Onset and persistence of depression in older people—results from a 2-year community follow-up study

Abstract

Background: baseline physical health, disability and social support have been shown to predict depression onset, but findings for persistence are inconsistent. For onset and persistence of depression, the effect of changes in these risk factors over time is unclear.

Objective: to use baseline factors and change in factors over time to predict onset and persistence of depression over a 2‐year period.

Methods: a prospective cohort study with index assessment and 2-year follow-up of patients initially aged ≥65 years registered with two South London practices (n = 1,164). Depression was defined by a score >5/15 on the 15-item Geriatric Depression Scale. Associations between risk factors and onset and persistence of depression were analysed using multiple logistic regression.

Results: the incidence of depression was 8.4%, while depression persisted amongst 61.2% of those depressed at baseline. Comparing onset and persistence suggested some common predictors: greater baseline depression score; and follow-up measures of poor general health and compromised social support. There was some evidence that pain and worsening disability were more important for depression onset. In contrast, low baseline belief in powerful others (health locus of control measure) predicted persistence only.

Conclusion: focusing on older people with increasing disability, pain, physical ill-health and compromised social support should help in both the prevention and recognition of onset of later-life depression. In older people with depression, those with the highest symptom scores and low belief in powerful others at baseline were more likely to develop chronic symptoms and could be targeted for more intensive treatment and support.

Introduction

Depression in older people is an important clinical and public health problem [1]. Cross-sectional studies demonstrate strong associations between disability [2–5], general health [2, 4, 5], social support [5, 6], socioeconomic factors [2, 4, 5], locus of control [2, 5] and prevalence of depression in older people. However, depression prevalence is determined by both its incidence and the rate of recovery or death. Factors influencing depression onset and persistence may differ, and assessment of this requires a longitudinal approach. Understanding the factors predicting onset may enable better prevention and recognition of depression in older people. Understanding factors predicting persistence may allow identification of subjects at the outset whose course may be prolonged and thus more effective targeting of interventions.

Longitudinal studies of depression onset in older people

A recent systematic review found that most studies suffered from incomplete follow-up and many lacked data on important risk factors [7]. However, there is evidence that baseline measures of depression score, poor physical health, disability and social isolation predict depression onset [8–11]. External locus of control [8] and pain [12] also had important effects, but were studied less. Risk factors can vary over time, but few studies assessed change in risk factors [7]. Those that did show declining health and increasing disability were important in predicting depression onset [13–15], but none controlled for baseline depression score or severity nor measured locus of control or pain.

Longitudinal studies of depression persistence in older people

Another systematic review concluded that all studies had some methodological limitations: most of them were based on small samples and results were inconsistent [16]. Further research has been published, but findings remain inconclusive about the roles of poor physical health [8, 17, 18], disability [11, 17, 19], social support [8, 11, 17, 18] and baseline depression score [8, 11, 17, 18] in determining depression persistence. External locus of control [8] and anxiety [18, 19] predicted persistence, but were not measured in other studies. Three studies examined change in risk factors over time; none found an effect of increasing disability on persistence, but findings regarding the effect of worsening health were inconsistent, and none controlled for baseline depression score or severity [19–21].

In summary, few studies measured a full range of risk factors; evidence is particularly lacking on pain, anxiety and locus of control. For onset of depression, there is consistent evidence on the effect of baseline factors, but data are lacking on the effect of change in risk factors. For persistence of depression, there are discrepancies between study findings in relation to both baseline factors and change in these over time.

Our study’s aim was to assess a broad range of baseline factors and change in factors over time and to use these to predict depression onset and persistence over a 2-year period.

Methods

Sampling and procedures

This was a prospective cohort study comprising an index assessment and 2-year follow-up of patients initially aged ≥65 years registered with two South London practices. Methods for the baseline survey (2000–2001) have been described previously [5]. Practices identified those with terminal illness or dementia (to exclude) and those it would be more appropriate to interview (e.g. frailty, poor vision, living in care homes). A postal survey was conducted for other subjects, and assistance in completion was offered. A response rate of 75% yielded 1,704 respondents. Two years later (2002–2003), practices identified subjects known to have died, moved away, developed terminal illness or dementia. Remaining subjects were re-contacted with a postal survey or interview as before (see Appendix 1 available as supplementary data on the journal website www.ageing.oxfordjournals.org).

