Intertwined depressive and cognitive trajectories and the risk of dementia and death in older adults: a competing risk analysis

Study population

The present study relies on data from the Health and Retirement Study (HRS), a nationally representative cohort that collected comprehensive socioeconomic and health information from US adults aged 50 years and older. The baseline survey was conducted in 1992, with subsequent follow-up visits every 2 years. Reinterview rates remained between 80% and 89%, and proxies (9%) were used when respondents were unable to complete the survey on their own. The HRS was supported by the National Institute on Aging (NIA U01AG009740) and was approved by the Institute for Social Research and Survey Research Center of the University of Michigan (IRB Protocol: HUM00061128).11 All participants provided written informed consent, and detailed information can be found at https://hrs.isr.umich.edu/data-products/restricted-data/available-products/11516.

In this study, we used the HRS 2010 as the baseline, with follow-ups until 2020. The cognitive and depressive trajectories were fitted using the HRS 1996–2010. Figure 1 shows the flowchart of this study. Of the 22 034 participants recruited in 2010, those who were aged <50 or had missing data on education (n=904); who had missing data on depressive symptoms or cognitive function in 2010 (n=1367); and who had prior all-cause dementia or AD (n=341) were excluded, leaving 19 422 analytical samples.

Request permission

Measurement of depressive symptoms

The depressive symptoms in the HRS were assessed using an 8-item subset of the Centre for Epidemiologic Studies Depression Scale (CESD-8) which consisted of ‘yes/no’ responses to items including: (1) Felt depressed; (2) Everything was an effort; (3) Sleep was restless; (4) I was happy; (5) Felt lonely; (6) I enjoyed life; (7) Felt sad; and (8) Could not get going. Participants who answered ‘yes’ were assigned as 1. The validity and reliability of CESD-8 have been confirmed among older adults.12 The final CESD score, ranging from 0 to 8, was calculated by summing up the participants’ responses to each item, with higher scores indicating more severe depressive symptoms. Participants with a CESD score of 4 or higher were defined as having depressive symptoms at baseline.13

Subsequently, the CESD scores in 2010 were calculated as z scores based on the mean (standard deviation (SD)) in 2010: (CESD scores: 1.50 (2.01)). Moreover, CESD scores from 1996 to 2008 were also converted to z scores based on the mean (SD) in 2010.

Measurement of cognitive function

The cognitive function in the HRS study was measured using a modified version of the Telephone Interview of Cognitive Status, which included episodic memory, serial-7 number subtraction questions and counting backward tests.14 Episodic memory was determined by the sum of immediate and delayed word recalls using 10 random words. The serial-7 number subtraction questions referred to five serial subtractions of 7 from 100 (0–5). For the counting backward test, participants had to count backward from 20 as fast as possible. Scoring was based on consecutive counts on the first attempt (two points) and successful counts on the second attempt (one point). The total cognitive function score ranged from 0 to 27, with a higher score indicating better cognitive function.

The cognitive scores in 2010 were then calculated as each-5-year-age-stratified and education-stratified z scores ((cognitive scores–means)/SDs) according to online supplemental table S1. Those with a cognitive z score in 2010 of <−1 were defined as having mild cognitive impairment at baseline. Subsequently, the cognitive scores from 1996 to 2008 were converted to z scores based on the means (SDs) in 2010.

Measurement of dementia and death

All-cause dementia and AD were defined by self-reported or proxy-reported physician diagnoses from 2010 to 2020. The vital status of participants was determined by extracting data from the respondent’s exit interview or the interview of their spouses/partners, which recorded the year and month of death.

Measurement of covariates

At baseline, information on age, sex (men, women), race (white/Caucasian, black or others), education (below high school, high school, college, university and above), household income (bottom, middle, top tertiles), marital status (married, single), smoking history (ever, never), drinking history (ever, never), vigorous exercise (never, ≤1 day per week, >1 day per week), weight status (normal (body mass index <25 kg/m²), abnormal (body mass index ≥25 kg/m²)) and a number of chronic diseases (hypertension, diabetes, cancer, lung disease, heart disease, stroke, psychiatric problems and arthritis, divided into 0, 1, and 2 or more) was collected through questionnaires. Participants who had missing information on race, marital status, smoking history, vigorous exercise, weight status and a number of chronic diseases were classified as ‘missing’.

