Introduction

Researchers and policymakers continue to be interested in the impact of the Coronavirus Disease 2019 (COVID-19) pandemic on mental health. Studies have identified declines in mental health over the course of the pandemic across the world, linked to concerns over infection, the consequences of lockdown and isolation measures, risks related to job insecurity and financial worries, and disruption in day-to-day activities [1–5]. While early findings drew on internet-based surveys during the pandemic [6–8], more recent research has examined pandemic experiences compared to information collected before the pandemic [2,9–15].

Different groups of the population face distinct challenges with respect to maintaining good mental health and how the pandemic impacted their lives. The prevalence of anxiety or depression has been found to decline with increasing age [16,17]. While older adults are at greater risk of adverse outcomes from exposure to COVID-19 and have been linked to greater worries about it [18], older age has been linked to better mental health in terms of anxiety and depression during the pandemic [19,20]. Nonetheless, older people did experience a deterioration in mental health over the course of the pandemic compared to before its onset [21].

Older adults are also not a homogeneous group, and the likelihood of living with conditions such as cognitive impairment or dementia increases with age. Around 6.7% of people aged 65+ in England were estimated to have dementia in 2015 [22]. For mild cognitive impairment, estimates for prevalence range from 5.0% to 36.7% depending on the various definitions and diagnostic criteria used in different studies [23]. Moreover, anxiety and depression are common neuropsychiatric symptoms among people with dementia or mild cognitive impairment [24–26]. One meta-analysis estimated the pooled prevalence of depression and anxiety among people with dementia at 39% each [26]; this compares to estimated prevalence rates of 13.3% for depression in the overall older population and between 1.2% and 15% for anxiety in community samples of older people [27,28]. Prevalence rates further vary according to age, for example, with estimates for depression of 17.1% among those aged 75+ and 30% to 50% for those aged 90+ [29]. Such neuropsychiatric symptoms may also predict conversion from mild cognitive impairment to dementia [30], although the evidence is mixed for anxiety and depression specifically [31,32].

Examining the mental health experiences of older people during the pandemic should, therefore, take cognitive function into consideration. Some research has already investigated this to an extent; a rapid review of evidence related to the impact of COVID-19 isolation measures on mental health among people with dementia found that most studies identified worsening behavioral and psychological symptoms of dementia (BPSD) [33]. Some of the research found these results through subjective assessments made by caregivers [34], qualitative interviews of people with dementia and their caregivers [35,36], and across international contexts [37].

Only limited evidence has examined mental health among people with dementia across the pandemic using quantitative measures that were also assessed prior to the pandemic [38]; quantitative longitudinal research on the impact of COVID-19 on people with dementia was a key direction for future research called for by the United Kingdom–based expert working group on dementia well-being and COVID-19 [39]. Most of the existing relevant research on cognitive function and mental health during the pandemic also does not differentiate between dementia and mild cognitive impairment, although one small study from Greece did examine this distinction with respect to pre-pandemic measures [40].

This study addresses these gaps in the evidence base by testing whether changes in depression and anxiety among older people during the COVID-19 were associated with cognitive impairment. Using data from the English Longitudinal Study of Ageing (ELSA), we examine changes in depression and anxiety from before the pandemic (2018/2019) across 2 time points during the pandemic (June/July and November/December 2020) with respect to 3 levels of cognitive function. We also investigate whether the associations between cognitive function and mental health varied according to key sources of sociodemographic inequalities related to wealth, education, geographic region, and multimorbidity.

Methods

Results

Table 1 provides descriptive statistics of our analytical sample for depression within the first assessment of the COVID-19 sub-study. The percentages provided reflect similar statistics for other time periods and outcomes of interest.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sociodemographic characteristics among the sample for depression at the first assessment of the ELSA COVID-19 sub-study (June/July 2020).

https://doi.org/10.1371/journal.pmed.1004162.t001

The analytical samples vary slightly by the outcome of interest, with 5,286 individuals included in analyses for depression and the single-item anxiety measure and 5,281 for the GAD-7 anxiety score. All figures reported below reflect the full model, adjusted for all covariates identified in the previous section, and vertical scales have been restricted to relevant outputs ranges to facilitate visual inspection.

