Study population

Potential participants for this study were recruited from all inhabitants aged 65 years or over recorded in national residents registration lists within two areas (one urban, one rural) of Kwangju, South Korea, in 2001. The study was approved by Chonnam National University Hospital Review Board. After sending a letter explaining the purpose of the study to all eligible older people, written informed consent was obtained from all participants.

Assessment procedures and measures

This study was conducted in two phases. Sixteen graduate-level research assistants, trained and supervised by the project psychiatrist, carried out home-based interviews with participants and their family members. This included a fully structured diagnostic interview for depression, information on vascular risk/disease and data on demographic characteristics, disability and cognitive function. At a second interview (attempted in all participants), further examination for vascular risk/disease and blood tests for APOE genotype were administered by two expert teams consisting of a psychiatrist, a senior nurse and a psychologist. At both stages, home visits were repeated on at least two occasions if no contact was made. The mean (s.d.) interval between the two interviews was 9 (5.4) days.

Statistical analysis

Vascular risk factors and the presence of clinical vascular disease were compared according to case-level depression in all participants at the first interview. All subsequent analyses were restricted to those who completed the two interviews and the blood tests. Potential associations between vascular risk/disease and depression (P<0.1 in univariate analysis) were analysed further using stepwise logistic or linear regression models to investigate confounding by demographic characteristics and mediation by disability (WHODAS II) and cognitive function (MMSE-K). Secondary analyses were carried out to investigate effect modification by cognitive function, using recommended cut-offs for the MMSE-K (Reference Park, Park and KoPark et al, 1991) for stratified analyses, and likelihood ratio tests for interaction terms within regression models. As a final procedure, analyses were repeated by excluding participants who reported any episode of depression occurring before age 60 years.

Recruitment

Of 1566 inhabitants aged 65 years or over identified from the registration lists, 1204 (77%) completed the first interview with research assistants. Of the remainder, contact could not be established with 195 (12%), 71 (5%) refused to participate, 55 (4%) had no fixed abode, 28 (2%) had changed address and 9 (1%) had died before the visit. No significant differences were observed in age (mean ages of 72.2 and 72.4, respectively) and gender ratio (58% and 62% female, respectively) between participants and non-participants.

Of the participants in the first interview, 732 (61%) participated in the second, more-intensive examination. Of non-participants at this stage, contact could not be established with 321 (27%), 92 refused (8%), 4 had died in the interval (<1%) and data were missing for 55 (5%). The principal apparent reason for attrition was that contact could not be established because the person was repeatedly away from home at the time of research visits. Participants at both interviews were less educated than those present at the first interview only (mean (s.d.) years of education were 3.4 (4.2) and 4.0 (4.4), respectively; P=0.035) and had lower cognitive function (mean (s.d.) scores on the MMSE-K were 23.3 (5.0) and 24.3 (5.1), respectively; P=0.002). However, no significant differences were observed between the participants and non-participants in terms of age (mean ages of 72.8 and 72.2, respectively), gender (59% and 57% female), disability (mean WHODAS II scores of 7.1 and 5.9) and GMS depression (14% and 13%). There were also no substantial differences in associations between vascular risk factors (ascertained at the first phase) and depression between all participants of the first phase and the subgroup who were present at the second phase. For example, the odds ratios (95% CI) for the associations between depression and stroke, hypertension and diabetes were 4.57 (2.74-7.62), 1.32 (0.93-1.88) and 1.35 (0.82-2.23), respectively, for all participants in the first phase (n=1204) and 3.68 (1.99-6.79), 1.48 (0.97-2.27) and 1.67 (0.94-2.99) in those present at the second phase (n=732). Further analyses therefore were restricted to the subgroup present at both interviews.

Univariate associations with depression

Depression was present in 101 (14%) of the 732 participants at the second interview. Univariate associations between depression and vascular disease/risk are summarised in Table 1. Depression was associated significantly with reported stroke and heart disease in the sample at both interviews. Depression was associated positively with reported hypertension and diabetes, with associations bordering on statistical significance. Depression was associated significantly with a lower level of HDL cholesterol and a higher level of LDL cholesterol.

