Previous research has highlighted a number of key life events that characterise the experience of later life, such as retirement, bereavement, unpaid caregiving, the development of long-term chronic conditions and hospital admission. Reference Robertson1 Given their potential to disrupt lives, such events may play a role in shaping older people’s mental health. Research on the associations between such events and depression has presented a mixed picture in many cases. The association between retirement and mental health is different depending on country setting, socioeconomic position and the type of retirement. Reference Mukku, Harbishettar and Sivakumar2–Reference Odone, Gianfredi, Vigezzi, Amerio, Ardito and d’Errico6 Evidence on the link between depression and the retirement of one’s spouse or partner is also mixed. Reference Szinovacz and Davey7,Reference Park and Kang8 The loss of a spouse or partner can be particularly challenging for people in later life, given the similarities between grief and depression in terms of psychological distress, with one meta-analysis asserting that bereavement is a prominent and consistent risk factor for depression among older people. Reference Cole and Dendukuri9 Other work found distinct trajectories of depression across the years before and after bereavement. Reference Galatzer-Levy and Bonanno10,Reference Maccallum, Galatzer-Levy and Bonanno11 Unpaid caregiving, which may be undertaken before the loss of a partner, has been associated with a higher risk of depression, although some research suggests this is limited to those providing intense levels of care (≥ 20 h per week) or those combining care and full-time work. Reference Bom, Bakx, Schut and Van Doorslaer12–Reference Bom15
In addition to the changing social circumstances related to retirement, caregiving and bereavement, health-related transitions can affect later life depression. Some research has found that older people with depression are more likely to develop certain long-term conditions such as diabetes, whereas others have identified a higher risk of developing depression among those with chronic illness and following its onset. Reference Karakus and Patton16–Reference Wilson-Genderson, Heid and Pruchno20 The bidirectional nature of the relationship between depression and other aspects of health extends to healthcare settings, with research identifying associations with the risk of hospital admission as well as the treatments received and their duration. Reference Prina, Cosco, Dening, Beekman, Brayne and Huisman21–Reference Osler, Mårtensson, Wium-Andersen, Prescott, Andersen and Jørgensen25 Depressive symptoms have also been linked to increased risk of falling and recurrence of falls. Reference Lam, Lee, Lalor, Stolwyk, Russell and Brown26–Reference Atlas, Kerse, Rolleston, Teh and Bacon28
Given the mixed (and sometimes lacking) evidence, our primary objective with this study was to examine the trajectories of depression among a representative cohort of people aged 50–69 before and after key later life events, estimating their impact over a long follow-up period. Our approach is informed by life course theory and the stress process model, which emphasise how transitions and stressors accumulate and interact over time to shape mental health trajectories, moderated by coping resources, social roles and contextual factors. Reference Szinovacz and Davey7,Reference Hill-Joseph29–Reference Upasen, Saengpanya, Awae and Prasitvej31 The originality of this study lies in its analysis of a broad range of such events within a single analytical framework, drawing on rich longitudinal data with repeated depression measures.
Method
Data and participants
Our research draws on data from the English Longitudinal Study of Ageing (ELSA). Reference Banks, Batty, Breedvelt, Coughlin, Crawford and Marmot32 ELSA has surveyed a representative sample of people aged 50+ across England every 2 years since 2002, covering topics such as health, finances and psychosocial well-being. We selected 6890 respondents aged 50–69 in Wave 1 (2002/2003) and followed them through to Wave 9 (2018/2019). Data on hospital admissions were obtained through linkage of ELSA to the Hospital Episode Statistics (HES) database. Ethical approval for data collection across the ELSA waves was granted from National Health Service (NHS) Research Ethics Committees under the National Research and Ethics Service, and written consent was obtained for linkage to the HES database. No additional ethics approval was required for the secondary data analysis in this study.
Measures
Our primary outcome of interest was depression, assessed at every wave of data collection. Depression was measured using the eight-item Center for Epidemiological Studies Depression (CES-D) scale, a validated and reliable instrument for assessing depression among older adults. Reference Karim, Weisz, Bibi and ur Rehman33 The scale draws on responses to eight yes/no questions to provides a score ranging from 0 to 8, with higher scores reflecting greater levels of depressive symptoms. This measure was then used to identify likely cases of clinical depression by constructing a binary measure where scores of 4 or more are classified as depression. Reference Steffick34
Key later life events of interest include retirement, the death of a spouse/partner (spousal bereavement), becoming an unpaid caregiver, developing a limiting long-term chronic condition (a self-reported assessment) and hospital admission because of a fall or because of other non-fall causes. The retirement of a spouse/partner was examined in addition to the respondent’s own retirement. Key events were identified when reported from Wave 2 (2004/2005) through to Wave 8 (2016/2017) to allow tests for changes in depression scores before and after the event.
