Socioeconomic Position and Premature Mortality in the AusDiab Cohort of Australian Adults

Abstract

Objectives. To determine the association of socioeconomic position indicators with mortality, without and with adjustment for modifiable risk factors.

Methods. We examined the relationships of 2 area-based indices and educational level with mortality among 9338 people (including 8094 younger than 70 years at baseline) of the Australian Diabetes Obesity and Lifestyle (AusDiab) from 1999–2000 until November 30, 2012.

Results. Age- and gender-adjusted premature mortality (death before age 70 years) was more likely among those living in the most disadvantaged areas versus least disadvantaged (hazard ratio [HR] = 1.48; 95% confidence interval [CI] = 1.08, 2.01), living in inner regional versus major urban areas (HR = 1.36; 95% CI = 1.07, 1.73), or having the lowest educational level versus the highest (HR = 1.64; 95% CI = 1.17, 2.30). The contribution of modifiable risk factors (smoking status, diet quality, physical activity, stress, cardiovascular risk factors) in the relationship between 1 area-based index or educational level and mortality was more apparent as age of death decreased.

Conclusions. The relation of area-based socioeconomic position to premature mortality is partly mediated by behavioral and cardiovascular risk factors. Such results could influence public health policies.

A relationship between lower socioeconomic position (SEP) and higher mortality has been demonstrated with a variety of individual SEP indicators: educational level,^1–3 occupational level,⁴ and income.⁵ In addition, differences in adverse health outcomes between people can partly be attributed to where they live in terms of the area’s social deprivation and the distance from residence to health services.^6–9

The relationship between individual- or area-based SEP and mortality can be partially attributed to the presence of a socioeconomic gradient in prevalence of cardiovascular disease (CVD) risk factors^10,11 such as smoking,¹² low physical activity, poor diet,¹³ and alcohol consumption.¹⁴ In analyses that adjusted for these factors, the association between individual-level SEP and mortality was generally attenuated but not completely eliminated,^1–3 and few studies used area-based SEP.^11,15 Long-term stress exposure can induce metabolic abnormalities via neuroendocrine autonomic stress, which can contribute to the association between SEP and health.^16,17

Most of the studies addressing the relationship between SEP and mortality have analyzed death occurring at any age. However, with increasing age, the contribution to death of each individual risk factor, including SEP, may diminish.^18,19 Thus, the associations of SEP with mortality at all ages or mortality at younger ages may be different. In the few studies that have reported a relationship between individual SEP indices and premature mortality,^20,21 working conditions and behavioral characteristics were the main mediators.

The relationship between area-based indices of SEP and premature mortality have rarely been assessed.¹¹ Such analyses may reveal leverage points for intervention to reduce the unequal burden of premature mortality across different areas of residence, such as implementing specific programs in community settings.

Our purpose was to determine the relationships of different SEP indicators (1 area-based SEP index, 1 geographical index, and 1 individual SEP measure) with premature mortality. We also aimed to examine the degree to which those relationships are explained by behavioral and CVD risk factors as potentially modifiable factors.

METHODS

The baseline methods and response rates of the Australian Diabetes Obesity and Lifestyle (AusDiab) study have been described in detail elsewhere.^22,23 Briefly, AusDiab was a national, population-based survey of 11 247 adults aged 25 years or older in 1999 and 2000. A stratified cluster sample was drawn from 42 randomly selected census collector district clusters across Australia. Information was collected via a brief household interview, followed by a biomedical examination. Of the eligible adults (n = 20 347), 70% completed the household interview, and 11 247 of these (55% of all eligible adults) completed the baseline biomedical examination.²² All participants provided written consent. We excluded 1793 participants with missing data on the variables of interest and 116 Aboriginals or Torres Straits Islanders (1.36% of the cohort) because of a different relationship between SEP and mortality in this group. Compared with the included participants (n = 9338), those excluded were slightly less educated and were more likely to be smokers and to live in the most disadvantaged areas. They also had a higher body mass index (BMI) and higher prevalence of diabetes; however, these differences were not large.

For analysis of premature mortality (death before age 70 years), we excluded a further 1244 because they were aged 70 years or older at baseline, leaving 8094 adults for this particular analysis.

Socioeconomic Status or Accessibility and Remoteness

Area-level socioeconomic status.