Table 1 Unadjusted associations between depression and vascular risk/disease

	No depression (n=631)	Depression (n=101)	Odds ratio or mean difference (95% CI)	P 1
Information on vascular risk/disease (n, %)
Previous stroke	34 (5)	18 (17)	3.68 (1.99 to 6.79)	<0.00 1
Heart disease	117 (18)	36 (34)	2.32 (1.48 to 3.64)	<0.00 1
Hypertension	198 (31)	42 (40)	1.48 (0.97 to 2.27)	0.069
Diabetes	66 (10)	17 (16)	1.67 (0.94 to 2.99)	0.078
Current smoker	254 (41)	48 (46)	1.05 (0.60 to 1.45)	0.29
Examination for vascular risk/disease (mean, s.d.)
Systolic blood pressure (mm Hg)	146.8 (26.0)	147.8 (30.1)	1.0 (−4.5 to 6.5)	0.73
Diastolic blood pressure (mm Hg)	87.7 (22.1)	88.4 (22.7)	0.7 (−3.9 to 5.3)	0.77
Blood glucose (mg/dl)	104.5 (50.2)	113.6 (66.3)	9.1 (−4.5 to 22.6)	0.19
Total cholesterol (mg/dl)	175.1 (34.1)	178.4 (39.7)	3.3 (−4.3 to 10.8)	0.45
High-density lipoprotein cholesterol (mg/dl)	48.7 (14.2)	43.9 (11.1)	− 4.8 (−7.8 to −1.8)	0.002
Low-density lipoprotein cholesterol (mg/dl)	96.0 (31.1)	105.0 (36.3)	9.0 (2.2 to 15.9)	0.023
Triglyceride (mg/dl)	153.0 (88.2)	147.5 (64.0)	− 5.5 (−24.0 to 12.9)	0.56
Body mass index (kg/m2)	22.5 (3.5)	22.8 (4.2)	0.23 (−0.51 to 0.98)	0.59
Waist/hip ratio	0.9 (0.1)	0.9 (0.1)	0.0 (−0.02 to 0.02)	0.24

With respect to other independent variables, depression was associated significantly with female gender (odds ratio (OR)=2.02; 95% CI 1.28-3.30), a greater number of physical illnesses (OR per illness reported=1.39; 95% CI 1.22-1.58) and lower scores on the MMSE-K (OR per single point decrease=1.05; 95% CI 1.01-1.09). Associations with increased age (OR per year increase=1.03; 95% CI 0.99-1.07) and lower education (OR per year decrease=1.03; 95% CI 0.99-1.08) were of borderline significance.

Univariate associations with APOE genotype

The APOE genotype frequencies were as follows: e2/2, 1.5%; e2/3, 9%; e2/4, 1%; e3/3, 71%; e3/4, 16%; e4/4, 1.5%. The APOE e4 allele was present in 132 participants (18%). Previous stroke was reported more frequently in participants with the e4 allele (11% v. 6%, P=0.067) and mean body mass index levels were significantly lower in those with e4 than in those without (mean (s.d.) 22.1 (3.6) and 22.8 (3.5) kg/m², respectively; P=0.041). There were no significant associations between presence of the e4 allele and any of the other factors measured (vascular risk/disease, depression, disability and cognitive function). The APOE e4 allele was associated overall with a more atherogenic lipid profile, but differences were not significant (P>0.1). Mean (s.d.) levels of cholesterol were 180 (39) mg/dl in those with the e4 allele and 175 (34) mg/dl in those without this allele. Respective levels (mg/dl) were: HDL cholesterol, 47 (14) and 48 (14); LDL cholesterol, 100 (34) and 97; triglycerides, 162 (105) and 150 (81).

Multivariate analysis of depression and vascular disease/risk factors

Results of regression analyses are summarised in Table 2. All associations between depression and vascular disease/risk were weakened by the inclusion of confounding variables, particularly level of disability. The positive associations with stroke and a lower level of HDL cholesterol remained significant, whereas those with other vascular factors were substantially diminished both in strength and significance. Further adjustment for the APOE e4 allele made no substantial difference. The odds ratio (95% CI) for the association between depression and previous stroke was 2.73 (1.37-5.59) in Model 3 (see Table 2) after further adjustment for the e4 allele. The B values for the association between depression and HDL/LDL cholesterol levels were -4.17 (-7.34 to -1.01) and 5.10 (-1.93 to 12.14) in Model 3 after further adjustment for e4.

