Specific Dimensions of Depression Have Different Associations With Cognitive Decline in Older Adults With Type 2 Diabetes

Depression is highly frequent in patients with type 2 diabetes, with twice the odds compared with people without type 2 diabetes. It is commonly associated with poorer diet, medication compliance, and exercise adherence as well as with more functional impairment (1). This is an expected effect of depression given its core symptomatology encompassing reduced level of energy and motivation to undertake a task, impaired concentration, and decreased self-esteem (2). As a result, comorbidity of depression and type 2 diabetes, compared with type 2 diabetes alone, is associated with poorer quality of life and increased mortality (1).

Depression is a heterogeneous ailment in terms of presentation and neurobiology as well as its response to treatment. Despite all being clustered under the umbrella of depressive disorder, patients can present with a variety of complaints, including dysphoria, apathy, anxiety (3), psychomotor retardation (4), and cognitive complaints (5). Although dimensions that specifically characterize depression in type 2 diabetes are yet to be delineated, apathy and anxiety are both frequent presentations in this population, the former being associated with poor glycemic control and cognitive decline (6).

Both depression (5) and type 2 diabetes (7) have been consistently associated with cognitive impairment in older adults. Yet, most studies have investigated depression as a unitary construct rather than as a heterogeneous conglomerate of symptoms, and there is a paucity of data on how various dimensions of depressive symptoms differentially affect cognition. In primarily people without type 2 diabetes, the depression dimensions of dysphoria, meaninglessness, and subjective cognitive impairment are associated with cognition (8). Functional disability, on the other hand, has been shown to be associated with depression dimensions of apathy, helplessness, and worthlessness, suggesting that different dimensions of depression may be associated with unique long-term clinical outcomes (9).

Despite the potential clinical value in the identification of relationships of specific dimensions of depression with longitudinal cognitive decline in type 2 diabetes, these associations are essentially unknown. To address this knowledge gap, we used data from 1,002 community-dwelling older adults participating in the Israel Diabetes and Cognitive Decline (IDCD) study and investigated the longitudinal associations of specific depression dimensions with decline in global cognition and in specific cognitive domains.

IDCD is a prospective longitudinal study investigating the relationship between type 2 diabetes and cognitive decline. It is a collaborative effort among the Mount Sinai School of Medicine, Sheba Medical Center, and Maccabi Healthcare Services (MHS), which is the second largest health maintenance organization in Israel. The IDCD research protocol was approved by the institutional review boards of all three institutions, and all participants signed informed consent. The IDCD study methods have been detailed elsewhere (10). Briefly, participants are ≥65-year-old patients with type 2 diabetes randomly selected from the ∼11,000 in the MHS diabetes registry, have normal cognition at baseline, and are living in the area of Tel Aviv, Israel. The diabetes registry has detailed information on diagnoses, medication, and laboratory results from patients who meet the following criteria: 1) HbA_1c >55.7 mmol/mol (7.25%) 2) glucose >200 mg/dL on two examinations >3 months apart, 3) purchase of antidiabetic medication twice within 3 months supported by an HbA_1c >47.4 mmol/mol (6.5%) or glucose >125 mg/dL within half a year, and 4) diagnosis of type 2 diabetes (ICD-9 code) by a general practitioner, internist, endocrinologist, ophthalmologist, or diabetes advisor supported by an HbA_1c >47.4 mmol/mol (6.5%) or glucose >125 mg/dL within half a year. Potential participants’ electronic medical charts are screened at MHS for baseline dementia (and its subtypes), mild cognitive impairment (MCI), cognitive impairment diagnoses, or physician notes suggesting cognitive impairment as well as for dementia medications. Participants are excluded if any of the above is present. A study physician then assesses the participants for dementia, and a neuropsychologist administers the cognitive battery. All clinical data are further discussed in a multidisciplinary consensus conference (which includes geriatric psychiatrists, neurologists, and neuropsychologists) to arrive at a diagnosis. Participants with dementia or MCI at baseline are excluded. Depression is not an exclusion criterion. The IDCD had 1,204 eligible participants at baseline; 1,002 were included in this study and had complete depression and covariate data.

