Baseline sample and measurements

A cross-sectional survey of a geographically defined population was carried out in 2001. The sampling procedure and measurements have been described previously (Kim et al, Reference Kim, Stewart and Shin2003a , Reference Kim, Stewart and Shin2004). In brief, 732 community residents aged 65 years or over, on national residents registration lists within two areas of Kwangju, South Korea, were interviewed. Interviews included a fully structured diagnostic interview for depression, examinations for vascular disease risk and formal assessment of disability and cognitive function.

Follow-up evaluation

Follow-up evaluation was carried out in 2003. The mean (s.d.) follow-up period was 2.4 (0.3) years. Diagnosis of depression (GMS-AGECAT) and information on new-onset vascular disease and risk (stroke, heart disease, hypertension and diabetes) during the 2-year interval were ascertained. The vascular examination, WHODAS II schedule and blood assays described above were also repeated. Responses to certain individual GMS questions at follow-up were also extracted for exploratory secondary analyses. These consisted of four psychological depressive symptoms (depressed mood, tearfulness, guilt and pessimism) and three vegetative symptoms (poor appetite, weight loss and sleep disturbance).

Statistical analysis

For the analysis reported here, participants with depression at baseline were excluded, and depression at follow-up was treated as the dependent variable for primary analyses. Baseline characteristics of participants and non-participants at follow-up were compared. Initial unadjusted analyses were carried out to investigate vascular factors associated with incident depression. For categorical independent variables, presence at baseline and onset during the follow-up period were considered separately. For continuous independent variables, baseline levels and changes in levels over the follow-up period were considered separately. Incident depression was compared across quintiles of baseline/change scores. Where curvilinear relationships were suggested, associations were tested entering both linear and quadratic terms for independent variable quintiles. Associations between independent and dependent variables were further analysed using stepwise logistic regression models to investigate confounding by demographic characteristics, and mediation by change in level of disability (WHODAS II) from base-line to follow-up.

Follow-up status

Of 732 participants at baseline, 631 (86.2%) did not have depression and formed the potential baseline population. Of these, 521 (83%) completed all evaluations at follow-up and comprised the study sample. Of the remaining 110, contact could not be established with 40 (36%), 23 (21%) had died, 21 (19%) refused to participate, 18 (16%) had changed address and 8 (7%) were too unwell to participate. Baseline characteristics of participants and non-participants at follow-up are compared in Table 1. There were no substantial differences in the demographic characteristics, disability and cognitive function between the two groups.

Table 1 Comparison of baseline characteristics between eligible participants who completed both examinations and those lost to follow-up

Baseline status	Completed follow-up (n=521)	Lost to follow-up (n=110)	Statistical coefficient 1	P
Age, years: mean (s.d.)	72.7 (5.5)	73.2 (5.7)	t=1.53	0.093
Gender, women: n (%)	287 (55.1)	65 (59.1)	χ2=1.1, d.f.=1	0.30
Education, years: median (IQR)	1 (0–6)	2 (3–9)	U=40 253	0.12
WHODAS II scores, median (IQR)	2 (0–7)	3 (0–8)	U=41 857	0.34
MMSE scores, median (IQR)	24 (21–27)	25 (20–28)	U=42 254	0.51

IQR, interquartile range; WHODAS II, World Health Organization Disability Assessment Schedule II; MMSE, Mini-Mental State Examination

