Associations of depressive symptoms and cortisol with cognitive performance among memory clinic patients

Abstract

Objective

Test the hypothesis that depressive symptoms are associated with cognitive performance and that cortisol levels may explain this association independently of Alzheimer’s Disease (AD) biomarker levels.

Design

Longitudinal observational study.

Setting

Memory clinic, Karolinska University Hospital, Stockholm, Sweden.

Participants

Consecutive patients (n = 162) who agreed to take part in the Cortisol and Stress in AD (Co-STAR) study during 2014-2017 and had data available for variables of interest.

Measurements

Participants rated their depressive symptoms using the Geriatric Depression Scale (GDS) and collected diurnal salivary cortisol samples at home. Cognitive performance was assessed by standardized cognitive tests in the following domains: memory, working memory, processing speed, perceptual reasoning, and general cognitive function. Dementia, mild cognitive impairment (MCI), and subjective cognitive decline (SCD) were diagnosed as part of the clinical work-up. We determined the associations between GDS and cognitive domain scores using linear regressions, including cortisol levels as covariates. We also tested if cerebrospinal fluid (CSF) AD biomarkers amyloid β42 (Aβ42) and tau proteins modified these associations.

Results

The GDS score was negatively associated with performance in working memory and processing speed, independently of cortisol levels. These associations were no longer significant after introducing AD biomarkers as covariates. Baseline GDS score was not associated with change in memory or processing speed at follow-up.

Conclusions

The underlying amyloid pathology may affect the association between depressive symptoms and cognitive performance in memory clinic patients.

1. Introduction

Cognitive impairment is a common condition in older adults (Karlamangla et al., 2014). According to the World Health Organization, over 55 million people worldwide live with severe cognitive impairment due to dementia and “every year, there are nearly 10 million new cases diagnosed with dementia” (World Health Organization, 2023). The number of people living with dementia is expected to rise to 74.7 million by 2030 (Prince et al., 2015). Therefore, it is important to identify and target modifiable risk factors for dementia. One such risk factor is represented by depressive symptoms, which are common in older adults with cognitive complaints (Auning et al., 2015; Seo et al., 2017). In fact, depression is considered a modifiable risk factor for dementia by the 2020 Lancet Commission on dementia prevention, intervention, and care (Livingston et al., 2020).

Depressive symptomatology, often assessed using standardized questionnaires for depression, has been recognized as a risk factor for dementia, especially in relation to subjective cognitive decline (SCD) (Brown et al., 2022; Grambaite et al., 2013; Seo et al., 2017; Wang et al., 2021; Zlatar et al., 2017) as well as mild cognitive impairment (MCI) (Auning et al., 2015; Grambaite et al., 2013; Sacuiu et al., 2016) during early stages of dementia. Studies have found this association in both community-dwelling (Ahn et al., 2021; Buckley et al., 2013; Buckley et al., 2016; Wang et al., 2021; Zlatar et al., 2014) and clinic-based (Eckerström et al., 2016; Zlatar et al., 2017; Zlatar et al., 2018) samples of older adults. The presence and severity of depressive symptoms are often linked to SCD. In older adults with high levels of CSF amyloid-β₄₂ (Aβ₄₂), more severe depressive symptoms have been associated with subjective memory complaints (Grambaite et al., 2013).

In a study among participants who were cognitively intact, had MCI, or had SCD, depressive symptoms were significantly associated with performance on tests of memory, executive function, psychomotor speed, and global cognition (Seo et al., 2017). Moreover, depressive symptoms moderated the SCD-cognition relationship such that only individuals with more severe depressive symptoms showed significant associations between SCD and objective cognitive performance (Seo et al., 2017). Similar results were obtained in other convenience samples (Buckley et al., 2013; Forbes et al., 2024).

In a previous study using the Co-STAR sample, we have demonstrated that a higher ratio between awakening cortisol and bedtime cortisol (AM/PM ratio) was associated with better results in general cognition and perceptual reasoning among participants with normal levels of CSF Aβ42 (Holleman et al., 2022). Additionally, we found an interaction between the ratio between awakening cortisol and bedtime cortisol levels (AM/PM ratio) and CSF Aβ42 in relation to general cognition, and an interaction between cortisol levels at awakening and CSF Aβ42 in relation to performance in relation to memory (Holleman et al., 2022).

However, little is known about the role of cortisol in the association between depressive symptoms and cognitive impairment in memory clinic populations. In a population survey, Potvin et al. found a relation between cortisol levels and risk of cognitive impairment that was modulated by affective symptomatology (Potvin et al., 2013). Also, altered cortisol levels have been reported in older individuals with cognitive complaints (Fiocco et al., 2006) and in patients with dementia with or without depression (Barca et al., 2019). Although certain longitudinal community studies have indicated a connection between incident depressive symptoms according to Geriatric Depression Scale (GDS) and brain accumulation of Aβ42 (Donovan et al., 2018; Harrington et al., 2017), studies of memory clinic populations have not found any association between AD biomarkers and depressive symptomatology measured by the GDS (Auning et al., 2015) or Cornell Scale for Depression in Dementia (Kramberger et al., 2012). Similarly, depressive symptoms using Beck Depression Inventory-II were not associated with brain amyloid deposition in a cross-sectional community study (Krell-Roesch et al., 2018).