Analytical approach
To assess the impact of each key later life event on depression trajectories, we used piecewise mixed-effects logistic regression models. This approach tests for changes in the probability of depression before each key event as well as a change in the rate of change after the event. Each event was modelled separately, and each model included a random intercept for time centred on the event wave, allowing individual-specific variation in depression trajectories. Separate linear slopes were modelled for the pre- and post-event periods, which captures discontinuity in the trend for depression with each event. Fixed effects included age at the time of the event, gender and ethnicity (White versus Black and minority ethnic), and two time terms capturing separate trajectories of depression before and after the event. Two additional models for each event included interaction terms for age and gender.
Analytical samples for each event comprise those participants who reported the event between 2004 and 2017 (i.e. at Wave 2 through to Wave 8) and who had at least one measurement for depression both before and after the event was reported (see Table 1 for numbers and extent of attrition because of these restrictions). For spousal bereavement and hospital admissions, only the first occurrence of the event was examined. For chronic conditions, becoming a caregiver and retirements, samples were restricted to those who also reported the event in subsequent waves. The detail available on hospital admissions from the HES database allowed a slightly wider timeframe, as admissions could be identified when occurring between 2003 and 2018, where respondents had been measured on depression in 2002 and 2019, respectively (based on interview year). Analyses were conducted using Stata 17.0 initially and then Stata 18.0 for macOS to produce the final figures. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.
Table 1 Descriptive statistics on prevalence of key events and analytical sample sizes (of 6890 original sample)
Results
Table 1 presents descriptive characteristics of the analytical samples, including prevalence of each event among the original sample of 6890 and the sample sizes used in each model. Attrition ranged between 1 and 3% for all events except those related to hospital admissions, where it relates to lack of consent for HES linkage as well as missing measurement of depression following the event, linked to study drop-out and/or respondent mortality. Given the nature of these missing data, we opted for complete case analysis without imputation.
The most frequent events reported were non-fall hospital admissions (70.1%), spousal retirement (36.1%) and own retirement (23.4%). Spousal retirement was more frequently reported by men (26.8%) than women (20.5%), and by White (23.9%) compared to Black and minority ethnic (10.1%) respondents (see Supplemental Material). Spousal bereavement was also more common among women than men at 11.9 and 5.4%, respectively.
We present the analytical results below by plotting the average probability of depression in the years before and following each event. Events are grouped by their impact on depression trajectories, first those with no significant changes in trajectory followed by those linked to improvements; none of our events were associated with worsening trajectories. We report odds ratios and 95% confidence intervals (95% CI) for the depression trajectory before each event and for the estimated change in trajectory following the event (i.e. the difference in the rate of change before and after the event). Complete results are included in the Supplement.
Events associated with no change in depression trajectories
Many key life events were not associated with any significant change in depression trajectory; these include developing a limiting chronic condition, becoming an unpaid caregiver, spousal retirement and hospital admissions because of a fall.
The probability of depression was not linked to a significant change over time before developing a limiting chronic illness (odds ratio: 1.05, 95% CI: 0.98–1.13), becoming a caregiver (0.93 (0.83–1.04)) or fall-related hospital admission 0.97 (0.93–1.01)). There was also no significant difference in the change following the events (0.99 (0.87–1.11), 1.04 (0.79–1.36)) and 1.06 (0.98–1.14), respectively), even though the graph for hospital admissions because of a fall suggests a shift in direction. These results can be viewed graphically in Fig. S1 of the Supplemental Material. The inclusion of interaction effects also provided no evidence for differences in the change in depression over time according to gender or age for these events.
As shown in Fig. 1, the change in the probability of depression was not significantly different following a spouse’s retirement (0.98 (0.90–1.07)). The probability of depression was, however, associated with a significant decline over time before spousal retirement (0.91 (0.86–0.97)); in other words, each year before a spouse’s retirement was associated with a 9% reduction in the odds of depression.
