Unhealthy lifestyles do not mediate the relationship between socioeconomic status and incident depressive symptoms; The Health ABC study

Abstract

Background

The relationship between low socioeconomic status (SES) and depressive symptoms is well described, also in older persons. Although studies have found associations between low SES and unhealthy lifestyle factors and between unhealthy lifestyle factors and depressive symptoms, not much is known about unhealthy lifestyles as a potential explanation of socioeconomic differences in depressive symptoms in older persons.

Methods

To study the independent pathways between SES (education, income, perceived income, and financial assets), lifestyle factors (smoking, alcohol use, body mass index, and physical activity), and incident depressive symptoms (CES-D 10 and reported use of antidepressant medication), we used 9 years of follow-up data (1997–2007) from 2,694 American black and white participants aged 70–79 from the Health, Aging, and Body Composition (Health ABC) study. At baseline, 12.1% of the study population showed prevalent depressive symptoms, use of antidepressant medication, or treatment of depression in the five years prior to baseline. These persons were excluded from the analyses.

Results

Over a period of 9 years time, 860 participants (31.9%) developed depressive symptoms. Adjusted hazard ratios for incident depressive symptoms were higher in participants from lower SES groups compared to the highest SES group. The strongest relationships were found for black men. Although unhealthy lifestyle factors were consistently associated with low SES, they were weakly related to incident depressive symptoms. Lifestyle factors did not significantly reduce hazard ratios for depressive symptoms by SES.

Conclusion

In generally healthy persons aged 70–79 years lifestyle factors do not explain the relationship between SES and depressive symptoms. (250)

OBJECTIVES

Socioeconomic gradients in health status are consistent across numerous physical health outcomes, settings, and age and sex strata (1–4). Socioeconomic gradients have also been found for mental health outcomes, such as depressive symptoms (5, 6). Explanations for these gradients include psychosocial factors, material resources, work characteristics, and adverse exposures in childhood (7–10). Lifestyle factors, such as smoking, excessive alcohol consumption, overweight and obesity, and a low physical activity pattern, may also contribute to socioeconomic differences in depressive symptoms (7, 8). Several studies have identified a relationship between low socioeconomic status (SES) and unhealthy lifestyle factors (11–14). Good evidence exists that unhealthy lifestyles elicit or exacerbate symptoms of depression (15–19). However, little is known about potential mediating effects of an unhealthy lifestyle on the socioeconomic gradient in depressive symptoms, particularly in older persons (8). Depressive symptoms are common in older adults and associated with highsocietal costs due to associated morbidity and high utilization of health services (20, 21). Since unhealthy lifestyles are amenable to change, indications for a mediating effect of an unhealthy lifestyle on SES-specific risks of incident depressive symptoms would offer new perspectives on prevention or treatment of this potentially debilitating mental disorder and the socioeconomic differences therein.

In the present study, nine years of follow-up data from the Health ABC study in American black and white men and women aged 70 to 79 years, initially free of depressive symptoms, will be used to study: 1) pathways between SES and lifestyle factors at baseline, 2) pathways between lifestyle factors at baseline and incident depressive symptoms, and 3) the relationship between SES at baseline and incident depressive symptoms. Additionally, the potentially mediating effect of unhealthy lifestyle factors on this relationship will be evaluated.

METHODS

Design and study population

The Health, Aging and Body Composition (Health ABC) study is a longitudinal cohort study. The study consists of 3,075 well-functioning black and white men and women aged 70–79 years. White participants were recruited from a random sample of Medicare (i.e. a federal system of health insurance for people over 65 years of age and for certain younger people with disabilities) beneficiaries residing in ZIP codes from the metropolitan areas surrounding Pittsburgh, Pennsylvania, and Memphis, Tennessee. Because of much smaller numbers, black participants were recruited from all age-eligible residents in these geographic areas. Participants were eligible if they reported no difficulty in either walking one quarter of a mile, going up 10 steps without resting, or performing basic activities of daily living. Participants were excluded if they reported a history of active treatment for cancer in the prior three years, planned to move out of the study area in the next three years, or were currently participating in a randomized trial of a lifestyle intervention. Baseline data were collected between April 1997 and June 1998 and included an in-person interview and a clinic-based examination, with evaluation of socioeconomic status and material adversity, body fat composition, lifestyle, clinical and subclinical diseases, and depressive symptoms.

