Introduction

The COVID-19 pandemic and its health, social and economic consequences initially led to a global increase in mental ill health in adults compared with prepandemic levels.1 2 Studies comparing rates of mental ill health before and during the COVID-19 pandemic report an increase in depression, anxiety and stress during the early stage of the pandemic.3–6 In addition to the adversities directly associated with the pandemic, such as infection and bereavement, the enforcement of nationwide lockdowns and restrictions have affected the lives of millions of people, including social and financial adversity due to limited social contact, increased work pressure and job loss.

To date, most longitudinal studies investigating trajectories of mental health during the pandemic have been limited to data collected in the early stages of the virus outbreak.2 3 5 7–9 Longer-term patterns are only starting to emerge. Beyond the initial phase of the pandemic, a number of studies have found a drop in mental health problems after the restrictions eased in summer and autumn 2020.2 10 For instance, large cohort data from the UK Household Longitudinal Study showed a drop in mental health problems until June 2020, though levels remained higher than before the pandemic.8 Data from The English Longitudinal Study of Ageing indicated that depression and anxiety levels increased from summer 2020 to November and December 2020, at least in older adults.9 This points to a possible long-term deterioration of depression and anxiety symptoms, but data on mental health trajectories in the longer term (ie, more advanced stages of the pandemic) are limited. It has also been shown that depression and anxiety rates and initial responses to the pandemic differ across different age groups.4 It is, therefore, important to investigate longer-term mental health trajectories across different age groups, accounting for prepandemic levels of mental ill health.

To fill these gaps, we investigated how a range of social, economic and physical factors were associated with trajectories of depressive symptoms in adults from May 2020 to March 2021, controlling for prepandemic mental health. In terms of predictors, we focused on previously identified risk factors for depressive symptoms in the early phase of the pandemic,3 5 that is, gender, financial difficulties, having a partner, living conditions, being in a vulnerable group and additional COVID-19-related factors (including adherence to social distancing measures). We tested if these factors were also predictors of depressive symptoms in the longer-term during the pandemic.

Methods

This study used latent growth modelling to assess individual trajectories of depression symptoms in adults to assess how depression scores developed over three time points during the COVID-19 pandemic. We then used these individual depression trajectories as a dependent variable and predicted depression trajectories from a range of variables that have previously been identified to be associated with depression during the COVID-19 pandemic.

Participants

This study used data from four existing, UK nationally representative, longitudinal cohorts within the COVID-19 Survey.11 The Millennium Cohort Study (MCS) follows the lives of around 19 000 children born 2000–2002 in England, Scotland, Wales and Northern Ireland.12 The Next Steps (NS) cohort includes around 16 000 people, born 1989–1990 in England.13 The British Cohort Study (BCS) recruited individuals born in a single week of 1970 in England, Scotland and Wales.14 The National Child Development Study (NCDS) includes 17 415 people born in England, Scotland and Wales within one week in 1958.15

Data were collected online before the end of the first UK national lockdown between 4 May 2020 and 30 May 2020 (wave 1; 16 784 participants), 10 September 2020 and 16 October 2020 when restrictions had been eased (wave 2; 24 247 participants), and 1 February 2021 and 21 March 2021 before the end of the third national lockdown (wave 3; 26 531 participants). Web survey non-responders in wave 3 were invited to take part via telephone (N=1597, 9%). The survey data are available to researchers in the UK from the UK Data Service (https://ukdataservice.ac.uk).

Due to the number of missing values and the nature of the data analysis (growth models need data from three time points), participants with total score of the depression and loneliness scales missing more than one wave were excluded, resulting in a total analytical sample of N=16 978. Sociodemographic information of the analytical sample is reported.

Measures

Symptoms of depression (dependent variable) during the pandemic were assessed with the same questionnaires across all cohorts. Prepandemic mental health (predictor) was assessed using different measures in different cohorts.

Outcome: depressive symptoms

The Patient Health Questionnaire (PHQ-2) was used to measure depressive symptoms across all cohorts. It comprises the first two items of the original 9-item PHQ-9.16 Items on the PHQ-2 are scored on a four-point Likert scale. The PHQ-2 has been found to be sensitive to change.16

