Cerebral small vessel disease is associated with a higher incidence of depressive symptoms in a general elderly population: the AGES-Reykjavik Study

Abstract

Objective

The vascular depression hypothesis postulates that cerebral small vessel disease (CSVD) leads to depressive symptoms via disruption of brain structures involved in mood regulation. However, longitudinal data on the association between CSVD and depressive symptoms are scarce. We investigated the association between CSVD and incident depressive symptoms.

Methods

Longitudinal data are from the AGES-Reykjavik study of 1,949 participants free of dementia and without baseline depressive symptoms (74.6 years/56.6% women). MRI markers of CSVD, detected at baseline (2002–2006) and follow-up (2007–2011), included white matter hyperintensity volume (WMHV), subcortical infarcts, cerebral microbleeds, Virchow-Robin spaces and total brain parenchyma volume. Incident depressive symptoms were defined by the 15-item Geriatric Depression Scale (GDS-15) score≥6 and(or) use of antidepressant medication.

Results

Depressive symptoms occurred in 10.1% of the participants. The association for a greater onset of depressive symptoms was significant for participants having a 1 standard deviation increase in WMHV over time, new subcortical infarcts, new Virchow-Robin spaces, a 1 standard deviation lower total brain volume at baseline, or a 1 standard deviation decreased total brain volume over time, after adjustments for cognitive function, socio-demographic and cardiovascular factors. Results were qualitatively similar when change in the GDS-15 over time was used as the outcome instead of incident depressive symptoms.

Conclusions

Most markers of progression of CSVD over time and some markers of baseline CSVD are associated with concurrently developing new depressive symptoms. This study supports the vascular depression hypothesis.

Introduction

Depressive symptoms are often present in older individuals (1–4), and are associated with an increase in morbidity and risk for mortality (5, 6). The pathobiology of late-life depressive symptoms is incompletely understood, but it has been suggested that cerebral small vessel disease (CSVD) is involved (7, 8). The vascular depression hypothesis postulates that CSVD leads to depressive symptoms via disruption of deep and frontal brain structures or their connecting pathways involved in mood regulation (7, 8).

However, longitudinal data (9–17) on the association between CSVD and depressive symptoms are limited and findings are mixed. Two previous studies (12, 13) found an association between markers of CSVD and incident depressive symptoms. In addition, three studies (9, 10, 15, 16) showed an association with increased severity or recurrence of depressive symptoms at follow-up, but not with incident depressive symptoms, whereas two other studies (11, 17) did not find any association with depressive symptoms. These mixed findings may be due to study differences in sample size, follow-up duration, method of depression assessment (clinical interview (12–15) versus questionnaire (9–11, 16, 17)), evaluation of different symptom clusters, and(or) the source of populations investigated (selected (9, 10, 12, 17) versus community-based samples (11, 13–16)). In addition, some studies did not adjust the results for potential important confounders, such as cognitive function (11, 13, 14, 17) and cardiovascular factors (12, 15, 17).

In view of the above, we investigated in a large well characterized cohort, the prospective association between, on the one hand, markers of CSVD (white matter hyperintensity volume [WMHV], subcortical infarcts, cerebral microbleeds, Virchow-Robin spaces and lower total brain parenchyma volume) and, on the other, incident depressive symptoms. We additionally investigated whether any such association was stronger for CSVD in brain regions involved in mood regulation (deep and frontal) as compared to other regions (temporal and occipitoparietal).

Methods

Participants

For the present study, we used longitudinal data from the Age, Gene/Environment Susceptibility (AGES)-Reykjavik Study. The AGES-Reykjavik Study is a population-based cohort study originating from the Reykjavik Study, as described fully elsewhere (18). Briefly, from 2002–2006, 5,764 surviving participants of the Reykjavik Study were examined. From 2007–2011 there was a follow-up examination of all surviving participants who agreed to participate (n=3,316). Reasons for not attending the follow-up examination included: death (n=1,039); refusal (n=1,198); and lost to follow-up (could not be contacted by any means; n=211). The AGES-Reykjavik Study was approved by the National Bioethics Committee in Iceland (approval number=VSN-00-063), and by the National Institute on Aging Intramural Institutional Review Board. After complete description of the study to the subject, written informed consent was obtained.