The present study uses 9 years of follow-up data, collected between 1997 and 2007. Data on socioeconomic status and lifestyle factors were collected at baseline. Data on depressive symptoms were collected at all years of follow-up, except for years seven and nine. In order to study the onset of depressive symptoms, respondents with depressive symptoms at baseline (short CES-D score >10 or use of antidepressant medication, n=263) or in the five years prior to baseline (Medicare hospitalization and outpatient data, n=63) were excluded. Furthermore, 31 persons with missing values on education and/or assets and 24 persons with missing data on lifestyle factors were excluded, leaving 2, 694 participants (mean age 73.6, SD=2.87, 1,339 men and 1,355 women) for the present analyses. All participants signed informed written consent approved by the institutional review boards of the clinical sites.

Measures

Depressive symptoms

Depressive symptoms were assessed at baseline (year 1) and at years 3, 4, 5, 6, 8 and 10, using the short form of the Center for Epidemiologic Studies-Depression (CES-D 10) scale (22, 23). This 10-item self-report scale is designed to measure depressive symptoms during the previous week. Scores ranged from 1 to 30, with higher scores indicating more depressive symptoms. A cut-off score of 11 was used as a criterion for depressive symptoms (22). In addition, self-reported use of antidepressant medication (prescribed for the treatment of depression or depressive symptoms) during the previous two weeks was recorded at years 1, 2, 3, 5, 6, 8, and 10. Participants were classified with depressive symptoms when the cut off score of 11 was crossed (n=702) and/or when use of antidepressant medication was reported (n=290).

Socioeconomic status

Four indicators of SES at baseline were used: education, family income, perceived adequacy of income, and ownership of financial assets. Categories for completed education were: less than high school, high school graduate, and college graduate and beyond. Net family income was defined as: wages, salaries, social security or retirement benefits, financial help from relatives, and rent from property. Five categories of family income from the year prior to interview (i.e. 1996–1997) were distinguished: < $10,000, $10,000–< $25,000, $25,000 – < $50,000, ≥$50,000 and missing (n=347). All analyses on family income were equivalized for the number of persons in the household. Furthermore, persons were asked to indicate whether they perceive their income adequate to meet essential needs. Categories for perceived adequacy of income were: poorly, fairly well, and very well. Finally, the number of financial assets a person reported was used as a SES measure. Assets included: money market account, saving bonds or treasury bills, home ownership or investment property or housing; a business or farm, stock or stock mutual funds, individual retirement (IRA) or KEOGH accounts (i.e. retirement plan for the self-employed and their employees), or other investments. Three categories were created: none, one or two, and three to seven (24).

Lifestyle factors

Lifestyle factors included smoking, alcohol use, physical activity, and body mass index (BMI). Three categories of smoking were created (i.e. never, former, current smoked). Categories of alcohol consumption were chosen according to national dietary guidelines (25, 26): abstinence, moderate drinkers (1–14 units per week), and heavier drinkers (>14 units per week).

Physical activity pattern was assessed using a modified leisure-time physical activity questionnaire (27). Energy expenditure doing household chores, care giving, walking and climbing stairs, and exercise and recreation was estimated by kilocalorie expenditure per week (28). Three categories were created: inactive, lifestyle active and exercise. Inactive as defined as reporting less than 1000 kcal/wk of exercise (i.e. less than the Surgeon General’s recommendations) and less than 2719 kcal/wk of total physical activity (representing the 25^th percentile of total physical activity for the Health ABC cohort). Lifestyle active was defined as reporting less than 1000 kcal/wk of exercise and more than 2718 kcal/wk of total physical activity per week. Exercises was defined as reporting 1000 kcal/wk or more of exercise alone (29).

BMI was calculated as weight in kilograms divided by height in meters squared (kg/m²). Three categories were created: <25 (because of small numbers this category also includes persons who were underweight), 25-<30, and ≥30.

Covariates

Sociodemographics included age, sex, race (black or white), study site (Memphis or Pittsburgh), and marital status (never married, previously married including widowers, and married). Presence of lung, heart and cerebrovascular disease, diabetes mellitus, osteoarthritis, and cancer at baseline was determined using standardized algorithms considering self-report, use of specific medications, and, when available, as for diabetes, the results of screening tests conducted in the cohort. A continuous measure of the number of the prevalent diseases (0–5) was used.

