Changes in mental well-being during the COVID-19 pandemic: A longitudinal study among Danes from 2019 to 2021

Numerous studies have shown that the COVID-19 pandemic had large influence on mental health during the first lockdown, but fewer studies have focused on the long-term influence on mental health. In a national longitudinal study, we examined mental well-being measured just before (fall 2019) and twice during (falls of 2020 and 2021) the COVID-19 pandemic. We utilized the Danish Health and Wellbeing Survey with questionnaires collected in 2019, 2020 and 2021 among the same study population consisting of 8179 persons. The outcome was mental well-being measured by the Short Warwick-Edinburgh Mental Well-Being Scale (SWEMWBS). Linear regression models were conducted to evaluate change in SWEMWBS from 2019 to 2021. The SWEMWBS distribution was similar in 2019, 2020 and 2021, although the distribution moved to lower scores in 2020 compared to 2019 and moved slightly to higher scores in 2021 compared to 2020. Mean SWEMWBS decreased from 24.8 (95%CI 24.7–25.0) in 2019 to 24.1 (24.0–24.2) in 2020 and increased to 24.4 (24.3–24.6) in 2021 (p < 0.001). The mean decrease from 2019 to 2020 and increase from 2020 to 2021 was strongest among women, persons below age 75 years, persons without depression and among persons with higher education and with employment. In conclusion, we find that mental well-being decreased from 2019 to 2020 and slightly increased from 2020 to 2021 without reaching the pre-pandemic level. These changes are statistically significant but small and support that COVID-19 may only have had a small long-term influence on mental health in the general population.


Introduction
The COVID-19 pandemic had pronounced influence on everyday life and the widespread use of quarantine, working and schooling from home, and the consequently social distancing have been reported to influence mental health (Ammar et al., 2021;Both et al., 2021). Longitudinal studies have reported that anxiety, depression, COVID-19 related stress and low mental health increased markedly during the first wave of COVID-19 (spring 2020) followed by a decrease in summer 2020 (Chandola et al., 2022;Daly and Robinson, 2022;Shevlin et al., 2021). However, mental health problems were still increased during the second wave of autumn and winter 2020 Robinson, 2021, 2022;Joensen et al., 2022) and into 2021 (Ellwardt and Prag, 2021;Vistisen et al., 2022). In general, the increase in mental health problems is observed most strongly for selected subgroups including youth (Daly and Robinson, 2022;Ellwardt and Prag, 2021) and persons aged 35-44 years (Vistisen et al., 2022), females (Daly and Robinson, 2022;Ellwardt and Prag, 2021;Joensen et al., 2022), people living alone (Ellwardt and Prag, 2021) or in denser households without direct access to outdoor spaces , high-income and education groups (Daly and Robinson, 2022), and those experiencing financial stressors or without work (Chandola et al., 2022;Ellwardt and Prag, 2021). In a meta-analysis of longitudinal studies, COVID-19 lockdowns had small, but statistically significant, effects on mental health symptoms, depression, and anxiety, while other mental health outcomes did not change. The results indicate substantial heterogeneity among studies, supporting that COVID-19 lockdowns may not have uniformly detrimental effects on mental health (Prati and Mancini, 2021). In another meta-analysis of 11 longitudinal studies from United Kingdom, mental health decreased from the pre-pandemic scores across three time points in the spring, autumn and winter of 2020 (Patel et al., 2022). The decrease was largest for females, higher educated and younger adults aged 25-34 years.
Danish studies have reported negative developments on psychological well-being during spring 2020 (Sønderskov et al., 2020a,b;Sønderskov et al., 2020a,b), but afterwards positive developments in autumn 2020 (Sonderskov et al., 2021;Joensen et al., 2022;Vistisen et al., 2022). One study reported that crisis-specific worries were stable during the lockdown in spring 2020 (Clotworthy et al., 2020). In a longitudinal study with information on mental health collected both before the COVID-19 lockdown (2018-February 2020) and again in April-May 2020, Petersen and colleagues report only minor increases in emotional distress during the COVID-19 lockdown compared to before the pandemic (Petersen et al., 2021). In another longitudinal study with repeated information, we have previously reported that mental well-being decreased from 2019 to 2020 suggesting a rather prolonged effect on mental well-being .
In this study, we extend the previous study  by examining changes in mental well-being in the period from just before the COVID-19 pandemic (autumn 2019) and twice during the pandemic (autumn 2020 and autumn 2021). Furthermore, we assess whether population subgroups experience different changes.

