Risk of stress/depression and functional impairment in Denmark immediately following a COVID-19 shutdown

Figure 1 provides a timeline of the Danish COVID-19-restrictions, the number of confirmed cases for March and start of April 2020, and the data collection window for the first wave of the survey [1, 21]. Denmark reported its first confirmed SARS-CoV-2 case on February 27, 2020 [1]. On March 11, Denmark initiated nationwide school closures and the closing of public institutions, as the cumulative number of confirmed infections had increased to 264. Lockdown measures were further strengthened over the following 6 days to include border closures, the closure of restaurants, malls and hairdressers, and general encouragements to work from home. Financial aid packages to businesses and furloughed employees were launched in the same six-day period. Unlike other European countries, Denmark did not introduce curfews, stay-home orders, or mandatory use of masks during the Spring lockdown. The Danish government began easing lockdown measures from April 15, 2020. Lockdown measures were continuously eased over the summer, and not re-introduced until September 2020. Based on this timeline, we defined the beginning of the lockdown measures as March 11, 2020.

We address the mentioned limitations and extend prior work by examining mental wellbeing in this low-COVID-19-burden country. There are three reasons why we still expect to observe effects of lockdown on mental wellbeing in this context. First, witnessing critical events, such as the growth of a pandemic, could impair wellbeing even when the event is not local, as when the 9/11 terrorist attack in the US caused an immediate 16% increase in the incidence rate of trauma- and stressor-related disorders in Denmark [22]. Relatedly, Ahrens and colleagues argue that physical distancing and lockdowns can be seen as “global macro-stressors” [19]. Second, a recent multi-country study found that people experiencing physical symptoms akin to those symptomatic of COVID-19 suffer worse mental health, implying that perceived impact of the pandemic could also be important [23]. And third, experiencing even a minor lockdown order could still affect people’s mental wellbeing, as lockdown for many implies social isolation and little to no (physical) contact with others, and as people had to adjust to new daily schedules including homeschooling and working from home.

To consider how measures of wellbeing and impairment changed following the imposed lockdown, we used a representative survey carried out by Statistics Denmark on behalf of the Capital Region of Denmark’s Mental Health Services during March–April 2020 (see Additional file 1 for a flow chart of the data collection). Using a random draw from the present population database of all Danish residents aged 18 and above, Statistics Denmark initially contacted 8300 people through personal digital postboxes that are linked directly to people’s unique social security numbers and used for communications between Danish residents and governmental institutions. All people were contacted at the same time in early March and could respond to the survey when convenient within the survey period (through to mid-April). Respondents answered through computer assisted web interviews (CAWI), with those initially failing to respond receiving prompts by message to their digital postbox. The response rate to the first wave of the survey was 34% (N = 2836); 1127 respondents completed the survey prior to the lockdown announcement and initiation on March 11, and 1709 respondents completed it after March 11. These numbers reflect respondents who provided valid responses to all items of our dependent variables. Respondents who completed the survey before and after March 11 were generally alike across the background characteristics, although the proportion of respondents age 60+ decreased and the proportion of respondents with children living at home increased slightly (see Additional file 1).

With permission from the Capital Region of Denmark’s Mental Health Services, we then carried out a follow up survey in July 2020, where the same respondents were re-interviewed. Again, invitations were sent out to everyone at the same time, early July, and respondents could respond to the survey when convenient throughout the month. Of respondents participating in the first wave, 1526 (54%) also participated in the second wave collected in July 2020, for whom we thus have repeated observations. Younger respondents and respondents with children living at home were less likely to participate in the second wave, as were respondents who experienced significant functional impairment in early March (which will likely cause us to underestimate a potential increase in functional impairment from March to July), but respondents in the second wave generally resemble respondents in early March (see Additional file 1). Answers to the survey can be linked at the individual level with administrative data from Statistics Denmark. In addition to survey data on mental wellbeing and functioning, our data therefore contain information on age, gender, living arrangement (single/in a relationship), whether respondents were living with any children in the home, region of residence (Nomenclature of Territorial Units for Statistics, level 2 ([NUTS-2]), and employment status (employed, unemployed, outside the labor force).

To capture wellbeing and experienced functional impairment, the survey included two validated measures—the WHO-5 and the WSAS—that both have distinct clinically relevant threshold values. A review finds that the WHO-5 remains among the most widely used questionnaires assessing subjective psychological well-being [24]. Since its first publication in 1998, the WHO-5 has been translated into more than 30 languages and has been used in research studies all over the world. The WHO-5 is a sensitive and specific clinical screening tool for risk of stress and depression that uses five items to capture risk of depression measured between 0 and 100, with each scale contributing 0–20 points. For this measure, 0 indicates the most severe depressive symptoms and 100 indicates no symptoms. The WHO-5 has strong construct validity as a unidimensional scale [24]. For Denmark, the population norm is established as 70 for adults [25]. For the WHO-5 index, we (consistent with work in clinical settings) used the established cutoff point at 50 to create a binary category for whether a respondent is at risk for depression and stress (i.e. WHO-5 < 50, [24]).