Table 2 Multivariate associations between depression and vascular disease/risk factors

Factor1	Unadjusted model	Adjusted
		Model 12	Model 23	Model 34
Stroke	3.68*** (1.99 to 6.79)	4.38*** (2.30 to 8.34)	2.56* (1.26 to 5.22)	2.59** (1.27 to 5.29)
Heart disease	2.32*** (1.48 to 3.64)	2.22** (1.40 to 3.52)	1.43 (0.83 to 2.43)	1.47 (0.86 to 2.52)
Hypertension	1.48 (0.98 to 2.27)	1.45 (0.94 to 2.24)	1.25 (0.79 to 1.96)	1.24 (0.79 to 1.95)
Diabetes	1.67 (0.94 to 2.99)	1.72 (0.95 to 3.09)	1.41 (0.75 to 2.64)	1.41 (0.75 to 2.64)
High-density lipoprotein cholesterol	− 4.83** (−7.82 to −1.84)	− 4.82** (−7.85 to −1.78)	− 4.30** (−7.45 to −1.14)	− 4.23** (−7.38 to −1.07)
Low-density lipoprotein cholesterol	9.05* (2.1 to 15.9)	5.76 (−0.96 to 12.5)	4.68 (−2.31 to 11.7)	4.58 (−2.42 to 11.6)

Stratification by APOE and MMSE-K (Table 3)

Associations between depression, stroke and lower HDL cholesterol were stronger in the presence of the e4 allele. However, interaction terms did not approach statistical significance (data not shown; P values for interaction terms >0.1). Associations between depression and vascular risk factors also were generally stronger in groups with borderline cognitive impairment (MMSE-K, 21-24), although again no individual interaction terms approached significance (Table 3; P values for interaction terms >0.1).

Table 3 Associations between depression and vascular risk/disease stratified by cognitive function (Korean version of the Mini-Mental State Examination, MMSE-K)

Factor1	Total sample (n=732)	MMSE-K score
		25-30 (n=333)	21-24 (n=222)	0-20 (n=177)
Stroke	4.38*** (2.30 to 8.34)	3.55* (1.23 to 10.2)	6.07** (1.67 to 22.0)	3.93* (1.30 to 11.9)
Heart disease	2.22** (1.40 to 3.52)	2.29* (1.10 to 4.78)	3.14** (1.38 to 7.17)	1.59 (0.66 to 3.86)
Hypertension	1.45 (0.94 to 2.24)	1.11 (0.55 to 2.25)	1.32 (0.59 to 2.99)	2.20* (1.01 to 4.76)
Diabetes	1.72 (0.95 to 3.09)	1.12 (0.40 to 3.12)	2.27 (0.79 to 6.50)	1.85 (0.64 to 5.34)
High-density lipoprotein cholesterol	− 4.82** (−7.85 to −1.78)	− 5.07* (−9.76 to −0.37)	− 7.07* (−12.85 to −1.28)	− 2.02 (−7.69 to 3.66)
Low-density lipoprotein cholesterol	5.76 (−0.96 to 12.5)	1.36 (−9.04 to 11.8)	5.06 (−7.85 to 18.0)	12.4 (−0.30 to 25.1)

Restriction by previous history of depression

Of the 732 participants at the second interview, a previous history of depression prior to age 60 years was reported by 16 (16%) of the 101 with current depression and by 17 (3%) of the 631 without current depression. No marked or consistent differences were found in the results when the above analyses were repeated in those without a past history of depression (data not shown).

Methodological issues

Previous community studies investigating these research questions have been limited by the use of brief screening instruments to define depression (Reference Stewart, Prince and RichardsStewart et al, 2001) or have focused on specific sub-populations (Reference Kim, Stewart and ShinKim et al, 2002). A strength of this study was that depression was ascertained using a diagnostic instrument that has been validated widely in a variety of international settings. In terms of sample representativeness, government registration lists represent a highly inclusive sampling frame for epidemiological research in South Korea because an accurate entry is required for many daily needs, including pension provision. The two-phase recruitment procedure potentially reduced the representativeness of the participants compared with the source population because of attrition between interviews, giving an overall response rate of <50%. In retrospect, a single-stage procedure would have been preferable. However, we feel that it is unlikely that the observed associations are explained by selection bias because neither vascular risk factor nor depression was associated with attrition. As well as this, the strength of association between the first-phase measures of vascular risk and depression were similar between the total sample and the analysed subgroup. It is therefore likely (although cannot be concluded definitely) that associations with factors measured at the second phase can be generalised to the source population.