Measures

Depression

Depression symptoms were measured at baseline and follow-up using the 15-item Geriatric Depression Scale (GDS-15) [22]. The cut-off >5 was taken to indicate depression (sensitivity 78%, specificity 82% for detecting major depression in UK older people) [23].

Risk factors for depression onset and persistence

These were identified from previous research [2, 4, 11, 13, 15, 24] and measured in both surveys to assess effects of change over time. The following areas were covered: physical health (including general health, pain, disability, vision and hearing impairments); psychological factors (anxiety and health locus of control); previous history of depression; social support (availability of support, satisfaction with support, presence of a confidante and loneliness); life events occurring during follow-up and socioeconomic factors (Appendix 2 in the supplementary data shows full details).

Analysis

Statistical analysis was performed using Intercooled STATA 6.0 software (Stata Corp, Union Station, TX, USA). Longitudinal analysis was restricted to those with complete GDS-15 scores at baseline and follow-up (n = 1,164). Outcome variables were depression onset and persistence, calculated comparing the GDS-15 scores across the two measurements, taking a cut-off of >5/15 at each time period to define depression. Onset was assessed from respondents not depressed at baseline (n = 945) and persistence from those depressed at baseline (n = 219).

Associations between risk factors and each outcome in turn were examined using multiple logistic regression, initially controlling for the effects of age, sex and practice, then additionally for baseline depression score. Tables 1 and 2 show the results for those variables with a statistically significant effect on either depression onset or persistence.

Variables with a significant effect on depression onset were entered into a forward stepwise logistic regression analysis to identify factors which remained significant at P = 0.05, for inclusion in a best-fit model. Age, sex and practice were locked into the model. The process was repeated for depression persistence. Direct comparisons of factors selected into models for depression onset and persistence are problematic due to sample size differences and also due to the different variables selected into a final model and thus adjusted for. To facilitate comparison, all variables identified as important predictors of either onset or persistence were included in logistic regression models, for both of these outcomes (Table 3).

Results

Response rate

After excluding subjects who had died (n = 151), moved away (n = 162), developed terminal illness (n = 35) or dementia (n = 40) since baseline, the survey 2 response rate was 94% (1,237/1,316). Attrition was related to baseline depression score, with the proportions scoring above the cut-off >5 (missing items for some subjects) as follows: responders 19% (224/1,183), non-responders 22% (16/72), moved 32% (48/148), excluded (illness or dementia) 50% (33/66) and died 45% (60/133) (chi-square = 79.9, P < 0.001).

Depression onset

At 2-year follow-up, 8.4% (79/945) of participants were newly classed as depressed. Baseline depression score was predictably associated with depression onset and other associations were controlled for this. Onset remained significantly associated with: increasing age; poor health (general health, disability, pain, chronic disease score, impaired vision, and high baseline anxiety score); compromised social support (loneliness, time spent alone, dissatisfaction with support, and conflict in relationships); financial vulnerability (cutting back on spending to pay bills); and stressful life events (serious illness or injury to self and moving into a care home). Importantly, for all health status and social support measures, survey 2 results were more strongly associated than baseline results. Also worsening health, pain and vision, and increasing disability and loneliness were all strongly associated with depression onset.

Independent predictors of depression onset selected by forward stepwise regression were: high baseline depression score; increase in disability between surveys; and follow-up measures of poor general health, pain, dissatisfaction with support and loneliness. There was also a weak gender effect, with men more at risk (Table 3).