We used 2 area-level measures of socioeconomic status: the Index of Relative Social Disadvantage (IRSD) and the Accessibility/Remoteness Index of Australia (ARIA). The IRSD is a measure of social deprivation of an area, and the ARIA is an index of geographic remoteness from large population centers. Because these 2 features may have important but independent influences on use of and access to care, it is important to consider both.

The IRSD, a component of the Socioeconomic Indexes for Areas,²⁴ characterizes the general SEP of census collection districts (the smallest geographic area, containing an average of 225 dwellings). It is a summary measure from a group of 20 variables (related to education, income, employment, family composition, housing benefits, car ownership, ethnicity, English language proficiency, and residential overcrowding) that display the dimensions of social disadvantage.²⁵ For these analyses, we based the IRSD scores on the 2001 census and divided the data into quintiles. Quintile 1 represents the most disadvantaged group and quintile 5 the least one.

The ARIA is a standard classification and index of remoteness or accessibility in Australia.²⁶ ARIA+ (its successor), used in these analyses, is a continuously varying index with values ranging from 0 (high accessibility) to 15 (high remoteness) as determined by road distance measurements from each locality (localities where the population is greater than 1000 persons) to the nearest service centers in each of 5 categories of service center based on population size. Using the index, we placed more than 12 000 populated localities into 1 of 5 groups: major urban areas, inner regional areas, outer regional areas, remote, and very remote. For these analyses, we collapsed the remote and very remote categories into 1 category.

Educational level.

We classified education into 4 categories on the basis of the highest educational qualifications received: (1) secondary school education; (2) trade or technical certificates; (3) associate degree, undergraduate diplomas, and nursing or teaching qualifications; and (4) bachelor’s degree and postgraduate qualifications.

Baseline Measures

The baseline variables used in the multivariate models, as described in the “Statistical Analysis” section, represent known risk factors for mortality for which there is considerable published evidence. Baseline data on age, gender, educational level, smoking status (never smoker, ex-smoker, or current smoker), and physical activity level (sedentary, insufficient, or sufficient) were collected via interview-administered questionnaires. Biochemical parameters (fasting plasma glucose, 2-hour plasma glucose after a 75-g oral glucose tolerance test, fasting serum triglycerides, total cholesterol and high-density lipoprotein [HDL] cholesterol levels) and anthropometrics were measured as previously described.²² We defined diabetes as taking hypoglycemic medication or having a fasting plasma glucose level of 7.0 millimoles per liter or higher or a 2-hour plasma glucose level of 11.1 millimoles per liter or higher. We defined hypertension as treatment with blood pressure–lowering medication or blood pressure of 140/90 millimeters of mercury or higher. We measured diet quality using the Dietary Guideline Index as a continuous variable.²⁷ Briefly, the Dietary Guideline Index was developed to reflect adherence to the Dietary Guidelines for Australian Adults,²⁸ which provides age- and gender-specific recommendations for the consumption of 5 core food groups (cereals, meats and alternative, fruits, vegetables, and dairy) and “extra foods.” We reduced the original Dietary Guideline Index from the original 15 components to 13 components, as measures of salt use or fluid intake were not available in this study.²⁹ We measured stress at baseline using the Perceived Stress Questionnaire (score range = 0–1).³⁰

Ascertainment of Mortality

Follow-up for mortality was until date of death or November 30, 2011 (for cause-specific mortality) or November 30, 2012 (overall mortality), whichever occurred first. We collected vital status and cause of death by linkage to the Australian National Death Index. We attributed deaths to CVD if the International Classification of Diseases, 10th Revision (ICD-10)³¹ code for underlying cause of death was I10 to I25, I46.1, I48, I50 to I99, or R96 and to cancer if the code was C00 to D48.

Statistical Analysis

We performed statistical analysis with Stata version 11.0 (StataCorp LP, College Station, TX). We analyzed 3 exposure variables: IRSD (quintiles), ARIA (4 classes), and educational level (4 classes).

We analyzed differences in proportions across socioeconomic groups with the Pearson χ² test and compared continuous variables using a 1-way analysis of variance test or a Kruskal–Wallis test for nonparametric distributions.

We used Cox regression models (with age as the time scale) to estimate the association between each indicator of SEP and time to death. For all indicators, we used the least-disadvantaged group as the reference. Our primary analysis focused on premature mortality, defined as death before age 70 years. We included participants if they were younger than 70 years at baseline and were censored either at death or upon reaching age 70 years. As a sensitivity analysis, we also tested the relationship between each indicator of SEP and mortality with different thresholds of age of death: at any age, before 65 years, before 75 years, and after 75 years.