The study was cross-sectional in design, which limits the extent to which cause and effect relationships can be clarified. It also raises the issue of information bias, although we feel that this is unlikely because the risk factors at the first phase were ascertained before the diagnostic interview for depression was administered and, furthermore, the data at the second phase were collected by a professional who did not have knowledge of the GMS findings. Information on vascular risk factors such as hypertension and diabetes relied on self-reported diagnoses and so corroboration by medical records was not feasible. There are certain other limitations in inference that should be considered: most ‘case’ participants will have had moderate levels of depression rather than the more severe forms that have been the focus for research in clinical samples; previous episodes of depression were ascertained by self-report, which is likely to have limited validity and the final restricted sample may not have had a truly ‘late-onset’ disorder; and differential rates of institutionalisation and mortality might conceivably have obscured the association between vascular risk and depression in this cross-sectional community study. Institutions were not sampled in this survey because there were none within the sampling areas. However, in Korea, provision of institutional care is limited and care at home is preferred by residents and their families, even in severe dementia. Mortality is therefore more likely to be a source of prevalence bias than institutionalisation.

Stroke and depression

Depression occurs frequently following stroke, and the association between previous stroke and depression observed in this sample has been reported in a similar Korean sample with cognitive impairment (Reference Kim, Stewart and ShinKim et al, 2002) and in other community studies (Reference Fuh, Liu and WangFuh et al, 1997; Reference Stewart, Prince and RichardsStewart et al, 2001). The association was reduced following adjustment for level of disability, but only partially so. This is consistent with findings in other populations (Reference Stewart, Prince and RichardsStewart et al, 2001) and suggests that other mediating pathways may be present, which may include the direct effect of an infarct on brain structure or chemistry but might also include functional impairment specific to stroke, or its psychological impact, which did not contribute to the disability scale.

Cholesterol subfractions and depression

We found no association between total cholesterol levels and depression, which was consistent with findings from some studies (Reference Brown, Salive and HarrisBrown et al, 1994; Reference McCallum, Simons and SimonsMcCallum et al, 1994; Reference Blazer, Burchett and FillenbaumBlazer et al, 2002), although depression has been found to be associated with lower cholesterol levels in others (Reference Morgan, Palinkas and Barrett-ConnorMorgan et al, 1993; Reference Partonen, Haukka and VirtamoPartonen et al, 1999), including a prospective study (Reference Cervilla, Prince and JoelsCervilla et al, 2000). One study of an elderly sample found depression to be associated with lower levels of LDL cholesterol (Reference Äijänseppä, Kivinen and HelkalaÄijänseppä et al, 2002). Another study of a younger sample (aged 31-65 years) found an association with higher levels of HDL cholesterol (Reference Horsten, Wamala and VingerhoetsHorsten et al, 1997). Our finding in the opposite direction might reflect racial or age-related differences in the aetiology and/or salience of vascular risk factors, but further research is required to clarify inconsistencies.