Depressive symptoms

Depressive symptoms were assessed on all participants, with the 15-item Geriatric Depression Scale (GDS-15; score range, 0–15) (19, 20). Incident depressive symptoms were defined as a predefined GDS-15 cut-off score of 6 or higher (19, 20) at follow-up and(or) new use of antidepressant medication (tricyclics, selective serotonin reuptake inhibitors, other nontricyclics and monoamino oxidate inhibitors) assessed from medication bottles brought to the clinic. Individuals were excluded for the present analysis if they had depressive symptoms at baseline (GDS-15 score of 6 or higher and(or) use of antidepressant medication at baseline).

Brain MRI measures

Image acquisition

All eligible participants were offered a high-resolution brain MRI acquired on a study-dedicated 1.5-T system (Signa Twinspeed, General Electric Medical Systems). The same imaging protocol was used in the 2002–2006 and 2007–2011 examinations, described elsewhere (21, 22), and included the following sequences: 3D spoiled-gradient recalled T1-weighted, proton density/T2-weighted fast spin-echo, fluid-attenuated inversion recovery (FLAIR) and T2*-weighted gradient-echo type echoplanar (GRE-EPI). All images were acquired to give full brain coverage with slices angled parallel to the anterior commissure–posterior commissure line in order to give reproducible image views in the oblique-axial plane.

Image analysis

Several markers of CSVD were evaluated. WMHV and total brain parenchyma volume (an indicator of cerebral atrophy) were computed automatically with a previously described image analysis pipeline (23) and were expressed as the percentage of total intracranial volume. Lower total brain parenchyma volume was considered to be a marker of CSVD, as CSVD leads to generalized loss of brain parenchyma via, amongst others, microinfarcts (24) and loss of white matter integrity (25). Subcortical infarcts were evaluated as described previously (21) and defined as brain parenchyma defects not extending into the cortex, with a minimum diameter of 4 mm and a signal intensity equal to cerebrospinal fluid on all pulse sequences (T2-weighted, proton density-weighted and FLAIR), and surrounded by an area of high signal intensity on FLAIR images. Parenchymal defects in the subcortical area with evidence of hemosiderin on the T2*-weighted GRE-EPI scan were labelled as resorbed hematomas and were excluded from the definition of subcortical infarcts. Virchow-Robin spaces were evaluated separately and defined as defects in the subcortical area without a rim or area of high signal intensity on FLAIR and without evidence of hemosiderin on the T2*-weighted GRE-EPI scan. Presence of Virchow-Robin spaces was considered to be a marker of CSVD, as they are associated with endothelial dysfunction, which may play a role in the pathogenesis of CSVD (26). Cerebral microbleeds were defined as focal areas of signal void within the brain parenchyma visible on T2*-weighted GRE-EPI scans and were identified as described previously (22).

Potential confounding variables

As described elsewhere (18), dementia case-finding was conducted at baseline and at follow-up according to a three step procedure. Diagnosis of dementia (all sub-types) was made according to the Diagnostic and Statistical Manual of Mental Disorders (4^th edition) by a panel that included a geriatrician, a neurologist, a neuropsychologist and a neuroradiologist. The following variables were assessed by questionnaire: education (categorized into primary, secondary and college/university education), smoking history (ever versus never), alcohol intake (high [>median] versus low consumers [<median]; median for women 3.2 and for men 8.0 g/week, respectively) and anxiety symptoms (presence versus absence). Presence of anxiety symptoms was defined by a positive response to any of the following questions: “In the past month, have you felt anxious or frightened?”; “Were there times lately that you felt anxious?”; “Are there special situations that make you anxious?”; and “Have you ever had attacks of fear or panic?”. Presence of anxiety symptoms is a potential confounder, because anxiety symptoms are frequently present in individuals with depression, and are associated with cerebrovascular disease independently of depression (27, 28). Gait speed, a measure of physical performance (29), was the time needed to walk 6 meters at a usual pace. Hypertension was defined as systolic pressure >140 mmHg, diastolic pressure >90 mmHg and(or) use of anti-hypertensive medication. Body mass index (BMI) was calculated as measured weight divided by height squared. Diabetes was defined as a self-reported doctor’s diagnosis of diabetes, use of blood glucose-lowering medication and(or) fasting blood glucose level ≥7.0 mmol/l. Coronary calcium score (categorized into sex-specific quartiles), a measure of atherosclerosis, was based on Computed Tomography. The Digit Symbol Substitution Test (DSST), a measure of psychomotor speech, and the Mini Mental State Examination (MMSE), a measure of global cognitive function, were also administered (18, 30).