As indicated in the latest report on dementia prevention in patients undergoing investigations at cognitive disorder clinics (Frisoni et al., 2023), it is important to identify modifiable risk factors and their interplay. Therefore, further research is needed to determine the relationship between depression, cortisol, and cognitive performance in individuals who seek or are referred to medical attention for cognitive decline. This study aims to test the hypothesis that depressive symptoms are associated with cognitive performance as reflected by results of standardized psychometric tests and that cortisol levels may explain this association among memory clinic patients, independently of the presence of CSF biomarkers of neuropathologic processes.

2. Methods and materials

2.1. Study design and participants recruitment

The sample and methods of the Cortisol and Stress in Alzheimer’s Disease (Co-STAR) study have been described in detail previously (Adedeji et al., 2023; Holleman et al., 2022).

Briefly, the Co-STAR study is an observational study focusing on patients referred to the Cognitive Disorders Clinic at the Karolinska University Hospital, Huddinge, Sweden to investigate stress-related factors, fluid biomarkers, and their associations with cognitive decline and daily life functioning.

The Co-STAR study included 187 patients referred to the memory clinic who were aged 45+ years and have received a diagnosis of SCD, MCI or dementia. Of these, the current study sample consisted of 162 patients (SCD=57, MCI=69, and dementias=36 of which 33 Alzheimer’s dementia, one subcortical vascular dementia, one frontotemporal lobe dementia and one unspecified dementia) who rated their depressive symptoms according to the GDS (response rate 86.6%). Patients were included if they did not suffer from any sensory impairments (e.g., visual, or auditory) that would compromise their ability to participate, did not suffer from conditions affecting cortisol output, and did not have any illnesses severely reducing their capacity to participate in the study and its follow-up assessment. Relevant information on medical history and medications was collected from medical journals.

2.2. Procedures at cognitive disorders clinic

Eligible participants who were included in this study had undergone routine assessments and additional Co-STAR specific assessments.

2.2.1. Routine assessments

At the baseline assessment, eligible patients had undergone routine clinical assessments, including information on demographic factors, medical examinations (physical, psychiatric, and neurologic), laboratory investigations (blood, CSF, brain MRI), and a comprehensive neuropsychological test battery.

2.2.2. Diagnosis of cognitive impairment

A clinical diagnosis of dementia was established at consensus meetings and was based on the International Classification of Diseases 10^th revision criteria (ICD-10) (World Health Organization, 2011). The diagnostic criteria for dementia in the ICD-10 include a decline in cognitive function that is severe enough to interfere with daily activities, along with detectable impairment in at least one cognitive domain including learning and memory, language, perceptual-motor skills, attention and executive function, or social behavior. Patients with cognitive complaints and objective cognitive impairment according to neuropsychological tests without fulfilling criteria for dementia received a diagnosis of MCI in accordance with Winblad et al. (Winblad et al., 2004). Thus, MCI diagnosis included the following: (i) the person is neither cognitively intact nor fulfills the criteria for dementia diagnosis; (ii) there is evidence of cognitive deterioration shown by either objectively measured decline over time and/or subjective report of decline by self and/or informant in conjunction with objective cognitive deficits; and (iii) activities of daily living are preserved and complex instrumental functions are either intact or minimally impaired. Those not fulfilling any of the criteria of MCI or dementia were diagnosed with SCD, i.e., subjective report of cognitive decline by self- and/or informant report as assessed during clinical interview.

2.2.3. Co- STAR specific assessments

In addition to the routine clinical assessments, Co-STAR participants answered questionnaires on a range of psychological and lifestyle factors and were provided with a kit for salivary cortisol sampling at home. Participants were instructed to provide saliva samples at six time points on two non-consecutive weekdays. Sampling time points were upon awakening (T1), at 30 minutes (T2) and 60 minutes (T3) after awakening, at 2:00 P.M. (T4), at 4:00 P.M. (T5), and before going to bed (T6). To avoid contamination, participants were asked not to eat or brush their teeth before sampling. Participants were also provided with journals to document the exact sampling time. Participants were also provided with questionnaires for assessment of depressive symptoms using the 15-items GDS.

2.2.4. Assessment of cognitive function

The neuropsychological test battery included 11 tests encompassing five cognitive domains: memory, working memory, processing speed, perceptual reasoning, and general cognitive function. Raw cognitive tests scores were transformed to Z-scores based on a cognitively healthy reference sample as described in detail previously (Holleman et al 2022). The adjusted Z-scores were averaged to create composite domain scores as follows:

• -The composite memory score was based on four tests: the Rey Auditory Verbal Learning test (verbal delayed recall) (Bowler, 2013), the Rey-Osterrieth Complex Figure test (visual delayed recall) (McKinlay, 2011), the Digit Symbol Substitution test from Wechsler’s Adult Intelligence Scale (WAIS) (incidental learning) (Bettcher et al., 2011), and the Hagman test immediate recall which was developed at the Karolinska University Hospital for the assessment of visual memory. Participants had to have completed at least two of the four tests to be included in the analyses.
• -A composite working memory score was calculated from WAIS Digit Span and Arithmetic tests.
• -Processing speed domain that was assessed with a single test, the WAIS Digit Symbol Substitution Test.
• -Perceptual reasoning was based on WAIS Block Design and WAIS Matrix Reasoning (Erdodi et al., 2017).
• -The general cognitive function score was obtained using four tests from the Wechsler Abbreviated Scale of Intelligence (WASI) i.e., Block Design, Matrix Reasoning, Similarities, and Information (Irby and Floyd, 2013). Two of these tests are related to verbal cognition (Similarities & Information), while the other two are related to non-verbal cognition (Ryan et al., 2003). Participants had to have completed at least one verbal and at least one non-verbal test to be included in the analyses.