Fig. 1 Trajectories in the probability of depression before and after a spouse’s retirement (overall total and stratified by gender) and the death of a spouse/partner.
With spousal retirement, we also found variations by gender, with women less likely than men to experience a decline in the probability of depression before their spouse’s retirement. Men were associated with a 15% reduction in the odds of depression each year before spousal retirement (0.85 (0.79–0.92)), which contrasts with no significant change among women (combined odds ratio of 0.96 (0.84–1.10), calculated from results for the slope before and interaction term: see Supplemental Material). The trajectory of depression was not, however, linked to a statistically significant change following spousal retirement for either men or women.
Events associated with improvements in depression trajectories
A statistically significant improvement in the trajectory of depression was found following spousal bereavement, one’s own retirement and non-fall hospital admissions.
The years before the loss of a partner or spouse were associated with a 41% increase in the odds of depression each year (1.41 (1.28–1.55)) (Fig. 1). Following the loss, the trajectory of depression changes direction and is of of a similar magnitude, being associated with a 48% annual reduction in odds (0.52 (0.44–0.61)). This yields a similar probability of depression for the same number of years before or after the death of a spouse, suggesting a kind of recovery or return to baseline. There were no significant interaction effects related to age or gender.
While we observed no significant change in the trajectory of depression because of spousal retirement, we did find a significant change following one’s own retirement (Fig. 2). Before retirement, there was a slight but statistically significant annual increase in the odds of depression (1.06 (1.001–1.12)). Following retirement, the change in trajectory was associated with a 15% reduction in the odds of depression (0.85 (0.78–0.92)).
Fig. 2 Trajectories in the probability of depression before and after one’s own retirement (overall and stratified by gender and age group).
The overall associations highlighted above hide underlying gender and age-related differences. After including interaction terms with gender, the change in trajectory following retirement was no longer statistically significant, which suggests that, for each gender, the trajectory of depression before their own retirement remained the same following the retirement. Nonetheless, men showed a 10% annual reduction in the odds of depression before their own retirement (0.90 (0.85–0.96)), in contrast, with no significant overall association among women (combined odds ratio: 1.02 (0.92–1.13)).
In interaction effects with the age of retirement, we found that later retirement was associated with a lower probability of depression, with a 47% reduction in the odds of depression for each additional year of age at retirement (combined odds ratio: 0.53 (0.41–0.70)). This is likely explained in part by a health selection effect. Figure 2 illustrates the nearly parallel trajectories for those retiring under the age of 65 and those retiring aged 65 and over, with the probability of depression lower for the older age group.
We also found evidence of an improvement in depression trajectory following hospital admissions because of reasons other than a fall (Fig. 3). Overall, the probability of depression was on a downward trajectory up to the non-fall hospital admission, with an 8% annual reduction in odds (0.92 (0.90–0.94)). This trajectory was attenuated following the non-fall admission, with the change in trajectories associated with a further 4% reduction (0.96 (0.94–0.99)).
Fig. 3 Trajectories in the probability of depression before and after a hospital admission because of causes other than falls (overall and stratified by gender and age group).
The changes in trajectories because of non-fall hospital admissions differed according to gender and age. Before including interactions, women were independently associated with around double the odds of depression (2.14 (1.69–2.72)). Statistically significant odds ratios were found for the slope change after non-fall hospital admission, being a woman and their interaction term; however, these effects attenuate each other, with the calculated confidence interval of the combined odds ratio crossing 1.0 (0.99 (0.92–1.07)). Nonetheless, the results do suggest an improved trajectory among men, as the change in trajectory following a non-fall hospital admission was associated with 9% lower odds (0.91 (0.87–0.95)) of depression.
There were also differences according to the age at which a hospital admission because of causes other than a fall occurred. Age at the time of admission was independently associated with a 7% reduction in the odds of depression (0.93 (0.91–0.95)). Introducing interaction effects highlights that older age was associated with a lower change in the trajectory of depression following a non-fall hospital admission; each additional year of age was associated with a 40% reduction in the change in the trajectory of depression (combined odds ratio: 0.60 (0.50–0.72)).
Discussion
This study investigated how depression trajectories change before and after common later life events. Statistically significant improvements were observed for spousal bereavement, one’s own retirement and non-fall hospital admissions, with reductions in the odds of depression of 48, 15 and 4%, respectively. In contrast, events such as limiting long-term illness onset, becoming an unpaid caregiver, spousal retirement and fall-related hospital admission showed no significant change in depression trajectories. This absence of statistically significant findings may reflect inconsistencies in previous research, which may stem from differences in study design, depression measurement or contextual factors such as social norms or welfare systems.
