Trajectories of Depressive Symptoms Before Diagnosis of Dementia: A 28-Year Follow-up Study

Key Points

Question  Does the course of depressive symptoms over adulthood in those who develop dementia differ from those who remain dementia free?

Findings  This cohort study found that depressive symptoms in late life but not midlife were associated with increased risk for dementia; analysis of depressive symptoms spanning 28 years showed them to emerge approximately a decade prior to dementia diagnosis. No substantive differences in depressive symptoms were apparent between those who went on to develop dementia and dementia-free persons 12 to 28 years prior to dementia diagnosis.

Meaning  The association between depressive symptoms and dementia in older adults may be primarily due to common causes or depressive symptoms being a feature of the preclinical phase of dementia.

Importance  Neuropsychiatric symptoms, depressive symptoms in particular, are common in patients with dementia but whether depressive symptoms in adulthood increases the risk for dementia remains the subject of debate.

Objective  To characterize the trajectory of depressive symptoms over 28 years prior to dementia diagnosis to determine whether depressive symptoms carry risk for dementia.

Design, Setting, and Participants  Up to 10 308 persons, aged 35 to 55 years, were recruited to the Whitehall II cohort study in 1985, with the end of follow-up in 2015. Data analysis for this study in a UK general community was conducted from October to December 2016.

Exposures  Depressive symptoms assessed on 9 occasions between 1985 and 2012 using the General Health Questionnaire.

Main Outcomes and Measures  Incidence of dementia (n = 322) between 1985 and 2015.

Results  Of the 10 189 persons included in the study, 6838 were men (67%) and 3351 were women (33%). Those reporting depressive symptoms in 1985 (mean follow-up, 27 years) did not have significantly increased risk for dementia (hazard ratio [HR], 1.21; 95% CI, 0.95-1.54) in Cox regression adjusted for sociodemographic covariates, health behaviors, and chronic conditions. However, those with depressive symptoms in 2003 (mean follow-up, 11 years) had an increased risk (HR, 1.72; 95% CI, 1.21-2.44). Those with chronic/recurring depressive symptoms (≥2 of 3 occasions) in the early study phase (mean follow-up, 22 years) did not have excess risk (HR, 1.02; 95% CI, 0.72-1.44) but those with chronic/recurring symptoms in the late phase (mean follow-up, 11 years) did have higher risk for dementia (HR, 1.67; 95% CI, 1.11-2.49). Analysis of retrospective depressive trajectories over 28 years, using mixed models and a backward time scale, shows that in those with dementia, differences in depressive symptoms compared with those without dementia became apparent 11 years (difference, 0.61; 95% CI, 0.09-1.13; P = .02) before dementia diagnosis and became more than 9 times larger at the year of diagnosis (difference, 5.81; 95% CI, 4.81-6.81; P < .001).

Conclusions and Relevance  Depressive symptoms in the early phase of the study corresponding to midlife, even when chronic/recurring, do not increase the risk for dementia. Along with our analysis of depressive trajectories over 28 years, these results suggest that depressive symptoms are a prodromal feature of dementia or that the 2 share common causes. The findings do not support the hypothesis that depressive symptoms increase the risk for dementia.

Alzheimer disease (AD), the most common form of dementia, is a progressive disorder. The histopathological hallmarks of AD begin decades prior to its clinical expression^1,2; a recent study estimated that amyloid-β deposits form over a period of more than 2 decades.³ The long preclinical phase of dementia has implications for the timing of interventions. There is growing consensus that interventions should target the earliest possible phase, perhaps the asymptomatic stage,^4,5 to be effective. The other major implication, which has received less attention to date, is the analytic framework used to identify putative risk factors for dementia. When measured in the years immediately prior to dementia diagnosis, these factors are likely to reflect common causes, the effects of preclinical disease (reverse causation), or prodromal changes rather than risk factors for dementia.

There is considerable research on the association of depression or depressive symptoms with dementia. A meta-analysis published in 2006 suggested that depression is associated with a 2-fold increased risk for AD,⁶ but studies covering the same period also concluded that depression may be a prodromal feature of dementia,^7-9 implying no causal effect of depression on dementia. Recent studies using repeat assessments of depressive symptoms have shown increasing symptoms to be associated with the risk for dementia.^10,11 In these studies, depression trajectories were assessed over 5 years in one study¹⁰ and over 11 years in the other study,¹¹ not long enough to cover the preclinical phase of dementia. Furthermore, the analytic strategy did not allow depressive symptoms in the follow-up period for dementia to be examined. Thus, whether depression is a risk factor for dementia or a symptom of an underlying neurodegenerative process could not be determined.

