Conversely, interstate travels T A B L E 2 Non-pharmaceutical interventions that were downgraded in the 10 weeks preceding the start of the RSV resurgence (for week >0.5% RSV detections) and number of weeks between changes in NPIs and the peak of the first RSV epidemic during the COVID-19 pandemic has been suggested as potential facilitating factor in Australia. 7,9 Although lower levels of internal travel were associated in RSV activity in the single NPI model, it was not included in the final model.
Overall, our analysis point toward a strong effect of school closures but probably did not sufficiently disentangled the effects of other NPIs.
Other factors than NPIs could contribute to the differences observed between countries. Meteorological factors have been found to be associated with RSV seasonality. 20 For RSV, the effect of population immunity seems more important. During the COVID-19 pandemic, RSV circulation was limited, and the susceptible fraction of the population increased through births and waning immunity, as RSV infections do not provide long-lasting protection. 21,22 Early models predicted the out-of-season resurgences that have been observed due to build-up in susceptible individuals. 23 Viral interference with other respiratory viruses may also play a role. In the few countries that reported both viruses' activity since the start of the COVID-19 pandemic, rhinovirus peaked before the RSV resurgence. [24][25][26][27] Negative interactions between RSV and rhinovirus have been previously reported 28,29 and may have contributed to delaying RSV resurgences after downgrading NPIs.
In addition to their impact on the circulation of respiratory viruses, negative socioeconomic impact related to the large-scale implementation of NPIs have been observed. 30 Consequently, implementing mandatory social distancing each winter to prevent the circulation of respiratory viruses is not realistic. However, models predicting future RSV seasons assumed an overall effect of NPIs on RSV transmission, whereas NPIs target different social groups. 23 According to these models, the magnitude of the changes in RSV seasonality is expected to increase with the duration of NPIs implementation. 23 As NPIs are still largely implemented, it is important to better understand their associations with RSV activity in order to accurately predict coming RSV seasons.
The main strengths of this study are its international perspective and the analysis of NPIs individually. This study also has some important limitations. First, we extracted RSV data from various surveillance systems. To limit the impact of changes in the number of tests done, we defined RSV seasons annually and used weekly proportion of detections. All pre-pandemic RSV seasons fell within the expected period of high RSV activity for the country. The AAP method may result in shorter epidemic periods than other methods, which could have resulted in an overestimation of the delay in onset T A B L E 3 Association between non-pharmaceutical interventions stringency levels and changes in RSV activity during the pandemic, defined as the difference between expected and observed proportions of weekly RSV detections, in linear mixed models, including hemisphere and country as random effects There were regular contacts between the sponsor and the study team to discuss the country selection and the data collection. The sponsor had no role in the data analysis or the preparation of the manuscript.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1111/irv.12998.

DATA AVAILABILITY STATEMENT
The data supporting the findings of this study were derived from publicly accessible resources. The list and links to these resources are available in Supplementary Table S1.