Analyses

All analyses were performed using SPSS, version 15.0.1 and were stratified by race and sex (education*race interaction: Wald χ² 10.87, df=2, p=0.004; income*sex interaction: Wald χ² 7.93, df=2, p=0.018; assets*race interaction: χ² 7.24, df=2, p=0.027). Differences in main characteristics between black and white men and women were determined using chi-square tests for categorical variables and t-test statistics for continuous variables. Cross tabulations of lifestyle factors among the different SES strata were calculated to study the association between SES and lifestyle factors.

To evaluate the relationship between lifestyle factors and incident depressive symptoms, Cox proportional hazard regression models were fitted, adjusting for age, site, marital status and, to avoid confounding and a possible selection effect on socioeconomic status, prevalent diseases. If a person crossed the CES-D cut off score of >10 or reported use of antidepressant medication (n=860), person time (in months since baseline) was set at the interview during which the person developed depressive symptoms. Persons surviving with no evidence of depressive symptoms were censored at the last follow-up measurement. Those who died with no evidence of depressive symptoms were censored at time of death (months from baseline), and those lost to follow-up were censored at their last follow-up measurement.

To determine if there were differences in (time to) incidence of depressive symptoms between SES groups, Cox Proportional Hazard regression models were fitted. The first model included age, site, marital status, and prevalent diseases at baseline. The second model was additionally adjusted for all lifestyle variables. A percentage reduction in hazard ratio from model 1 was computed using: (HR_model1-HR _{model 2})/(HR_{model 1}-1)*100%. Mediation was considered present when inclusion of the lifestyle variable(s) in the logistic regression analyses caused at least a 10% decrease in relative risk compared to the original relative risk associated with SES on incident depressive symptoms (30). The proportional hazards assumption was investigated by testing the constancy of the log hazard ratio over time by means of log-minus-log survival plots and interactions with time (log transformed). According to the tests, the proportional hazard assumption was not violated. Results for each interaction term are as follows: education * T_: Wald χ² = 3.39, df=1, p=0.067; income * T_: Wald χ²=0.694, df=1, p=0.405; assets*T_: Wald χ²=1.05, df=1, p=0.307; perceived income* T_1.14, df=1, p=0.285.

RESULTS

Table 1 shows how covariates, baseline SES factors and lifestyle factors were distributed among black and white men and women. Black participants were less well educated, reported less income than needed, and fewer assets compared with white participants. In general, women reported lower incomes and fewer assets, compared to men. Furthermore, black men and women were significantly more likely than whites to be obese and to report unhealthy lifestyles, such as smoking and being inactive. However, they more often reported alcohol abstinence, while white men more often reported heavier drinking. In general, black and white men were more likely to have ever smoked in life and to drink alcohol, while women were more likely to be inactive.

Table 1.

Distribution of covariates, baseline SES, and lifestyle factors among study participants^*

