Impact of the Coming‐of‐Age Day and ceremony on the risk of SARS‐CoV‐2 transmission in Japan: A natural‐experimental study based on national surveillance data

Abstract Background Quantifying the impact on COVID‐19 transmission from a single event has been difficult due to the virus transmission dynamics, such as lag from exposure to reported infection, non‐linearity arising from the person‐to‐person transmission, and the modifying effects of non‐pharmaceutical interventions over time. To address these issues, we aimed to estimate the COVID‐19 transmission risk of social events focusing on the Japanese Coming‐of‐Age Day and Coming‐of‐Age ceremony in which “new adults” practice risky behavior on that particular day. Methods Using national surveillance data in Japan in 2021 and 2022, we conducted difference‐in‐differences regression against COVID‐19 incidences by setting “new adults” cases as the treatment group and the cases 1 year younger or older than these “new adults” as the control group. In addition, we employed a triple differences approach to estimate the risk of holding the Coming‐Age ceremony by using a binary variable regarding the presence or absence of the ceremony in each municipality. Results We estimated the relative risks (RRs) of the Coming‐of‐Age Day as 1.27 (95% confidence interval [CI] 1.02–1.57) in 2021 and 3.22 (95% CI 2.68–3.86) in 2022. The RR of the Coming‐of‐Age ceremony was also large, estimated as 2.83 (1.81–4.43) in 2022. Conclusions When planning large social events, it is important to be aware of the unique risks associated with these gatherings, along with effective public health messages to best communicate these risks.


| INTRODUCTION
As of March 31, 2022, over 6,400,000 confirmed cases and 27,000 associated deaths have been reported in Japan due to coronavirus disease (COVID-19). 1 As a non-pharmaceutical intervention (NPI), many countries have implemented public event bans and group size restrictions for social gatherings. NPIs are reportedly associated with a reduced number of COVID-19 cases and deaths. [2][3][4] However, quantifying the impact from a single intervention or event is difficult, and there is a lack of relevant evidence. One reason for this is the challenge in setting controls as counterfactuals due to COVID-19 transmission dynamics (lag from exposure to reported infection, nonlinearity arising from person-to-person transmission, and the effects of NPIs having different effects over time). 5 To address these issues and estimate the risk of infection following social events, we focused on the Japanese Coming-of-Age Day ("Seijin-No-Hi"), a national holiday celebrated on the second Monday of January. Around this day, the Coming-of-Age ceremony ("Seijin-Shiki") is conducted for the age cohort who have turned or will turn 20 years old between April 2 of the previous year and April 1 of the current year (new adults). Many "new adults" celebrate with their friends at an izakaya (Japanese-style pub) after attending the ceremony. 6,7 This unique setting provides a natural environment to quantify the impact of the social gathering on the transmission risk among this population.
In this study, we aimed to estimate the impact of the Coming-of-Age Day on the risk of COVID-19 transmission in 2021 and 2022 by a difference-in-differences (DID) analysis using the number of infections among the new adults (20-year-old age cohort) before and after the Coming-of-Age Day; the number of infections among persons 1 year younger or older than the new adults (19-and 21-year-old cohort, respectively), whose attributes are similar to those of the new adults, served as the control group. Some municipalities conducted the Coming-of-Age ceremony at a time other than the second Monday of January or postponed the ceremony due to the COVID-19 pandemic.
Therefore, we used a triple-difference approach 8 to quantify the impact of conducting the ceremony on the risk of SARS-CoV-2 transmission.

| Data sources
In Japan, all COVID-19 cases are registered in the national-level Health Center Real-time Information-sharing System on COVID-19 (HER-SYS). 9 We obtained the daily number of newly reported cases by age (date of birth) and symptom onset date in each municipality using HER-SYS. Since our target population included individuals celebrating the Coming-of-Age Day, we stratified age cohorts by the date of birth between April 2 and April 1 (grade cohort in the Japanese education system, where the school year starts in April). We also collected statistics such as population 10 and population density 11 in each municipality as covariates and examined the respective website of all 1916 municipalities to confirm whether the Coming-of-Age ceremony was conducted in 2022. We created a binary variable for the occurrence of the ceremony between January 8 and 10. The status of the ceremonies in 2021 could not be obtained.
2.2 | DID for quantifying the impact of the coming-of-age day We used DID to quantify the impact of the Coming-of-Age Day on the reported number of cases among new adults (20-year-old age cohort) who were set as the treatment group. Individuals 1 year younger or older than the new adults (19-and 21-year-old age cohort, respectively) were set as the control group. Over time, secondary transmission from the treatment to the control group was expected. Therefore, we set the outcome as the cumulative number of cases over 4 days rather than the daily number of new cases to include only cases exposed to the Coming-of-Age Day as the incubation period (duration between exposure and symptom onset) for COVID-19 is approximately 3-5 days. [12][13][14] The quasi-Poisson regression model with several covariates was used to estimate the impact of the Coming-of-Age Day using the following equations: We also conducted the same analysis dichotomizing municipalities on "Seijin-Shiki" status (conducting or not conducting the ceremony).