Analytic sample

Of the 3,316 participants who attended the follow-up examination, 709 had missing MRI data and another 195 had missing data on depressive symptoms at baseline or at follow-up. Missing MRI data was due to contraindications (n=278), refusal/nonattendance (n=360), or technical reasons (no qualitatively acceptable data available for all necessary sequences; n=71). In the remaining 2,412 participants, 138 were excluded because of a diagnosis of dementia at baseline (n=31) or follow-up (n=107). Finally, participants were excluded with presence of depressive symptoms at baseline (n=325). The final study sample thus consisted of 1,949 participants. Participants excluded for the present analysis were more likely to be older (75.6 versus 74.6 years), female (60.8 versus 56.6%), less educated (for primary school or less: 23.3 versus 18.9%) and to have diabetes (12.4 versus 9.1%), hypertension (82.0 versus 76.9%) and(or) stroke (9.1 versus 5.2%) (P-value for all<.05).

Statistical analysis

The percentage of missing values on potential confounders was minimal (maximum, 1.7%). We imputed those data using multiple imputation chained equations (10 datasets) (31). WMHV was logarithmically transformed to normalize its skewed distribution.

The statistical analysis proceeded in several stages. Logistic regression analysis was used to estimate the association between markers of baseline CSVD (baseline WMHV, expressed per 1 higher standard deviation [per +1 SD], presence of any (coded 0,1) subcortical infarcts, cerebral microbleeds and Virchow-Robin spaces and total brain parenchyma volume, expressed per 1 lower SD [per −1 SD]) and incident depressive symptoms. Analyses were repeated looking prospectively at change in, or progression of, markers of CSVD as the determinant and development of depressive symptoms over the same period. These markers were an increase in WMHV from baseline values (per +1 SD), any incident (coded 0,1) subcortical infarcts, cerebral microbleeds and Virchow-Robin spaces, and a decrease in total brain parenchyma volume from baseline values (per −1 SD).

We repeated the above analyses for a priori selected brain regions: deep (sub-cortical) areas (internal and external capsules, thalamus, striatum, hippocampus and amygdala combined), and frontal, temporal and occipitoparietal lobes. To efficiently summarize the pathology in each of these regions, we created a dichotomous composite score. This was calculated by assigning one point per CSVD marker based on the following cut-offs: regional WMHV, quartile 4 versus quartiles 1 to 3; subcortical infarcts, cerebral microbleeds and Virchow-Robin spaces, ≥1 versus 0 lesion(s) per region; and regional brain parenchyma volume, quartile 1 versus quartiles 2 to 4. The points for each marker were combined to compute a dichotomous composite score per region, which indicated high (≥2 points) or low (0 or 1 point(s)) burden of CSVD. A separate composite score was computed for baseline and progression over time of CSVD, respectively.

In addition, linear regression was used to evaluate the association between markers of CSVD and change of the GDS-15 score over time as the outcome.

All models were adjusted for the following potential confounders: baseline age, sex, DSST score, MMSE score, education level, presence of anxiety symptoms, gait speed, alcohol use, smoking, diabetes, BMI, hypertension, coronary calcium score, head coil and follow-up time (model 1); and additionally for baseline GDS-15 score (model 2). The composite scores for each region were additionally mutually adjusted for each other, with the exception of the scores for the frontal and deep brain region, which were not adjusted for each other, because both regions are thought to be involved in mood regulation (7, 8). We did not adjust the results for total disease load, because total disease load includes the load per investigated region (as indicated by the composite scores), and, thus, can be considered an overadjustment (32).

To test the robustness of the results, several sensitivity analyses were done. Logistic regression analyses were repeated with incident depressive symptoms defined only as GDS-15 score of 6 or higher as the outcome, irrespective of new use of antidepressant medication. To minimize the potential confounding effect of stroke, analyses were repeated after excluding individuals with baseline stroke or incident stroke during follow-up. In addition, to assess the possibility that depressive symptoms lead to CSVD (reverse ‘causality’), analyses were repeated with baseline presence of depressive symptoms as the exposure variable (we did not exclude individuals with baseline depressive symptoms; n=325) and markers of progression of CSVD over time as the outcome. All analyses were conducted with PASW Statistics (version 21).