Baseline demographic variables include age, sex, and years of education, which are collected at the IDCD study baseline. Body measurements (weight and height); vital signs, including systolic and diastolic blood pressure; and blood examinations are received from MHS at baseline. Height is measured without shoes in cm, and weight is measured with light clothing in kg. Blood pressure is measured from the left arm while seated. Morning blood examinations, measured after 10 h of fasting, include glucose, HbA_1c, total cholesterol, HDL, LDL, triglycerides, and creatinine. Duration of diabetes was defined as the time from entry into the MHS diabetes registry and the IDCD baseline (11). A physical activity index was determined by the number of various physical activities (e.g., swimming, jogging or brisk walking, dancing, spinning, light exercise) performed over the previous 2 weeks using a simplified version of the Minnesota Leisure Time Activity Questionnaire (12). Finally, metformin use (ever/never) up to the baseline of the IDCD study was also a covariate.

All study participants underwent a comprehensive neuropsychological battery at baseline and at approximately every 18 months thereafter. The battery included 12 neuropsychological tests as follows: episodic memory, Alzheimer Disease Assessment Scale word list immediate recall, delayed recall, and recognition; attention/working memory, Diamond Cancellation and Digit Span forward and backward; executive functions, Trail Making Tests A and B and Digit Symbol Substitution Test; and language/semantic categorization, similarities subscale of Wechsler Adult Intelligence Scale-Revised, animal fluency, and Boston Naming Test. Z-scores were calculated for each neuropsychological test (the difference between the individual test score and the baseline average test score was divided by the baseline SD). Before normalization, logarithmic transformation was performed for Trail Making Tests A and B, and the sign was corrected so that when change scores were created, higher values were always better. A domain-specific Z-score was calculated as the difference between the individual domain Z-score (average of the Z-scores across all the neuropsychological tests within the domain) and the baseline average domain-specific Z-score and then divided by the SD of the baseline domain-specific Z-score. The global Z-score is defined as the difference between the individual average domain Z-score (average of the four domain Z-scores) and the baseline average domain Z-score and then divided by the SD of the individual baseline domain Z-score.

The 15-item version of the Geriatric Depression Scale (GDS) was used for measurement of depression at every visit. The GDS (13) is a self-report depression scale with dichotomous yes-no responses to the questions, and higher scores reflect more severe depression. It originally constituted 30 items, but a shorter 15-item scale (GDS-S or GDS-15) has also been validated (14) and used in large-scale studies. It is particularly useful for older adults with type 2 diabetes because it does not focus on somatic symptoms of depression, which may be confounded by neuropathy and motor limitations (15).

A number of studies have conducted a factor analysis of the GDS as an attempt to identify subdimensions of depression. Adams et al. (16) used confirmatory factor analysis of the 30-item GDS using a community-based sample of older adults and identified five dimensions of geriatric depression, including dysphoric mood, withdrawal-apathy-vigor (WAV), hopelessness, worry, and cognitive impairment (15). We adopted their dimensions for GDS-15 as follows: dysphoric mood constituted items 1, 3, 4, 5, 7, 11, and 15; WAV, items 2, 9, and 13; anxiety, item 6; memory complaint, item 10; and hopelessness, items 8, 12, and 14 (Supplementary Table 1). We chose the above constructs because they were validated in similar community-dwelling older adults with no prior known diagnosis of cognitive disorders (16).

The sample description is summarized as n (%) for categorical variables and mean (SD) or median (interquartile range [IQR]) for continuous variables. The response to some of the GDS items were reversed such that a value of 1 represented depression. Because the number of GDS items differed across the five depression dimensions, to facilitate a fair comparison among dimensions, the depression score for each dimension was further standardized for each patient by dividing the number of positive responses by the total number of available items in that dimension. This step resulted in a final depression score for each dimension that ranged from 0 to 1, with 0 indicating absence of any symptom and 1 indicating presence of all symptoms.

We examined the longitudinal associations of the five depression dimensions with overall cognition and each of the four cognitive domains, respectively, using the random coefficients models. The models assumed that the trajectory of the cognitive data since baseline was linear within the study period, with a unique intercept and slope for each participant. Furthermore, the participant intercepts and slopes were correlated and followed a bivariate normal distribution. This assumption was deemed appropriate after examining the spaghetti plot of the individual cognitive Z-score profiles over time. We adopted a conservative approach and included all five depression dimensions within the same model. To ensure that there is no significant collinearity and that it is appropriate to include all five depression dimensions in a single model, we examined the variance inflation factor for each of the depression dimensions and all were <1.45 (far lower than a cutoff of 10), indicating no collinearity (17).

