Depressive Symptoms Longitudinally Mediate the Effect of Hyperglycemia on Memory Decline in Type 2 Diabetes

Type 2 diabetes (T2D) increases risk for cognitive decline and dementia in older adulthood (1,2), with chronic hyperglycemia representing a primary pathophysiological mechanism underlying T2D-related effects on cognitive health (3). In samples of older adults with T2D, higher levels of glycated hemoglobin (HbA_1c) have been prospectively linked to incident cognitive impairment (1), particularly memory decline (4,5), although findings have been equivocal (for reviews, see the articles by Ganmore and Beeri (6) and Palta et al. (7)). Identifying potential pathways that link HbA_1c to memory aging in older adults with T2D may help clarify inconsistent findings and point to potential targets to reduce T2D-related cognitive morbidity.

Depressive symptoms may represent one pathway underlying hyperglycemia-related memory decline in T2D. Systematic reviews and meta-analyses provide robust evidence of T2D and/or high HbA_1c as a risk factor for elevated depressive symptoms and the onset of clinical depression in older adults (8–10). Higher levels of HbA_1c may increase T2D self-care burden, which may negatively impact mood (10), and hyperglycemia-induced fatigue may contribute to activity disengagement and low mood (10,11). In turn, worse depressive symptoms have been linked to memory impairment and decline (12,13), although studies in T2D-specific samples are relatively sparse (14,15). Depressive symptoms may influence cognition, especially memory, via behavioral and stress-related pathways involving increases in glucocorticoid secretion, which damage brain structures such as the hippocampus that subserve memory functions (16–20).

Evidence for the potential mechanistic role of depressive symptoms in diabetes-related memory decline has been suggested by disparate bodies of literature. Longitudinal studies have prospectively linked HbA_1c to depressive symptoms (21,22), and separate investigations have linked depression and depressive symptoms to poor episodic memory and dementia (12,13), including in individuals with T2D (15). Depressive symptoms have been shown to mediate the longitudinal association between cardiometabolic dysregulation and global cognitive function (23). However, T2D-specific associations and associations with episodic memory, a cognitive domain commonly implicated in T2D that is also a primary determinant of dementia risk, remain unexplored. Thus, additional research is needed to clarify whether depressive symptoms may longitudinally mediate the association between T2D-specific hyperglycemic effects on episodic memory change.

Of note, although authors of a previous study of older adults with type 1 diabetes and T2D observed a mediating role of depressive symptoms in T2D-related memory dysfunction, the study’s cross-sectional design precludes interpretations of directional associations (24). Indeed, a smaller body of literature suggests a different direction of effects because depression may contribute to elevated HbA_1c and incident T2D (25). It is plausible that for older adults with T2D, clinically significant depression or depressive symptoms may lead to decreased diabetes self-care, which, in turn, may be associated with higher levels of HbA_1c. Additionally, HbA_1c levels in older adults with T2D may be influenced in part by cognitive impairment (26). For instance, poor memory may lead to lapses in self-care activities (e.g., forgetting to take medication or monitor glucose), which may worsen glycemic levels. Discordant patterns of association among HbA_1c, depressive symptoms, and memory in T2D highlight the critical need for longitudinal studies to test hypothesized models as well as alternative models with opposite directions of association. Furthermore, examining depressive symptoms as a mediator of T2D-specific hyperglycemic effects on memory change may aid in clarifying mechanisms underlying T2D-related cognitive morbidity and help in the early identification of older adults at risk for accelerated cognitive decline. Additionally, understanding the potentially mediating role of depressive symptoms may help tailor the design and timing of interventions for older adults.

Thus, in this study, we used a longitudinal mediation framework to examine whether depressive symptoms mediate the prospective association between HbA_1c and subsequent memory change in older adults with T2D from the U.S.-wide Health and Retirement Study (HRS). We hypothesized that incident elevations in depressive symptoms would at least partially mediate the association between HbA_1c and 6-year changes in memory.

