Trajectories of depressive symptoms and associated patterns of cognitive decline

The global population aged 60 years or older is rapidly growing, with associated challenges including declining cognitive abilities, which are linked to mortality, functional impairment, dementia, increased healthcare costs, and reduced quality of life. Depression is frequently co-occurring with cognitive impairment in older adults and is particularly associated with elevated dementia risk. The temporal nature of the depression–cognition association remains unclear: depression may contribute causally to later cognitive impairment, may reflect prodromal phases of neurocognitive disorders, or be a psychological response to perceived cognitive changes. Proposed mechanisms include neurotoxic effects of depression (e.g., reduced brain-derived neurotrophic factor), increased amyloid and tau pathology in those with a history of depression, and shared cerebrovascular comorbidities and microvascular changes. This study aimed to examine whether distinct longitudinal trajectories of depressive symptoms are associated with different patterns and rates of cognitive decline in a large, multinational cohort of middle-aged and older adults.

Prior work shows mixed evidence on whether depression accelerates cognitive decline and for whom. Biological links include reduced BDNF in depression, increased hippocampal plaques and tangles in individuals with lifetime depression, and microRNAs associated with late-life depressive symptoms also linked to cognitive decline. Epidemiological studies have variously found that depression can precede cognitive impairment, reflect prodromal dementia, or be a reaction to cognitive changes. Studies of depressive symptom trajectories have identified groups such as low, increasing, moderate declining, and high, with differing associations to cognition and dementia risk. For example, in the English Longitudinal Study of Ageing, persistent depressive symptoms were associated with steepest cognitive decline; in Health ABC and Rotterdam cohorts, high and increasing, or specifically increasing depressive symptom trajectories, were associated with higher dementia risk. Trajectories of cognitive decline themselves may be linear or non-linear; prior analyses (e.g., Yale Precipitating Events Project, Cambridge City over 75 Cohort) identified heterogeneous, often accelerating patterns, underscoring the need for flexible modeling to capture fluctuating decline. This study builds on these findings by jointly considering depressive symptom trajectories and detailed patterns of cognitive change in a large, diverse European cohort.

Design and data source: Prospective cohort study using the Survey of Health, Ageing and Retirement in Europe (SHARE), waves 1, 2, 4, 5, 6, and 7 (2004–2017), with approximately biennial follow-ups. Participants: Community-dwelling adults aged ≥50 years (and their spouses), from 18 countries. Ethical approvals obtained; informed consent collected. Analytical sample: From 139,556 individuals with ≥1 SHARE interview, exclusions were applied for only one interview (n=49,301), insufficient depressive symptom data (n=17,266), insufficient cognitive data (n=1,056), age <50 years (n=2,497), and misaligned baseline measures (n=370), yielding 69,066 participants (mean age ~64 years; 55% women). Median follow-up was 6 years (range 1–12). At least three measures were available for 56% (EURO-D, verbal fluency) and 77% (immediate and delayed recall). Measures: Depressive symptoms were assessed with the EURO-D scale (12 items; score 0–12; higher indicates more symptoms) in waves 1, 2, 4, 5, 6, 7. Cognition was assessed with three measures: verbal fluency (animal naming in 1 minute; count), immediate recall (0–10 words immediately after list), and delayed recall (0–10 words at end of session). Alternate test versions were used across waves to limit practice effects. Covariates: Sociodemographic (age, sex, education per ISCED-97 categories, household net worth, job status, family status, numbers of children/grandchildren) and health-related factors (treatment for depression/anxiety, limitations in instrumental activities of daily living, number of chronic diseases, BMI, physical inactivity, smoking history, alcohol use, maximal grip strength). Time-varying covariates (e.g., BMI, physical activity, grip strength, net worth, marital and job status, IADL, chronic diseases, treatment) were retrieved per wave. Deriving depressive symptom trajectories: Latent growth mixture modeling (LGMM) with a latent basis growth model (freely estimated slopes at each time point) identified latent classes. Model selection followed recommended criteria; a 4-class solution was selected: constantly high (4.3%), constantly low (71.9%), decreasing (13.9%), increasing (9.9%). Baseline covariates were included via a 3-step approach in multinomial regression to predict class membership without altering class structure. Statistical analysis of cognition: Descriptive statistics compared baseline characteristics across depressive trajectories. For cognitive change patterns, smoothing splines mixed effects models (p-spline basis) visualized potentially non-linear trajectories by depressive class. Rates of cognitive decline were estimated using linear mixed effects models with random intercepts (individuals) and random slopes (time in years since baseline), stratified by depressive trajectory to obtain class-specific time coefficients. Primary models adjusted for baseline age, sex, education, and country of origin; extended models additionally adjusted for socioeconomic, behavioral, and clinical covariates. Outcomes were examined using raw scores and wave-specific z-score standardization to compare magnitudes across domains. Secondary analyses stratified by baseline age (<65 vs ≥65) and by region (CEE, Scandinavia, Southern Europe, Western Europe, Israel). Granger causality tests assessed temporal precedence between changes in cognition and depressive symptoms. Analyses used R 3.6.0 and Mplus 8.4; p<0.05 considered significant.