In addition, we included the time (in months) and depression dimensions interaction to evaluate whether the depression dimension is associated with cognitive decline. A negatively significant interaction with time indicates a greater decline of cognition for every unit change in the depression score. We applied two models: Model 1 adjusted for the sociodemographic factors, including age, sex, and education, and model 2 additionally adjusted for cardiovascular risk factors (i.e., duration of type 2 diabetes, cholesterol, creatinine, HbA_1c, triglycerides, systolic and diastolic blood pressure, BMI), which have been shown to be associated with cognitive decline and dementia (18). We also adjusted for physical activity (19,20) and the use of the antidiabetic medication metformin, which may be associated with improved mood and cognitive function (21,22). Statistical analysis was conducted using SAS 9.4 software (SAS Institute, Cary, NC). A two-sided P ≤ 0.01 was defined as the significance level in all statistical tests to adjust for five cognitive outcomes.

The study included 1,002 IDCD study participants. The mean age of the participants at baseline was 71.6 (SD 4.6) years, and 41.1% were female. Participants had, on average, 13.2 (SD 3.6) years of education and a Mini Mental State Examination score of 28.0 (SD 1.5), consistent with normal cognitive status. The median duration of follow-up was 48 (IQR 24, 54) months (Table 1). Overall, the sample had a GDS score of 2.3 at baseline, reflecting a relatively low number of depressive symptoms. The most commonly reported GDS-15 items at baseline among the IDCD participants were item 2 (drop activities) at 41.02%; item 9 (stay home), 33.05%; item 13 (energy), 23.19%; and item 6 (something bad will happen), 22.2% (Fig. 1). See the GDS item distribution across each of the depression domains in Supplementary Table 2.

Supplementary Table 3 describes the cross-sectional association of the depression dimensions with cognitive function at baseline (i.e., main effects of depression dimensions in models 1 and 2); analyses of each depression dimension were adjusted for all other dimensions. The coefficient represents the average difference in cognitive Z-score from no symptoms to full symptoms (depression dimension score from 0 to 1). At baseline, adjusting for sociodemographic factors (model 1), all dimensions of depression, except for anxiety, were associated with some aspect of cognition. Greater dysphoric mood was associated with worse global cognition (P < 0.001), executive function (P < 0.001), and language (P = 0.009) but not with attention/working memory and memory. Similarly, a higher score in the WAV dimension was associated with worse language function (P = 0.003). Greater hopelessness was associated with worse global cognition (P = 0.003). Participants who reported poorer memory compared with their peers had worse global cognition (P = 0.001) and episodic memory (P < 0.0001).

The magnitude of the coefficients was generally reduced in the fully adjusted model (model 2). The associations between dysphoric mood (P = 0.019) and WAV (P = 0.015) with language were attenuated and did not reach significance at the 0.01 level. However, the association of more hopelessness with worse global cognition remained significant (P = 0.004).

Table 2 presents results of the relationships of depression dimensions with longitudinal cognitive decline (i.e., the interaction between depression dimensions and time in the model). The negative coefficient represents the rate of decline in cognitive Z-scores per month. The association of dysphoria with decline in executive function was significant in model 1 (P = 0.009), but in the fully adjusted model, it did not withstand adjustment for multiple comparisons (P = 0.11). Similarly, the association between WAV and decline in global cognition, which was significant in model 1 (P = 0.004), was attenuated in the fully adjusted model (P = 0.028). Nevertheless, the association of WAV with faster decline in executive function (P = 0.002, model 1) (Table 2) remained significant in the fully adjusted model and withstood adjustment for multiple comparisons (P = 0.004, model 2) (Table 2 and Fig. 2). Anxiety, hopelessness, and subjective memory complaint were not associated with faster decline in any of the cognitive domains or in global cognition.

Previous longitudinal studies have shown that depression is a risk factor for cognitive decline and dementia in general populations (5) and specifically among patients with type 2 diabetes (23), who have a high prevalence of depression and are at high risk for cognitive decline and dementia. Our study of 1,002 participants expands these findings by providing evidence for specific depression-related dimensions that may differentially affect cognitive decline in older adults with type 2 diabetes.