The HRS is a representative sample of ∼20,000 people from across the United States, who are aged ≥51 years and who have been followed and interviewed biennially since 1992, with sample refreshment every 6 years (https://hrsdata.isr.umich.edu). Data are collected by phone or during face-to-face interviews by trained staff. In 2006, the HRS initiated an enhanced face-to-face interview, which included blood-based data collection from half of the participants, randomly selected, and the other half was eligible to participate in 2008. For both sample halves, blood-based data were collected again 4 years later (i.e., 2010 for participants seen in 2006; 2012 for participants seen in 2008). Cognitive and survey data were collected biennially from all participants.

In the present study, we extracted data from HRS participants between 2006 and 2014. To maximize sample size, data obtained from random sample halves (i.e., from the present study’s 2006 and 2008 cohorts) were combined to form our baseline time point (T1), data from 2010 and 2012 were combined to form a second time point (T2), and data from 2012 and 2014 were combined to form a third time point (T3). At T1 (2006/2008), participants who completed blood-based data collection, self-reported a T2D diagnosis, had HbA_1c ≥5.7% (≥39 mmol/mol), and did not have missing covariate data were included in analyses. HbA_1c ≥5.7% was selected because it is the threshold for nonnormal glycemic level (27). Self-reported T2D diagnosis was ascertained with responses (yes/no) to “Has a doctor ever told you that you have diabetes or high blood sugar?” All participants provided written informed consent; all study procedures were approved by the University of Michigan institutional review board. This research was conducted in accordance with the Declaration of Helsinki.

Episodic memory is assessed biennially with a 10-word list-learning task shown to have adequate validity (28) and reliability (coefficient = 0.85). Participants recall the words immediately and after a 5-min delay. We calculated z scores from raw scores on the immediate and delayed trials using means and SDs at T1. Next, z scores were averaged into a composite to improve the reliability of the episodic memory outcome. For the random sample halves of the participant cohort whose HbA_1c was measured in 2006 or 2008 (i.e., the present study’s two baseline cohorts), we modeled three time points of episodic memory data obtained over 6 years (T1: 2006/2008; T2: 2010/2012; T3: 2012/2014). Episodic memory was modeled as a continuous variable.

HbA_1c was used to index levels of blood glucose and measured in capillary blood via dried blood spot expressed onto specially treated filter paper that ultimately was shipped to laboratories for analysis. The process was constructed so that no special temperature control was needed to preserve the values of specimens. Details of HbA_1c collection and assay procedures have been previously described (29). Per HRS recommendations, we used HbA_1c values equivalent to venous blood values in the National Health and Nutrition Examination Survey. Analyses were conducted using HbA_1c as a continuous variable (%). Primary analyses used HbA_1c data collected at T1. Sensitivity analyses additionally included HbA_1c data collected at T2.

Depressive symptoms (a continuous variable) were assessed with the eight-item Center for Epidemiologic Studies Depression scale (30) modified to a yes/no format. Higher scores reflect more depressive symptoms. In the present study, we included two time points of depressive symptoms assessed 4 years apart (T1 and T2).

Covariate data were self-reported and obtained at T1. Age was participants’ age in years. Sex was dichotomized (reference: male). Education was self-reported in years (0–17). Wealth was the sum of assets (e.g., stocks, bonds) minus debts (e.g., loans, mortgages). Race and ethnicity were dummy coded into mutually exclusive categories (non-Latinx Black, Latinx of any race, and non-Latinx of any other race), with the largest group (non-Latinx White) as the reference. Chronic disease comorbidities was the sum of the following physician-diagnosed (per self-report) conditions: arthritis, hypertension, lung disease, cancer, and cardiovascular problems (continuous variable). General medication adherence was operationalized using the question “Do you regularly take your prescription medication?” and responses (yes/no) were dummy coded. T1 year (2006/2008) was dummy coded (reference: 2006).

Analyses were conducted using Mplus, version 8. Full information maximum likelihood was used to handle missing data, including attrition and death. Full information maximum likelihood is a theory-based approach to missing data that includes participant data in the analysis regardless of whether they responded to every question in every wave. Descriptive information on missing data is provided in Supplementary Table 1. Model fit was evaluated with commonly used indices: comparative fit index, root-mean-square error of approximation, and standardized root-mean-square residual. Comparative fit index >0.95, root-mean-square error of approximation <0.06, and standardized root-mean-square residual <0.05 determined adequate model fit (31).