	Men			Women			Men vs Women
	White (N=836)	Black (N=503)	p-value (df)	White (N=717)	Black (N=638)	p-value (df)	p-value (df)
Covariates
Mean age (SD)	73.9 (2.92)	73.5 (2.79)	0.008 (df=1337)	73.6 (2.77)	73.3 (2.92)	0.166 (df=1314.8)	0.004 (df=2692)
Study site (Pittsburgh, %)	48.8	50.5	0.548 (df=1)	44.9	54.1	0.001 (df=1)	0.911 (df=1)
Marital status (Married, %)	73.7	62.0	<0.001 (df=3)	44.8	23.2	<0.001 (df=3)	<0.001 (df=3)
≥1 prevalent chronic diseases, %	68.8	63.0	0.018 (df=4)	67.8	59.2	0.012 (df=5)	0.025 (df=5)
Mean CES-D score at baseline (SD)	2.09 (2.18)	2.37 (2.47)	0.036 (df=933.7)	2.70 (2.44)	2.64 (2.48)	0.629 (df=1343)	<0.001 (df=2658.6)
Education, %
Low	14.0	49.3	<0.001 (df=2)	10.2	37.5	<0.001 (df=2)	<0.001 (df=2)
Middle	25.7	24.1		42.3	35.3
High	60.3	26.6		47.6	27.3
Income, %
<1<0.001	1.0	16.7	<0.001 (df=4)	6.1	26.8	<0.001 (df=4)	<0.001 (df=4)
10.000–<25.000	25.0	43.7		31.2	44.0
25.000–<50.000	38.0	23.3		34.2	14.1
≥50.000	26.9	8.2		14.2	1.9
Missing	9.1	8.2		14.2	13.2
How well does income fit need? %
Poorly	1.9	8.7	<0.001 (df=3)	1.8	11.0	<0.001 (df=3)	0.001 (df=3)
Fairly well	36.5	54.1		36.3	60.3
Very well	58.7	34.8		57.2	25.7
Missing	2.9	2.4		4.7	3.0
Assets, %
0	5.9	21.1	<0.001 (df=2)	12.1	31.0	<0.001 (df=2)	<0.001 (df=2)
1–2	28.2	59.4		28.9	54.5
3–7	65.9	19.5		59.0	14.4
Smoking, %
Current	4.9	20.5	<0.001 (df=2)	7.7	12.2	0.017 (df=2)	<0.001 (df=2)
Former	65.7	49.3		32.8	32.4
Never	29.4	30.2		59.6	55.3
Alcohol consumption
Heavier	13.4	9.3	<0.001 (df=2)	4.9	1.7	<0.001 (df=2)	<0.001 (df=2)
Abstinence	35.9	53.5		47.1	69.1
Moderate	50.7	37.2		48.0	29.2
Physical activity, %
Inactive	16.4	28.2	<0.001 (df=2)	23.0	26.5	0.009 (df=2)	<0.001 (df=2)
Lifestyle active	42.3	50.7		57.5	60.0
Exercise	41.3	21.1		19.5	13.5
BMI, %
Obese (≥30.0)	19.0	25.8	0.001 (df=2)	16.6	44.4	<0.001 (df=2)	<0.001 (df=2)
Overweight (25.0 –<30.0)	51.3	41.6		38.6	35.6
Normal weight (<25.0)	29.7	32.6		44.8	20.1

^*Differences between groups were tested using 2-tailed t-tests for continues measures, and Chi-square tests for categorical variables.

All indicators of SES were consistently associated with lifestyle factors (Table 2). Black and white men and women from lower SES groups were more likely to report unhealthy lifestyle factors at baseline, such as smoking, inactivity, and obesity. These associations were strongest in black men and women. More than 35% of black women with a low educational level reported an inactive physical activity pattern, in contrast to only 18% of black women with higher educational levels. Opposite associations for alcohol intake were found, with white men from high SES groups reporting the highest alcohol intake. For example, 16% of white men with high levels of education reported heavier alcohol intake, in contrast to only 8% of white men with lower levels of education.

Table 2.

Lifestyle factors by socioeconomic status

	Total (N=2694)
	Current smokers	Heavy drinkers	Sedentary lifestyle	Obese BMI≥30
Education, %
Low	14.6	4.9	31.2	31.9
Middle	9.7	5.6	21.9	28.5
High	8.2	10.8	18.5	19.9
	P<0.001; df=4	P<0.001; df=4	P<0.001; df=4	P<0.001; df=4
Income, %
<10.000	18.9	2.6	34.5	35.2
10.000–<25.000	12.3	7.2	24.7	29.8
25.000–<50.000	9.6	8.4	17.9	20.3
≥50.000	2.4	12.6	15.0	17.1
Missing	6.9	5.6	26.7	27.7
	P<0.001; df=8	P<0.001; df=8	P<0.001; df=8	P<0.001; df=8
How well does income fit need? %
Poorly	21.0	8.4	34.3	37.1
Fairly well	12.1	6.1	22.9	28.2
Very well	7.7	9.2	21.0	21.5
Missing	4.5	4.5	25.8	29.2
	P<0.001; df=6	P<0.001; df=6	P<0.001; df=6	P<0.001; df=6
Assets, %
0	14.3	5.2	28.4	37.7
1–2	13.0	6.4	25.9	28.4
3–7	6.2	9.6	17.7	18.5
	P<0.001; df=4	P<0.001; df=4	P<0.001; df=4	P<0.001; df=4

Note: p-values and df are based upon Pearson χ² tests.

Over a period of 9 years time, 860 participants (31.9%) developed depressive symptoms. Table 3 shows the relationship between lifestyle factors and incident depressive symptoms over 9 years. While the magnitude of the relationships was fairly consistent across all the four race-sex groups, significant relationships were found for white women only. In this group, inactive physical activity pattern and being a current smoker predicted depressive symptoms, even after adjusting for age, site, marital status, and prevalent diseases.