Results

The mean age of the study population at baseline was 74.6 years and 56.6% were women (Table 1). In total, 10.1% (n=197) of the participants had incident depressive symptoms, of whom 38.1% (n=75) had a GDS-15 score of 6 or higher and 70.6% (n=139) had started using antidepressant medication (22 used tricyclics, 85 selective serotonin reuptake inhibitors and 43 other antidepressant medication). The mean time between the baseline and follow-up examination was 5.2±0.2 years.

Table 1.

Characteristics of both the total study population (n=1,949) and according to incident depressive symptoms

	Total study population (n=1,949)	SD, IQR or n	Individuals without incident depressive symptoms (n=1,752)	SD, IQR or n	Individuals with incident depressive symptoms (n=197)	SD, IQR or n
General baseline characteristics
Age (years)	74.6	4.6	74.5	4.6	75.2	4.9
Women	56.6	1,103	55.4	971	67.0	132
Education level
- Primary	18.9	367	18.5	324	21.9	43
- Secondary	51.3	996	51.3	897	50.5	99
- College/University	29.9	580	30.1	526	27.6	54
Presence of anxiety symptoms	26.7	521	24.8	435	43.7	86
Gait speed (m/s)
- Baseline	0.98	0.92–1.13	0.98	0.91–1.13	0.92	0.84–1.08
- Change over time	−0.09	0.23	−0.08	0.22	−0.13	0.25
Digit Symbol Substitution Test score
- Baseline score	33	10	33	10	31	11
- Change over time	−3	5	−3	5	−4	6
Mini Mental State Examination score
- Baseline score	28	27–29	28	27–29	28	26–29
- Change over time	−1	−3–1	−1	−3–1	−2	−4–1
Smoking status
- Ever	55.9	1,090	55.2	967	62.4	123
- Never	44.1	859	44.8	785	37.6	74
Alcohol use (g/week)	3	0–16	3	0–16	3	0–16
Body mass index (kg/m2)	27.2	4.0	27.1	4.0	27.5	4.3
Diabetes	9.1	177	9.1	160	8.6	17
Hypertension	76.9	1,499	76.9	1,347	77.2	152
Coronary calcium score (Agatston score)	212	28–744	211	30–751	193	26–693
Stroke	5.0	98	4.9	85	6.6	13
Brain MRI markers
Total white matter hyperintensity volume (ml)
- Baseline volume	11	6–21	11	6–20	12	6–22
- Change over time	6	8	5	8	7	8
Subcortical infarcts
- Baseline presence of any infarct(s)	7.1	139	6.7	117	11.2	22
- Any new infarct(s) over time	4.0	77	3.6	63	7.1	14
Cerebral microbleeds
- Baseline presence of any microbleed(s)	17.3	337	17.2	301	18.3	36
- Any new microbleed(s) over time	17.9	348	17.6	308	20.3	40
Virchow-Robin spaces
- Baseline presence of any space(s)	15.9	309	15.8	277	16.2	32
- Any new space(s) over time	2.6	51	2.2	39	6.1	12
Total brain parenchyma volume (ml)
- Baseline volume	1099	104	1100	104	1087	106
- Change over time	−31	23	−31	24	−35	20

Data are presented as percentage of participants (n), mean ± standard deviation (SD) or median (interquartile range, IQR).

The results of the analysis with markers of baseline CSVD showed that subcortical infarcts and a lower total brain parenchyma volume were statistically significantly associated with a higher incidence of depressive symptoms, after adjustment for potential confounders (Table 2, model 1). Further adjustment for baseline GDS-15 scores did not materially change these results (model 2).

Table 2.