The objective of our study is to characterize trajectories of depressive symptoms starting at dementia diagnosis using a backward timescale over 28 years and compare them with changes in depressive symptoms over the same period in those free from dementia. A secondary objective is to assess whether dementia risk is higher in those with chronic/recurring depressive symptoms in midlife and late life.

The Whitehall II study is an ongoing cohort study of 10 308 persons (6895 men and 3413 women), aged 35 to 55 years at study recruitment in 1985.¹² Participants responded to a questionnaire and underwent a structured clinical evaluation, consisting of measures of anthropometry and cardiovascular and metabolic risk factors and diseases. Follow-up assessment including postal questionnaire, and clinical examinations have taken place approximately every 5 years. A postal questionnaire–only wave was also undertaken in between these waves. Participants provided written informed consent. Participant consent and research ethics approvals (University College London ethics committee) are renewed at each contact; the latest approval was by the Joint University College London/University College London Hospitals Committee on the Ethics of Human Research (Committee Alpha; reference 85/0938). Data analysis for this study was conducted from October to December 2016.

Depressive symptoms were self-reported on 9 occasions (1985, 1989, 1991, 1997, 2001, 2003, 2006, 2007, and 2012) using the 30-item General Health Questionnaire (GHQ-30),¹³ a well-established screening questionnaire for nonpsychotic psychological distress, largely depression, suitable for use in general population studies. Response options to 30 questions are not at all, no more than usual, rather more than usual, and much more than usual. The binary scoring method was used (the 2 least symptomatic answers scoring 0 and the 2 most symptomatic answers scoring 1), and so the total score ranged from 0 to 30. Scores of 5 or more defined caseness or the presence of depressive symptoms.¹⁴ Depressive symptoms were also assessed using the 20-item Center for Epidemiologic Studies Depression Scale (CES-D)¹⁵ on 3 occasions (2003, 2007, and 2012). The CES-D is a 20-item inventory of the National Institute of Mental Health Center for Epidemiological Studies to assess the frequency and severity of depressive symptoms using a standard cutoff score of 16 or greater. Relative performances within the cohort of GHQ-30 and CES-D against the interviewer-administered revised Clinical Interview Schedule as criterion for detecting a depressive episode were similar; sensitivity and specificity were 78% and 83% for the GHQ-30 and 89% and 86% for the CES-D, respectively.¹⁶

Chronic/recurring GHQ depressive symptoms were defined in 2 ways: during the early phase of follow-up using data from 1985, 1989, and 1991, and in the late phase using data from 1997, 2001, and 2003.

We used comprehensive tracing of electronic health records for dementia ascertainment using 3 databases: the national Hospital Episode Statistics database, the Mental Health Services Data Set, and the mortality register. The UK National Health Service uses in-house codes mapped onto International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes for dementia. The National Health Service provides most of the health care; the Hospital Episode Statistics database and Mental Health Services Data Set are national databases with information on both inpatient and outpatient care, with the latter also including data on care in the community. Record linkage until March 31, 2015, identified 322 cases of dementia: 172 cases were first recorded in the hospitalization register, 142 in the mental health register, and 8 in the mortality register.

The validity of dementia cases in our study is supported by modeling changes in the global cognitive score, composed of tests of memory, reasoning, and phonemic and semantic fluency administered to the participants in 1997, 2003, 2007, and 2012.¹⁷ These results show accelerated decline in global cognitive score in the 8 to 10 years before dementia diagnosis, as has been shown in studies that use a gold-standard dementia ascertainment procedure (eFigure 1 in the Supplement).¹⁸

Sociodemographic factors included age, sex, race/ethnicity (white and nonwhite), marital status (married/cohabiting vs other), education (no formal education, lower secondary school, higher secondary school, university, or higher degree), and occupational position, a 3-level variable related to salary, social status, and level of responsibility at work.¹²

Health behaviors included smoking (current, former, and never smoker), alcohol consumption (categorized as no/occasional, moderate [1-21 alcohol units per week for men and 14 alcohol units per week for women], and heavy [≥21 units per week for men and 14 units per week for women]), physical activity (hours per week of moderate or vigorous physical activity), and frequency of fruit and vegetable consumption (less than once daily, once daily, or more than once daily).