1. t-test, χ²-test or Mann–Whitney U-test, as appropriate

Descriptive data

Of the 521 participants, 63 (12%) incident cases of depression were identified. Preexisting stroke, heart disease, hypertension and diabetes had been identified in 21 (4%), 99 (19%), 141 (27%) and 47 (9%) participants, respectively. New-onset cases of stroke, heart disease, hypertension and diabetes were found in 20 (4%), 31 (6%), 57 (11%) and 26 (5%) participants, respectively. Mean (s.d.) baseline vascular examination measurements were as follows: systolic blood pressure 146 (25) mmHg; diastolic blood pressure 88 (23) mmHg; blood glucose 106 (54) mg/dl; total cholesterol 175 (35) mg/dl; HDL cholesterol 49 (15) mg/dl; LDL cholesterol 95 (32) mg/dl; triglycerides 156 (89) mg/dl. Mean (s.d.) changes in levels from baseline to follow-up were as follows: systolic blood pressure +2 (28) mmHg; diastolic blood pressure -3 (24) mmHg; blood glucose +2 (54) mg/dl; total cholesterol +6 (32) mg/dl; HDL cholesterol -1 (13) mg/dl; LDL cholesterol +14 (36) mg/dl; triglycerides +15 (91) mg/dl; and WHODAS II score +1 (12) units.

Associations between vascular disorders and incident depression

Unadjusted associations between vascular disorders and incident depression are summarised in Table 2. Incident depression was significantly associated with heart disease at baseline and with stroke occurring during the follow-up period. Associations with baseline or incident hypertension or diabetes were all positive, but were weak and not statistically significant.

Table 2 Associations between self-reported vascular risk factors and incidence of depression (n=521)

Self-report	Pre-existing risk factor			Incident risk factor
	OR (95% CI)	P	OR (95% CI)	P
Stroke	1.6 (0.6–4.4)	0.37	2.3 (1.0–5.1)	0.040
Heart disease	2.1 (1.3–3.3)	0.003	1.1 (0.5–1.1)	0.83
Hypertension	1.1 (0.7–1.8)	0.59	1.5 (0.8–2.8)	0.16
Diabetes	1.4 (0.7–2.7)	0.38	1.0 (0.4–2.5)	0.99

Associations between incident depression and lipid levels are summarised in Table 3. For baseline total cholesterol, a significant U-shaped association with incident depression was observed, with highest risk in those with lower or higher baseline levels. This pattern was similar but not significant for LDL cholesterol. For HDL cholesterol, a strong and significant linear association was found between decreasing levels and increased incidence of depression. For changes in lipid levels over the follow-up period, U-shaped associations were suggested for total and LDL cholesterol, with highest risk in participants with declining or increasing levels. However, associations fell just short of statistical significance. No associations were found between incident depression and baseline blood pressure or blood glucose, nor with changes in these factors over the follow-up period (data not shown).

Table 3 Unadjusted associations between lipid levels and incidence of depression

Quintiles of exposure	Incident depression (%) by quintiles of independent variables
	Total cholesterol	HDL cholesterol	LDL cholesterol	Triglycerides
Baseline score
1 (lowest)	21.3	22.3	19.0	15.7
2	15.4	19.0	17.3	13.9
3	10.0	12.2	15.0	15.0
4	10.7	10.0	9.9	13.8
5 (highest)	18.3	9.7	13.8	17.3
P (linear)	0.41	0.005	0.16	0.81
P (quadratic)	0.034	0.71	0.63	0.58
Change in score
1 (decrease)	24.0	14.0	20.0	19.6
2	14.8	15.4	15.4	13.7
3	11.3	14.0	12.0	12.5
4	14.9	18.0	13.7	16.3
5 (increase)	20.0	23.1	24.0	22.2
P (linear)	0.65	0.21	0.71	0.62
P (quadratic)	0.072	0.56	0.082	0.96

HDL, high-density lipoprotein; LDL, low-density lipoprotein

Multivariate analysis

Results of logistic regression analyses are summarised in Table 4. In general, odds ratios were not changed substantially following each adjustment. In the fully adjusted model, independent associations were found with baseline heart disease and lower level of HDL cholesterol, and the association with incident stroke bordered on statistical significance. The quadratic term for the relationship between quintiles of baseline total cholesterol and incident depression also remained significant after adjustment for all factors listed in Table 4 (unadjusted odds ratio=1.32, P=0.009; adjusted odds ratio=1.29, P=0.015).