Predictors

Health and socioeconomic information. Based on previous findings,1 3 5 17 various variables were used to assess their association with depression trajectories during the COVID-19 pandemic. Sociodemographic variables included gender, and whether participants experienced financial difficulties at any point during the three waves (1=living comfortably, 5=finding it very difficult). COVID-19-related variables included whether participants had been infected by COVID-19 (1=yes, confirmed by a positive test, 0=no; excluded: yes, based on strong personal suspicion or medical advice; unsure), had been hospitalised due to a COVID-19 infection, had been asked to shield (people who were clinically extremely vulnerable to severe illness from COVID-19 in the UK, for example, due to having a respiratory illness, were sent a letter with information on how to best protect themselves from the virus, and could get a shielding note for their employer), and how much they adhered to social distancing (0=not at all, 10=fully compliant). Increased risk due to COVID-19 included being a keyworker/essential worker (‘Are you a Key worker, or has your work been classified as critical to the Covid-19 response?’), being pregnant and having a long-term illness (eg, asthma, obstructive pulmonary disease, diabetes, heart disease or cancer). The current living situation was assessed as having access to a garden/backyard at any point during the pandemic, cohabitating with a partner (no=0, yes=1) at the beginning of the pandemic, living with children 4 years old or younger, number of people in household, and number of rooms. Whether participants had a partner (no=0, yes=1) at any time during the pandemic was also considered as a predictor (online supplemental tables 1 and 2).

Loneliness: All participants completed a four-item version of the original three-item University of California, Los Angeles (UCLA) Loneliness Scale18 with an additional item (‘How often do you feel lonely?’). Items are measured on a four-point Likert scale, with higher scores indicating higher levels of isolation and loneliness.

Prepandemic mental ill health and Well-being: Mental health problems and mental well-being were assessed with different measures in the different cohorts before the pandemic.

MCS and BCS cohort participants completed the seven-item version of the Warwick-Edinburgh Mental Well-being Scale prior to the COVID-19 waves. It is measured on a five-point Likert scale, with higher scores indicating higher levels of well-being.

NCDS cohort participants completed a nine-item version,19 measured on a five-point Likert scale, with higher scores indicating higher levels of mental well-being.

MCS cohort members completed the Kessler-6 (K-6) at age 17. The K-6 assesses symptoms of depression and anxiety, that is, ‘psychological distress’.20 The K-6 is a shorter six-item version of the original K-10, measured on a five-point Likert scale (0–4), with a clinical cut-off of ≥13.

NS cohort members completed the General Health Questionnaire-1221 at age 25, which detects mental health problems. It is measured on a four-point scale but is coded into yes/no, resulting in a range of 0–12, with scores >3 indicating clinical levels.

The nine-item version of the Malaise Inventory (yes/no) was completed by the BCS cohort at age 46 and by the NCDS members at age 55, measuring mental health problems, clinical cut-off is ≥4.22

Statistical analyses

To avoid exceeding acceptable levels of missing data to impute, cases where more than one value was missing for the PHQ or the loneliness scale were excluded. For the remaining cases, 15% of values were missing and imputed, using multiple imputations. A repeated measures mixed-model analysis of variance (ANOVA) was used to investigate and describe differences in depression scores between each of the waves (wave 1–wave 2 and wave 2–wave 3 per cohort: 2×4 = 8 post hoc tests) and across cohorts (MCS, NS, BCS, NCDS: 6 post hoc tests per wave). N=21 post hoc t-tests were Bonferroni corrected (for this correction, the alpha threshold 0.05 is divided by the number of post hoc tests conducted and is adjusted accordingly) and were therefore only considered significant if p<0.002. Participants that only participated in wave 3, and were therefore excluded from our analyses had higher mean depression values (mean=3.42, SD=1.67) than participants that were included in several waves (mean=3.18, SD=1.54; t(21307)=−8.92, p<0.001, d=−0.15).

After looking at changes in mean depressive symptoms from one wave to the next, the overall trajectory of depressive symptoms was taken into account and predicted from a number of independent variables in each cohort. Non-linear (quadratic) latent growth curve models were calculated for the mental health scales, that is, depression and loneliness across the pandemic, respectively, to gain indicators of their long-term trajectories (Lavaan package in R).23 Intercepts and slopes for each participant were extracted to track how symptoms of depression and loneliness changed across the COVID-19 pandemic (online supplemental table 3).

Linear regressions were conducted for each of the cohorts with depression slopes (trajectories) as the dependent variable, and health variables, socioeconomic factors, prepandemic mental health, loneliness intercepts and loneliness slopes as independent variables. Combined design weights and non-responder weights were used to make the samples representative of the nationally representative UK cohorts they were part of. Variables that had fewer than 25 non-zero cases were excluded from the regression analyses (online supplemental table 1).24 Regressions were conducted for each cohort respectively due to the different prepandemic mental health measures.