2.2.5. Assessment of depressive symptoms

Depressive symptoms were reported by the participants using the 15-item GDS (Sheikh and Yesavage, 1986), which was designed to detect depressive symptoms in older adults. The scale includes both positive and negative statements, and respondents are asked to indicate whether each statement applies to them or not (dichotomous response format “Yes” score 1 or “No” score 0). It is relatively easy to use, and it takes around 5-10 minutes to administer. The 15 items are summed up to give a final GDS score from 0 to 15, with a higher score indicating an increased load of depressive symptoms.

2.2.6. Assessment of burnout syndrome

History of burnout syndrome was assessed as part of a standardized self-reported questionnaire. Participants were asked if they have ever been on sick leave for a period longer than one month due to severe stress (Yes/No).

2.2.7. Follow-up

Fifty-seven participants who completed the GDS at baseline participated in the follow-up cognitive assessment (mean follow-up time 2.7 ± 0.7 years; range 1.2 to 4.3 years from baseline). Only tests of memory and processing speed were administered at follow-up since they were deemed sensitive to change. Follow-up data on memory and processing speed were available for 51 and 47 participants, respectively. The raw scores and domain composite score in memory at follow-up were handled similarly as the baseline scores (see 2.2.4). Score changes were computed for the processing speed test and for the memory domain composite by subtracting baseline Z-score from follow-up Z-score.

2.3. Ethical considerations

All participants signed informed consent for participation in the Co-STAR study. The study was approved by the Regional Ethical Review Board (Stockholm, 2014/524-31/1).

2.4. Statistical analyses

The two-day averaged cortisol raw measures across time-points T1-T6 were winsorized to include extreme values (outliers) transformed to values at 3SD of the mean of the sample. The raw values thus obtained showed significant right skewed distributions >1.0, normal Q-Q plots (Kolmogorov-Smirnov test p<0.05). Thus, cortisol values at time points T1 through T6 were normalized using natural logarithmic transformation to allow parametric statistical analysis. CSF T-tau and phosphorylated-tau (P-tau) were also normalized using natural logarithmic transformation due to skewness >1.0 (Kolmogorov-Smirnov test p<0.05). All these resulting variables were non-skewed.

To address any potential bias related to responding to the GDS self-reported questionnaire, we conducted tests to compare the groups with and without GDS. For continuous parameters with normal distributions (age, years of education, and logarithmically transformed cortisol parameters), we used ANOVA with F-statistic to test for differences across groups. To analyze differences in proportions (sex and clinical parameters), we used Fisher's exact statistic. We applied the same statistical tests to analyze differences between groups in demographic and clinical characteristics to provide a description of the available GDS sample. A cut-off score of GDS 5 or above was applied (Yesavage and Sheikh, 1986).

Linear regression models and selection of explanatory variables

We tested the associations between the continuous GDS score (main explanatory variable) and each cognitive domain score (dependent continuous outcome variable) using linear regression unadjusted and adjusted for confounders, for salivary cortisol level and for CSF AD biomarkers.

The models were adjusted for established confounding factors (age, years of education, sex, and health-related factors (i.e., history of burnout syndrome, use of steroids or cortisone medication)). Although the presence of APOE-ε4 genotype could confound associations with cognitive function (Small et al., 2004), DNA APOE-genotype was available only in 42.8 % of the sample (43.8% of GDS responders). Thus, we were unable to fit reliable regression models to include APOE-ε4 genotype. Moreover, we have discarded the use of antidepressants and sedatives as covariates due to significant associations (collinearity) with burnout syndrome (Fisher’s exact test p<0.05).

To reduce the number of explanatory variables we computed the following salivary cortisol parameters: the cortisol awakening response (CAR) across time points T1-T2 using the area under the curve with respect to increase-formula, overall diurnal cortisol level (AUC)_T1-6 for time points T1-T2-T4-T5-T6 using the area under the curve with respect to ground-formula (Pruessner et al., 2003), and the AM/PM ratio based on cortisol measurements at time points T1 and T6. Time point T3 was discarded when we computed CAR and AUC_T1-6 due to a large proportion of invalid data points for both measurement days (missing data 51 individuals, 27.3% of the sample). Skewed data for the overall diurnal cortisol were normalized using zero skewness logarithmic transformation AUC_t1-6 – constant (k=13.77), while the cortisol ratio AM/PM was normalized using zero skewness logarithmic transformation (T1/T6) + constant (k=1.29) (Holleman et al., 2022).