Our study addressed two key questions: How do depression trajectories change around common later life events, and which events are associated with recovery or risk? These questions are central to identifying critical windows for targeted screening, early intervention or mental health support services, with important implications for clinical care and preventative strategies. Overall, our findings contribute to the literature by using a unified analytical approach to assess changes in depression trajectories before and after a broad set of later life events in health, social and family domains, offering insights into when older adults may be most vulnerable – or resilient – in relation to depression risk.
Depression and retirement
Our findings on one’s own retirement and depression trajectory carry particular implications in the context of the rising state pension age (SPA) in the UK. Our findings demonstrate a worsening trajectory of depression before retirement, followed by a significant improvement after retirement. This pattern is similar to that found in one US study, which reported that depression increased the likelihood of retirement. Reference Segel-Karpas, Ayalon and Lachman35 However, our models do not allow causal inference and cannot disentangle whether pre-retirement declines in depression stem from working conditions, anticipation of retirement or other factors.
Interestingly, the depression trajectories associated with one’s own retirement were notably different between men and women. Men were linked to an improving trajectory over time, with no significant change following retirement. In contrast, women showed no significant change in depression either before or after retirement. These differences are particularly relevant given policy reforms to women’s SPA during the study period and likely reflect ongoing inequalities in the labour market, raising questions of why women do not experience similar reductions in depression before retirement and beyond.
Age at retirement further shaped associations. Those retiring at 65 or older had consistently lower probabilities of depression than those retiring earlier, although their trajectories were parallel. This may imply a role for depression in driving early retirement, although most women in the analyses would have reached SPA before 65. Alternatively, it could be that other factors that influence early retirement also contribute to higher levels of depression. The findings nonetheless underscore the potential importance of identifying older workers with depression and offering adequate preventative support for people up to and upon retirement.
Explaining unexpected or counterintuitive findings
The improvement in depression trajectory following spousal bereavement may initially seem counterintuitive, given the expected emotional toll of losing a partner. Yet, earlier research identified a distinct trajectory of high pre-loss depression followed by a decrease. Reference Galatzer-Levy and Bonanno10,Reference Maccallum, Galatzer-Levy and Bonanno11 Other research suggests that social capital and wealth moderate the health effects of partner loss, more strongly benefitting men. Reference Kung36 Moreover, our analysis does not take into account the nature of the spouse/partner’s death, while our assessment of bereavement is taken from a loss occurring within a 2-year window. More acute depressive responses to sudden partner loss may be missed in biennial interviews, while gradual declines in a partner’s health may prompt earlier emotional adjustment. In such cases, surviving partners may experience a kind of relief or recovery following the partner’s death.
The lack of any observed change in depression trajectory after chronic illness onset may relate to respondents’ life stage. Previous work suggests that chronic illness onset has a more acute impact on depression among younger adults, while older adults may be more resilient because of different expectations, personal experiences of illness and more capacity to cope through the concept of mastery. Reference Hill-Joseph29 Mastery, the self-perception of being able to cope and manage, is argued to be the primary psychological mechanism through which depression is affected by chronic illness. Reference Turner and Lloyd30 The association between depression and chronic illness has also been found to be attenuated or to disappear when controlling for subjective health status, which was not included in our models. Reference Park, Park, Yang and Chung37 The use of a self-reported assessment for chronic illness may also have introduced error, although this is partly compensated for by looking at health-related events in the separate analyses using hospital data.
More broadly, the lack of worsening trajectories may reflect resilience among the study cohort. Reference Upasen, Saengpanya, Awae and Prasitvej31,Reference Li, Ge, Dong and Jiang38,Reference Zhai, Wang, Liu and Zhang39 Most events examined can be considered stressors or shocks, largely unexpected events associated with negative experiences. For some respondents, even retirement may have been an unexpected, negative experience, as involuntary labour market exit remains an issue among older workers in the UK. Individual differences in coping skills and social support likely shape emotional responses to these events. The fact we found no associations with worsening depression could suggest that, on average, respondents had sufficient resilience to cope with these later life stressors when it comes to their impact on depression.