Table 4 Adjusted associations between vascular factors and incidence of depression. Odds ratios derived from logistic regression models are displayed with 95% confidence intervals

Association	Pre-existing heart disease	Incident stroke	Lower HDL cholesterol 1
Unadjusted	2.1 ** (1.3–3.3)	2.3 * (1.0–5.1)	1.3 ** (1.1–1.6)
Adjusted for age, gender, education	2.4 ** (1.5–4.0)	2.4 * (1.0–5.4)	1.3 ** (1.1–1.6)
plus change in disability (WHODAS II)	2.3 ** (1.4–3.8)	2.3 (1.0–5.2)	1.3 ** (1.1–1.6)
plus baseline cognitive function (MMSE)	2.2 ** (1.3–3.7)	2.2 (1.0–5.0)	1.3 * (1.1–1.6)

HDL, high-density lipoprotein; WHODAS II, World Health Organization Disability Assessment Schedule II; MMSE, Mini-Mental State Examination

1. Entered in decreasing quintiles

* P < 0.05

** P < 0.01

Secondary analyses

Of the 521 followed-up participants who were not depressed at the baseline, 7 (1%) were taking antidepressants. No marked or consistent differences were found in the results when the above analyses were repeated excluding this group. Results were also not changed if baseline WHODAS II score was entered, rather than change from baseline to follow-up.

The prevalence at follow-up of four psychological depressive symptoms were: depressed mood 42%, tearfulness 31%, guilt 14%, and pessimism 23%; and of three vegetative symptoms were: poor appetite 33%, weight loss 27% and sleep disturbance 42%. In a secondary analysis, pre-existing heart disease was significantly associated with depressed mood, tearfulness and guilt (P<0.05) and was associated with pessimism at borderline levels of statistical significance (P=0.060). Incident stroke was significantly associated with all four of these symptoms (P<0.05). No significant associations were found between either of these factors and the three vegetative symptoms (P>0.20). Lower baseline HDL cholesterol levels, on the other hand, were significantly associated with all seven symptoms (P<0.05). The quadratic term for total cholesterol was significantly associated with all but five outcomes (P<0.05) and was associated with the remaining two at borderline levels of significance (for guilt P=0.064; for pessimism P=0.071).

Methodological issues

Most previous community studies investigating the association between vascular risk factors and depression in older people have been cross-sectional in design (Reference Stewart, Prince and MannStewart et al, 2001; Reference Kim, Stewart and ShinKim et al, 2004). This limits the extent to which causal relationships can be clarified, particularly for factors such as blood pressure and lipid levels, which may well be altered in depressed states. Previous prospective studies have been limited in number and also in the use of brief screening instruments to define depression (Reference Lyness, King and ConwellLyness et al, 2000), or by the specific nature of the cohorts analysed (Reference Cervilla, Prince and Rabe-HeskethCervilla et al, 2004). Strengths of our study were that prospective data on both depression and potential risk factors were obtained from a community population, and that depression was ascertained using a widely validated diagnostic instrument. The follow-up rate was reasonable and not apparently differential with respect to risk factors of interest. A limitation of the study was that information on some vascular factors such as stroke and heart disease relied on self-reported diagnoses and corroboration by medical records was not possible. Neuroimaging was also not feasible in this population. Furthermore, it should be borne in mind that most participants will have had moderate levels of depression, and the results may not be generalisable to secondary care clinical samples with more severe syndromes.