Thus, we built five separate sets of linear regression models using each of the cognitive domains as the dependent variable for one set of analyses as follows:

• -Model 1 was a bivariate regression with GDS score as the only independent variable.
• -Model 2 included GDS score along confounders age, years of education, sex, history of burnout syndrome, and the use of steroids or cortisone medication.
• -Model 3 included GDS score along diurnal cortisol levels CAR, AUC_T1-6 and AM/PM, one at a time, as additional explanatory variables along with significant confounder variables from model 2. Since the significant results of the models did not change when all covariates were used, only the results of the complete model 3 with GDS score and all three cortisol parameters are presented.
• -Model 4 tested if AD CSF biomarkers affected the associations between GDS score and cognitive performance by introducing as continuous explanatory covariates Aβ₄₂, T-tau and P-tau along GDS score, diurnal cortisol variables and those variables that were significantly associated with each of the cognitive domain score in model 3.

Interaction terms were built between GDS score (main explanatory variable) and each of those other explanatory variables that were significantly associated with cognitive scores according to model 4 (i.e, interaction GDS score*CSF biomarkers in relation to working memory domain). We used logarithmic transformation of the interaction terms if skewness >1.0 to normalize the data and improve the accuracy of the statistical analyses. We then retested the associations between the continuous GDS score (main explanatory variable) and the cognitive domain score (dependent continuous outcome variable) using stepwise backward linear regression models (standard criterion: probability-of-F-to-remove ≥ 0.1, standard listwise deletion) adjusted for confounders and for salivary cortisol level. This model has the advantage of reducing the chances of overfitting the data, assessing the joint predictive ability of all variables (Chowdhury and Turin, 2020).

Furthermore, in an ancillary analysis, we tested the association between GDS score and working memory after we stratified the sample by normal (≥ 550 ng/ml) or pathologic (< 550 ng/ml) CSF Aβ42 level according to the clinical laboratory cut-offs at the Karolinska University Hospital during the time of the study. Years of education was introduced as a confounder, in line with initial results from adjusted linear regressions. Also, CAR, AUG_t1-6 and AM/PM ratio, T-tau and P-tau proteins were introduced as covariates in these sub-group analyses using linear regression models. Models were also run within strata of cognitive impairment, i.e., among those impaired (MCI and dementias) and in SCD, except for the group SCD with pathologic CSF Aβ42 that was too small to fit a regression model (n=2).

Finally, we explored the association between the GDS score at baseline and change in domain scores at follow-up regarding memory domain and processing speed. The linear regression models were adjusted for salivary cortisol levels, i.e, CAR, AUC_T1-6 and AM/PM. Due to reduced sample size at follow-up and based on results from cross-sectional linear regressions at baseline, we have only used three confounders in these analyses, i.e., age at follow-up, sex, and years of education.

All tests were considered significant at alpha levels < 0.05, two-tailed. Statistical analyses were run using IBM SPSS Statistics for Windows, Version 28.0.1.0. Armonk, NY: IBM Corp.

3. Results

3.1. Demographic and clinical characteristics of the study population

Table 1 presents the demographic and clinical characteristics of the participants who had a valid GDS score. The average GDS score was 5.7 ± 4.0 (95%CI 5.1 to 6.3) with a median score of 5 (interquartile range (IQR) 7.0, 1^st Quartile 2 and 3^rd Quartile 9), indicating mild depressive symptomatology in this sample. The GDS score was deemed to be normally distributed with skewness value 0.478. In the group with GDS scores ≥ 5, there were more women, the participants were younger, and a self-reported history of burnout was more frequent. Additionally, antidepressant and anxiolytic medication usage was higher in this group. However, to ensure statistical power in our relatively small clinical sample, we included a continuous GDS score as the predictor variable in further analyses.

Table 1. Demographic and clinical characteristics of GDS participants (N=162).