Table 3.

Hazard Ratios (HR) and 95% confidence intervals (CI) for incident depressive symptoms by lifestyle factors^*

	Men				Women
Cases (%)a	White (N=836) 229 (27.4%)		Black (N=503) 154 (30.6%)		White (N=717) 260 (36.3%)		Black (N=638) 217 (34.0%)
	Wald (p) df=1	HR (95% CI)	Wald (p) df=1	HR (95% CI)	Wald (p) df=1	HR (95% CI)	Wald (p) df=1	HR (95% CI)
Smoking
Current	0.18 (0.668)	1.16 (0.59–2.26)	1.04 (0.308)	1.28 (0.80–2.05)	5.36 (0.021)	1.65 (1.08–2.53)	0.06 (0.812)	1.05 (0.68–1.63)
Former	2.04 (0.154)	1.24 (0.92–1.67)	0.78 (0.377)	1.18 (0.82–1.70)	0.52 (0.469)	1.11 (0.84–1.45)	0.56 (0.454)	0.89 (0.67–1.20)
Neverb		Ref		Ref		Ref		Ref
Alcohol consumption
Heavier	2.18 (0.140)	0.71 (0.45–1.12)	0.50 (0.478)	0.78 (0.40–1.54)	0.15 (0.700)	0.87 (0.44–1.73)	0.74 (0.390)	0.54 (0.13–2.21)
Abstinence	0.09 (0.769)	1.04 (0.79–1.39)	0.44 (0.508)	0.89 (0.64–1.25)	1.41 (0.234)	1.17 (0.90–1.51)	0.17 (0.683)	0.94 (0.70–1.26)
Moderateb		Ref		Ref		Ref		Ref
Physical activity
Inactive	1.75 (0.185)	1.29 (0.88–1.89)	1.83 (0.176)	1.37 (0.87–2.18)	7.32 (0.007)	1.74 (1.17–2.60)	1.93 (0.165)	1.37 (0.88–2.13)
Lifestyle active	0.81 (0.368)	1.14 (0.86–1.53)	0.45 (0.501)	1.16 (0.76–1.76)	2.38 (0.123)	1.32 (0.93–1.88)	0.12 (0.728)	0.93 (0.63–1.39)
Exerciseb		Ref		Ref		Ref		Ref
Body Mass Index
Obese (BMI≥30)	1.30 (0.253)	1.24 (0.86–1.79)	1.60 (0.206)	1.30 (0.87–1.96)	0.01 (0.911)	0.98 (0.70–1.38)	2.02 (0.156)	1.31 (0.90–1.89)
Overweight (25–<30)	0.22 (0.638)	0.93 (0.68–1.26)	0.01 (0.911)	0.98 (0.67–1.43)	0.13 (0.721)	0.95 (0.73–1.25)	0.03 (0.869)	1.03 (0.70–1.52)
Normal (<25)b		Ref		Ref		Ref		Ref

^*Adjusted for age, site, marital status, and prevalent diseases

^aPersons with incident depressive symptoms;

^bCategory of reference in analysis

Table 4 shows the relationship between SES and incident depressive symptoms for white and black men and women. For white men, hazard ratios for incident depressive symptoms, adjusted for all covariates, were significantly higher in the lowest income groups and those reporting inadequate income. These relationships remained statistically significant after adjustment for all lifestyle factors. In black men, adjusted hazard ratios of depressive symptoms by income were even higher. Low education and fewer assets were also significant predictors of depressive symptoms in black men.

Table 4.

Hazard Ratios (HR) and 95% confidence intervals (CI’s) of incident depressive symptoms according to SES strata in men and women.

