The influence of repeated mild lockdown on mental and physical health during the COVID-19 pandemic: a large-scale longitudinal study in Japan

The mental and physical effects of repeated lockdowns are unknown. We conducted a longitudinal study of the influence of repeated mild lockdowns during two emergency declarations in Japan, in May 2020 and February 2021. The analyses included 7,893 people who participated in all online surveys. During repeated mild lockdowns, mental and physical symptoms decreased overall, while loneliness increased and social networks decreased. Subgroup analyses revealed that depression and suicidal ideation did not decrease only in the younger age group (aged 18-29) and that younger and middle-aged people (aged 18-49), women, people with a history of treatment for mental illness, and people who were socially disadvantaged in terms of income had higher levels of mental and physical symptoms at all survey times. Additionally, comprehensive extraction of the interaction structure between depression, demographic attributes, and psychosocial variables indicated that loneliness and social networks were most closely associated with depression. These results indicate that repeated lockdowns have cumulative negative effects on interpersonal interaction and loneliness and that susceptible populations, such as young people and those with high levels of loneliness, require special consideration during repetitive lockdown situations.

( The onslaught of the coronavirus disease 2019  remains severe as of June 2021 1 . Many countries have implemented lockdowns to prevent the spread of infection. While these lockdowns have been effective in reducing the spread of infection, many negative psychological effects have been reported, including a high prevalence of psychiatric symptoms, such as depression and anxiety [2][3][4][5][6][7] .
Previous studies reporting the negative effects during lockdown have been limited to cross-sectional surveys 3,4,6 and follow-up studies under single lockdown situations 5,7 . There have been no longitudinal data demonstrating the effects of repeated lockdowns. Therefore, the question of whether multiple lockdowns have a negative influence on symptoms, or whether the influence can be reduced as people become accustomed to lockdown circumstances remains 8 . Considering that lockdowns have a significant impact on people's lives and economic activities 9 , it can be assumed that repeated lockdowns have different effects on people's mental health and interpersonal interaction styles than single lockdowns. Since repeated lockdown policies have already been implemented in many parts of the world and may continue to be implemented in the future due to new outbreaks, it is essential to elucidate the influence of repeated lockdowns on mental health.
Japan is one country that has implemented repeated lockdowns. The Japanese government declared a state of emergency on 7 April 2020 and 8 January 2021 and implemented a "mild lockdown" 6 , an unenforceable request for self-restraint ( Figure 1). During the first mild lockdown, a significant increase in psychological distress and depression was reported, despite the lack of legal enforcement of selfrestraint 6,10,11 , the effect of the second mild lockdown on those participants is unclear. 1 -M a r -2 0 1 -A p r -2 0 1 -M a y -2 0 1 -J u n -2 0 1 -J u l -2 0 1 -A u g -2 0 1 -S e p -2 0 1 -O c t -2 0 1 -N o v -2 0 1 -D e c -2 0 1 -J a n -2 1 1 -F e b -2 1 1 -M a r -2 1 1 -A p r -2 1 1st mild lockdown 7 Apr -25 May 2020 2nd mild lockdown 8 Jan -21 Mar 2021 Phasing out of the declaration of a state of emergency Wave 1 of the survey (11-12 May 2020) Wave 2 of the survey (24 Feb -1 Mar 2021) Number of reported cases of  In this lockdown context in Japan, longitudinal studies of mental health characteristics are critical to understanding the influence of repeated lockdowns. Such analyses will provide important fundamental data for improving lockdown approaches that still permit maximum infection prevention and maintain mental health.
The influence of lockdown is also particularly prevalent among certain groups, such as women, the young, the socially disadvantaged in terms of income, and those with a history of mental illness 6,12,13 .
Because the effects of repeated lockdowns may be specific to these groups, a detailed examination of these associations would provide information regarding what groups require particular attention.
In this study, we examined the influence of repeated lockdowns by conducting a large-scale panel survey of the mental health of people in urban areas under two lockdowns due to declared states of emergency in Japan.

Participant characteristics
Of the 11,333 participants in the first wave of the survey, 7,893 participated in the second wave (response rate: 69.6%) ( Table 1). The mean age of those who participated in both surveys was 49.6 years (SD=13.7), with the 30-49 age group the most highly represented (41.8%). Of the respondents, 65.6% were married and 68.2% were employed. Compared to those who participated in the first wave only (N = 3,440), those who participated in both were older and had a larger proportion of men and married people (Supplementary Table S1). In addition, those who participated in both experienced greater loneliness and smaller social networks, while fewer people exceeded the cut-off points for depression (PHQ9 ≥ 10) and psychological stress (K6 ≥ 5) and reported suicidal ideation in the first survey (Supplementary Table S1).