Our cross-sectional results showed that after adjusting for sociodemographic, cardiovascular, and diabetes-related factors and for metformin use and physical activity, all dimensions of depression, except anxiety, were associated with worse cognitive function in at least one domain. Dysphoria was associated with impairment in language, executive function, and global cognition at baseline. Among studies that have investigated the association of depression and cognitive impairment, very limited studies looked into the effect of dysphoria as a distinct subsyndrome of depression. Consistent with our results, others have reported a significant association between a higher score in a dysphoria dimension (on the basis of the GDS-30) and poorer immediate/delayed memory and language in a community-based sample of adults >40 years of age (8). Given that dysphoria represents the core feature of depression and constitutes the major content of all depression scales, one may anticipate the association of the dysphoria dimension with the cognitive functions to mirror those of depression. Along this line, we previously reported an association of depression with worse executive function, language, and overall cognition in IDCD participants (24). Our results were largely consistent with the cross-sectional studies that show an association between depressive symptoms and cognitive impairment (5).

Our longitudinal results showed that the apathy dimension was the strongest predictor of cognitive decline, which withstood adjustment for metabolic risk factors and for multiple comparisons. Participants reporting greater apathy had a faster rate of decline in global cognition and executive function (Fig. 2). Although there was a suggestion of association of apathy with greater decline in episodic memory (P = 0.04), this did not withstand adjustment for multiple comparisons. GDS items 2, 9, and 13, which are clustered in our study as the apathy dimension, were the most commonly reported items among our participants. These items were previously described as the GDS-3A subscore and used as an inferred apathy scale (25). Bertens et al. (25) characterized the GDS-3A in two cohorts of community-dwelling elderly and suggested a cutoff of ≥2 for the definition of apathy. In their two study cohorts, 16.6% and 10.9% of subjects scored ≥2 on the GDS-3A scale. Among IDCD participants, 26.2% scored ≥2 on GDS-3A, which met the suggested threshold for apathy. This higher prevalence of apathy in our study population is consistent with previously reported higher apathy in individuals with type 2 diabetes (6).

Apathy is a behavioral syndrome characterized by a reduction in motivation, behavioral and cognitive retardation, and reduced emotional reactivity, leading to decreased goal-oriented activities (26). In samples not necessarily focused on people with diabetes, apathy has been shown to strongly predict cognitive decline (27). Moreover, in patients with MCI, apathy was shown to be associated with a higher rate of conversion to Alzheimer disease 1 year later (28). It is an important clinical presentation across many neuropsychiatric disorders, including stroke, traumatic brain injury, and neurodegenerative disorders, that leads to significant functional impairment yet with limited effective treatments. Across different disorders, apathy was found to be consistently associated with alterations in frontostriatal circuits and, in particular, dorsal anterior cingulate cortex and ventral striatum, which includes the nucleus accumbens, all regions implicated in motivation (29). Apathy is also a common feature of small vessel disease and is suggested to be the result of insult to the reward networks through reduction in the integrity of white matter (30). White matter hyperintensities are markers of small vessel disease, chronic microperfusion ischemic demyelination, and neuronal damage (31). Both in population studies (32) and in patients with Alzheimer disease (33), white matter hyperintensities have been found to be correlated with higher apathy. Cognitive domains most consistently affected by small vessel disease are executive function and processing speed (34), although associations with episodic memory have also been reported (35). Similar to our results, others also have shown that in the population of patients with small vessel disease, higher baseline apathy rather than depression predicted dementia (36). Our cohort of older adults with type 2 diabetes similarly showed strong associations of apathy with greater decline in executive function. Patients with type 2 diabetes have a greater degree of cerebrovascular disease both in large (37) and in small vessels (38), suggesting it as a potential underlying mechanism for the associations we have found.

The GDS-15 item on anxiety (item 6) inquires about fear of something bad happening, which is consistent with general anxiety. Our results showed that answering yes to this item was not associated with significant cognitive deficit at baseline or with cognitive decline over time. Gulpers et al. (39) conducted a meta-analysis on the longitudinal studies of the association of anxiety and cognitive decline and found that anxiety is a significant predictor of dementia in older adults age >80 years from a community-based sample, about a decade older than our sample. Moreover, anxiety was not a predictor of dementia among a population with clinical MCI, consistent with our results.