	Men						Women
Cases (%)	White (N=836) 229 (27.4%)			Black (N=503) 154 (30.6%)			White (N=717) 260 (36.3%)			Black (N=638) 217 (34.0%)
	Model 1a HR (95% CI) Wald (p)	Model 2b HR (95% CI) Wald (p)	% red	Model 1a HR (95% CI) Wald (p)	Model 2b HR (95% CI) Wald (p)	% red	Model 1a HR (95% CI) Wald (p)	Model 2b HR (95% CI) Wald (p)	% red	Model 1a HR (95% CI) Wald (p)	Model 2b HR (95% CI) Wald (p)	% red
Income
1 (Low)	3.08 (1.12–8.53) 4.71 (0.030)	3.37 (1.20–8.42) 5.35 (0.021)	+	5.02 (2.09–12.05) 13.06 (<0.001)	4.96(2.05–11.98) 12.67 (<0.001)	1.5	1.27 (0.68–2.39) 0.55 (0.460)	1.12 (0.59–2.12) 0.13 (0.723)	55.6	1.09 (0.46–2.56) 0.04(0.844)	0.98 (0.41–2.35) 0.002 (0.966)	>100
2	1.43 (1.00–2.05) 3.80 (0.051)	1.40 (0.98–2.02) 3.35 (0.067)	7	3.40 (1.47–7.87) 8.19 (0.004)	3.42(1.46–7.97) 8.07 (0.005)	+	1.51 (0.99–2.30) 3.58 (0.059)	1.39 (0.90–2.15) 2.25 (0.134)	23.5	1.06 (0.46–2.43) 0.02 (0.890)	0.96 (0.41–2.23) 0.009 (0.924)	>100
3	1.16 (0.83–1.62) 0.72 (0.398)	1.16 (0.83–1.62) 0.74 (0.389)	0	2.00 (0.83–4.84) 2.37 (0.124)	2.04(0.84–4.94) 2.48 (0.115)	+	1.33 (0.88–2.03) 1.84 (0.175)	1.27 (0.83–1.93) 1.20 (0.273)	18.2	0.70 (0.29–1.71) 0.60(0.438)	0.65 (0.27–1.59) 0.90 (0.344)	+
4 (high)c	Ref	Ref		Ref	Ref		Ref	Ref		Ref	Ref
Education
Low	1.03 (0.68–1.55) 0.018 (0.893)	1.00(0.66–1.50) 0.001 (0.983)	33.3	2.31 (1.52–3.51) 15.48 (<0.001)	2.31 (1.52–3.54) 15.22 (<0.001)	0	1.84 (1.27–2.66) 10.32 (0.001)	1.74 (1.19–2.54) 8.06 (0.005)	11.9	1.84 (1.27–2.68) 10.22 (0.001)	1.83 (1.24–2.70) 9.40 (0.002)	1.2
Middle	0.96(0.70–1.32) 0.055 (0.815)	0.93 (0.68–1.27) 0.22(0.638)	+	1.45 (0.89–2.38) 2.20 (0.138)	1.43 (0.87–2.36) 2.01 (0.156)	4.4	1.16 (0.89–1.51) 1.13 (0.288)	1.17 (0.89–1.52) 1.29 (0.257)	+	2.15 (1.49–3.11) 16.80 (<0.001)	2.20 (1.51–3.21) 16.64 (<0.001)	+
Highc	Ref	Ref		Ref	Ref		Ref	Ref		Ref	Ref
Perc income
Poorly	2.73 (1.39–5.38) 8.42(0.004)	2.71(1.37–5.36) 8.18 (0.004)	1.2	1.72 (0.97–3.05) 3.43 (0.064)	1.60(0.89–2.88) 2.44 (0.119)	16.7	0.91 (0.34–2.47) 0.03 (0.854)	0.78 (0.29–2.13) 0.23 (0.631)	+	2.16 (1.38–3.37) 11.45 (0.001)	2.11 (1.34–3.32) 10.29 (0.001)	4.3
Fairly well	1.32 (1.00–1.74) 3.91 (0.048)	1.29 (0.98–1.70) 3.30 (0.069)	9.4	1.34 (0.93–1.93) 2.52 (0.112)	1.33 (0.92–1.92) 2.30 (0.129)	2.9	1.15 (0.89–1.49) 1.10 (0.294)	1.11 (0.86–1.44) 0.66 (0.416)	26.7	1.30 (0.93–1.81) 2.37(0.124)	1.29 (0.92–1.80) 2.17 (0.141)	3.3
Very wellc	Ref	Ref		Ref	Ref		Ref	Ref		Ref	Ref
Assets
0	0.75 (0.40–1.43) 0.75(0.386)	0.71 (0.37–1.36) 1.07(0.300)	+	1.70 (1.02–2.86) 4.07 (0.044)	1.60 (0.95–2.70) 3.10 (0.078)	14.3	0.97 (0.65–1.46) 0.02 (0.896)	0.88 (0.58–1.34) 0.35 (0.555)	+	1.54 (0.98–2.44) 3.43 (0.064)	1.50 (0.94–2.39) 2.88 (0.090)	7.4
1–2	0.97(0.71–1.31) 0.05(0.817)	0.93(0.69–1.26) 0.20(0.652)	+	1.53 (0.99–2.38) 3.69 (0.055)	1.48 (0.95–2.29) 3.03 (0.082)	9.4	1.10 (0.83–1.45) 0.43 (0.512)	1.03 (0.77–1.36) 0.03 (0.866)	70	1.53 (1.01–2.34) 3.93(0.047)	1.50 (0.98–2.30) 3.47 (0.062)	5.7
3–7c	Ref	Ref		Ref	Ref		Ref	Ref		Ref	Ref