In mediation analyses, we compared the mortality risk of the most disadvantaged group and the least disadvantaged. Individual-level risk factors included as potential mediators were smoking status, physical activity, dietary quality, diabetes, hypertension, waist circumference, HDL cholesterol and triglyceride levels, and stress. Gender and education were considered as confounders.

We adjusted Cox regression models as follows: model 1—age (time scale) and gender; model 2—age, gender, and education (or IRSD); model 3—age, gender, and behavioral variables (smoking status, physical activity, diet quality); model 4—model 3 and stress; model 5—age, gender, and the combination of all CVD risk factors (waist circumference plus diabetes, hypertension, HDL cholesterol, triglycerides); and model 6—the final fully adjusted model including all previous variables. For model 5, we modeled the risk factors separately at first and then combined. When model 6 tested the effects of IRSD or ARIA, we adjusted for educational level; when it tested the link between educational level and mortality, we adjusted for IRSD. Since ARIA was collinear to IRSD, we did not include them in a model together.

We used a 2-sided P value (P < .05) to determine statistical significance. We found that the proportional hazard assumptions for Cox regression models were not violated. There was no significant interaction effect between IRSD, ARIA, or educational level and age or gender (all P > .02).

We calculated the percentage reduction in relative risk of mortality attributable to risk factors using an equation of Lynch et al.,³² which provides an estimate of the explanatory contribution of risk factors to inequalities in mortality:

where HR = hazard ratio, and risk factors are smoking status, physical activity, diet quality, presence of diabetes, hypertension, waist circumference, HDL cholesterol, triglycerides, and stress.

To characterize uncertainty in the estimated percentage risk reduction, we employed bootstrapping techniques,³³ using 1000 random bootstrap samples with replacement and taking into account the clustered design of the survey. When the lower confidence level estimated by bootstrapping was below zero, we arbitrarily replaced its value with zero, and when the upper level was higher than 200, we reported the results as “> 200.”

RESULTS

People living in the most disadvantaged areas had less favorable risk factors than people in the least disadvantaged areas and higher prevalence of diabetes and hypertension (Table 1). During a median (interquartile range) follow-up of 12.6 years (interquartile range = 12.2–13.1), there were 991 deaths (10.6%), comprising 286 cardiovascular deaths (3.1% of the cohort), 325 cancer deaths (3.5%), and 273 deaths (2.92%) from other causes; there were 106 deaths without any code (1% of the cohort, 11% of deaths). Mean age of death was 78.0 ±12 years overall and 59.3 ±8.8 years for people who died before age 70 years. Mortality rates were higher among those living in the most disadvantaged quintile versus the least disadvantaged areas (Table 1).

TABLE 1—

Characteristics of the Population, by Quintile of Index of Relative Social Disadvantage at Baseline: The Australian Diabetes Obesity and Lifestyle Study, 1999–2012