Memory at T3 was regressed onto depressive symptoms at T2, which was regressed onto HbA_1c at T1. All autoregressive paths (e.g., from memory at T2 to memory at T3) were included. Within the same model, cross-sectional associations among HbA_1c, depressive symptoms, and memory were examined by regressing memory at T1 onto depressive symptoms and HbA_1c at T1 and by regressing depressive symptoms at T1 onto HbA_1c at T1. All variables were regressed onto covariates. “Indirect effects” reflect the product of all regression coefficients within a given pathway from HbA_1c to memory, independent of covariates. “Direct effects” reflect associations between HbA_1c and memory, independent of mediators and covariates. “Total effects” reflect the sum of direct and indirect effects, corresponding to the association between HbA_1c and memory, independent only of covariates.

A series of sensitivity analyses were conducted to test the robustness of associations (Supplementary Table 2). First, we conducted analyses (models 1A and 1B) to examine the directionality of associations. In model 1A, the exposure and mediator were swapped, with depressive symptoms (at T1) as the exposure and HbA_1c (at T1 and T2) as mediators. Next, in model 1B, we tested whether memory at T1 predicted change in HbA_1c from T1 to T2, using latent difference score analysis (32).

Second, we examined models (cross-sectional and longitudinal paths) that modified inclusion criteria for HbA_1c (models 2A and 2B) and T2D diagnosis (model 2C). In model 2A (n = 1,298), we used a higher cutoff for HbA_1c at T1 (sensitivity analysis cutoff: HbA_1c ≥6.5% (≥48 mmol/mol) versus primary analysis cutoff: HbA_1c ≥5.7% (≥39 mmol/mol)). Although somewhat arbitrary, HbA_1c ≥6.5% was selected because it is above the typical clinical target (absent comorbidities and other complexities) and represents the criterion for T2D diagnosis (27). In model 2B (n = 2,560), participants were those who self-reported T2D diagnosis regardless of HbA_1c level at T1 (i.e., no cutoff for HbA_1c). In 2C (n = 5,682), the analytic sample included all participants (i.e., with and without T2D) who had at least prediabetes HbA_1c levels (≥5.7% (39 mmol/mol)) at T1 (27). Third, analyses excluded hypertension and cardiovascular conditions as covariates as they may be in the causal path from T2D to cognition. In model 3A, hypertension was excluded; in model 3B, cardiovascular conditions were excluded; and in model 3C, hypertension and cardiovascular conditions were excluded.

The datasets generated and analyzed in this study are available for download from the HRS (https://hrs.isr.umich.edu).

Participant inclusion and characteristics of the 2,155 individuals in the present study are provided in Fig. 1 and Table 1. Mean age was 69 years and mean years of education was 12. Approximately 55% of participants were women, 19% were non-Latinx Black, and 14% were Latinx participants. Bivariate correlations among exposure, mediator, and outcome variables are provided in Supplementary Table 3.

The longitudinal mediation model fit well: root-mean-square error of approximation was 0.052 (90% CI 0.034–0.071), comparative fit index was 0.993, and the standardized root-mean-square residual was 0.009. Standardized estimates for paths are provided in Fig. 2. Data on the total effects, direct effects, and specific indirect effects of interest are provided in Table 2.

As depicted in Fig. 2, concurrent associations showed that higher HbA_1c at T1 was associated with more depressive symptoms at T1, which, in turn, was associated with lower episodic memory performance at T1. Independent of depressive symptoms, there was a significant, direct negative effect of HbA_1c on concurrent memory. Longitudinal associations from the same model showed that 4-year change in depressive symptoms longitudinally mediated the association between HbA_1c and episodic memory 6 years later at T3. Specifically, HbA_1c at T1 was associated with more depressive symptoms at T2, controlling for depressive symptoms at T1. In turn, depressive symptoms at T2 predicted lower episodic memory at T3, controlling for episodic memory at T1 and T2. Overall, depressive symptoms accounted for 19.0% of the total effect of HbA_1c on 6-year memory decline independent of covariates. After accounting for indirect effects, there was no significant direct effect of HbA_1c on episodic memory at T3.