Material and methods
The Danish Health and Wellbeing Survey is the Danish part of the European Health Interview Survey (EHIS). In this study, we included the wave from autumn 2019, which we re-invited both in autumn 2020 and autumn 2021 (The Danish Health and Wellbeing Survey) (Rosendahl Jensen et al., 2021).
EHIS covers topics as health status (self-rated health, chronic conditions, mental health, etc.), health behavior (smoking, alcohol consumption, etc.), healthcare utilization (use of healthcare services), and social and demographic characteristics.
Danish residents have a personal identification number which is used throughout administrative registers, including the Civil Registration System (Pedersen, 2011). From the Civil Registration System, 14,000 persons aged 15 years or more were randomly selected and invited to the survey in autumn 2019. Data were collected via a self-administered questionnaire and 6629 persons (47%) completed the questionnaire. We re-invited all from the 2019 survey, who were still alive and living in Denmark in mid-August 2020, to participate in the follow-up survey (n = 13,554). We excluded persons with specific reasons for non-response in 2019 (e.g., 'hard' refusers, persons with severe cognitive impairment, and persons not understanding Danish) (n = 80). Thus, 13,474 persons aged 16 years or older were re-invited in 2020. In all, 6712 persons (50%) completed the self-administered questionnaire in 2020. The same procedure was conducted in autumn 2021 with an additional sample of 388 young persons aged 15-16 years to reestablish the age-distribution from 2019 (n = 13,584). We excluded persons with specific reasons for non-response in 2020 (n = 14). Thus, 13,570 persons aged 15 years and older were invited in 2021 whereof 5384 persons (40%) completed the questionnaire.
Data in 2019 were collected September-December 2019 and data in 2020 and 2021 were collected September-November. Denmark had a lockdown starting in March 2020 with restricted travel and dictated closing of schools, shops, working places, and cultural institutions with an opening starting in April 2020 in some areas gradually increasing in May and June (Statens Serum Institut, 2023;Hellmann et al., 2023). In June to August of 2020, only minor restrictions were still in place, but a general request to work from home was not lifted. By August 2020, the number of COVID-19 cases increased, and some restrictions were reintroduced. During September and October 2020, restrictions were reintroduced both regarding travel restrictions and the number of persons that people could engage with. In the period where the 2021 survey was conducted, COVID-19 was no longer considered a critical societal threat mainly due to a high proportion of vaccination of the Danish population and therefore most restrictions were removed.

Outcome measure
We measured mental well-being, the outcome of this study, using the short version of the Warwick-Edinburg Mental Wellbeing Scale (SWEMWBS) in 2019, 2020 and 2021. We used the metric scores of SWEMWBS, where higher scores of SWEMWBS reflect higher well-being (range 7-35). We included SWEMWBS as a continuous variable.

Explanatory variables
We included several explanatory variables to describe the development in SWEMWBS from 2019 to 2021. The questions reflected dimensions of demography (sex and age), mental and somatic diseases, and socio-economic factors. Information on all explanatory variables were obtained during follow-up (2019, 2020 and 2021).
Depression was assessed by the eight-item Patient Health Questionnaire depression scale (PHQ-8), which is an established valid diagnostic depression measure also useful in general population surveys (Kroenke et al., 2009). We included PHQ-8 as a binary variable (a total score of ≥10) representing a clinically relevant depression.
A standard question on long-standing illness or health problem was also included in the questionnaire: 'Do you have any long-standing illness or long-standing health problem (by long-standing we mean illnesses or health problems which have lasted, or are expected to last, for 6 months or more?'). The response categories were 'yes' and 'no'.
Educational level was divided into three categories (elementary school, upper secondary or vocational education, and higher education). Information on educational level was based on data from the Education Register (Jensen and Rasmussen, 2011) where we included highest attained education. Persons with missing information (1.0%) were categorized as elementary school.
Occupational status was included in five categories (employed, unemployed, transfer payment (student compensation or sick leave), pensioner or other). Information was based on the Employment Classification Module (Petersson et al., 2011) of the most important place of employment. We used information for the calendar year before each survey.

Weighting
All analyses were weighted to account for non-response in each survey wave. When calculating weights for each survey year, we estimated the probability of participating among those invited in each survey by age in five-year categories and sex. The weights were calibrated to sum to the number of observations included and had a range of 0.57-2.90 with an interquartile range of 0.71-1.25. Younger persons and males had higher weights and middle-aged (55-74 years) had lower weights. All analyses were weighted meaning that the results reflect a population with the same age-and sex-distribution as the invited population.