	GDS score < 5 (N=75)	GDS score ≥ 5 (N=87)
	N (%)
Women	37 (49.3)	59 (67.8)*
Current smoker	9 (12.9)	8 (9.6)
APOE-ε4 carrier†	12 (32.4)	14 (41.2)
Burnout syndrome	14 (18.9)	36 (41.9)**
Steroid or cortisone medication	7 (9.6)	8 (10.0)
Antidepressant medication	9 (12.0)	35 (40.7)***
Anxiolytic medication	2 (2.7)	12 (14.1)*
Dementias	18 (24.0)	18 (20.7)
MCI	27 (36.0)	42 (48.3)
SCD	30 (40.0)	27 (31.0)
	mean ± SD (N)
Age	65.3 ± 8.0 (75)	60.7 ± 7.5 (87)***
Years of education	14.0 ± 3.2 (75)	13.7 ± 3.2 (87)
Cognitive domain (Z-score) at baseline
Memory	-1.2 ± 1.5 (66)	-1.1 ± 1.3 (71)
Working Memory	-0.5 ± 0.9 (57)	-0.9 ± 0.8 (56)**
Processing Speed	-0.8 ± 1.2 (63)	-0.8 ± 1.3 (65)
Perceptual Reasoning	0.1 ± 0.7 (53)	0.1 ± 0.9 (50)
General cognitive function	-0.8 ± 0.9 (66)	-0.8 ± 1.0 (64)
Cognitive domain (Z-score) at follow-up
Memory	-0.2 ± 1.0 (27)	-0.3 ± 0.9 (30)
Processing speed	-0.6 ± 1.2 (27)	-0.3 ± 1.2 (28)
Cognitive domain score change ‡
Memory	0.1 ± 0.6 (25)	0.3 ± 0.7 (26)
Processing speed	0.05 ± 0.6 (25)	0.03 ± 0.6 (22)
Cortisol parameters log-transformed
T1-awakening cortisol	2.1 ± 0.6 (71)	2.1 ± 0.6 (78)
T2 30-min awakening cortisol	2.3 ± 0.5 (70)	2.4 ± 0.5 (76)
T3 60-min awakening cortisol	2.1 ± 0.5 (59)	2.0 ± 0.6 (66)
T4 2 PM-cortisol	1.3 ± 0.6 (71)	1.1 ± 0.6 (78)
T5 4 PM-cortisol	0.9 ± 0.6 (71)	0.9 ± 0.9 (78)
T6-bedtime cortisol	0.2 ± 1.0 (70)	0.3 ± 1.0 (78)
AM/PM ratio	2.1 ± 0.8 (70)	2.1 ± 0.6 (78)
¶CAR	0.5 ± 1.2 (70)	0.7 ± 1.5 (75)
AUCT1-6	4.1 ± 0.6 (70)	4.1 ± 0.6 (75)
CSF-biomarkers log-transformed
¶Aβ42 (ng/L)	721.3 ± 273.3 (66)	731.9 ± 258.3 (73)
T-tau	5.9 ± 0.5 (66)	5.7 ± 0.6 (73)
P-tau	3.9 ± 0.4 (66)	3.8 ± 0.5 (73)

GDS Geriatric Depression Scale; MCI Mild cognitive impairment; SCD Subjective cognitive decline; WASI Wechsler Abbreviated Scale of Intelligence; CAR Cortisol awakening response; AUC Area under the curve.; MMSE Mini Mental State Examination.

^†APOE ε4-carriers missing n=91 (56.2% of the sample) (missing 38 out of 75 in GDS < 5 and 53 out of 87 in GDS ≥ 5).

^‡Score change was computed by subtracting baseline Z-score from Z-score at follow-up.

T1-T6 represent time points for salivary cortisol sampling (two-day averages winsorized at 3SD of the mean of the sample). The AM/PM ratio was based on T1/T6. All cortisol parameters, except ^¶CAR, were normalized using log-transformations due to significant skewness. CSF-biomarkers, except ^¶Amyloid (A)β42, were also log-transformed due to skewness.

Fisher's exact statistic was used to test differences in proportions regarding sex and clinical parameters, and ANOVA with F-statistic for continuous parameters with normal distribution (age, education, cognitive domain’s Z-scores and score change, and log-transformed cortisol parameters). All tests were considered significant at alpha levels < 0.05, 2-sided bivariate analyses: P-values < 0.05*, < 0.01**, < 0.001***.

The demographic and clinical characteristics of the GDS responders (n=162) were not different from those who did not complete GDS (non-responders) (n=25) (Supplementary table S1).

3.2. The relation between depressive symptoms and cognitive performance

A higher GDS score was associated with lower scores in the working memory and processing speed domains in the unadjusted linear regressions and in the adjusted Model 2 (Table 2). These associations remained significant in the fully adjusted models including salivary cortisol parameters (Model 3) (Table 2). No associations were found between GDS score and cognitive performance in other cognitive domains. Moreover, no associations could be demonstrated between GDS score and any of the cognitive domains in the adjusted models that included AD biomarkers Aβ42, T-tau and P-tau proteins (Model 4) (Table 2).

Table 2. The associations between GDS score, diurnal salivary cortisol levels and CSF biomarkers with the cognitive domains.