Health status included diabetes (fasting glucose level ≥126.1 mg/dL [to convert to millimoles per liter, multiply by 0.0555], a 2-hour postload glucose level ≥200 mg/dL, reported physician-diagnosed diabetes, or use of diabetes medication); clinically assessed cardiovascular disease including coronary heart disease (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes: I20-I25) and stroke (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes: I60-I64); cardiovascular disease medication; and antidepressant use.

Associations between GHQ-30 caseness and participant characteristics in 1985 and 2003 were examined using the t test and χ² test. Two sets of analyses were performed, described here. As there were no sex differences in effect estimates (all P for interaction between .13 and .98), men and women were combined in the analyses.

The association of GHQ-30 caseness and incidence of dementia was modeled using Cox regression; participants were censored at date of record of dementia, death (to account for competing risk of mortality), or March 31, 2015, whichever came first. These analyses were first undertaken using GHQ-30 (and covariates) drawn from 1985 and repeated using data from 1991, 1997, and 2003, with mean follow-up of 26.6, 21.7, 16.3, and 11.1 years, respectively. We also examined the association of CES-D caseness in 2003 and incidence of dementia. All analyses were first adjusted for sociodemographic measures (model 1), then also for health behaviors (model 2), and finally for health covariates (model 3).

We then examined the association of chronic/recurring depressive symptoms (never, once, or ≥2 times) over the early study period (1985, 1989, and 1991) and late study period (1997, 2001, and 2003) with incidence of dementia; covariates in these analyses were drawn from 1991 and 2003, respectively. Follow-up for dementia began in 1991 for early-phase and 2003 for late-phase depressive symptoms.

Trajectories of GHQ-30 depressive scores (range, 0-30) over 28 years were modeled using a backward timescale such that year 0 (index date) was year of dementia for dementia cases, year of death for those who died during the follow-up, and March 31, 2015 (end of follow-up), for all others. The GHQ-30 score in each of the 28 years (year 0 to year −28) was modeled using mixed-effects models with the intercept and slope as random effects.¹⁸ Dementia (coded as 1 or 0) and its interaction with slope terms (time, time-squared, and time-cubed to allow for nonlinear change) were added to the model to test for differences in GHQ-30 trajectories between those with dementia and all others. This modeling strategy implies that year 0 (the index date) was the intercept in the analysis and the beta associated with the dementia term yielded the difference in GHQ-30 score between those with and without dementia diagnosis. We examined whether the terms dementia by time, dementia by time-squared, and dementia by time-cubed improved fit of the model using the Wald test. Analyses were adjusted for age at year 0, sex, race/ethnicity, education, year of birth (5-year categories), time-dependent occupational position, and marital status. These analyses on GHQ-30 trajectories were repeated using the CES-D score (range, 0-60) within a 12-year time window prior to dementia diagnosis.

Analyses were undertaken using STATA version 14 (StataCorp) for analysis. A 2-sided P < .05 was considered statistically significant.

Of the 10 308 participants recruited to the study, data on GHQ-30 were available on 10 189 participants in 1985 and 6728 participants in 2003; the study design flowchart is in eFigure 2 in the Supplement. Of the 3461 participants lost to follow-up over this period, 15.3% had died and 0.4% had a dementia diagnosis before 2003. Those without data at the 2003 assessment were more likely to be older (44.9 vs 44.7 years in 1985; P = .04), to be women (41.1% vs 29.4%; P < .001), and not have a university degree (79.6% vs 70.9%; P < .001). Cases of dementia accrued mainly between 1995 and 2015, with 73% of cases recorded in the last 5 years of follow-up. Increasing age (hazard ratio [HR] for 1 year greater age at study baseline, 1.21; 95% CI, 1.19-1.24), female sex (HR, 1.58; 95% CI, 1.27-1.96), and education less than secondary school diploma (HR, 1.76; 95% CI, 1.41-2.19) were associated with a higher risk for dementia. Table 1 shows sample characteristics of participants in 1985 and 2003 as a function of GHQ-30 caseness.

Table 2 shows that the association of GHQ-30 caseness in 1985, 1991, 1997, and 2003 with incidence of dementia only evident for the 2003 measure of depressive symptoms (HR, 1.72; 95% CI, 1.21-2.44; model 3). These analyses are based on maximum available data, although analysis on those with depressive symptoms data at all measures yielded similar results (HRs [95% CIs] of 1.14 [0.79-1.63], 0.93 [0.60-1.44], 1.40 [0.93-2.09], and 1.86 [1.27-2.73] with GHQ-30 caseness measures in 1985, 1991, 1997, and 2003, respectively; number with dementia = 146, total number = 5783). Table 2 also shows that in fully adjusted analyses, CES-D caseness was associated with a higher risk for dementia (HR, 2.28; 95% CI, 1.53-3.39), with the mean follow-up being 11.1 years. Further analysis showed associations of dementia with late-life (mean age, 70 years) but not midlife (mean age, 50 years) GHQ-30 depressive symptoms (eTable 1 in the Supplement).