Mental health under repeated mild lockdown in the whole sample
Depression, suicidal ideation, psychological distress, and physical symptoms were significantly lower in the second than the first wave (all ps < 0.001; Table 2). Alternatively, in the second wave, loneliness increased significantly (b = 0.24, 95% CI = 0.15 to 0.33, p < .001) and social network decreased significantly (b = -0.43, 95% CI = -0.54 to -0.33, p < .001). The percentage of patients who exceeded the cut-off score for GAD, measured only in the second wave, was 8.5% (Table 2).

Mental health under repeated mild lockdown with a focus on subgroups
The first-and second-wave data for all subgroups are included in the Supplementary Materials (Supplementary Tables S2-S4).

Estimated rate of depression
There was a significant interaction between gender and wave, and between each age group and wave, for the estimated rate of depression as assessed by the PHQ-9 (Gender: Wald χ2(1) = 7.49, p = .006; Age: χ2(3) = 24.70, p < .001). Both men and women displayed a reduction in depression rate in the second wave (Supplementary Table S2). However, this reduction was significantly smaller among women than men (b = 0.18, 95% CI = 0.52 to 0.32, p < 0.001). Women also had a significantly higher estimated rate of depression in both the first and second waves than men (wave 1: OR = 1.30, 95% CI = 1. 15

Suicidal Ideation
There was a significant interaction between gender and wave, and between each age group and wave, for suicidal ideation (Gender: Wald χ2(1) = 17.87, p < 0.001; Age: χ2(3) = 26.27, p < 0.001). Although suicidal ideation was reduced in the second wave for both men and women (Supplementary Table S2), the degree of reduction in suicidal ideation was significantly smaller among women than men (b = 0.27, 95% CI = 0.14 to 0.39, p < 0.001; Figure 3). Women also reported significantly more suicidal ideation than men in the second wave (OR = 1.29, 95% CI = 1. 15 Table S2). Additionally, suicidal ideation was significantly lower in both waves among those aged >65 than the other age groups (all ps < 0.001).
Stress was significantly higher among women (OR = 1.55, 95% CI = 1.42-1.70, p < 0.001), those with a history of treatment for mental illness (ORs = 2.79 to 6.87, all ps < 0.001), and those with an income of < ¥8 million (ORs = 1.26 to 1.98, all ps < 0.005) compared to men, those without a history of treatment for mental illness, and those with an income of >¥8 million, respectively (Supplementary Tables S2-S4).

Somatic Symptoms
There was a significant interaction in terms of age and income group (Age: Wald χ2 (3) Table S4 Physical symptoms were higher among women (b = 0.87, 95% CI = 0.63 to 1.11, p < 0.001) and those with a history of treatment for mental illness (bs = 3.14 to 6.15, all ps < 0.001) than men and those without a history of treatment for mental illness. (Supplementary Tables S2 and S3).
Those <65 years of age (bs = 2.40 to 4.55, all ps < 0.001), those with a history of treatment for mental illness (bs = 2.06 to 4.11, all ps < 0.001), and those with an income of < ¥8 million (bs = 1.06 to 4.55, all ps < 0.001) had significantly higher loneliness than those over 65 years of age, those without a history of treatment for mental illness, and those with an income of >¥8 million, respectively (Supplementary Tables   S2-S4).
The social network was smaller for those aged 30-64 years (bs = -1.74 to -1.37, all ps < 0.001), those with a history of treatment for mental illness (bs = -1.34 to -2.69, all ps < 0.001), and those with an income of <¥8 million (bs = -1.08 to -5.17, all ps < 0.001) than for those aged >65, those without a history of treatment for mental illness, and those with an income of >¥8 million (Supplementary Tables S2-S4).

Exhaustive interaction of factors associated with depression
The final convergence results of the extracted cluster structures are shown in Figure 4 and Supplementary Table S5.