ResultsDepression scores across the pandemic by generation

A repeated measures ANOVA with depression scores as the dependent variable showed a main effect for time F(2, 33 948)=235.95, p<0.001, eta2=0.01, a main effect for cohort F(3, 16 974)=709.91, p<0.001, eta2=0.11, and a significant interaction between time and cohort F(6, 33 948)=12.29, p<0.001, eta2=0.002. Across all time points, mean depression scores increased with decreasing age (cohort is used as a proxy for age here), with significant differences across all cohorts. Within cohorts, only depression scores in the MCS cohort decreased from wave 1 to wave 2 (standardized mean difference (SMD)=0.10, 95% CI 0.06 to 0.14). Depression scores increased from wave 2 to wave 3 across all cohorts with a small but consistent effect size (SMD=0.12, 95% CI 0.11 to 0.13; online supplemental table 4, figure 1). Participants who scored above the clinical cut-off for depression increased further from wave 1 to wave 3 with the largest difference in the youngest cohort (MCS=12%) and the smallest increase in the 1970 BCS cohort (4%; online supplemental table 1).

Figure 1

Means (±SE) of depressive symptoms across the three COVID-19 waves, that is, May 2020, September/October 2020, February/March 2021 differed by cohort. Cohort members in their early 20s (MCS) had the highest scores with an initial decrease at wave two and an increase at wave 3. Members of the NS cohort (early 30s), BCS cohort (early 50s) and NCDS cohort (early 60s) experienced no decrease in symptoms of depression but these also increased from wave 2 to wave 3. ***P<0.001. BCS, British Cohort Study; MCS, Millennium Cohort Study; NCDS, National Child Development Study; ns, not significant; PHQ-2, Patient Health Questionnaire 2.

Predictors of depression trajectories

Regressions models accounting for loneliness and pre-pandemic mental health explained more variance in depression trajectories (MCS: R2=0.18, F=12.95, p<0.001; NS: R2=0.21, F=36.69, p<0.001; BCS: R2=0.22, F=25.60, p<0.001; NCSID: R2=0.18, F=45.94, p<0.001) compared with models that only included socioeconomic and health predictors (MCS: R2=0.06, F=5.42, p<0.001; NS: R2=0.08, F=14.19, p<0.001; BCS: R2=0.08, F=11.13, p<0.001; NCSID: R2=0.03, F=9.80, p<0.001). Predictors remaining significant in the MCS cohort included chronic illness, being a keyworker (protective factor), loneliness level at beginning of the pandemic, and loneliness trajectory (all p<0.05; table 1). In the NS cohort, significant predictors included financial difficulties, chronic illness, having a child under 4 years (protective factor), compliance with social distancing measures (protective factor) being pregnant (protective factor), loneliness level at beginning of the pandemic, loneliness trajectory and prepandemic mental health problems (all p<0.05; table 2). In the BCS cohort, significant predictors included financial difficulties, chronic illness, being a keyworker (protective factor), number of people in a household, loneliness level at beginning of the pandemic, loneliness trajectory, prepandemic mental health problems and prepandemic well-being (protective factor)—all p<0.05 (table 3). In the NCDS cohort, significant predictors included chronic illness, loneliness level at beginning of the pandemic, loneliness trajectory, prepandemic mental health problems and prepandemic well-being (protective factor)—all p<0.05 (table 4). Gender was no longer significant as a predictor when loneliness and prepandemic mental health were entered into the regression.

Table 1

Regression predicting depression trajectories in the age group in their 20s

Table 2

Regression predicting depression trajectories in the age group in their 30s

Table 3

Regression predicting depression trajectories in the age group in their 50s

Table 4

Regression predicting depression trajectories in the age group in their 60s

Discussion

We investigated trajectories of depressive symptoms beyond the early stages of the COVID-19 pandemic in four large UK adult cohorts of different ages. We found a long-term increase in depression symptoms across all age groups. Severity of depressive symptoms initially decreased in those aged 20 from the lockdown in May 2020 to when restrictions had eased in September 202. However, they increased to even higher levels during the lockdown in February/March 2021, compared with the May 2020 lockdown. This small but consistent increase was found across all the included generations. In addition to the large increase in clinically relevant mental health symptoms after the onset of the pandemic, rates of clinically relevant depression symptoms increased further across the cohorts from spring 2020 to spring 2021, with a third of young people showing clinically relevant depression symptoms by spring 2021.