Associations between markers of baseline and progression over time of cerebral small vessel disease and incident depressive symptoms

Determinants	Model	Incident depressive symptoms
		Odds ratio	95%CI	P-value
White matter hyperintensity volume (% ICV)
- Per +1 SD volume at baseline	1	1.04	0.89; 1.21	.67
	2	1.02	0.88; 1.19	.78
- Per +1 SD volume change over time	1	1.24	1.09; 1.42	.002
	2	1.21	1.06; 1.39	.007
Subcortical infarcts
- Baseline presence of any infarct(s) vs. no infarcts	1	1.90	1.15; 3.14	.012
	2	1.83	1.10; 3.05	.021
- Any new infarct(s) over time vs. no new infarcts	1	2.39	1.28; 4.48	.007
	2	2.31	1.21; 4.39	.011
Cerebral microbleeds
- Baseline presence of any microbleed(s) vs. no microbleeds	1	1.15	0.77; 1.72	.50
	2	1.10	0.73; 1.66	.64
- Any new microbleed(s) over time vs. no new microbleeds	1	1.31	0.92; 1.87	.16
	2	1.36	0.98; 1.86	.12
Virchow-Robin spaces
- Baseline presence of any space(s) vs. no spaces	1	1.09	0.73; 1.61	.69
	2	1.08	0.70; 1.66	.71
- Any new space(s) over time vs. no new spaces	1	3.44	1.71; 6.91	.001
	2	3.75	1.83; 7.68	<.001
Total brain parenchyma volume (% ICV)
- Per −1 SD volume at baseline	1	1.23	1.05; 1.45	.012
	2	1.23	1.04; 1.45	.017
- Per −1 SD volume change over time	1	1.30	1.08; 1.57	.006
	2	1.32	1.09; 1.59	.004

Model 1: adjusted for baseline age, sex, Digit Symbol Substitution Test score, Mini Mental State Examination score, education level, presence of anxiety symptoms, gait speed, alcohol use, smoking, diabetes, body mass index, hypertension, coronary calcium score, head coil and follow-up time. Model 2: additionally adjusted for baseline 15-item geriatric depression scale (GDS-15) score.

Abbreviations: CI: confidence interval; SD: standard deviation; ICV: intracranial volume.

In addition, the results of the analysis with markers of progression of CSVD over time showed that an increase in WMHV over time, incident subcortical infarcts and Virchow-Robin spaces and a decrease in total brain parenchyma volume over time were statistically significantly associated with a higher incidence of depressive symptoms (Table 2, models 1 and 2).

The composite scores of baseline (Figure 1, Panel A) and the change in pathology over time (Panel B) in the deep brain region was most strongly and statistically significantly associated with a higher incidence of depressive symptoms. In addition, the composite baseline score of the frontal brain region was statistically significantly associated with a higher incidence of depressive symptoms.

Figure 1.

Associations between brain region-specific composite scores of baseline (panel A) and progression over time (panel B) of cerebral small vessel disease and incident depressive symptoms.

Composite scores indicate high versus low burden of cerebral small vessel disease per region. The composite scores were mutually adjusted for each other (except for the frontal and deep brain region which were not associated for each other) and for all potential confounders. For further explanation: see text.

The results of the analyses with change of the GDS-15 score over time as the outcome were qualitatively similar to the results of the analyses with incident depressive symptoms (Table 3, models 1 and 2). The association for a greater GDS-15 score over time was statistically significant for higher baseline WMHV, an increase in WMHV over time, baseline and incident subcortical infarcts, incident cerebral microbleeds, baseline, incident Virchow-Robin spaces and a decrease in total brain parenchyma volume over time (Table 3, models 1 and 2).

Table 3.

Associations between markers of baseline and progression over time of cerebral small vessel disease and change of the GDS-15 score over time

Determinants	Model	Change of GDS-15 score over time
		β coefficient	95%CI	P-value
White matter hyperintensity volume (% ICV)
- Per +1 SD volume at baseline	1	0.11	0.04; 0.18	.003
	2	0.11	0.04; 0.18	.001
- Per +1 SD volume change over time	1	0.16	0.09; 0.24	<.001
	2	0.20	0.13; 0.26	<.001
Subcortical infarcts
- Baseline presence of any infarct(s) vs. no infarcts	1	0.27	−0.01; 0.54	.05
	2	0.34	0.08; 0.59	.01
- Any new infarct(s) over time vs. no new infarcts	1	0.34	−0.02; 0.70	.07
	2	0.40	0.06; 0.74	.02
Cerebral microbleeds
- Baseline presence of any microbleed(s) vs. no microbleeds	1	−0.03	−0.22; 0.15	.73
	2	−0.02	−0.18; 0.17	.98
- Any new microbleed(s) over time vs. no new microbleeds	1	0.20	0.02; 0.38	.03
	2	0.18	0.01; 0.36	.04
Virchow-Robin spaces
- Baseline presence of any space(s) vs. no spaces	1	0.19	−0.00; 0.38	.05
	2	0.18	0.00; 0.36	<.05
- Any new space(s) over time vs. no new spaces	1	0.80	0.37; 1.22	<.001
	2	0.74	0.34; 1.13	.001
Total brain parenchyma volume (% ICV)
- Per −1 SD volume at baseline	1	0.04	−0.04; 0.12	.32
	2	0.06	−0.01; 0.14	.11
- Per −1 SD volume change over time	1	0.08	0.02; 0.15	.02
	2	0.08	0.01; 0.15	.02