Statistical analysis
Descriptive statistics included number, proportions, means, 95% confidence intervals, and chi-square tests (for categorical data). To examine developments in SWEMWBS from 2019 to 2021, a general linear mixed model with repeated measurements was used. The main interest was whether mean SWEMWBS changed in the period from 2019 to 2021 which was tested by an overall type-3 test supplemented with two pairwise comparisons (change from 2019 to 2020 and change from 2020 to 2021). Furthermore, to evaluate whether the change was different between subgroups, an interaction between time and subgroup was tested using an overall type-3 test. All analyses were adjusted for sex and age (included as a continuous variable, and as second-and thirddegree polynomials). To account for correlation due to repeated measurements within the same person, we used generalized estimation equations (GEE) models with a Gaussian distribution and an unstructured correlation matrix. As a sensitivity analysis, we included respondents in all three years to evaluate the influence of explanatory variables only measured in 2019. Analyses were carried out using SAS version 9.4.

Ethical aspects
Participation in the surveys was voluntary which was informed to the invitees. In Denmark, register and questionnaire studies do not require approval by committees on biomedical research ethics according to Danish legislation. The surveys were approved by SDU Research & Innovation Organization (RIO), which evaluates and approves all scientific projects at University of Southern Denmark according to the Data Protection Regulation.

Results
The number of participants was similar in 2019 and 2020 but decreased in 2021 (Table 1). The number of persons included in the analysis with valid SWEMWBS responses was 8179. The number of persons participating one time was 2,772, the number with two responses was 2324 and 3083 had three responses. Thus, the total number of observations in the repeated regression model was 16,669. More women participated in all years (Table 1). Depression increased slightly (p = 0.002), long-standing illness decreased slightly (p < 0.001) and proportion of pensioners increased (p < 0.001) in the period, while educational level (p = 0.04) was stable.
The distribution of mental well-being moved to lower well-being in 2020 compared to 2019 and moved to higher well-being in 2021 although lower level than 2019 (Fig. 1).
The mean SWEMWBS score decreased statistically significantly from 24.8 in 2019 to 24.1 in 2020 and increased to 24.4 in 2021 (overall type-3 test p < 0.001) ( Table 2). Pairwise tests showed statistically significant decrease from 2019 to 2020 and increase from 2020 to 2021 (both tests: p < 0.001).
When evaluating subgroup specific changes, females experienced a slightly larger decrease from 2019 to 2020 and larger increase from 2020 to 2021 (p = 0.048). In the analyses by age groups there was statistically significant interaction between time and age groups (p = 0.005) reflecting decrease from 2019 to 2020 and increase from 2020 to 2021 among persons below age 75 years and increase from 2019 to 2020 and stable development from 2020 to 2021 among persons aged 75+ years.
When stratifying by depression, a decrease from 2019 to 2020 was only observed among persons without depression. Among persons with depression, a stable or slight increase in SWEMWBS was observed from 2019 to 2021. The test for interaction between time and depression was statistically significant (p < 0.001). There was no difference when stratifying by long-standing illness (p = 0.31).
In the analyses of educational level, the decrease in SWEMWBS from  (2) 86 (1) 72 (1) (1) Chi-square test was performed to assess the relationship between the baseline characteristics and time. Persons with missing values on specific characteristic were not included. L.C. Thygesen et al. 2019 to 2020 and increase from 2020 to 2021 were larger for persons with highest educational level compared to lower educational level (p = 0.002). In terms of occupational status, the decrease in SWEMWBS from 2019 to 2020 was larger for employed and self-employed persons (p = 0.002). When restricting the analyses to persons who responded to all three surveys in 2019-2021, the patterns were mostly similar although the mean SWEMWBS were in general slightly higher (Supplementary Table 1). Although the patterns were similar, the test for interaction between time and sex was not statistically significant (p = 0.20). The increase in mean SWEMWBS was larger among persons with depression in this analysis compared to the analysis presented in Table 2. The difference between educational groups was not as large in this restricted sample and the interaction test between time and educational group was not statistically significant (p = 0.43). The development among persons above age 75 years was different in this analysis since this group had a decrease in SWEMWBS in the whole period.