| Triple-difference analysis for quantifying the impact of the Coming-of-Age ceremony
We further estimated the detailed impact of the Coming-of-Age ceremony by conducting the triple-difference analysis, which incorporates the binary variable of whether or not the ceremony was conducted.
The quasi-Poisson regression model with the following equation was used: where S a,t,i is a binary variable regarding the conducting of the ceremony (1: municipality with the ceremony; 0: municipality without the ceremony). The impact of conducting the ceremony can be estimated with the coefficient γ 7 :

| Sub-analysis and sensitivity analysis
In the sub-analysis, we stratified each municipality into three groups by population size. We conducted DID and triple-difference analyses to determine whether the impacts of the Coming-of-Age Day and ceremony differed by population size.
We conducted the analysis in two different ways for the sensitivity analysis. Since some municipalities conducted the Coming-of-Age ceremony on the day before the Coming-of-Age Day, we set the cumulative incidence of cases 4 days before and after the Coming-of-Age Day as the outcomes. Although the number of cases by date of onset is less affected by weekly bias than the number of cases by date of diagnosis, some weekend influence on the onset data may be present. To further reduce the effect of this bias, we set the seven-day cumulative number of cases as the outcome instead of the four-day cumulative number of cases.

| Test for parallel trends
An essential assumption of DID estimation is that without treatment, the trend in outcomes would be the same for the treatment and con- Age interaction terms, we examined whether the pre-Coming-of-Age Day trends were parallel. We considered P < 0.05 as a potentially significant interaction effect.

| Ethics approval
The Ethics Review Board of the National Institute of Infectious Diseases reviewed this study; no ethical approval was necessary because this study was conducted for public health purposes using national surveillance data.

| RESULTS
We  the Coming-of-Age ceremony, the RR of these municipalities was higher than that of municipalities that did not conduct the event (Table 1). medium-and small-sized municipalities, respectively) ( Table 2). In addition, the triple-difference analysis showed that for all three population groups, the impacts of conducting the Coming-of-Age ceremony were large and significant, with RRs of 1.99-5.22 (Table 3).
Our sensitivity analysis showed similar trends for the aforementioned RRs. However, we observed two different results from the main analysis. First, when we set the day before the Coming-of-Age Day as the treatment day, the estimated RRs were smaller (e.g., RR of the Coming-of-Age Day in all municipalities in 2022 was 2.59 [95% CI 2.14-3.13]). Second, the RR for the seven-day cumulative incidence was greater than that of the four-day cumulative incidence in 2021, whereas the results were the opposite in 2022 (Tables S1-S3).

| DISCUSSION
The present study showed the impact of the Coming-of-Age Day, a national holiday, on the population, specifically targeting only the 20-year-old age cohort in Japan. Our results indicated that the  pandemic was increasing (2022). The results could be attributed to epidemic expansion beginning in densely populated areas of large cities. 16 Notably, compared with the sensitivity analysis results, the RR of the cumulative number of cases in 4 days was greater than that of the cumulative number of cases in 7 days in 2022. In contrast, the converse was true in 2021. One possible explanation may be the differ- Although these populations are not the target of the celebrations, some individuals likely got infected from the 20-year-olds infected due to the Coming-of-Age Day. Moreover, some cases in the control group may have been exposed to risks related to the Coming-of-Age Day (e.g., having dinner with a 20-year-old sibling and serving a T A B L E 2 Relative risk (RR) of the Coming-of-Age Day by municipality size (municipalities categorized into three groups according to population size); for 2022, RRs are also shown stratified by whether or not the Coming-of-Age-ceremony was held 20-year-old in a restaurant). Therefore, we referred to our results as impacts, not "causal effects." Furthermore, to deal with the issue to the best extent possible, we set the outcome in the main analysis as the cumulative number of cases for a limited number of days (4 days).
In addition, cases reported in a municipality that did not conduct the Coming-of-Age ceremony may have participated in the ceremony in their hometown; hence, the RR of the Coming-of-Age ceremony could have been underestimated. Moreover, our study employed the number of cases based on the onset date as the outcome. Although the analysis would have been more valid if the infection dates were known, we used the onset date due to the limitations in the surveillance data. There could be recall bias and social desirability bias when assessed based on disease onset data. 25 Furthermore, there may have been differences in health-seeking behavior between the treatment and control groups. If the treatment group, the new adults, had been more concerned and more likely to see a clinician or had been instructed to see a clinician with any concern, it is possible that the