Cognitive domain	Explanatory variables	Model 1		Model 2		Model 3		Model 4
		B (95% CI); β	P-value	B (95% CI); β	P-value	B (95% CI); β	P-value	B (95% CI); β	P-value
Memory		N=137		N=127		N=118		N=110
	GDS score	0.036 (-0.023, 0.096); 0.103	0.232	-0.045 (-0.108, 0.019); -0.125	0.170	-0.055 (-0.122, 0.012); -0.152	0.104	-0.019 (-0.077, 0.038); -0.053	0.501
	CAR					0.180 (-0.009, 0.370); 0.155	0.062	0.110 (-0.056, 0.277); 0.097	0.191
	AUCT1-6					-0.290 (-0.679, 0.099); -0.139	0.143	-0.273 (-0.625, 0.079); -0.133	0.127
	AM/PM ratio					-0.101 (-0.475, 0.273); -0.050	0.594	-0.165 (-0.500, 0.170); -0.083	0.332
	Aβ42							0.002 (0.001, 0.003); 0.411	< 0.001
	T-tau							-1.273 (-2.774, 0.227); -0.494	0.095
	P-tau							1.392 (-0.364, 3.149) 0.441	0.119
Working memory		N=113		N=107		N=99		N=91
	GDS score	-0.051(-0.089, -0.012); -0.240	0.010	-0.062 (0.109, - 0.015); -0.294	0.010	-0.064 (0.114, - 0.013); -0.294	0.014	-0.027 (-0.071, 0.016); -0.127	0.218
	CAR					-0.064 (-0.218, 0.090); -0.084	0.412	-0.047 (-0.197, 0.103); -0.066	0.533
	AUCT1-6					0.003 (-0.294, 0.299); 0.002	0.986	0.042 (-0.253, 0.337); 0.033	0.780
	AM/PM ratio					-0.021 (-0.302, 0.261); -0.146	0.884	0.051 (-0.233, 0.335); 0.045	0.719
	Aβ42							0.001 (0.000, 0.002); 0.253	0.030
	T-tau							-1.348 (-2.551, -0.145); -0.817	0.029
	P-tau							1.597 (0.187, 3.006); 0.802	0.027
Processing speed		N=128		N=121		N=112		N=102
	GDS score	0.008 (-0.048, 0.064); 0.026	0.769	-0.069 (-0126, -0.012); -0.214	0.018	-0.062 (-0.121, -0.002); -0.188	0.042	-0.018 (-0.075, 0.040); -0.055	0.547
	CAR					0.057 (-0.114, 0.228); 0.054	0.510	0.058 (-0.105, 0.222); 0.059	0.480
	AUCT1-6					0.234 (-0.104, 0.572); 0.127	0.173	0.247 (-0.091, 0.584); 0.141	0.150
	AM/PM ratio					0.313 (-0.016, 0.642); 0.176	0.062	0.285 (-0.042, 0.612); 0.165	0.087
	Aβ42							0.000 (-0.001, 0.001); 0.067	0.463
	T-tau							-1.484 (-3.033, 0.064); -0.659	0.060
	P-tau							1.442 (-0.369, 3.252); 0.518	0.117
Perceptual reasoning		N=103		N=95		N=91		N=86
	GDS score	-0.009 (-0.049, 0.031); -0.044	0.658	-0.008 (-0.054, 0.037); -0.041	0.715	-0.007 (-0.055, 0.040); -0.034	0.770	0.011 (-0.032, 0.054); 0.057	0.601
	CAR					0.029 (-0.119, 0.178); 0.043	0.697	0.105 (-0.049, 0.260); 0.157	0.179
	AUCT1-6					-0.024 (-0.373, 0.325); -0.018	0.892	0.093 (-0.261, 0.448); 0.075	0.603
	AM/PM ratio					0.208 (-0.102, 0.518); 0.182	0.186	0.309 (-0.018, 0.636); 0.279	0.064
	Aβ42							0.000 (-0.001, 0.001); 0.045	0.706
	T-tau							-0.503 (-1.691, 0.685); -0.345	0.402
	P-tau							0.261 (-1.122, 1.643); 0.148	0.708
General cognitive function		N=130		N=121		N=112		N=105
	GDS score	0.000 (-0.042, 0.041); -0.001	0.990	-0.035 (-0.076, 0.005); 0.084	0.087	-0.029 (-0.072, 0.014); -0.120	0.178	0.005 (-0.033, 0.044); 0.023	0.785
	CAR					-0.017 (-0.149, 0.115); -0.020	0.801	0.000 (-0.123, 0.122); -0.001	0.995
	AUCT1-6					-0.042 (-0.293, 0.209); -0.029	0.741	-0.032 (-0.269, 0.205); -0.024	0.790
	AM/PM ratio					0.081 (-0.158, 0.319); 0.059	0.504	0.176 (-0.047, 0.399); 0.139	0.120
	Aβ42							0.001 (0.000, 0.001); 0.168	0.046
	T-tau							-0.597 (-1.590, 0.396); -0.348	0.235
	P-tau							0.572 (-0.584, 1.729); 0.272	0.328

Linear Regression Model: B Unstandardized beta; β standardized beta. CI Confidence Interval. GDS Geriatric Depression Scale. The cognitive domain score was the continuous dependent variable in all models. Independent covariates in: Model 1) GDS score (bivariate model); Model 2) GDS score and five confounding variables represented by age, sex (male 0/ female 1), years of education, burnout syndrome (dichotomous variable no 0/yes 1), current use of steroid or cortisone medication (dichotomous variable no 0/yes 1) (6 covariate model); Model 3) three other explanatory variables were added to the previous model representing diurnal salivary cortisol levels (cortisol awakening response CAR; diurnal cortisol AUC_T1-6; AM/PM ratio); and Model 4) GDS score, CSF biomarkers (Aβ₄₂, ln T-tau and ln P-tau) and significant covariates according to the results in Model 3 (i.e., years of education in all models; age was also added in models using memory and processing speed as dependent variables; sex and age were added in the model using general cognitive function as the dependent variable). All independent variables were continuous unless stated otherwise, with normal/normalized distributions. P-values were considered significant at 2-sided alpha < 0.05..