Depression persistence

Nearly two-thirds [61.2% (134/219)] of those depressed at baseline remained depressed at follow-up. Persistence was strongly predicted by index depression score. After controlling for this, the following also remained predictive: male gender; practice 2 status; self-reported previous depression; baseline health locus of control measure, low belief in powerful others; baseline anxiety; and follow-up measures of poor general health, loneliness and dissatisfaction with support.

Independent predictors of depression persistence selected by forward stepwise regression were: high baseline depression score; practice 2 status; low belief in powerful others; and follow-up measures of loneliness, dissatisfaction with support and poor general health. There was also a weak gender effect, with men more likely to remain depressed (Table 3).

Discussion

Statement of principal findings

The following factors predicted both depression onset and persistence: baseline depression score and follow-up measures of poor general health and compromised social support. Male gender was also weakly associated with both outcomes. There was some evidence that worsening disability and pain at follow-up were more important for depression onset, and practice status more important for depression persistence, although the effects of these factors on both outcomes were all in the same direction. Only belief in powerful others had an effect in different directions for onset and persistence, a low belief in powerful others predicted depression persistence.

Study strengths and weaknesses

The main study strengths are the population-based sample, the completeness of follow-up and the broad range of variables assessed. Incomplete follow-up and lack of information on important risk factors were major problems identified previously [7, 16].

A further strength is in the way that factors predicting depression onset and persistence were compared. These groups have very different sample sizes and direct comparison of odds ratios selected on the basis of P-values could therefore be misleading, as the power for analysing depression onset is greater than that for persistence. It is also important when making comparisons to do so when the same variables are included for both onset and persistence. Previous studies have tended to ignore these problems [8, 11, 17]. We have taken factors selected as important for predicting each outcome and put them all into models for predicting each in turn.

A problem common to all longitudinal studies of depression in older people is that the main reasons for attrition (death, terminal illness or dementia) are all related to baseline depression score [8, 11, 17]. This unavoidable selective loss of the most frail may lead particularly to underestimation of depression persistence.

We studied depression symptoms rather than depression diagnoses. However, the GDS-15 is well validated for detecting depression in older people [23], and depressive symptoms, even in the absence of a depressive syndrome (sub-syndromal depression), significantly impact on health and functioning [25].

Depression is a relapsing illness; however, symptoms were only assessed at the beginning and end of the follow-up. Beekman et al. [8] suggest a bimodal distribution, with either a short self-limiting or a chronic course. More frequent symptom measurement is required to elucidate this further.

Subjects with depression may report their health and other factors more negatively (symptom attribution) which could lead to spurious associations between risk factors and depression onset and persistence. This is a problem with self-report measures, and applies to all of the above longitudinal studies. We used a range of variables to assess each area, e.g. for health we included more objective variables such as number of diseases, visual impairment and activities of daily living (disability score) as well as subjective measures such as general health and pain. The consistency of associations seen across a range of variables strengthens our study findings.

Our study is similar to all of the above studies in examining the natural history of late-life depression, without controlling for ongoing treatment. A consistent finding was that the majority (80–90% in most studies) of older people with depression were not on antidepressant treatment [8, 11, 18, 20]. Studies that examined the effect of medication [19, 20] and mental health visits [20] found neither predicted recovery from depression.

Comparison of our findings with other studies

Depression onset

Our onset rate of 8.4% (79/945) is comparable to studies with follow-up periods of 1 year 12.0% [11], 2 years 6.0% [26], 13.1% [13] and 3 years 9.7% [8] despite different depression measures, and highlights an important clinical and public health burden. Factors predicting depression onset were broadly similar to those shown by other studies: baseline depression score [8–11], poor physical health [8–11], disability [8–11], social isolation [8–11] and pain [12]. Additional factors documented as important by others, baseline anxiety [14] and move into institutional care [15], as well as a measure of financial strain, were associated with onset, but not selected into our final model. After controlling for other variables, increasing age no longer predicted onset, in accord with others’ findings [8, 10, 11, 13, 14], but a weak association between male gender and onset was found, which has not been observed before [8, 10, 13–15]. In line with other studies examining change in variables over time [13–15], we showed that declining health and increasing disability predicted depression onset; we also had the advantage of controlling for the effects of baseline depression score. Whereas Kennedy et al. [13] found declining health and increasing disability overshadowed social support in explaining depression onset, we found that social isolation measures were equally important.