Heart disease, stroke and depression

Reported heart disease was significantly associated with incidence of late-life depression in the present study, consistent with other research (Reference Hippisley-Cox, Fielding and PringleHippisley-Cox et al, 1998). Depression following myocardial infarction has been associated with higher mortality, and it is likely that the true association between cardiovascular disease and depression is underestimated because of differential mortality. Associations between previous stroke and depression are also well recognised in community samples (Reference Fuh, Liu and WangFuh et al, 1997; Reference Stewart, Prince and MannStewart et al, 2001; Reference Kim, Stewart and ShinKim et al, 2004). In our study the association was stronger for stroke occurring during the follow-up period than for that reported at baseline. This is likely to reflect a reduction in risk with increased duration since exposure, but might also be accounted for by differential attrition. As with previous cross-sectional studies, we found that this association was not entirely explained by level of disability (Reference Stewart, Prince and MannStewart et al, 2001; Reference Kim, Stewart and ShinKim et al, 2004) which, as for heart disease, may reflect either specific biological causal pathways or unmeasured psychological factors (such as the impact of the diagnosis). In this respect, it is noteworthy that the association (in contrast to that of lipid levels) appeared to be principally explained by cognitive rather than vegetative symptoms, which supports the latter causal pathway over the former.

Total cholesterol levels and depression

Previous studies investigating the association between total cholesterol levels and depression in later life have reported conflicting findings. Some have found significant associations with low cholesterol levels (Reference Morgan, Palinkas and Barrett-ConnorMorgan et al, 1993), while others have found no associations (Reference Blazer, Burchett and FillenbaumBlazer et al, 2002; Reference Kim, Stewart and ShinKim et al, 2004). In this prospective study we found that risk of depression at follow-up was associated with both high and low total cholesterol levels at baseline. Incident depression also appeared to be associated with both rising and falling total cholesterol levels over the follow-up period, although these associations fell below conventional levels of statistical significance. These findings are likely to represent mixed processes. High cholesterol, as a risk factor for arterial disease, may also be a risk factor for depression. Low cholesterol, on the other hand, is associated with frailty and poor health (Reference Corti, Guralnik and SaliveCorti et al, 1997), which in turn are important risk factors for depression. Hypocholesterolaemia might also have direct risk effects, for example through reduced serotonin levels (Reference EngelbergEngelberg, 1992). Unfortunately data were not collected on cholesterol-lowering agents. However, we believe that the use of these agents is rare in this population and is unlikely to explain our observed associations.

Vascular risk factors and depression

In a previous cross-sectional analysis of the baseline population for this study, we had found that low HDL cholesterol levels were associated with depression (Reference Kim, Stewart and ShinKim et al, 2004). In this prospective analysis, low HDL cholesterol also predicted incident depression. Furthermore, there were no apparent changes in HDL cholesterol levels in participants who became depressed, reducing the likelihood of reverse causality as an explanation. White matter abnormalities on neuroimaging are found more frequently than expected in people with late-life depression (Reference de Groot, de Leeuw and Oudkerkde Groot et al, 2000), supporting vascular aetiology in at least some cases. However, population-based studies have frequently failed to find associations between depression and conventional vascular risk factors such as hypertension and diabetes (Reference Jones-Webb, Jacobs and FlackJones-Webb et al, 1996; Reference Rajala, Keinanen-Kiukaanniemi and KevelaRajala et al, 1997; Reference Stewart, Prince and MannStewart et al, 2001). Where associations have been found, it is unclear whether these are specific to the disorders themselves or explained by well-recognised associations between depression and worse physical health. The association with an atherogenic lipid profile is consistent with a vascular basis for late-life depression, but the weak or absent associations with reported hypertension or diabetes (or with recorded blood pressure or blood glucose) are not supportive. A neuropathological study also found that late-life depression was associated with arterial atheroma rather than microvascular disease (Reference Thomas, Ferrier and KalariaThomas et al, 2001), and it is possible that vascular mechanisms underlying late-life depression have a different pattern to those implicated in stroke and vascular dementia. However, it is also possible that associations are explained by earlier unrecalled affective states which increase risk for both cardiovascular pathology and late-life depression. Vascular disease, vascular risk factors and depression are common conditions in older populations, and are associated with high levels of morbidity. Our findings provide some epidemiological support for vascular factors in the aetiology of late-life depression (Reference Alexopoulos, Meyers and YoungAlexopoulos et al, 1997). However, prospective data over a longer follow-up period (possibly from early adulthood) are likely to be required before firm conclusions can be drawn concerning the direction of association between these two processes.