Statistical analysis

The baseline characteristics of included participants were described as medians with interquartile ranges for continuous variables given their non-normal distribution and frequency and percentage (%) for categorical variables. To compare characteristics, differences in continuous or ordered categorical variables were assessed by Kruskal-Wallis tests while those of binary or unordered categorical variables were assessed by χ² tests.

Based on the ‘rms’ package in R,15 the Cox proportional hazard regression-based restricted cubic spline (RCS) was used to investigate the dose-risk associations (hazard ratio (HR) and 95% confidence interval (CI)) of depressive and cognitive z scores in 2010 with new-onset all-cause dementia, AD and all-cause death from 2010 to 2020. The RCS used four knots at prespecified locations according to the percentiles of the distribution of exposure: 5%, 25%, 50% and 75%. Subsequently, the mild cognitive impairment-stratified dose-risk associations of depressive z scores in 2010 with new-onset all-cause dementia, AD and all-cause death from 2010 to 2020, and depressive symptoms-stratified dose-risk associations between cognitive z scores in 2010 and these outcomes were also explored using the RCS. The interactions between depressive and cognitive z scores were assessed using their multiplication term, that is, ‘depressive z scores|cognitive z scores’. The p values of the multiplication term less than 0.05 referred to significant interactions. All these associations were fully adjusted for age, sex, race, education, household income, marital status, smoking history, drinking history, vigorous exercise, weight status and number of chronic diseases. Furthermore, stratified analyses of men and women were conducted to explore the sex differences in these associations.

Group-based trajectory modelling (GBTM) was further conducted to analyse the independent trajectories of depressive symptoms and cognitive function from 1996 to 2010. In this study, participants with four or more time-point depressive and cognitive data were used to conduct GBTM, based on the ‘proc traj’ procedure in the SAS.16 Two-order and three-order trajectory models were set to explore the 14-year developmental trajectories of depressive symptoms and cognitive function. Models with lower Bayesian information criteria, lower Akaike’s information criterion and higher entropy were selected. Furthermore, the fully adjusted associations (subdistribution HR (sHR) and 95% CI) of independent depressive and cognitive trajectories with new-onset all-cause dementia, AD and all-cause death were investigated using the subdistribution hazard regression. Subdistribution hazard regression is a type of competing risk analysis introduced by Fine and Gray, which is preferable when estimating actual risks and prognosis.17 In this study, we viewed all-cause death as the competing event of all-cause dementia and AD to conduct the subdistribution hazard regression, based on the ‘cmprsk’ package in R.18 Sensitivity analysis based on complete-case analysis was further conducted to verify the robustness of the data imputation.

Subsequently, the GBTM was used to identify the joint trajectories of depressive symptoms and cognitive function from 1996 to 2010. The orders of depressive and cognitive trajectories were the same as those selected in the independent trajectory analysis. The fully adjusted associations of the joint trajectories with new-onset all-cause dementia, AD and all-cause death after 2010 in total participants, men, and women were also investigated using subdistribution hazard regression.

Reporting of this study was done under Strengthening the Reporting of Observational Studies in Epidemiology guidelines. Analyses were performed using SAS (V.9.4, SAS Institute, licensed to Peking University School of Public Health-R&D LIC) and R statistical software V.4.2.3 (R Project for Statistical Computing). All analyses were two-sided.

Main findings

In this cohort study, we found that rapid cognitive decline was more common in older adults with lower depressive symptoms. Interestingly, older adults who showed significant disparities between depressive status and cognitive status, such as no depressive symptoms but poor cognitive function, exhibited comparable or even higher risks of dementia and all-cause death when compared with those who had long-term comorbid depressive symptoms and cognitive decline, particularly evident among women.