Model 1: adjusted for baseline age, sex, Digit Symbol Substitution Test score, Mini Mental State Examination score, education level, presence of anxiety symptoms, gait speed, alcohol use, smoking, diabetes, body mass index, hypertension, coronary calcium score, head coil and follow-up time. Model 2: additionally adjusted for baseline 15-item geriatric depression scale (GDS-15) score.

Abbreviations: CI: confidence interval; SD: standard deviation; ICV: intracranial volume.

Sensitivity analyses

Results were qualitatively similar when we repeated the analyses with incident depressive symptoms defined only as a GDS-15 score of 6 or higher, irrespective of new use of antidepressant medication (online data supplement, Table S1). When we excluded individuals with baseline stroke or incident stroke during follow-up (n=152), results did not materially change (online data supplement, Table S2). Presence of depressive symptoms at baseline was not statistically significantly associated with markers of progression of CSVD (online data supplement, Table S3).

Discussion

The present study investigated the association between markers of CSVD and incident depressive symptoms and had two main findings. First, various markers of CSVD were associated with a higher incidence of depressive symptoms. This association was statistically significant for an increase in WMHV over time, baseline and incident subcortical infarcts, incident Virchow-Robin spaces, a lower total brain parenchyma volume at baseline and a decrease in total brain parenchyma volume over time. These results were independent of cognitive function, education level, physical performance, anxiety symptoms and cardiovascular factors. In addition, results were qualitatively similar when change in the GDS-15 score over time was used as the outcome instead of incident depressive symptoms. Second, CSVD located in the deep brain region was, as compared to other brain regions, more strongly associated with a higher incidence of depressive symptoms. To our knowledge, this is the clearest demonstration to date that CSVD is a risk factor for depressive symptoms.

Our findings are in accordance with previous cross-sectional studies (13, 14, 33) which consistently have shown an association between markers of CSVD and depression. In addition, two previous, smaller longitudinal studies, the 3City-Dijon study (13) and the LADIS (12), found an association between higher WMHV at baseline and a higher incidence of depressive symptoms after a follow-up of 4 and 3 years, respectively. Furthermore, the SMART-Medea study (9) found an association between lacunar infarcts in deep white matter tracts and an increased severity and more fluctuating course of depressive symptoms, in particular motivational/apathy related-symptoms of depression, during a follow-up of 3.5 years. In addition, the Cardiovascular Health Study (16) and a neuroimaging substudy of the Rotterdam Study (15) showed an association between higher WHMV and(or) cerebral infarcts on the one hand and worsening and recurrence of depressive symptoms on the other after 4 years of follow-up. In contrast, two other longitudinal studies, a post-hoc analysis of a clinical trial (17) and the Baltimore Longitudinal Study of Aging (11), did not find any association between WMHV at baseline and incident depressive symptoms. The latter studies were, however, relatively small (n<550) (11, 17) and(or) had a relatively short follow-up duration (<3 years) (17), which may have led to an underestimation of the association between CSVD and depressive symptoms. The present study extends previous research because of its large population-based sample of older individuals, long follow-up duration, the comprehensive brain MRI assessment of a wide spectrum of CSVD markers determined at baseline and at follow-up, and the extensive characterization of participants which enabled us to adjust for a series of potential confounders.

CSVD may lead to depressive symptoms via damage to deep and frontal brain structures involved in mood regulation (7, 8). In accordance, the present study found that CSVD in the deep brain region was, as compared to other regions, more strongly associated with a higher incidence of depressive symptoms, although the 95%CI of the OR for disease in the deep brain region did overlap with those of other regions.