^aAdjusted for age, site, marital status and prevalent diseases;

^bAdjusted for the variables listed under ^a and for smoking, drinking, physical activity and BMI;

^cCategory of reference in analysis. Note: Each Wald Statistic and p value was based on a Wald test with 1 df

For white women, hazard ratios for incident depressive symptoms, adjusted for all covariates, were significantly higher in those with low education. When additionally adjusted for lifestyle factors, this relation reduced but remained statistically significant. For black women, hazard ratios were higher for middle and low educated women. Also in black women, compared to those who perceived their income as ‘very well’ and had three to seven assets, those with poorly perceived income and fewer assets had higher rates of depressive symptoms.

Overall, no relevant reductions (≥10%) of the hazard ratios were found when lifestyle factors were introduced into the model.

In additional analyses, depression was defined by means of the CES-D score only. Moreover, a composite measure of unhealthy lifestyle factors (based on a count of the number of these factors) was calculated to test our hypotheses. Similar results were obtained. Compared with respondents who were included in our analysis sample (N=2,694), those who were not (n= 381; because of prevalent depressive symptoms, use of medication, history of depression, or missing values on relevant variables) were significantly more likely to be female (Pearson χ² = 12.86, df= 1, p<0.001). Of 1,411 participants, no complete data on CES-D scores or use of medication at follow-up measurements was available (i.e. due to premature death or missing values). Attrition and missing values were higher for black men (60.6%, Pearson χ²=74.94, df=3, p<0.001), persons with lower SES (61.9%, Pearson χ²= 79.57, df=2, p<0.001), persons with depressive symptoms at baseline (62.6%, Pearson χ² = 15.92, df=1, p<0.001) and persons with unhealthy lifestyle (e.g. 66.9% of current smokers, Pearson χ² = 72,41, df=3, p<0.001). Sensitivity analyses, in which we imputed scores as having depressive symptoms (i.e. to assess the most extreme effect of missing data), showed slightly altered hazard ratios but no differences in significance and the role of lifestyle factors.

CONCLUSION

This study in older black and white men and women showed that low SES predicted an increased incidence of depressive symptoms over nine years. In black men, socioeconomic differences in depressive symptoms were especially strong. Although unhealthy lifestyle factors were inversely associated with SES levels at baseline in all groups, unhealthy lifestyle factors were only weakly related to depressive symptoms during follow-up. Furthermore, lifestyle factors did not substantiallyreduce the hazard ratios for depressive symptoms by SES.

Our results confirm the presence of a social gradient in depressive symptoms among initially healthy older persons in their seventies. We showed not only considerable socioeconomic differences in the incidence of depressive symptoms, but also racial and sex differences herein. The absolute prevalence and incidence of depressive symptoms was highest in white women (see table 1 and 4). In contrast, black men with low incomes had a risk of incident depressive symptoms in excess of five times the risk in black men with high incomes.

We also showed differences in the relevance of the individual SES measures in predicting depressive symptoms. Income was an important predictor of incident depressive symptoms in men but not women, whereas in women educational level was important. Comparing black and white men, perceived income seemed to have more importance predicting depressive symptoms in white men, while a more absolute measure on the number of financial assets had more predictive power in black men. Perceived income and the number of financial assets seemed to have more importance in predicting depressive symptoms in black women, compared to white women. The relationship between SES and depressive symptoms was only weakly mediated by lifestyle factors. This finding agrees with earlier studies examining the role of similar lifestyle factors on the relationship between SES and depressive symptoms (7, 8). For example, Koster et al found that lifestyle factors explained less than 5% of the association between SES and incident depressive symptoms in Dutch older adults (8). Furthermore, in the Whitehall II study among male and female civil servants in London, UK, Stansfeld et al found that lifestyle factors were an important explanatory factor for the gradient in physical functioning but not for the gradient in depressive symptoms (7).