We first present results showing estimated depression scores across the 3 measured time points for the 3 categories of cognitive function (Fig 1). We find that the estimated depression score was significantly different across the 3 cognitive function groups prior to the pandemic. The score for those with no impairment was 1.17 (95%CI: 1.12 to 1.22) compared to 1.39 (1.29 to 1.49) for those with mild impairment and 1.81 (1.53 to 2.10) for those with dementia. Scores increased over time through the pandemic, with statistically significant increases between June/July 2020 and November/December for those with no impairment, going from 1.84 (1.79 to 1.90) to 2.03 (1.96 to 2.10), and those with mild impairment (from 1.89 (1.77 to 2.00) to 2.16 (2.02 to 2.30)). There was no statistically significant change in the score among those with dementia. In June/July and November/December, the differences among cognitive function groups were also no longer statistically significant.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Depression score (CES-D) over time by cognitive function (estimated scores and 95% confidence intervals, adjusted for all available covariates).

https://doi.org/10.1371/journal.pmed.1004162.g001

Turning to the anxiety score measured using the GAD-7 scale, Fig 2 shows that the estimated score for people with dementia was higher just after the start of the pandemic in June/July 2020 than for the other cognitive function groups, at 4.59 (3.66 to 5.52) compared to 2.99 (2.72 to 3.26) for those with mild impairment and 2.79 (2.67 to 2.90) for those with no impairment. There was no statistically different change for people with dementia during the pandemic, but the estimated average score did rise by November/December for the mild impairment and no impairment groups, reaching 3.57 (3.28 to 3.86) and 3.02 (2.90 to 3.14), respectively, from 2.99 (2.72 to 3.26) and 2.79 (2.67 to 2.90).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Anxiety score (GAD-7) over time by cognitive function (estimated scores and 95% confidence intervals, adjusted for all available covariates).

https://doi.org/10.1371/journal.pmed.1004162.g002

Although the single-item measure of anxiety is different from the GAD-7 scale, it provides added insight here with respect to differences in anxiety before and during the pandemic. Like with GAD-7, we see in Fig 3 a statistically significant increase in the average estimated rating between June/July and November/December for those with no impairment, going from 2.65 (2.57 to 2.74) to 2.85 (2.77 to 2.95). Both the no impairment and mild impairment groups demonstrated a significant increase in anxiety rating between 2018/2019 and November/December 2020; the score for those with no impairment increased from 2.20 (2.11 to 2.28) to 2.85 (2.77 to 2.95), while for the mild impairment group, it went from 2.48 (2.30 to 2.66) to 3.14 (2.95 to 3.33). The apparent change for those with dementia was similar to the other groups, but wide confidence intervals yield no statistically significant differences over time.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Single-item anxiety rating over time by cognitive function (estimated ratings and 95% confidence intervals, adjusted for all available covariates).

https://doi.org/10.1371/journal.pmed.1004162.g003

While the CES-D and GAD-7 scales allow us to examine changes in the average estimated scores, changes in scores alone cannot inform whether these are clinically significant changes or more related to the general unease most people faced due to the unprecedented and unpredictable nature of the pandemic. To explore this, we estimated results based on clinically recognized cutoff values for the 2 scales (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Likely clinical depression and anxiety over time by cognitive function (estimated probabilities and 95% confidence intervals, adjusted for all available covariates).

https://doi.org/10.1371/journal.pmed.1004162.g004

With respect to anxiety, we find no statistically significant differences over time in the probability of likely generalized anxiety disorder as measured using GAD-7, although point estimates suggest a possible decline in probability for those with dementia from 0.165 (0.108 to 0.223) to 0.080 (0.035 to 0.126).

Regarding depression, we find statistically significant increases in the probability of likely clinical depression for those with no impairment and those with mild impairment, looking from before the pandemic to during it. The probability for those with mild impairment was higher than for those with no impairment before the pandemic, at 0.139 (0.120 to 0.159) compared to 0.110 (0.099 to 0.120). These increased significantly after the start of the pandemic by June/July, then reaching 0.234 (0.204 to 0.263) and 0.206 (0.191 to 0.222) by November/December 2020. While there was an increasing trajectory in the point estimates for those with dementia, the differences in estimated probabilities were not statistically significant.