Subsequent models tested a series of sensitivity analyses. All models fit well. The first set of models queried directionality of associations. In model 1A, the exposure and mediator were swapped to allow for the possibility that HbA_1c may mediate the association between depressive symptoms and memory. HbA_1c did not mediate cross-sectional (β = −0.003; SE = 0.001; P = 0.083) or longitudinal associations between depressive symptoms and memory, and depressive symptoms at T1 did not predict HbA_1c at T2 (β = −0.020; SE = 0.026; P = 0.445). Next, in model 1B, the association between episodic memory at T1 and change in HbA_1c from T1 to T2 was tested, given previous research suggesting that baseline level of cognitive functioning may predict future glucose levels. The latent difference score model showed that episodic memory at T1 was not associated with HbA_1c change from T1 to T2 (β = −0.010; SE = 0.025; P = 0.696).

The second set of sensitivity analyses modified inclusion criteria for HbA_1c (models 2A and 2B) and T2D diagnosis (model 2C). In 2A (n = 1,298), when a higher HbA_1c cutoff (≥6.5% [≥48 mmol/mol]) was used, patterns of association were similar to the primary model, with a numerically larger effect for longitudinal mediation (specific indirect effect: β = −0.009; SE = 0.004; P = 0.021). In 2B (n = 2,560), in which no HbA_1c cutoff was used, patterns of association were comparable to findings from the primary model. However, in this sample that comprised 15.8% of participants with T2D with euglycemia, the negative direct effect of T1 HbA_1c on episodic memory at T1 independent of T1 depressive symptoms that was observed in the primary model was attenuated and not statistically significant (β = −0.033; SE = 0.017; P = 0.061). In model 2C (n = 5,682), all participants with HbA_1c ≥5.7% (≥39 mmol/mol) at T1 regardless of T2D diagnosis were included. Patterns of associations persisted; additionally, after accounting for indirect effects, there was a direct effect of HbA_1c at T1 to memory at T3 (β = −0.032; SE = 0.012; P = 0.007) that was independent of depressive symptoms. In the third set of sensitivity analyses, hypertension and/or cardiovascular conditions (models 3A, 3B, and 3C) were excluded, as covariates showed almost identical patterns of association to those in the primary model.

Findings from this U.S.-wide longitudinal study showed that depressive symptoms longitudinally mediated the effect of HbA_1c on poorer episodic memory 6 years later in older adults with T2D. Indeed, depressive symptoms may represent a biobehavioral mechanism that links chronic hyperglycemia to T2D-related cognitive decline. Even after adjusting for other variables related to cognition, including sociodemographics, chronic health comorbidities, and medication adherence, depressive symptoms accounted for 19% of HbA_1c-related effects on longitudinal memory scores. Thus, this study provides compelling evidence for depressive symptoms as a potential target of interventions to reduce the significant cognitive morbidity prevalent in the rapidly growing population of older adults with T2D.

The indirect effect of HbA_1c on 6-year memory decline through increases in depressive symptoms extends previous work on the role of depression as a potential mechanism underlying cognitive morbidity in older adults with T2D. Specifically, previous studies have linked HbA_1c and/or T2D to greater depressive symptoms and the onset of depression (8–10,21,22). Separate studies have linked depression and/or depressive symptoms to increased dementia risk, worse cognitive function, and accelerated cognitive decline (12–15). Previous findings from a small, New York–specific sample of older adults show that depressive symptoms cross-sectionally mediated the association between diabetes diagnosis and cognition (24). However, the study’s findings were limited by its exclusive use of concurrent assessments, which preclude the interpretation of potential directional links, and by its sample, which included participants with type 1 diabetes. Results from a longitudinal study of two independent midlife and late-life cohorts indicate that depressive symptoms mediate the relationship between cardiometabolic dysregulation and cognitive decline (23). In that study, cardiometabolic dysregulation was operationalized as a single variable comprising six dichotomously defined indicators, including metabolic syndrome components and C-reactive protein level >3 mg/L, and the unique association between glucose and cognition was not estimated. We extend the specificity of prior research on overall cardiometabolic effects on cognition by suggesting that the distinct association between hyperglycemia and memory changes may be linked by incident increases in depressive symptoms.