Strengths, limitations and future research
Our study has several strengths. ELSA offers some of the best observational data available for our study objectives, with rich longitudinal data across nine waves linked to hospital data. The ELSA sample is representative of adults aged 50–69 in England at baseline (2002/2003), and depression was assessed repeatedly using a validated measure. These design features enabled a robust examination of patterns over time not possible in cross-sectional or shorter-term studies. Future analyses should assess whether similar patterns are observed in more recent cohorts.
A key limitation relates to sample size. Statistically significant findings largely relate to those events more frequently reported, yielding larger analytical samples for each regression analysis. Smaller samples for other events may have limited power to detect associations. On the other hand, the lack of associations may underscore the impact of resilience, which could partly explain why the statistically significant results were small in magnitude, that is, depression around later life events may be characterised more by resilience than large shifts. Still, the small magnitude of observed changes may reflect limited clinical relevance, highlighting the importance of considering both statistical and practical significance when interpreting changes in depression trajectories.
In addition, small sample sizes restricted our ability to include additional potential confounders, especially those related to socioeconomic or health status. While such variables are known correlates of depression and could have attenuated some of the results, including them in these event-centred models risked conflating the impact of the life event itself – especially where such variables may themselves be influenced by the event. We thus opted for a parsimonious and consistent modelling strategy focused on change around each transition. Future research should consider alternative approaches to overcome such limitations related to sample size and confounding.
We noted above the liminal restriction in our data, that is, measures are taken every 2 years, with key events noted as occurring when reported at a given wave. We are unable to identify the exact timing of events before the interview, which is likely important given that depression is assessed at the interview. We could also not account for experiencing multiple events simultaneously, which could have important impacts for mental health and depression. We also note that, while the eight-item CES-D is a validated scale for assessing depression risk among older adults, results could have differed if based on other psychometric scales or clinically assessed diagnoses.
Finally, our modelling of depression trajectories before and after events used linear functions, which assume a constant rate of change. This could potentially mask acute, short-term spells of depression. Future research should explore nonlinear approaches in estimation.
Summary and Implications
This study examined how depression trajectories change in response to key later life events. We found significant improvements following spousal bereavement, one’s own retirement and hospital admission because of causes other than a fall. No changes in trajectory were associated with developing a limiting long-term illness, becoming an unpaid caregiver or following spousal retirement or a hospital admission because of a fall. Gender- and age-related differences were found for the trajectories following one’s own retirement and hospital admission because of causes other than a fall. For both, an improved trajectory was identified for men but there was no statistically significant change among women. Higher age at the time of the event was associated with a lower probability of depression for those who retired, while it was associated with a smaller change in the trajectory of depression following hospital admission.
These results suggest a relative resilience among older adults in England, with no events linked to worsening trajectories of depression. However, the absence of worsening trends does not imply an absence of need, and trajectory changes did not account for the level of depression at a given event. For example, the greatest improvement was observed following spousal bereavement, but the probability of depression was highest at the time of this event compared to the other events. Support therefore also has a role to play among people at risk of experiencing these key events, aligning with public health priorities emphasising the role of prevention.
Supplementary material
The supplementary material is available online at https://doi.org/10.1192/bjp.2025.10426
Data availability
Data for the English Longitudinal Study of Ageing (ELSA) are available for free upon registration through the UK Data Service (https://doi.org/10.5255/UKDA-Series-200011). The linkage to Hospital Episodes Statistics data is provided in agreement with NHS England, which does not allow onward sharing; these data are accessible only to approved members of the ELSA research team. Analytic code and detailed research materials related to the analyses presented here are available upon request to the authors.
Acknowledgements
The English Longitudinal Study of Ageing (ELSA) 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. We would also like to acknowledge Prof Andrew Steptoe, Principal Investigator of ELSA, and the Centre for Ageing Better, who provided support to another project that led to this study.
Author contributions
P.Z. contributed to the design of the main project, with input from E.-J.S. to refine the current study parameters. E.I. and B.B. conducted analyses and data management for the study. B.B. drafted the manuscript, with P.Z., E.I. and E.-J.S. providing support and feedback to arrive at the final manuscript.
Funding
This work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2020S1A5A2A03045511). The English Longitudinal Study of Ageing (ELSA) is funded by the National Institute on Aging in the United States (RO1AG017644) and a consortium of UK government departments coordinated by the National institute for Health and Care Research (NIHR Ref: 198-1074).
Declaration of interest
None.
eLetters
No eLetters have been published for this article.