Vascular risk factors and depression

Depression in later life has been found to be associated with changes suggestive of cerebrovascular disease, such as white matter hyperintensities on neuroimaging (Reference Steffens, Helms and KrishnanSteffens et al, 1999; Reference de Groot, de Leeuw and Oudkerkde Groot et al, 2000). The vascular depression hypothesis also proposes a syndrome where severe depression and vascular disease co-exist. In this analysis, we did not set out to test the existence of such a syndrome, but to investigate well-recognised risk factors for cerebrovascular disease as potential risk factors for depression. If there is a vascular aetiology for depression in late life, risk factors for cerebrovascular disease such as hypertension and diabetes also may be important population-level risk factors for depression because of high prevalence rates in many populations. However, there is currently little evidence from community samples to suggest strong causal relationships (Reference Jones-Webb, Jacobs and FlackJones-Webb et al, 1996; Reference Rajala, Keinanen-Kiukaanniemi and KevelaRajala et al, 1997; Reference Stewart, Prince and RichardsStewart et al, 2001) and, where associations have been found, it is unclear whether these are specific to disorders affecting the vasculature or whether they are accounted for by well-recognised associations between depression and general poor health and disability (Reference Prince, Harwood and ThomasPrince et al, 1998). Hypertension and diabetes were associated with increased likelihood of depression in our study at borderline levels of statistical significance. However, these associations were substantially reduced in strength after adjustment for the WHODAS II score. This suggests that the impact of these disorders on depression may be mediated through associated general disability rather than specific vascular effects. However, it should be borne in mind that ‘adjustment’ by disability in a cross-sectional study such as this one may obscure associations with any potential risk factor for depression because depression and reported disability are strongly interrelated.

The association with decreased HDL cholesterol therefore does suggest a more atherogenic lipid profile in people with late-life depression, but other conventional risk factors for stroke were not associated with depression. Interestingly, a pathological study found late-life depression to be associated with arterial atheroma rather than microvascular disease (Reference Thomas, Ferrier and KalariaThomas et al, 2001). It is therefore possible that specific vascular pathways may underlie the disorder and there may be other vascular risk factors for depression (such as cerebral haemodynamics, blood pressure regulation and inflammatory processes) that were not investigated in this study. However, the direction of causality underlying cross-sectional associations is unclear because there is strong evidence from prospective research that depressive symptoms earlier in life are associated with increased risk of stroke (Reference Jonas and MussolinoJonas & Mussolino, 2000). Associations in later life may be the consequence of complex interrelationships between vascular risk and affective state over a long period (Reference Stewart, Chiu, Ames and KatonaStewart, 2002).

Interactions with cognitive impairment and APOE genotype

The association between vascular risk factors and depression might conceivably be explained by well-recognised associations between vascular risk factors and dementia, and between depression and later dementia (Reference Stewart, Chiu, Ames and KatonaStewart, 2002). This would predict stronger associations between depression and vascular risk factors in people at risk of cognitive decline (e.g. those with the APOE e4 allele or those with cognitive function that is already impaired). Recently, a study found no interactions between APOE e4 allele and cholesterol level in associations with depression in cross-sectional and longitudinal analyses in a biracial elderly community sample (Reference Blazer, Burchett and FillenbaumBlazer et al, 2002). To our knowledge, there has been no study to investigate effect modification by APOE genotype for other vascular risk factors and late-life depression. However, we found no evidence of substantial confounding or effect modification by APOE e4 allele for the association of interest. All vascular risk factors were associated more strongly with depression in the presence of borderline (compared with normal) cognitive function (MMSE-K, 21-24). These findings provide partial support for our hypothesis and suggest that there might be an interface between vascular disease, depression and early dementia. However, individual interactions were not demonstrable at conventional levels of statistical significance and further research is required to address this question.

Public health implications

Our findings support others in showing associations between late-life depression and clinical stroke, that were only partially explained by level of disablement. However, we found little evidence for associations with other vascular risk factors such as hypertension or diabetes. From a public health perspective regarding the prevention of depression, these support a focus on post-stroke populations and on reducing levels of general disability rather than on specifically targeting groups with vascular risk factors. The direction of causation for the association between depression and an atherogenic lipid profile cannot be concluded. However, at the very least it suggests that attention should be paid to cardiovascular risk profiles and the prevention of adverse vascular outcomes in older adults with depression.

CLINICAL IMPLICATIONS

1. ▪ Previous stroke and lower levels of high-density lipoprotein cholesterol were associated significantly with late-life depression.

2. ▪ Associations between other established risk factors for cerebrovascular disease, such as hypertension or diabetes, and late-life depression were not evident.

3. ▪ Associations, where present, were stronger in the presence of borderline cognitive impairment, although not to a statistically significant extent.

LIMITATIONS

1. ▪ The focus was on moderate levels of depression rather than the more severe forms that have been the focus of research in clinical samples.

2. ▪ There was an important level of attrition between interviews, although this was not associated with exposure or outcome measures of interest.

3. ▪ The study was cross-sectional in design, limiting inferences regarding the direction of causation.