Chronic hyperglycemia may increase depressive symptoms via behavioral and physiological mechanisms. The manifold demands of managing diabetes represent a significant burden that may negatively impact mood (10) For instance, at higher levels of HbA_1c, the therapeutic regimens required to achieve positive outcomes may become more complicated and demanding. Additionally, HbA_1c-related increases in depressive symptoms may be due to adverse psychological experiences (e.g., existential concerns, sense of loss) associated with living with T2D and its complications, which can persist to later life. Indeed, previous work has shown that patients with diabetes report greater hopelessness and suicidal ideation in comparison with other internal medicine outpatients, with stronger associations between T2D and those depressive symptoms in older patients (33).

Physiological mechanisms linking hyperglycemia to depressive symptoms may involve inflammatory pathways: T2D is associated with higher levels of proinflammatory cytokines (34), which, in turn, is associated with more depressive symptoms (35). Additionally, impaired energy metabolism stemming from dysregulations in insulin signaling contributes to hyperglycemia-related fatigue (11), which may lead to depressive symptoms directly by lowering mood and/or indirectly through fatigue-associated activity disengagement (10,11). Given that fatigue also represents a symptom of depression, research is needed to clarify the distinct contribution of T2D-related fatigue to depressive symptoms.

Mechanisms underlying the association between depressive symptoms and memory decline may involve psychological and/or physiological pathways. For example, depression-related effects on memory decline may occur via a hypothalamic-pituitary-adrenal axis dysregulation cascade. Mood-induced increased glucocorticoid secretion and dampened negative feedback may accelerate neuronal apoptosis and impede neurogenesis in memory-dependent brain structures (17–19). Furthermore, depressive symptoms are associated with systemic inflammation, which may explain depression-related dementia risk; however, because these physiological disturbances are also associated with T2D, it is challenging to disentangle variance contributed by depression, T2D, and their comorbidity.

In this study, the direct effect of HbA_1c on longitudinal memory decline (i.e., independent of depressive symptoms) was in the expected direction, though not statistically reliable (Table 2). In sensitivity analysis model 2C, which included all participants with HbA_1c ≥5.7% (≥39 mmol/mol) (i.e., regardless of T2D diagnosis), there was a significant direct effect of HbA_1c on longitudinal memory independent of depressive symptoms. In a separate sensitivity analysis (model 2B) comprising all participants with self-reported T2D (of whom 16% had euglycemia), there was no direct effect of HbA_1c on concurrent cognition independent of depressive symptoms. Taken together, these association patterns align with prior research that has suggested that hyperglycemia (rather than T2D diagnosis alone) directly drives poor cognitive function observed in adults with T2D. Independent of depressive symptoms, one potential mechanism may involve hyperglycemia-related disruptions to brain network topology. For instance, a prior neuroimaging study observed that higher HbA_1c levels were associated with worse global efficiency and connection paths in white matter neural networks, potentially due to hyperglycemia-related neurotoxicity (36).

Previous studies within the HRS have observed an association between baseline HbA_1c and episodic memory change independent of depressive symptoms (4,5). One reason for differences in results between those studies and ours may be that those studies adjusted for depressive symptoms at baseline and did not account for longitudinal changes in depressive symptoms. Prior work also differed by inclusion criteria and operationalization of episodic memory. A previous study within the HRS used scores from a proxy-reported questionnaire for participants who were too impaired to participate in the memory assessments (4). Although including proxy-completed responses may address attrition-related bias in missing memory data, the authors were only able to use scores from non-Latinx Black and non-Latinx White participants and not Latinx participants (who composed approximately 14.0% of the present study’s sample). Of note, the inclusion of Latinx individuals in studies of T2D is critical not only to ameliorate their underrepresentation in research studies but also to ensure that findings are generalizable to this segment of the population who are at increased risk for T2D.