Discussion
The aim was to describe the mental well-being in a longitudinal study with self-reported data measured just before and twice during the COVID-19 pandemic in autumns of 2020 and 2021. We observed a statistically significant but small decrease in mental well-being from 2019 to 2020 and a smaller increase from 2020 to 2021 even though not to the same level as 2019. This decrease and increase were strongest among women, persons younger than 75 years, persons without depression, persons with highest educational level and persons in employment.
Few other studies have shown the long-term development on mental health through 2020 and 2021. In a Danish study among young people, Joensen and colleagues reported decrease in mental well-being and increase in loneliness during the first lockdown (spring 2020) followed by a modest deterioration during the second lockdown (autumn and winter 2020/21) compared to pre-pandemic measures .
Vistisen and colleagues showed that in September 2021, the well-being among Danes had returned to pre-COVID-19 level except among persons aged 35-44 years (Vistisen et al., 2022). In a U.S. survey, Daly and Robinson reported that anxiety symptoms increased markedly during the onset of the COVID-19 pandemic (spring 2020) and decreased quickly during summer and autumn of 2020 although still slightly increased in December 2020 compared to pre-pandemic measures (Daly and Robinson, 2021). Furthermore, they reported that psychological distress increased during the second lockdown (January 2021) compared to pre-pandemic measures (Daly and Robinson, 2022). Ellwardt and Prag reported that approximately 40% of participants in the UK Household Longitudinal Study experienced elevated distress in the whole period from early 2020 to mid-2021 (Ellwardt and Prag, 2021). In a meta-analysis, mental health decreased from the pre-pandemic period to spring, autumn and winter of 2020 (Patel et al., 2022). In general, these other longitudinal studies are in agreement with our results of long-term influence on mental health into late 2020 and 2021 compared to pre-pandemic measures, although this deteroriation was more modest compared to the larger decrease during the first lockdown.
Our results also support different subgroup specific developments in mental health with the general pattern that higher educated, employed and persons without depression experience larger decrease in mental well-being from 2019 to 2020 and larger increase from 2020 to 2021 compared to other subgroups. These different patterns are partly mirrored in other publications supporting that high-income and education groups experience larger decrease in mental health (Daly and Robinson, 2022;Patel et al., 2022), but on the other hand that those experiencing financial stressors or without work have stronger decrease (Chandola et al., 2022;Ellwardt and Prag, 2021). Surprisingly, among persons with depression a slight increase in mental wellbeing was observed from 2019 to 2021, which we previously reported from 2019 to 2020 . This result was also reported among young people in another Danish study  and in a study from the Netherlands among adults with bipolar disorder reported less psychiatric symptoms during COVID-19 compared to before the Table 2 Mental well-being at time 2019, 2020 and 2021 and the difference between subgroups among persons with at least a response to one survey (n(persons) = 8,179, n (observations) = 16,669). Weighted by non-response weights. (1) Linear regression adjusted for sex and age. The p-value for the overall type-3 test of the influence of time.
(2) Linear regression adjusted for sex and age. The p-value is an overall type-3 test of interaction between time (2019, 2020 and 2021) and the stratification variable.
pandemic (Orhan et al., 2021). The mean change in mental wellbeing from 2019 to 2020 was 0.7 points followed by an increase of 0.3 points from 2020 to 2021. Although there is no agreed upon minimal important level of change in SWEMWBS, one study proposes that changes in SWEMWBS from 1 to 3 points have clinical relevance (Shah et al., 2018) indicating that the change was smaller than clinically relevant in the current study. For some subgroups, e.g., persons aged 45-74 years, persons with higher education and persons employed, the decrease from 2019 to 2020 was 1 point or more (Table 2).
Our study has several strengths, including a rather large sample with three measures of mental well-being among the same study population. The prospective collection of data is a main strength of our data. Furthermore, the data collection was conducted during autumn all years limiting the influence of seasonal variations in mental health. Our study may be influenced by selection bias since the response proportion was rather low (2019: 47%; 2020: 50%; 2021: 40%). To limit the influence of selection bias, we weighted the analyses by response weights based on age-and sex-distribution of the invited sample and included persons only responding once or twice.
In conclusion, our study reports a statistically significant but small decrease in mental wellbeing during the first COVID-19 lockdown and a small increase from 2020 to 2021 in a population-based study. This decrease was most pronounced among persons with higher education, employed and persons without depression.

Data sharing
Data are linked to administrative registers and can only be accessed through affiliation with University of Southern Denmark. The authors welcome any contacts regarding collaboration. The questionnaires (in Danish) are accessible at www.sdu.dk/da/sif/forskning/projekter/b etydningen_af_covid_19_krisen.