In discrepancy with the results of the linear regression Model 4 for working memory, in a stepwise backward linear regression introducing the natural logarithm-transformed interaction term between GDS and CSF Aβ42 (probability-of-F-to-remove ≥ 0.1; listwise deletion N=87) along the GDS, salivary cortisol parameters (i.e., CAR, AUG_t1-6 and AM/PM ratio), CSF neuropathologic biomarkers, and years of education as covariates, GDS was associated with working memory domain score (GDS unstandardized B -0.055, 95%CI -0.098 to -0.012, standardized β -0.262, p=0.012) along with the CSF neuropathologic biomarkers (Aβ42 unstandardized B 0.001, 95% CI 0.000 to 0.002, standardized β 0.229, p=0.044; T-tau unstandardized B -1.330, 95% CI -2.437 to -0.222, standardized β -0.839, p=0.019; and P-tau unstandardized B 1.466, 95% CI 0.161 to 2.771, standardized β 0.765, p=0.028), while the interaction term ln GDS*Aβ42 was not associated with the outcome (p=0.704). No significant associations were found between the working memory domain score and the interactions terms GDS*T-tau (p=0.302) or GDS*P-tau (p=0.398), nor did these terms modified the association between GDS and working memory domain as demonstrated using linear regression Model 4.

Thus, we chose to stratify the sample by CSF Aβ42 level, to further test the associations between GDS and the performance in the working memory domain.

3.3. The relation between depressive symptoms and working memory performance in the analyses stratified by CSF Aβ42

In patients with normal Aβ42 level, there was a negative association between GDS score and working memory domain score that no longer reached significance (standardized β -0.246, p=0.058). The relationship between GDS and working memory was not affected by the inclusion of the cortisol parameters, T- and P-tau in the regression model (Table 3). The results did not change when analyses were stratified by cognitive impairment (Supplementary Table S2).

Table 3. The associations between GDS score and working memory domain score in the stratified sample by CSF Aβ42 level.

CSF Aβ42 level (N)	Explanatory variables	Unstandardised B (95% CI)	Standardised β	P-value
Normal (64)	GDS score	-0.052 (-0.105, 0.002)	-0.246	0.058
≥ 550 ng/ml	CAR	0.004 (-0.164, 0.172)	0.006	0.963
	AUCT1-6	0.074 (-0.310, 0.457)	0.057	0.703
	AM/PM ratio	0.106 (-0.235, 0.447)	0.096	0.537
	T-tau	-1.089 (-2.733, 0.555)	-0.602	0.190
	P-tau	1.352 (-0.596, 3.300)	0.610	0.170
Pathologic (27)	GDS score	0.045 (-0.040, 0.129)	0.220	0.281
< 550 ng/ml	CAR	-0.388 (-0.780, 0.004)	-0.490	0.052
	AUCT1-6	0.231 (-0.278, 0.739)	0.216	0.355
	AM/PM ratio	-0.338 (-0.942, 0.266)	-0.279	0.256
	T-tau*	-2.190 (-4.231, -0.149)	-1.229	0.037
	P-tau*	2.971 (0.710, 5.231)	1.538	0.013

GDS Geriatric Depression Scale. CSF Aβ42 Cerebrospinal fluid amyloid β42. Linear regression models using independent covariates: GDS score, salivary cortisol levels (cortisol awakening response CAR; diurnal cortisol AUC_T1-6; AM/PM ratio) and the CSF biomarkers (ln T-tau and ln P-tau) (all representing explanatory variables), and years of education (significant confounder according to logistic regressions Model 2 through 4). All explanatory variables were continuous with normal/normalized distributions. P-values were considered significant at 2-sided alpha < 0.05: * P < 0.05.

In those with pathologic levels of Aβ42, there was no association between GDS and working memory domain score (standardized β 0.220, p=0.281). The performance in working memory domain, was associated with T- and P-tau (Table 3). However, after we excluded SCD (n=2) from this group, we found a significant negative association between CAR and working memory domain (unstandardized B -0.415, 95%CI -0.773 to - 0.057, standardized β -0.636, p=0.026), independently of tau-proteins (Supplementary Table S2).

3.4. The relation between depressive symptoms and change in cognitive performance at follow-up

In this clinical sample, we did not detect a significant relationship between depressive symptoms according to GDS score at baseline and change in cognitive performance at follow-up in memory (N=51: unstandardized β 0.009, 95% CI -0.046 to 0.064, standardized β 0.061, P-value=0.740) or processing speed (N=47: unstandardized B -0.021, 95% CI -0.072 to 0.031, standardized β -0.151, P-value=0.420) using the covariates age at follow-up, years of education and sex. Introducing baseline measurements of cortisol parameters CAR, AUG_t1-6 and AM/PM ratio as additional covariates in these models did not change the results.

4. Discussion

Our study aimed to examine the relationship between depressive symptomatology, salivary cortisol levels, and cognitive domains among memory clinic patients. The main finding of our study indicates that individuals with higher GDS score had poorer performance in the working memory and processing speed domains, independently of their cortisol levels. To further investigate this relationship, we conducted stratified analyses by CSF Aβ42 levels. However, we could not demonstrate significant associations between the GDS score or cortisol parameters and working memory score in patients with normal levels of CSF Aβ42 in this small sample. In individuals with pathologic CSF Aβ42 and cognitive impairment (MCI and dementias), we found that CAR, and not GDS score, was associated with performance in the working memory domain. Overall, our findings highlight the importance of considering both depressive symptoms and AD biomarkers when evaluating cognitive function in memory clinic patients.