Discussion
The results of this study indicate that the overall psychological and physical symptoms decreased during the second mild lockdown compared to the first. During the first mild lockdown due to the declaration of the first ever state of emergency in Japan, there were major economic and social disruption impacts 14 , which resulted in a significant increase in depression and stress. In contrast, during the second mild lockdown, these symptoms may have been reduced to some extent as a result of adaptation to various environments and life changes caused by the COVID-19 pandemic.
However, at both time points of the study, social network levels indicated social isolation (LSNS-6 < 12) 15 , which was even more pronounced during the second mild lockdown. Furthermore, loneliness scores increased in the second wave compared to the first wave. These considerations indicate that repeated mild lockdowns cause cumulative changes in interpersonal interactions and that subjective loneliness is maintained and exacerbated.
A striking result when focusing on subgroups was that younger people (18-29 years old) displayed the highest levels of depression and suicidal ideation, and unlike other groups, these indicators did not decrease during the second wave. Additionally, the younger age group demonstrated the smallest decrease in stress and physical symptoms in the second wave of the survey. Women also indicated higher levels of depression, suicidal ideation, and increased loneliness compared to men. These results are consistent with previous reports of deteriorating mental health among young people during the pandemic 5,7,16 , indicating that younger people are more vulnerable to the negative impacts of repeated mild lockdown. Further, considering that the estimated rate of depression among younger people in this study (22.8%-24%) was significantly greater than that of the general population in Japan in the past (7.9%) 17 , the present results may indicate that young people experience difficulties with their current situation (e.g. various obstacles related to school and work, financial difficulties due to inability to work), and concerns about the future (e.g. uncertain career paths and employment status) during lockdown situations. Given these considerations, immediate support for young people is highly desirable during repeated lockdown situations. For example, support is needed from educational institutions, companies, and the government by guaranteeing educational and employment opportunities, providing mental health care, and providing information to assuage their concerns.
In addition, those with a history of treatment for mental illness and those who were socially disadvantaged (especially those with incomes <¥2 million) had higher levels of depression, suicidal ( ideation, stress, physical symptoms, and loneliness, and smaller social networks at both waves. These findings are consistent with previous research 18,19 and indicate that these populations are also more susceptible to negative effects of repeated lockdowns. The COVID-19 pandemic has also been reported to cause various economic crises, resulting in lower wages, cutbacks, and general job losses in various industries 9 , making life increasingly difficult. Therefore, during repeated lockdown situations, a detailed understanding of the difficulties of people with these characteristics is important, and additional social support from social workers and local governments should be considered.
Next, the comprehensive extraction of interactions among variables associated with depression demonstrated that the intensity of loneliness and the size of social networks were most closely related to mental health during both mild lockdown situations. Loneliness and social networks have been considered risk factors associated with mental and physical health 20,21 , and have been identified as important factors affecting mental health during the COVID-19 pandemic 4,22 . Considering these previous studies and our results indicating that repeated mild lockdowns cumulatively reduced social networks and maintained or exacerbated loneliness, we believe that the health status of people who display reduced social networks and high loneliness requires special attention in these situations.

Limitations
In this study, causal factors, such as which approach mitigates or exacerbates the effects of repeated mild lockdowns, have not yet been examined. Therefore, it is necessary to examine the effects of psychosocial variables such as coping strategies and lifestyles in the future.
Additionally, since this study used only an online survey method, the mental health of groups without online access has not yet been examined. Further, since mental health indicators were based on self-reports rather than clinical diagnoses, they may not necessarily correspond to objective assessments by mental health professionals. Therefore, it is necessary to widely use non-online methods in the future to accumulate findings that can be generalised to a larger population.
Finally, those who did not participate in the second wave of the survey had worse mental health compared to those who participated in both waves. Therefore, it should be noted that the present results using only those who participated in both waves may underestimate the impact of repeated lockdowns.

Conclusion
During the two rounds of mild lockdown in Japan, depression, suicidal ideation, stress, and physical ) symptoms decreased overall in the second compared to the first wave, while loneliness increased and social networks decreased. These results suggest that repeated mild lockdown exacerbates social isolation and has a cumulative negative effect on loneliness. In addition, depression and suicidal ideation were highest among younger individuals, and as a striking result, did not decrease in the second wave only in the younger age group. Furthermore, women, those with a history of treatment for mental illness, and those from socially disadvantaged backgrounds displayed poorer mental and physical health in both waves.
These results indicate that these populations are particularly vulnerable to the negative effects of repeated mild lockdowns. Additionally, loneliness and social networks, in particular, were closely related to mental health at both time points. Given the potential for social isolation and loneliness to be exacerbated by repeated mild lockdowns, these results suggest that during a pandemic situation, people who exhibit these characteristics may require special attention. In the future, it will be essential to conduct continuous research on the health status of potentially vulnerable populations using a combination of various measures, to elucidate protective and risk factors, and to develop support systems tailored to individuals' difficulties.