Inequalities in mental health and cognitive function

To test for systematic inequalities in the association between cognitive function and mental health, we first constructed models controlling only for age and gender. We found no significant three-way interactions, indicating the rate of change in our outcomes was not significantly distinct for people with cognitive impairment across distinct social, economic, and health groups.

We did, however, identify some significant two-level interactions that provide further insights into the inequalities in mental health experienced by people with cognitive impairment (mild or dementia) during the pandemic. Models were subsequently estimated using all covariates featured in those reported above. We found significant interactions linking education to anxiety. For depression, we found significant results related to wealth and multimorbidity (Fig 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Significant inequalities relating mental health and cognitive function (estimated scores and 95% confidence intervals, adjusted for all available covariates).

https://doi.org/10.1371/journal.pmed.1004162.g005

Looking at wealth and depression, the results show that, prior to the pandemic, those with cognitive impairment in the poorest third of wealth had an estimated depression score of 1.62 (1.48 to 1.76), making them worse off compared to the other groups. This contrasts to those without impairment in the richest two-thirds, who scored 1.13 (1.07 to 1.19) on depression, positioning them better than either group with cognitive impairment; those with impairment in the richest two-thirds had an estimated depression score of 1.33 (1.20 to 1.45).

During the pandemic, however, we see notable changes, especially among those with no impairment in the poorest third of wealth, whose scores change from 1.25 (1.16 to 1.34) in 2018/2019 to 2.01 (1.91 to 2.12) in June/July. Their scores become much more similar to those with impairment also in the poorest third, who scored 2.03 (1.87 to 2.19) in June/July 2020. During the pandemic, the score for those with no impairment in the richest two-thirds was significantly lower than those for the poorest groups, at 1.76 (1.69 to 1.82) in June/July and 1.93 (1.85 to 2.01) in November/December. In other words, it appears that being in the poorest third of wealth is a stronger driver than cognitive function for estimated depression scores during the pandemic.

Turning to multimorbidity, the results illustrate that those with cognitive impairment and multimorbidity scored worse on depression than the groups without cognitive impairment prior to the pandemic, with a score of 1.62 (1.46 to 1.77); this contrasts with a score of 1.24 (1.14 to 1.34) for those with multimorbidity but no impairment and 1.15 (1.09 to 1.20) for those without either multimorbidity or impairment. Just after the start of the pandemic, they were only slightly worse off than those with neither cognitive impairment nor multimorbidity, with scores increasing to 2.04 (1.87 to 2.22) and 1.80 (1.74 to 1.86), respectively. By November/December, this difference had grown further, with scores of 2.35 (2.13 to 2.57) and 1.98 (1.90 to 2.06), respectively, but there were still no significant differences for those with only one of either cognitive impairment or multimorbidity.

Across the 2 assessments of the COVID-19 sub-study, there was a notable difference in anxiety score by education. In June/July, the only significant difference in anxiety score was between those with cognitive impairment and low education (3.28 (2.92 to 3.63)) and those with no impairment and high/medium education (2.73 (2.60 to 2.86)). This difference persisted by November/December, widening to 3.84 (3.47 to 4.21) and 3.01 (2.88 to 3.15), respectively, while those with no impairment and low education were also significantly lower in anxiety score, at 2.99 (2.73 to 3.25), than those with impairment and low education.

Discussion

Using a representative sample living in private households in England, we have found that depression and anxiety worsened during the pandemic compared to before it for people with mild cognitive impairment or no impairment, whereas no statistically significant increases were found for those with dementia. We also found that differences according to cognitive function that existed before the pandemic were no longer present by June/July 2020, indicating a convergence in mental health across cognitive function groups during the pandemic. In November/December 2020, there also were no statistically significant differences among cognitive groups. Taken together, our findings provide partial support for our hypotheses on the relationship between mental health and cognitive impairment over the course of the pandemic.