Alternative explanations for socioeconomic, racial and sex differences in depressive symptoms may have a psychosocial background. For example, exposure to major life stressors and low levels of social support and social integration are good candidates for mediation between SES and depressive symptoms (31, 32). Moreover, it has been suggested that the experience of low social status itself can program a ‘defensive’ phenotype, which is marked by sustained activation of stress-related autonomic and neuro-endocrine responses (33). These stress responses may contribute to the onset of depressive symptoms later in life, independently of lifestyle factors. Further research on explanation for socioeconomic differences in depressive symptoms in otherwise healthy older persons is recommended to develop effective interventions in the near future. However, because some interventions aimed at reducing depressive symptoms might only benefit the higher socioeconomic groups (34), we should focus on the modifiability of the factor of interest, and whether the intervention will actually be effective in reducing the socioeconomic gradient in depressive symptoms.

Strengths and limitations

The strengths of this study include the use of a relatively long follow-up period and the ability to exclude cases with prevalent depressive symptoms at baseline. These features allowed us to study the causal pathway between SES and depressive symptoms and to exclude a potential effect of depressive symptoms on SES (i.e. reverse causation or selection effect (35)). Several methodological considerations may, however, affect the interpretation of the results of this study.

First, the definition of depression is classically a fluctuating condition. The concept of ‘incidence’ might not fit well with its natural history. We have used a relatively healthy older population, initially free of mobility limitations, depressive symptoms during the five years prior to baseline, and use of antidepressant medication. There was however, limited data available on episodes of depression in between visits and lifetime depression. Because of underreporting of a remitting illness, it is possible that some depressive symptoms cases have been missed in our analyses. To address this, we included the use of antidepressant medication in our definition of depressive symptom, with the recognition that these medications are occasionally used for other indications. Additional analyses in which we only included the CES-D criteria yielded similar results. Still, it is possible that depression earlier in life might have interfered in the process of status attainment. It is unclear how residual confounding by a history of depression may have affected the results of our study.

Second, the study relied on self-reports for data on SES, lifestyle factors, and depressive symptoms. Individuals with a general tendency towards negative perceptions of material well-being (e.g. perceived income, assets) may also over-report depressive symptoms (36). This may have led to an overestimation of the presented relations. However, by excluding persons with prevalent depressive symptoms at baseline from the analyses, this potential bias should be minimal (37, 38).

Third, because of the observational nature of this study and the fact that we have only examined the effect of lifestyle factors at baseline, we cannot exclude the possibility that the inclusion of lifestyle modifications into the models could perhaps have further attenuated the impact of low SES on depressive symptoms. Furthermore, controlling for prevalent disease at baseline may be considered as over-adjustment and might also have attenuated the effect of lifestyle factors. In this respect, pathways linking lifestyle factors to depressive symptoms, mediated by disease might also be plausible. Additional analyses, in which we have looked at the role of prevalent and incident diseases in separate models, however, yielded similar results with regard to the lifestyle factors.

Fourth, our research may be limited by potential selection biases, which have been described in the Results section. Moreover, the goal of recruitment for the Health ABC study was not to have a representative sample but rather enrollment of a high-functioning cohort to examine onset of functional limitations. Still, however, the prevalence rate of depressive symptoms at baseline (n=326, 12.1%, based on CES-D, use of antidepressant medication and treatment of depression in the five years prior to baseline) is comparable with other rates of depression, ranging from 11 to 15%, that have been reported in American older persons (39, 40).

Conclusion

This study showed considerable socioeconomic, racial and sex differences in the incidence of depressive symptoms in a relatively healthy subsample of older persons in the United States. Black men with low perceived income in particular were at heightened risk of depressive symptoms. The mediating effect of lifestyle factors on the relationship between SES and incident depressive symptoms was weak. Further research on alternative explanation for socioeconomic differences in depressive symptoms in older persons is recommended to help develop effective interventions.