Those with chronic/recurring GHQ-30 depressive symptoms (≥2 times) in the early phase of the study (1985, 1989, and 1991) did not have a higher risk for dementia (model 3; HR, 1.02; 95% CI, 0.72-1.44) (Table 3). However, those with chronic/recurring depressive symptoms (≥2 times) over the later waves of the study (1997, 2001, and 2003) had a higher risk for dementia (model 3; HR, 1.67; 95% CI, 1.11-2.49).

The trajectory of GHQ-30 depressive symptoms, modeled as a continuous variable, over 28 years was different (Wald χ² = 126.85; P < .001) in those with dementia compared with all participants without dementia (Figure); the differences in GHQ-30 score between these 2 groups are shown in eTable 2 in the Supplement. These results showed an accelerated increase in depressive symptoms in the decade prior to dementia diagnosis; differences in depressive symptoms became apparent 11 years (difference, 0.61; 95% CI, 0.09-1.13; P = .02) before dementia diagnosis and became more than 9 times larger at the year of diagnosis (difference, 5.81; 95% CI, 4.81-6.81; P < .001) (eTable 2 in the Supplement). The CES-D score trajectory, modeled up to 12 years before dementia diagnosis, also showed an increase in depressive symptoms in the years prior to dementia diagnosis (Figure; eTable 2 in the Supplement). The robustness of the shape of the depressive symptoms trajectory was confirmed in analysis using cubic regression splines (eFigure 3 in the Supplement).

In this study of 10 189 men and women, depressive symptoms in late life but not midlife were associated with a higher risk for dementia. In effect, this association was evident only when depressive symptoms were measured in the decade preceding dementia diagnosis. Even chronic/recurring midlife depressive symptoms, assessed in the early years of the study, were not associated with an increased risk for dementia. The retrospective trajectory of depressive symptoms over 28 years shows the emergence of depressive symptoms a decade before dementia diagnosis, differences that went on to amplify more than 9 times at dementia diagnosis. Taken together, these findings are consistent with the hypothesis that depressive symptoms are a prodromal feature of dementia or that the 2 share common causes. Thus, our findings do not support the hypothesis that depressive symptoms increase the risk for dementia.

In up to 50% of people with AD or dementia, depression is comorbid.^19,20 At older ages, markers of AD pathology²¹ and cognitive impairement²² have been shown to be associated with subsequent depressive symptoms and the presence of depressive symptoms with subsequent accelerated cognitive decline.²³ It has been hypothesized that depression increases the risk for dementia through hyperactivity of the hypothalamic-pituitary-adrenal axis²⁴; prolonged exposure to glucocorticoids in persons with depression would lead to hippocampal atrophy and development of dementia.²⁵ However, there is limited support for this hypothesis in relation to hippocampal and amygdalar volume²⁶ or other neuropathologic markers.²³

Data from meta-analyses of prospective studies have found depressive symptoms in late life to be associated with an approximate 2-fold increase in the risk for dementia.^6,27 Nonetheless, these findings are compatible with several alternative explanations: a causal effect of depressive symptoms, depression being a prodromal feature, and common causes. To test the first of these explanations, studies have used long follow-ups, most recently, 17 years²⁸ and 24.7 years.²⁹ It has been suggested that severe^30,31 or repeated depressive episodes³² carry risk, although our findings do not support this hypothesis. The analytical approach in these studies is time to event analysis, such as Cox regression where even in studies with very long mean follow-up, dementia occurring in the first few years of follow-up is likely to influence estimates of associations with depressive symptoms. Furthermore, the remitting and relapsing nature of depressive symptoms is not accounted for in such analyses. Recent use of a 2-step design where latent class trajectory models on repeat data on depression are used to categorize individuals and then time to event analysis for assessment of risk for dementia shows higher risk for dementia in those with increasing levels of depression.^10,11 Even with this design, depressive symptoms in the follow-up for dementia are not taken into account in Cox regression, a limitation addressed in our analysis of the 28-year trajectory of depressive symptoms, which models them over the entire follow-up.