Before this study, relatively little was known about how persistent mental health changes are as a consequence of COVID-19, and how mental ill health affects different age groups. While previous available evidence indicated that symptoms of depression initially increased but decreased or stabilised in summer 2020,2 25 this study demonstrates that the winter lockdown 2020/2021 was associated with an increase in depression symptoms in the UK beyond the effects of the first lockdown. While findings of resilience are encouraging,25 it appears that repeated COVID-19-related measures and/or the ongoing pandemic have a detrimental impact on mental health across all age groups 20–62 years.

Data from the four cohorts showed that the younger the cohort, the higher symptoms of depression were throughout the data collection waves. Depression symptoms were highest in individuals aged 20, followed by those aged 30, 50 and 62. This mirrors findings in the early stages of the pandemic,26 and is consistent with research in other samples.5

Of note, across all age groups, feeling lonely was the strongest factor in explaining the increase in depression symptoms across the pandemic. This was true for both the level of loneliness participants experienced at the start of the pandemic, and the trajectory of feeling lonely across the pandemic. Interestingly, the degree of compliance with social distancing measures was not associated with increased depression symptoms in any of the cohorts, and even served as a protective factor in those aged 30. Thus, it appears that feeling lonely is a much more powerful predictor of depression than the actual adherence to social distancing measures in adults. This implies that feeling connected is not necessarily dependent on face-to-face contact in adults and could be enhanced by other means. This is in line with a recent study showing that higher quality, more face to face or video/phone social contact is associated with lower depression scores.27 It is interesting that social distancing measures acted as a protective factor in those in their 30s. One possibility is that working from home and flexible working arrangements benefited this age group, including young, well-functioning families. It should be noted that this study did not assess effects of social distancing on the children, however.

Chronic illness was associated with increasing symptoms of depression across all age groups. Individuals with certain long-term illnesses (eg, diabetes) have been identified as a high-risk group for becoming severely unwell if infected by COVID-19.28 Increasing depression in this subgroup may be due to the heightened health concerns during the pandemic. It is likely that chronic health conditions were associated with higher prepandemic rates of comorbid mental health problems, such as anxiety and depression. However, when in our model, we accounted for prepandemic mental health problems, long-term illness still remained significantly associated with increases in depression symptoms across the pandemic. These results are in line with a recent study showing that individuals with health problems have high levels of disease anxiety regarding COVID-1929; this might apply to depression levels as well.

Financial difficulties were also significantly associated with an increase in depression scores across the pandemic in those aged 30 and 50. This is in line with previous findings and reflects the economic hardship COVID-19 has caused for many, including job loss, zero hour contracts and income cuts.3

Prepandemic mental health difficulties were associated with an increase in depression symptoms across age groups 30–62, corroborating previous findings,3 while mental well-being was a protective factor in older adults. This again points to the importance of taking prepandemic mental health into account where possible.5

We also found a few protective factors, for instance working as a keyworker in those aged 20 and 50. This is a surprising finding that should be further explored, and might be related to the early access to vaccination, at the last data collection wave, offered to keyworkers, or to the resilience developed in particularly challenging settings. Furthermore, for participants aged 30, pregnancy and having a small child was a protective factor, although with small effect sizes. While previous data showed that living with young children during the pandemic was a risk to mental health,5 our data suggest that the presence of a future or current child could be a protective factor. The possible underlying reasons deserve additional investigation.

While being female has been shown to be one of the risk factors for poorer mental health in general, and during the COVID-19 pandemic in particular,3 5 26 an increase in depression scores was not associated with being female when feelings of loneliness were controlled for. It has to be noted that the findings do not speak to the baseline differences in depression levels that exist between males and females to begin with, but highlight how depression scores developed during the pandemic. Future research should explore the relationship of social isolation, feelings of loneliness and depression in females during the pandemic more closely.

Crowding has previously been identified as a risk factor,30 and living with more household members was associated with increasing depression scores in people aged 50 only.

Strengths and limitations

Strengths of this study include a large sample size, the possibility to analyse data from different age groups, and the longitudinal design. Moreover, prepandemic mental health data were available, and this is an important factor to consider when investigating the mental health effects of the COVID-19 pandemic.

While a strength of this study is that prepandemic mental ill health was accounted for in the analyses, the questionnaires that assessed mental health difficulties differed across the cohorts. This also limits the comparability of clinically relevant mental health symptoms before and after the onset of the pandemic. Moreover, these were collected at varying time points prior to the pandemic, and are therefore limited in their ability to reflect the mental health status of cohort members immediately prior to the pandemic, although the results were consistent across the cohorts. Participants who were only recruited in wave 3 were excluded from our analysis and had higher depression values than participants who had participated in several waves. Our analyses may, therefore, underestimate population depression values at wave 3.