Study design, participants, and data collection.
This longitudinal study was conducted between 11 May and 12 May 2020 (wave 1) and between 24 February and 1 March 2021 (wave 2). Both surveys were conducted just before the emergency declaration was phased out, and each was set up to assess the psychological impact of a mild lockdown of approximately one month on participants (full elaboration in Figure 1). Participants were recruited by The data from the first wave of the survey in this study are part of the dataset used in previous publications 6,11 , and details of this data can be found in these papers.
A link to the online survey was distributed to those who wished to participate in the study, and the survey was conducted online. All participants completed the anonymous survey voluntarily and provided informed consent online. The online survey form was designed so that all items were required to be answered before proceeding to the next step, thus, there were no missing data in this study.

Measures
Depressive symptoms were assessed using the Japanese version of the Patient Health Questionnaire-9 (PHQ-9) 23 ; a score of >10 was considered to indicate a high likelihood of depression 23 and was used as a cut-off point in this study. Suicidal ideation was assessed using item 9 of the PHQ9 and coded as a binary variable with "not at all" as "no suicidal ideation" and other responses as "suicidal ideation" Psychological distress was measured using the Japanese version of the K6 24 . A threshold of 5 points, commonly used for screening mild to moderate mood/anxiety disorders, was adopted as the cut-off point for psychological distress 25 . Somatic symptoms were measured using the Japanese version of the Somatic The Japanese version of the Generalised Anxiety Disorder scale-7 (GAD-7) 28 was used to assess symptoms of generalised anxiety disorder. Scores above 10 are considered to indicate a moderate level of anxiety 29 and were used as a cut-off point in this study. Since the GAD-7 was used only in the second + wave, it was used to describe the characteristics of people in the second wave.
These self-report measures are more extensively described in Supplement A in the Supplementary Materials.
Next, to determine the influence of mild lockdowns on groups that have been previously identified as vulnerable to the effects of lockdown 6,12,13 , demographic attributes such as age, gender, and income were collected, as well as whether or not individuals were currently being treated for mental illness and whether or not they had a history of treatment for a mental illness.

Statistical analysis
Analyses were performed on a dataset of 7,893 participants from both waves of the survey. First, comparisons were made using t-tests and chi-square tests for the data in the first wave for those who participated in both surveys (N = 7,893) and those who only participated in the first wave survey (N = 3,440).
Next, for each outcome variable, a generalised estimating model (GEE) was constructed to test for changes in the variable between waves for the whole sample. This approach is applicable to binary variables and variables other than normally distributed, and has excellent analytical precision, even for longitudinal data with high within-subject correlation 30 . In the GEE approach, binomial logit modelling was used for the binary outcome variables (PHQ-9 cut-off score, suicidal ideation, and K6 cut-off score), and linear Gaussian identity modelling was used for the continuous outcome variables (physical symptoms, loneliness, and social network).
Additionally, GEE models were used to examine differences in changes in outcome variables in subgroups focused on gender (men, women), age (18-29, 30-49, 50-64, >65), income (< ¥2 million (approximately £15000), ¥2-4 million, ¥4-6 million, ¥6-8 million, >¥8 million), and treatment for mental illness (currently in treatment, past treatment, past treatment and still in treatment, no treatment). For analysis of the whole sample, binary outcome variables were analysed using a binomial logit GEE model, and continuous outcome variables were examined using a linear Gaussian identity GEE model. Significant interactions between each subgroup and wave, or main effects for each subgroup, were reported. When analysing the subgroups of income groups, we excluded 1781 participants who answered that they did not know.
To visualise the exhaustive interaction structure of the variables most closely associated with depression, we used nonparametric Bayesian co-clustering 31 for gender, age, income, history of treatment for mental , illness, depression, suicidal ideation, loneliness, and social networks. Fifteen thousand iterations based on the Bayesian optimisation principle were performed to calculate the log-peripheral likelihood, which indicates the fit of the model. Finally, the model with the highest log-perimeter likelihood was adopted.
The significance level for all tests was set at α = 0.05, two-tailed. The analysis was performed using