• Four depressive symptom trajectories were identified: constantly low (n=49,660; 71.9%), constantly high (n=2,999; 4.3%), increasing (n=6,828; 9.9%), decreasing (n=9,579; 13.9%).
• Baseline cognition: Constantly low depressive symptoms had the highest baseline cognitive scores (immediate recall 5.4±1.7; delayed recall 4.0±2.0; verbal fluency 21.0±7.4). Constantly high had the lowest (4.2±1.8; 2.7±2.0; 15.8±7.2). Increasing and decreasing groups had similar baseline cognition (~4.9±1.8 immediate recall; ~3.4±2.1 delayed recall; ~18.5–18.6±7.4–7.5 verbal fluency).
• Pattern of cognitive change: Smoothing spline models showed near-linear decline in cognition for the constantly high and increasing depressive symptom groups, versus highly fluctuating (non-linear) patterns for constantly low and decreasing groups.
• Rates of decline (primary models adjusted for age, sex, education, country): The steepest decline across all domains occurred in the increasing depressive symptoms group, followed by constantly high. Constantly low and decreasing groups showed smaller declines; for decreasing symptoms, recall tests showed no significant decline in some models.
• Z-score standardized declines (β per year; 95% CI): • Immediate recall: Constantly high −0.03 (−0.04, −0.03); Constantly low −0.02 (−0.02, −0.02); Increasing −0.04 (−0.04, −0.04); Decreasing −0.02 (−0.02, −0.01). • Delayed recall: Constantly high −0.03 (−0.03, −0.02); Constantly low −0.02 (−0.02, −0.02); Increasing −0.04 (−0.04, −0.03); Decreasing −0.02 (−0.02, −0.02). • Verbal fluency: Constantly high −0.03 (−0.03, −0.03); Constantly low −0.02 (−0.02, −0.01); Increasing −0.04 (−0.04, −0.04); Decreasing −0.01 (−0.02, −0.01).
• Multivariable adjustment (adding socioeconomic, behavioral, clinical covariates) attenuated decline rates but most remained significant. Exceptions: recall declines in the constantly high group and verbal fluency decline in the decreasing group were no longer statistically significant in fully adjusted models.
• Secondary analyses: Across age strata (<65, ≥65) and regions, increasing symptoms were associated with the fastest cognitive decline; exception was Israel, where constantly high symptoms showed the fastest decline.
• Temporal ordering: Granger causality tests suggested changes in cognitive functioning precede changes in depressive symptoms across cognitive domains.
• Sociodemographic/health profile: Compared with constantly low symptoms, the constantly high, increasing, and decreasing groups were more likely women and had worse socioeconomic and health profiles.

The study demonstrates that both the pattern (linear vs fluctuating) and rate of cognitive decline differ by longitudinal depressive symptom trajectories. Individuals with increasing depressive symptoms experienced the most pronounced and linear cognitive decline, suggesting that late-life escalation of depressive symptoms may be a manifestation of underlying neurodegenerative processes rather than a causal driver. Constantly high depressive symptoms also tracked with linear decline, supporting potential neurotoxic or vascular contributions of chronic depression to cognitive deterioration. In contrast, constantly low and decreasing depressive symptom trajectories exhibited fluctuating, generally milder cognitive change, implying relative preservation of cognitive health and/or reduced exposure to neurotoxic effects of sustained depression. Granger analyses indicated that cognitive changes precede depressive symptom changes, aligning with the hypothesis that emerging cognitive impairment may trigger depressive symptoms in late life. These findings refine the depression–cognition relationship by identifying subgroups with distinct decline profiles and suggest that unique biological mechanisms may underlie these trajectories, with implications for targeted monitoring and intervention.

In a large, multinational cohort, four depressive symptom trajectories were identified that corresponded to distinct patterns and rates of cognitive decline. The greatest and most linear decline occurred among individuals with increasing depressive symptoms, followed by those with persistently high symptoms; decreasing and persistently low symptoms were associated with smaller, fluctuating declines. These results highlight the importance of tracking depressive symptom trajectories to identify individuals at heightened risk for accelerated cognitive decline. Public health and clinical strategies aimed at reducing depressive symptoms—particularly preventing late-life increases—may help mitigate cognitive decline. Future research should: (1) apply more robust causal modeling (e.g., autoregressive latent trajectory models) to clarify directionality; (2) investigate biological underpinnings of trajectory-specific decline; (3) evaluate whether effective depression treatment can slow cognitive decline, especially in the increasing-symptom subgroup; and (4) develop tailored interventions based on trajectory membership and regional/contextual factors.

• Assessment intervals (~2 years) may miss within-individual fluctuations in depressive symptoms and cognition.
• EURO-D is self-reported; cognitive changes may influence symptom reporting (measurement bias).
• SHARE participants tend to be healthier and more educated than the general population, potentially underestimating depression prevalence and cognitive decline magnitude (selection bias, limited generalizability).
• Cognitive outcomes were limited to verbal fluency and episodic memory (immediate/delayed recall), not covering all cognitive domains.
• Directionality analyses used Granger causality; more sophisticated longitudinal causal models may be needed to fully elucidate the depression–cognition relationship.