Other underlying mechanisms may, however, explain the observed associations. First, it has been suggested that the association between CSVD and depressive symptoms exists because late-life depressive symptoms represent an early manifestation of (vascular) dementia (34). For the present study, however, we excluded individuals with dementia at baseline or at follow-up. In addition, results were adjusted for scores on the DSST and the MMSE, tests that evaluate multiple cognitive functions (35). Second, other factors may be independently related to both CSVD and depressive symptoms, such as anxiety, cardiovascular factors and stroke. The associations between CSVD and incident depressive symptoms were, however, independent of anxiety symptoms and cardiovascular factors. In addition, the associations did not materially change when we excluded individuals with stroke. Third, we cannot exclude the possibility that the observed associations reflect reverse causation. Indeed, previous studies (36, 37) have shown an association between depression and incident cardiovascular disease, including cerebrovascular disease. Although it is not fully understood how depression might lead to vascular disease, possible mechanisms include low-grade inflammation, endothelial dysfunction, platelet dysfunction and unfavorable lifestyle habits (38). It is unclear, however, why depression would lead to vascular disease in specific brain regions, e.g. the deep brain region involved in mood regulation. In addition, presence of depressive symptoms at baseline was not statistically significantly associated with markers of progression of CSVD over time in the present study, although the 95%CIs of the effect estimates do not exclude the possibility of such an association. Fourth, it has been suggested that associations between CSVD and depression may be (partially) attributable to apathy (39). Apathy overlaps with depression, but may be a distinct syndrome (40). In the present study, we did not evaluate apathy and this issue needs further study.

The present study showed that most markers of progression of CSVD over time were associated with incident depressive symptoms, but only some markers of baseline CSVD. This may be due to the design of the present study with exclusion of individuals with depressive symptoms at baseline. This may have led to an underestimation of the association between baseline CSVD and development of depressive symptoms, but not between progression of CSVD over time and depressive symptoms, because individuals with depressive symptoms at baseline were most likely those with the strongest association between lifetime accumulation of CSVD (which is reflected by baseline CSVD) and depressive symptoms.

We analyzed depressive symptoms both as a dichotomous and a continuous outcome. The results of these analyses were qualitatively similar, except that more associations were statistically significant with change in the continuous GDS-15 score than with the dichotomous incident depressive symptoms variable. This difference may be due to the fact that, in general, analyses with a continuous outcome have higher statistical power than analyses with a dichotomous outcome. Indeed, studying depression on a continuum has the merit that not only information on extremes is used, but that all available information is exploited.

There are a number of limitations to the present study. First, incident depressive symptoms were assessed by questionnaire and use of antidepressant medication, but not by a structured interview. Therefore, no information was available on clinical depression. Nevertheless, the sensitivity and specificity of questionnaire measures as compared to a depression diagnosis based on a structured interview are high (>80%) (20). Yet, the prevalence of depressive symptoms is greater, in particular in older individuals (2, 4). Furthermore, late-life depressive symptoms, even in the absence of a diagnosis of a major depressive disorder, are associated with a greatly increased morbidity and mortality risk (5, 6). Second, misclassification of incident depressive symptoms may have occurred because antidepressant medication is also prescribed for other reasons. The results were, however, qualitatively similar when GDS-15 scores alone were used as the outcome. Third, the present study is the first to evaluate the association between brain region-specific composite scores and depressive symptoms, and further study is, therefore, needed to confirm the present findings (e.g. using voxel-based morphometric analysis). Fourth, a limitation of the analysis with markers of progression of CSVD over time as the determinant is that progression of CSVD and incident depressive symptoms occur in the same time interval and cannot be assigned to a given time point within this interval. Finally, we used cerebral atrophy and Virchow-Robin spaces as markers of CSVD. Cerebral atrophy is, however, an indirect measure of vascular disease and is also strongly determined by other factors, in particular the process of neurodegeneration. We therefore cannot exclude the possibility that the association between lower total brain parenchyma volume and depressive symptoms is due to factors other than CSVD. In addition, the etiology of Virchow-Robin spaces is currently incompletely understood, and this issue requires further study.

In conclusion, the present study shows that most markers of progression of CSVD over time and only some markers of baseline CSVD are independently associated with a concurrent development of higher incident depressive symptoms. From a clinical point of view, this association is important as it suggests that CSVD is a target for treatment and prevention strategies of late-life depression. Further study is needed to elucidate which factors contribute to CSVD and whether such factors can be therapeutic targets for late-life depression.