Consistent with previous literature, the finding that chronic hyperglycemia may lead to poor mental and cognitive health may have important clinical implications. Screening for depressive symptoms and cognitive impairments may be important for treatment planning among older adults with T2D (27), and screening for depressive symptoms should be addressed during all care visits. Older adults with clinically elevated symptoms may benefit from referrals to mental health professionals, whereas those who exhibit cognitive difficulties may benefit from referrals to specialty cognitive clinics or formal neuropsychological assessment (27). Additional resources for family members and/or caregivers should be considered for those with worsening depressive symptoms or diabetes control. Preliminary evidence suggests that higher levels of social support may attenuate the association between glycemic level and cognition in older adults with T2D (37). Although additional research is needed to clarify specific mechanisms underlying those observed effects, theorized models of social support implicate pathways involving depressive symptoms (38). Future research on such nonpharmacological approaches to diabetes care may be especially useful for older adults because of their age-related vulnerability to polypharmacy-related adverse effects (27).

As with other longitudinal cohort studies, this study is limited by its use of self-report measures of depressive symptoms and T2D diagnosis, which may have underestimated the association between HbA_1c and depressive symptoms (10) and the prevalence of T2D (39). Meta-analytic findings show that the T2D–depression association is stronger when depression is assessed using diagnostic interviews versus questionnaires, likely due to the diagnostic imprecision and greater measurement error of self-reported questionnaires (10). Because the HRS does not routinely assess antidepressant use, we were unable to examine the contribution of antidepressant use to observed associations. Another limitation is the use of just one cognitive domain. Future studies should examine associations with other domains of cognition (e.g., executive function). Furthermore, we did not consider T2D duration, which should also be included in future studies, given its association with cognitive outcomes.

Additionally, given our research question on baseline glycemic level in relation to changes in depressive symptoms and subsequent memory decline, we examined only one time point of HbA_1c. Furthermore, the lack of available HRS data precludes examining whether changes in HbA_1c (T1 to T2) predict subsequent changes in depressive symptoms (T1 to T3) and subsequent memory decline (T1 to T4). Although the HRS was not designed to collect comprehensive data on medical history, future studies may improve the operationalization of T2D by incorporating medication use, medical chart review, and health care access. Indeed, individuals with T2D may not report taking medication, because of inadequate clinical care, whereas those at risk for T2D may be prescribed oral hypoglycemics prophylactically (40). In the present study, sensitivity analyses revealed that of the 12,333 participants with HbA_1c data at T1 (2006/2008), 423 (3.42%) who did not report T2D met the HbA_1c criterion (≥6.5%) for diagnosis. Additionally, self-reported general medication adherence was higher than estimates from other studies. Although we adjusted for other factors (e.g., education, financial resources including debt, health comorbidities) that can influence medication adherence, future studies should incorporate comprehensive, standardized assessments of adherence.

Strengths of this study include multiple waves of episodic memory and depressive symptoms data in order to characterize HbA_1c associations with changes in depressive symptoms and cognition. In the present study, we featured a large sample of older adults with T2D who were recruited across the United States and a comprehensive set of covariates. A particularly important strength of this study is its rigorous methodological approach, which examined longitudinal mediation within a structural equation modeling framework and a broad set of sensitivity analyses to test the robustness of results.

In conclusion, findings from this U.S.-wide longitudinal study highlight the potential mechanistic role of depressive symptoms in linking hyperglycemia to T2D-related memory decline. Reducing depressive symptoms may represent one method to decrease the adverse effects of hyperglycemia on memory aging in later life and may minimize the cognitive morbidity prevalent in the rapidly growing population of older adults with T2D.

Funding. The HRS is sponsored by the National Institute on Aging (grant U01AG009740) and is conducted by the University of Michigan. The study also was supported by the Alzheimer’s Association (grant AARFD-22-924846) and a pilot award from the Columbia Center for Interdisciplinary Research on Alzheimer's Disease Disparities (National Institute on Aging, grant P30AG059303).

These funding sources had no role in the study’s design and conduct or the manuscript’s development and publication.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. A.Z.K., V.L.E., and L.B.Z. were involved in the conception and design of the study and the analysis and interpretation of the results. A.Z.K. wrote the first draft of the manuscript, and all authors edited, reviewed, and approved the final version of the manuscript. A.Z.K. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in poster form at the 2021 annual meeting of the Gerontological Society of America, Phoenix, AZ, 10–14 November 2021.