Existing research has established the positive associations of increasing depressive symptoms and accelerated cognitive decline with subsequent dementia and death.7 8 Depressive symptoms and cognitive decline are both associated with changes in brain structure and function. The reduced cortical thickness and surface area can contribute to cognitive decline.19 Depressive symptoms, on the other hand, affect brain regions involved in emotional regulation and cognitive processing.20 All of these are precursors or risk factors for dementia. Furthermore, depressive symptoms and cognitive decline can lead to chronic stress and trigger an inflammatory response.21 Elevated inflammation and oxidative stress can further increase the risk of dementia by damaging neurons and promoting the formation of harmful proteins in the brain.22 23 Additionally, depressive symptoms and cognitive decline can result in social isolation and a lack of mental stimulation, which, in turn, can contribute to cognitive decline and heighten the risk of developing dementia.24

This study further underscored the significance of cognitive decline in older adults without depressive symptoms as a comparable or even more potent risk factor for subsequent dementia and death. Older adults with depressive symptoms have been found to display abnormal levels of neurotransmitters, such as dopamine, norepinephrine and serotonin, in their brains, which have a consequential impact on their brain function.20 25 In addition, those with depressive symptoms are more susceptible to chronic stress, sleep disorders and other issues that may negatively impact brain health.21 26 Therefore, cognitive decline in them may be more closely associated with the depressive symptoms themselves and the biological changes they elicit, rather than being directly related to the development of dementia. On the other hand, those without depressive symptoms may not experience these depression-related biological changes. Consequently, if they experience cognitive decline, it may be more attributable to age-related and gene-related biological changes, including neuronal apoptosis and brain atrophy, which may be more directly associated with the development of dementia.27 In addition, older adults with depressive symptoms may seek medical attention or make lifestyle changes due to their symptoms and thus reverse their accumulated risk factors and reduce their subsequent risk of dementia and death. Conversely, those without depressive symptoms may not display any apparent symptoms before their cognitive decline occurs, thus potentially missing the optimal treatment window, which may also help explain why the cognitive decline in older adults without depressive symptoms is associated with a higher risk of dementia and death.

Notably, the interactions between depressive symptoms and cognitive function in associations with dementia exhibited greater significance in women compared with men. Women are more susceptible to experiencing depressive symptoms and are typically more attuned to their mental health status than men.28 As mentioned above, depressive symptoms can contribute to cognitive-related problems, such as attention and memory problems. Consequently, women with depressive symptoms may be more inclined to recognise cognitive decline and seek medical intervention. However, in women without depressive symptoms, cognitive decline may be less noticeable, leading to a progressive deterioration of cognitive function. In addition, the decline in oestrogen levels after menopause may expedite cognitive decline and neurodegeneration in women, thereby amplifying their risk of developing dementia.29 Finally, sex differences in social roles may also make sense. For instance, women typically undertake more caregiving responsibilities at home,30 which can result in greater stress and reduced opportunities to engage in cognitive activities, potentially increasing the risk of cognitive decline and dementia.

To the best of our knowledge, this is the first and most comprehensive population-based cohort study to explore the interactions and joint trajectories of depressive symptoms and cognitive function in their associations with all-cause dementia, AD and all-cause death in older US adults. In contrast to prior evidence, the current study is groundbreaking in examining the dynamic interactions between depressive symptoms and cognitive function before new-onset dementia and death. Specifically, we identified a high-risk population characterised by significant disparities between depressive status and cognitive status. Furthermore, 15 common joint trajectories of depressive symptoms and cognitive function were identified in this study. The complex statistical methods, including RCS curves, GBTM and competing risk analysis, guarantee the robustness of our conclusions. The sex-stratified analysis in this study helps explore sex differences and ensures the reliability of the findings. The large sample size and excellent population representation of the HRS further ensure the extrapolation of our findings.

Limitations

Several limitations of our study need to be acknowledged. First, the diagnosis of dementia and AD relied on self-reported or proxy-reported physician diagnoses, which may result in an underestimation of the prevalence. Second, we were not able to adjust for certain potential confounders due to data limitations, such as waist circumference. Lastly, the apolipoprotein E allele was not adjusted in our analysis to retain a larger sample size.

Implications

In conclusion, our findings suggest that older adults with significant disparities between depressive and cognitive status, such as low depressive symptoms but poor cognitive function, are at comparable or even higher risk of subsequent dementia and all-cause death, relative to those with long-term comorbid depressive symptoms and poor cognitive function. It is crucial to recognise the importance of depressive and cognitive trajectories and their interactions in the early prevention of dementia and all-cause death among older adults.