Our multi-item scale of anxiety (GAD-7) provides insight only during the pandemic, as we unfortunately had no pre-pandemic comparable measure. With this measure, we did identify significant differences in anxiety between those with no impairment and those with dementia during the pandemic. The contrast with the results using the single-item measure may stem from an underlying distinction in the measured concepts, i.e., the single-item captures individual self-assessments of being anxious, whereas GAD-7 captures a broader perspective on generalized anxiety. Given potential concerns over the ability of respondents with cognitive impairment to provide reliable self-assessments, it may be that the results from a multiple-item measure like GAD-7 are more robust than a single response for identifying differences between cognitive groups.

With GAD-7, the mild impairment group also demonstrated significant change, being similar to the levels of those with no impairment in June/July 2020 but more similar to those with dementia by November/December 2020. This may relate to a deterioration in cognitive function among those with mild impairment during this time that stimulated increases in generalized anxiety. It may also suggest that those with mild impairment encountered anxiety-inducing challenges during the latter part of 2020 that other groups did not. In addition, although there was a significant difference according to cognitive function in November/December using GAD-7, the estimated scores do show a converging trend, possibly supporting the results from the single-item measure.

This convergence may also explain how our results sit against those from other research that found no change in anxiety during the pandemic among the adult UK population overall and lower levels for older age groups [15]. That research and others assessed anxiety looking at proportions with clinical scores on GAD-7 (i.e., ≥10), which may attenuate the results and subsequent conclusions, partly explaining why findings related to the overall older population in England vary from ours accounting for cognitive function [21].

Another key insight from this work relates to the increase in likely clinical cases of depression for those with no impairment or mild cognitive impairment during the pandemic compared to 2018/2019. This highlights the potential increase in demand on mental health services that might be expected moving forward. Moreover, this sits alongside the finding that the level of likely clinical cases among those with dementia has not declined, so any challenges in service delivery that existed before the pandemic will likely persist.

Alongside these potential pressures on service delivery for mental health, we should also note that other research suggests that the negative impact of the pandemic on mental health was less acute for older adults compared to other age groups. This has been found in Scotland with respect to depression and anxiety [54] along with other UK-based studies [13,55]. It has been argued that this stems from a less pronounced difference in mental health among older adults comparing before and after the pandemic [56]. Our results may partially support this argument, at least with respect to recognizing little difference among those with dementia pre- and post-pandemic.

Our findings also stand in contrast to other work in the UK looking at mental health with respect to the pandemic. One study found that the negative impact on mental health extended through the initial months of the pandemic and started to improve from July 2020 [57], whereas another found a recovery trajectory in anxiety among UK adults from April [4]. Moreover, a large meta-analysis of longitudinal cohort studies found that the changes in anxiety attributed to the pandemic appeared short-lived, with peaks appearing around March/April 2020 and declining by July, although there was still a small increase in depression between May and July [14].

Our data during the pandemic were collected in June/July and November/December 2020, yet we did not find a significant improvement between these time points. In fact, we found evidence for worsening mental health for those with no impairment or mild impairment. One possible explanation is the specific timeframes in which our data were collected. The assessment in June/July may have missed the initial spike in poor mental health reported in other studies. Our second assessment took place in November/December; this is later than those examined in the studies mentioned above. It also coincides with the second lockdown imposed in the UK in November, which was followed by easing and then further tightening of restrictions in December. These shifts may have had somewhat similar effects on mental health as the first lockdown. This may be further supported by evidence suggesting deterioration of mental health in the UK between July/August 2020 and February 2021 [15].

The differences in other results and ours may also relate to the various ways depression and anxiety have been assessed, for example, average scores on a scale, the proportions based on clinical cutoffs, or other measures used elsewhere like the 12-item General Health Questionnaire (GHQ-12). While there are alternative measures for depression and anxiety that have been used in other studies, CES-D has been used since the inception of ELSA as the main measure of depression and to facilitate international comparisons with other ageing cohort studies like the Health and Retirement Study based in the United States, while GAD-7 has been introduced for similar reasons. Both measures have also demonstrated good performance and are useful as self-reported assessments [58–61]. Moreover, our study focuses on cognitive impairment, making distinctions between dementia and mild cognitive impairment. Evidence related to these differences from longitudinal studies that include pre-pandemic assessments is limited; some studies suggest that changes in mental health observed among people with cognitive impairment were more related to expected changes linked to the impairment rather than the pandemic itself [38,40].