The need to assess the risk for dementia associated with earlier-life depressive symptoms, before 60 years of age, was emphasized in a recent review.²⁰ The evidence from these studies is inconclusive as studies show no association with early-life depression,³³ and when they do, the use of a 1-item measure of depressive symptoms,³⁴ self-report of previous depression,³⁵ or associations seen only in men³⁶ do not allow firm conclusions to be drawn. Our results with depressive symptoms assessed once (Table 2), 3 times before age 60 years (Table 3), or at age 50 years (eTable 1 in the Supplement) provide no robust evidence of increased risk for dementia. A complementary analytic framework, reflected in our analysis of depressive trajectories over 28 years, up to the year of dementia diagnosis is novel and suggests that a rapid increase in depressive symptoms over 10 years before dementia diagnosis may be the primary explanation for the association of depressive symptoms and dementia at older ages. One previous study used a similar analytic approach in a cohort of adults 65 years and older followed up for 14 years to assess prodromal changes in AD and showed the emergence of depressive symptoms 8 years before diagnosis.¹⁸

Besides the prodromal explanation, it is possible that risk factors common to both depression and dementia explain the observed association between these conditions. Impairment in memory, sleep disturbances, and impaired social functioning are common to both conditions and common pathophysiological pathways, such as neurodegeneration, inflammation, vascular risk factors, and hypothalamic-pituitary-adrenal axis dysregulation, may well explain their association. Irrespective of the nature of the association, the comorbidity of depressive symptoms and dementia is well established and needs to be taken into consideration in the care of patients with cognitive impairment or dementia. For now, it is important to determine whether treatment for depression improves cognitive functioning.

Our findings need to be considered in light of the study’s strengths and limitations. The major strength of this study is the use of data on depressive symptoms covering a period of 28 years, which allowed us to assess both the risk associated with symptoms early and late in life and to model their trajectories over 28 years prior to dementia diagnosis. Despite the limitations of GHQ-30, the association between late-life depressive symptoms and dementia in this study is similar to that reported in the literature.⁶ Furthermore, the analysis of trajectories shows its ability to reflect changes in depressive symptoms over time. A limitation of the study is ascertainment of dementia being based on linkage to electronic health records. While the specificity of cases with this method is likely to be high, the sensitivity is undoubtedly low as only half of patients living with dementia have a documented diagnosis.⁴ As dementia ascertainment in our study was independent from timing and report of depressive symptoms, major bias due to undetected dementia is unlikely. We were unable to examine the subcategories of dementia due to small numbers but previous reports show similar findings for the association of depression with dementia subtypes.^26,33

Depression is common at older ages and often comorbid with many chronic diseases, and it is associated with greater risk for mortality, higher health care costs, and disability. However, our study provides no support for the hypothesis that depressive symptoms increase the risk for dementia. The observed association between the 2 is likely to be due to common causes or the effects of preclinical dementia.

Accepted for Publication: March 2, 2017.

Corresponding Author: Archana Singh-Manoux, PhD, INSERM U1018, Centre for Research in Epidemiology and Population Health, Hôpital Paul Brousse, Bât 15/16, 16 Avenue Paul Vaillant Couturier, 94807 Villejuif Cedex, France (archana.singh-manoux@inserm.fr).

Published Online: May 17, 2017. doi:10.1001/jamapsychiatry.2017.0660

Author Contributions: Drs Singh-Manoux and Sabia had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Singh-Manoux, Dugravot, Ebmeier, Kivimäki, Sabia.

Acquisition, analysis, or interpretation of data: Singh-Manoux, Fournier, Abell, Ebmeier, Kivimäki, Sabia.

Drafting of the manuscript: Singh-Manoux.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Singh-Manoux, Dugravot, Sabia.

Obtained funding: Singh-Manoux, Ebmeier, Kivimäki.

Administrative, technical, or material support: Singh-Manoux, Abell.

Study supervision: Singh-Manoux, Kivimäki.

Conflict of Interest Disclosures: None reported.

Funding/Support: The Whitehall II study is supported by grants R01AG013196 and R01AG034454 from the US National Institute on Aging, grant MRC K013351 from the UK Medical Research Council, and grant RG/13/2/30098 from the British Heart Foundation. Dr Kivimäki is supported by the Medical Research Council and NordForsk.

Role of the Funder/Sponsor: The funding bodies had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank all of the participating civil service departments and their welfare, personnel, and establishment officers; the British Occupational Health and Safety Agency; the British Council of Civil Service Unions; all participating civil servants in the Whitehall II study; and all members of the Whitehall II study team. The Whitehall II Study team comprises research scientists, statisticians, study coordinators, nurses, data managers, administrative assistants, and data entry staff who make the study possible.