Our main finding supports the results from some previous studies which also found working memory impairments among older adults with late life depression (Koenig et al., 2014). Studies of young adults with major depressive disorder have demonstrated a depression-specific deficit in updating emotional content in working memory (Joormann and Gotlib, 2008; Joormann et al., 2011; Levens and Gotlib, 2010; Yoon et al., 2014). However, other clinical studies examining depressive symptoms in older patients have reported impairments in episodic memory and executive function (Ahn et al., 2021), and poor performance in tests of objective memory, executive function, psychomotor speed, and global cognition (Seo et al., 2017). Clinical studies of minor and major depression have revealed cognitive deficits in executive function, memory, and attention (McDermott and Ebmeier, 2009). The reason why we did not find an association between the GDS and memory may be attributed to methodologic differences such as the samples studied, and the assessments used to detect depression. Firstly, our study included patients referred to the cognitive disorders’ clinic, without controls, while earlier clinical studies have compared patients diagnosed with depression with cognitively intact individuals free of depression. This approach may explain the larger panorama of cognitive impairments associated with depression in the latter studies. Secondly, we have used the self-reported GDS to assess affective symptomatology in our sample. The GDS scores indicated only mild depressive symptomatology which may have reduced the chance to find associations between more severe depressive symptoms and cognitive impairment in our study. Another study among patients with subjective cognitive complaints showed that GDS-15 scores did not correlate with cognitive test performances in word list memory, visual memory, and working memory (Grambaite et al., 2013). Furthermore, a study in a clinical sample with major depression found that the self-reported GDS scale was less accurate in detecting depressive symptomatology in the presence of cognitive impairment (Sacuiu et al., 2019).

We found that AD CSF biomarkers were associated with working memory performance. Therefore we stratified our sample by the level of CSF Aβ42, which is regarded as the main culprit in the pathologic AD cascade (Hardy and Selkoe, 2002).Thus, our secondary finding is that GDS and amyloid level in CSF interact in relation to working memory, such that the association remains (at a p <0.1) in those with normal Aβ42, but not in those with abnormal Aβ42, independently of cortisol levels. Additionally, we showed an association between CAR and working memory in those with pathologic CSF Aβ42. Earlier studies have reported impairment in working memory among depressed adults at different ages (Joormann and Gotlib, 2008; Joormann et al., 2011; Koenig et al., 2014; Levens and Gotlib, 2010; Yoon et al., 2014). However, these studies did not analyze CSF Aβ42 levels. Our observations could be explained by the fact that levels of CSF Aβ42 in neurodegenerative disease may modulate the expression of the depressive phenotype as patients suffering from incipient dementias may lose their self-awareness and fail to report depressive symptoms on GDS (Sacuiu et al., 2019). Neuroinflammation-mediated dysregulation of the HPA axis (Ahmad et al., 2021), which is often present in individuals with abnormal Aβ42 levels, may overshadow the association between depression and cortisol levels. Further studies are needed to fully understand the role of assessing CAR in cognitively impaired individuals with mild depressive symptoms.

Our study has several strengths, including its representativeness of memory clinic patients, which included three diagnostic groups with different cognitive levels (SCD, MCI and AD), detailed assessments of neuropsychological features and CSF AD biomarkers. We have also used repeated measures of cortisol at consecutive days, which allowed more exact estimates of the diurnal cortisol profile, and various questionnaires on demographic factors. However, it also has limitations, including a small sample size which limited the statistical power. The effects were only significant for some models whereas the associations lost statistical significance in the fully controlled models, for which the number of participants was even smaller. Another issue is that the beta weights are small, indicating low contribution to explaining the variance, although they were statistically significant. Another limitation is that cross-sectional analyses limited our ability to understand the temporal relationship between the associations. However, the longitudinal analyses we were able to carry out did not show a relationship between GDS and cognitive changes at follow-up. Although the use of self-reported data regarding depressive symptoms was considered the best method for younger and more educated samples (Enns et al., 2000), it has been shown to be less effective in detecting depression among patients of advanced age (Sayer et al., 1993) and among those cognitively impaired (Sacuiu et al., 2019). This aspect might also have affected the possibility to find associations between depression and cognitive performance in our sample of older adults from a memory clinic. However, there were no differences in demographic and clinical characteristics between those who completed GDS and those who did not. Furthermore, the self- and informant reports of cognitive decline were assessed during clinical interviews, without following a standardized study procedure or including a self-reported scale on cognitive decline. Therefore, we relied only on results from standardized psychometric testing, which further diminished the number of participants in some analyses. Also, the large number of statistical tests increased the probability of type I error.

In conclusion, we found that depressive symptomatology was an independent risk factor for working memory and processing speed impairments among patients from a memory clinic, independently of the level of salivary cortisol. However, levels of CSF Aβ42 may interact with depressive symptoms such that their association with cognition is no longer significant when amyloid pathology is present according to the threshold of CSF amyloid positivity.