The analyses using the binary measures of likely clinical depression or anxiety give some insight into clinical relevance, particularly with respect to the result that the probability of likely clinical depression among those with no impairment or mild impairment was significantly higher during the pandemic than in 2018/2019. For the continuous measures, however, there is no consensus on what level of change in score constitutes clinical significance, although some work suggests, with respect to longer versions of CES-D, that somewhere between a 11% and 17% change across the scale would represent a minimal clinically important difference (MCID) [62,63]. With respect to CES-D, this would range 0.88 to 1.36, which suggests a clinically significant difference between 2018/2019 and November/December 2020 for those with no or mild impairment. One study of GAD-7 suggests an MCID of 4 points, a threshold not reached in our findings [64].

Across our analyses, this study has several strengths. Our data are drawn from a longitudinal sample of people aged 50+ in England, allowing us to examine outcomes at 3 distinct time points across the same people. Indeed, the COVID-19 sub-study achieved a notably high response rate for each wave of data collection and longitudinally. This strengthens our ability to generate robust results even during the public health crisis and social restrictions caused by the pandemic. We are also able to employ validated measures of depression and anxiety in addition to comparing the latter with a single-item measure to assess changes from before the pandemic. Finally, the breadth of the ELSA data allowed us to incorporate a wide range of adjustment variables to control for confounding in our models.

The strength of our findings is possibly limited by the relatively small number of people categorized with dementia. The wide confidence intervals this yielded may, in fact, hide true differences that exist but are masked in the statistical results. Yet, our study also includes people categorized with mild cognitive impairment, strengthening the analytical perspective in relation to cognitive function. We also cannot know if people with dementia were less likely to respond to the survey despite high response rates overall. However, our study has much lower attrition between the 2 assessments during the pandemic than most of the other studies reported in 1 systematic review of mental health before and during the pandemic [14]. This could indicate that stability in mental health found elsewhere is influenced by the attrition of those with worse or declining mental health over time.

The implications for our findings can be generalized to people with cognitive impairment or dementia living in the community but not to those who live in care homes or other institutional settings. This implies that our findings do not apply to the 40% of older UK adults with dementia living in such places [65]. However, this is a strength of our study, as fewer studies of people with dementia have been conducted in a way to reflect those living in the general community. Moreover, despite any potential limitations, ELSA and its related sub-studies have been an invaluable resource for examining the experiences of older people before and during the pandemic, along with a more robust assessment of cognitive function, potentially being the best evidence for this to date.

Our findings highlight 2 key implications for future public health responses in England as we move through the post-pandemic phase. First, mental health services will need to be supported and adequately resourced to meet the predicted increased demand that will come from older adults, especially those not living with cognitive impairment or dementia. This is underscored by the potential demand from other age groups, who may have experienced worse impacts on mental health than what we have identified in our cohort.

Second, given we saw little significant change in mental health outcomes among those with dementia, we must recognize their need for support will continue to exist. In the near term at least, the challenges in delivering this support are less likely to relate to the cognitive impairment itself but to questions of accessibility and availability, especially if the supply of support is diverted to those with no impairment. Policymakers and care practitioners will, therefore, need to ensure that people with dementia have equal access to measures to support their mental health.

To conclude, this study has improved our understanding of the way that mental health changed for people with cognitive impairment and dementia due to the pandemic. We find evidence for increasing levels of depression and anxiety during the pandemic among those with mild cognitive impairment or no cognitive impairment, but not among those with dementia. Compared to before the pandemic, mental health has become more similar across cognitive function groups, suggesting a convergence that may impact future demand for support services.

Supporting information

Acknowledgments

The English Longitudinal Study of Ageing was developed by a team of researchers based at University College London, NatCen Social Research, the Institute for Fiscal Studies, the University of Manchester, and the University of East Anglia.