The WSAS is a functional impairment measure designed to measure a patient’s perceived functional impairment following health problems across five items [26]. Mundt and colleagues’ paper (British Journal of Psychiatry 2002) of the WSAS finds that this instrument is a reliable and valid measure of impaired functioning. Although not originally intended for non-clinical populations, the WSAS displays valid psychometric properties across different patient populations that cover both mild and more severe (psycho-)somatic [27‐29] and mental health [26, 30‐32] conditions. Furthermore, the WSAS captures a dimension of impairment distinct from depression [32]. Although not validated in a Danish version, it has previously been used both in Danish clinical and research settings [30]. WSAS measures impairment on a scale from 0 (no impairment) to 40 (most severe impairment). Given that we study a non-clinical sample, we use the cutoff point at a WSAS score of 10, with any score above 10 indicating at least significant functional impairment with or without additional psychopathologies (WSAS > 10 [26]).

Our analytical strategy exploits the fact that data collection took place across the announcement and initiation of lockdown in Denmark in March 2020. First, we use ordinary least squares linear probability models to compare the outcomes between respondents who answered the survey before and after lockdown began. We adjust all models for gender (indicator variable for female, male as reference category) and age (indicator variables for 10-year age groups with Ages 40–49 as reference category) and use t tests to evaluate differences between respondents who answered the survey before and after lockdown began. Next, in fully adjusted regression models, we control for NUTS2-region of residence (Capitol Region of Copenhagen as reference category), labor market status (employed, unemployed, outside the labor force (reference)), relationship status (single, married/cohabiting (reference)), and whether respondents had children at home (no children at home as reference category), and again use t tests to evaluate the statistical significance of differences between respondents who answered the survey before and after lockdown began.

The impact of lockdown could differ according to the home environment. Persons living with family, for instance, may experience relative more social interaction than would persons not living with family during the imposed lockdown. To explore this possibility, we performed sub-sample analyses that compare single individuals to individuals who are living with a partner, as well as sub-sample analyses that compare people living without children in the home to people living with children in the home, and test for differences between subgroups using Chow-tests (which test whether regression coefficients are different for split data sets; in our case, we split the data by family type).

In addition, to fully leverage our data structure with re-interviews in July 2020, we then compared the reported levels of risk of depression and stress and significant functional impairment measured prior to lockdown in March (we thus drop respondents who responded during lockdown) to the levels experienced by the same persons in July accounting for repeated measures of the same individuals with clustered standard errors. The latter exercise captures the development in the outcomes across the first wave of COVID-19 in Denmark. Here, we again adjust for control variables in the regression but because we use within-individual variation across survey waves this inclusion is not essential (e.g., the age of the individual does not change substantially from March to July). All calculations were carried out using Stata 15/MP.

We then performed several robustness checks. First, we evaluated whether our choice of thresholds in the outcome variables (WHO-5 < 50 and WSAS > 10) affect inference. Second, because our sample is not fully identical to the Danish population on characteristics such as gender and age, we replicate main results using population weights provided by Statistics Denmark instead of adjusting for variables. Third, selective survey participation across the lockdown in March and across the two survey waves (if, for example, people who experience increased mental distress due to the lockdown are less likely to participate than before the lockdown or they are more likely to not respond to survey wave 2) could invalidate our results (Additional file 1 showed some sign of such selection from wave 1 to 2, although not to any discernable degree from early to late March, on observed characteristics). To address this, we use the within person changes in response between March and July. As there was very little change in lockdown measures in July, all returning respondents recompleted the survey under identical lockdown circumstances. If our main pattern of results between respondents who answered prior to and during lockdown in March persist once we take into account individual change up to the post-lockdown July wave, it would indicate results are robust to differential selection in the first wave of response across the lockdown period, and results would thus at least be internally valid. Fourth, the items of the WHO-5 ask respondents to consider their experiences during the preceding 2 weeks. To account for the possibility that respondents answering within 2 weeks after lockdown have to consider both time before and after lockdown, as a robustness check we weighted answers given in the 2 weeks after lockdown with the amount of time since the lockdown announcement. If people considered a full two-week horizon it would mean that our main estimates of the impact of the lockdown order in March will be biased toward zero—that is, our main estimates would be conservative. Last, because our main results focus on respondents with children living at home, and because there may be different requirements and worries associated with having children at different ages at home, we checked whether results differ by age of the children (which we obtained from the general registers).