The IDCD study included, at baseline, participants with normal cognition of whom 10% reported subjective memory complaint. In the current study, self-reported memory complaint was found to be related to poorer performance on episodic memory, while longitudinally it was not associated with cognitive decline in any of the domains, including episodic memory. There has been a growing interest in subjective memory complaints as a possible screening tool for identifying subtle changes in cognition and a marker for preclinical stages of dementia. Some longitudinal studies have shown that baseline subjective memory complaint is associated with an increased risk of dementia (40). It has also been shown to be associated with biomarkers of Alzheimer disease, including amyloid deposition (41), cerebral hypometabolism (42), cortical volume loss (43), and altered functional connectivity (44). Yet, consistent with our findings, numerous studies did not find relationships of subjective memory complaint with cognitive decline or neuropathology (45).

Our results on the differential effect of depression dimensions on cognitive outcomes emphasize the heterogeneity of depression as a clinical construct rather than as a single entity. The variability of depression presentation have been linked to different brain substrates (46) and needs to be taken into account not only with regard to the diagnosis and treatment choices but also in terms of the factors affecting longer-term outcome. Our results highlight the importance of clinical screening for apathy among older adults with type 2 diabetes given its high prevalence, its relationship to poorer type 2 diabetes management and self-care (47), and long-term cognitive decline.

One of the shortcomings of the study pertains to the limitations of GDS-15 as a brief self-report scale of depression. Using a more in-depth and objective assessment of depression may help to more accurately identify subtle and specific symptoms, such as apathy. The GDS-15 is a widely used screening tool of depression in large-scale studies and is particularly useful in the context of type 2 diabetes since it is less focused on the somatic symptoms (15) commonly present in individuals with type 2 diabetes as a result of comorbidities not necessarily related to depression, such as neuropathy. The mean baseline GDS-15 score in our study was 2.3, which alludes to a relatively low number of depressive symptoms in this sample because patients with depression are generally less likely to consent to the study. The associations found may become stronger as a greater number of depressive symptoms occur. Moreover, there was a small correlation (r = −0.11; data not shown) between the number of follow-ups and number of depressive symptoms, possibly suggesting that individuals with more depressive symptoms tend to drop out more frequently. However, our mixed models used estimation and maximization steps that attempt to use the available data to estimate missing values from existing values, minimizing this limitation. Furthermore, baseline characteristics in those who completed only baseline and those who also completed additional follow-ups were very similar (Supplementary Table 3).

The IDCD study is focused on type 2 diabetes, and thus, results cannot be generalized to other populations. The magnitude of the effect of apathy on cognitive decline is relatively small. However, we highlight that this sample is based on individuals with initially normal cognition with relatively low levels of depressive symptoms. It is anticipated that this effect becomes cumulatively more pronounced as time progresses and greater cognitive impairment occurs. While the larger prevalence of the apathy dimension allows for better statistical power, it does not affect the magnitude of the β-coefficient. In our study, the β-coefficient of apathy for global and executive function decline is large compared with most of the other β-estimates, suggesting that the effect is real and not due to the larger prevalence of the apathy dimension. Finally, reverse causality cannot be ruled out because impaired cognition may result in a greater degree of depressive symptoms. Strengths of this study include the large number of older adult subjects with a well-validated diagnosis of type 2 diabetes, the wealth of detailed information on long-term cardiovascular and diabetes related data rather than self-reported measures, and the broad cognitive battery, which permitted calculation of specific cognitive domains and examination of their unique associations with different depression dimensions.

Acknowledgments. The authors thank Marina Nissim, Milan Italy.

Funding. This work is funded by National Institutes of Health grants R01-AG-034087, AG-053446, AG-051545, and AG-043878 to M.S.B. and P50-AG-05138 to M.S. The authors are also grateful for the generosity of the LeRoy Schecter Foundation and to Dr. Marina Nissim

Duality of Interest. M.S. serves as a board member of the Alzheimer’s Association, International Psychogeriatric Association, and National Association of Veterans’ Research and Education Foundations; is a consultant on trial design for Eisai, Avenir, vTv, Biogen, BioXcel, and F. Hoffmann La Roche; and has other relationships with and has served on the data safety monitoring board for Syneos. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. L.S. contributed to the study design, interpreted data, and drafted/revised the manuscript. R.R.-S. contributed to the study design, data collection, and review/revision of the manuscript. H.-M.L. and X.L. contributed to the data analysis and interpretation. M.S. reviewed/revised the manuscript and contributed to the discussion. A.H. contributed to the study design, data collection, and review/revision of the manuscript. M.S.B. contributed to the study design, data analysis and interpretation, and drafting/revision of the manuscript. L.S. 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 Alzheimer’s Association International Conference 2020, Virtual, 27–31 July 2020.