Depression persistence

Our rate of 61.2% (134/219) for depression persistence is comparable to studies using different depression measures, with follow-up periods of 1 year 63.2% [11], 2 years 46.0% [20] and 3 years 50.4%, [8] 51.7% [19]. Such high levels of persistence for a disorder that significantly impairs older people’s health and functioning again highlights an important public health issue. Baseline depression score predicted persistence, in line with some previous studies [17, 18], but not others [8, 11]. We found a weak association between male gender and persistence, in contrast to other studies with no association [8, 11, 17–19]. Lack of social support predicted persistence, in accordance with Prince et al. [11], but in contrast to other studies [8, 17–19]. Additional factors documented as important by others, baseline anxiety [18] and previous history of depression [19], were associated with persistence, but not selected into our final model. In line with other studies examining change in variables over time, increasing disability did not predict persistence [19–21]. Previous studies of the effects of worsening health had been inconsistent [19–21]; our findings, which controlled for baseline depression score, showed no significant association with persistence. However, for both increasing disability and worsening health our confidence intervals do not rule out an effect.

A novel finding was that those with a strong baseline belief in powerful others (doctors and modern medicine) were less likely to remain depressed at follow-up. This builds on our cross-sectional results, where strong beliefs in both powerful others and internality (controlling your own health) were protective, whereas a strong belief that chance or fate controlled your health increased depression risk [5]. Our findings demonstrate the importance of longitudinal studies in aiding understanding, as belief in powerful others had an effect that went in different directions for onset and persistence of depression. One other longitudinal study has examined locus of control (not specifically health related) and found a strong belief in fate was related to depression persistence [8]. These intriguing findings suggest that the role of health locus of control in later-life depression deserves further study.

Implications

Focusing on older people with increasing disability, pain, physical ill-health and social isolation should help in both the prevention and recognition of onset of later-life depression. In older people with depression, those with the highest symptom scores and low belief in powerful others at baseline were more likely to develop chronic symptoms and could be targeted for more intensive treatment and support from the outset. Attention to poor general health and loneliness during follow-up is also important. Apart from the health-belief findings, which need further confirmation, these conclusions resonate with recent national guidance on managing depression in older people. These stress the importance of prevention [1, 27], early detection [1, 27], stepped care with more intensive treatment for greater severity of illness [27] and a multifaceted approach to management [1, 27].

Key points

• Onset of depression occurred in 8% of older people over 2 years of follow-up and depression persisted in 61% of those depressed at baseline.

• Common predictors of onset and persistence of depression were baseline depression score and follow-up measures of poor general health and compromised social support.

• Pain and worsening disability were more important for depression onset, whilst low baseline belief in powerful others predicted persistence only.

• Focusing on older people with increasing disability, pain, physical ill-health and compromised social support should help in the prevention and recognition of onset of later-life depression.

• In older people with depression, those with the highest symptom scores and low belief in powerful others at baseline were more likely to develop chronic symptoms and could be targeted for more intensive treatment and support.

Ethical approval

Approval for both the baseline and follow-up studies was given by the Wandsworth Local Research Ethics Committee.

Declaration of sources of funding

The BUPA Foundation funded the study. They played no role in the design, execution, analysis and interpretation of data, or writing of the study.

Conflict of interest

No conflict of interest to declare.

Acknowledgements

We thank the two practices involved (Bridge Lane Health Centre, Battersea, London and Benhill & Belmont Practice, Sutton, Surrey) and all the participants who gave their time to respond to the surveys. Thanks also to Dr Sunil Shah and Mrs Pit Rink for their involvement early on in the study, particularly with questionnaire development, and to Professor Anthony Mann for his advice on measuring depression and anxiety symptoms in older people.

References