	Quintilea
Characteristic	1, No. (%) or Mean ±SD(n = 1922)	2, No. (%) or Mean ±SD(n = 1883)	3, No. (%) or Mean ±SD(n = 184)	4, No. (%) or Mean ±SD(n = 1828)	5, No. (%) or Mean ±SD(n = 1861)	P
Age, y	52.5 ±15.3	51.7 ±14.4	51.1 ±14.6	50.0 ±13.2	52.2 ±13.5	< .001
Men	824 (43)	826 (44)	855 (46)	826 (45)	857 (46)	.01
Socioeconomic status
Educational level
Secondary school qualification	1064 (55.4)	816 (43.3)	722 (39.1)	618 (33.8)	522 (28.1)
Trade, technician’s certificate	574 (30.0)	602 (32.0)	604 (32.8)	559 (30.6)	475 (25.5)
Associate and othersb	169 (8.8)	212 (11.3)	232 (12.6)	257 (14.1)	305 (16.4)
Bachelor’s degree or postgrad diploma	115 (6.0)	253 (13.4)	286 (15.5)	394 (21.6)	559 (30.0)	< .001
Occupational status
Professional	281 (14.6)	485 (25.8)	512 (27.7)	674 (36.9)	808 (43.4)
White collar	187 (9.7)	279 (14.8)	243 (13.2)	266 (14.6)	237 (12.7)
Blue collar	1454 (75.7)	1119 (59.4)	1089 (59.1)	888 (48.6)	816 (43.9)	< .001
Behavioral factors
Smoking
Current smoker	419 (21.8)	331 (17.6)	279 (15.1)	220 (12.0)	183(9.8)
Ex-smoker or nonsmoker	1503 (78.2)	1552 (82.4)	1565 (84.9)	1608 (88.0)	1678 (90.2)	< .001
Physical activity
Sedentary	406 (21.1)	360 (19.1)	332 (18.0)	273 (14.9)	211 (11.3)
Insufficient	593 (30.9)	583 (31.0)	597 (32.4)	562 (30.7)	544 (29.2)
Sufficient	923 (48.0)	940 (49.9)	915 (49.6)	993 (54.3)	1106 (59.4)	< .001
Exercise, h/wk	4.1 ±5.1	4.3 ±5.5	4.4 ±5.5	4.8 ±5.5	5.1 ±5.6	< .001
Diet quality (possible range = 0–130)	82.5 ±15.0	83.8 ±14.4	84.1 ±14.4	84.5 ±13.7	85.8 ±13.7	.001
Cardiometabolic variables
Body mass index, kg/m2	27.5 ±5.3	27.3 ±5.0	27.0 ±4.9	26.6 ±4.7	26.1 ±4.4	< .001
Waist circumference, cm	92.5 ±14.1	92.0 ±13.7	91.4 ±13.9	90.1 ±13.5	87.5 ±13.2	.001
Hypertension ≥ 140/90 mm Hg or treated	691 (35.0)	654 (34.7)	579 (31.4)	540 (29.5)	560 (30.1)	.01
Diagnosed diabetes mellitus	217 (11.3)	170 (9.0)	125 (6.8)	110 (6.0)	127 (6.8)	< .001
Fasting plasma glucose, mmol/L	5.6 ±1.4	5.7 ±1.3	5.6 ±1.1	5.6 ±1.0	5.5 ±1.2	< .001
2-h plasma glucose, mmol/L	6.6 ±2.7	6.6 ±2.5	6.3 ±2.3	6.1 ±2.1	6.1 ±2.2	< .001
Total cholesterol, mmol/L	5.6 ±1.1	5.7 ±1.1	5.6 ±1.0	5.7 ±1.0	5.6 ±1.1	.45
High-density lipoprotein cholesterol, mmol/L	1.4 ±0.4	1.4 ±0.4	1.4 ±0.4	1.4 ±0.4	1.5 ±0.4	.001
Triglycerides, mmol/L	1.6 ±1.1	1.6 ±1.1	1.5 ±1.0	1.5 ±1.1	1.4 ±1.0	.001
Outcome
No. of deaths (mortality ratec)	283 (12.5)	199 (8.8)	178 (7.9)	150 (6.6)	181 (7.9)	< .001
Death at < 65 y (n = 136)	37	23	17	29	30
Death at 65–75 y (n = 176)	42	39	33	29	33
Death at ≥ 75 y (n = 679)	204	137	128	92	118

^aQuintile 1 is most disadvantaged and quintile 5 is least disadvantaged.

^bThis educational level category includes associate degree, undergraduate diplomas, and nursing/teaching qualifications.

^cMortality rate is expressed in deaths per 1000 person-years.

Relationships Between SEP Indicators and Mortality

As determined by IRSD score, people who lived in the most disadvantaged areas (quintile 1) had a higher risk of premature mortality than those in the least disadvantaged areas (quintile 5; age- and gender-adjusted HR = 1.48; 95% confidence interval [CI] =1.08, 2.01; Table 2). Although the rates of death differed between the extreme categories (quintile 1 vs quintile 5), there was no gradient across quintiles. The strength of the relationship between disadvantage and mortality was greater among those who were overweight (for quintile 1 vs quintile 5, HR = 1.45; 95% CI = 1.09, 1.93) and those who were obese (for quintile 1 vs quintile 5, HR = 1.82; 95% CI = 1.15, 2.87). The strength of the relationship between disadvantage and mortality increased very slightly as the age threshold for premature deaths was reduced: for quintile 1 versus quintile 5, hazard ratios for mortality at any age, before age 75 years, and before age 70 years were 1.35, 1.39, and 1.48, respectively. All hazard ratios were significant except for mortality before age 65 years (Table A, available as a supplement to the online version of this article at http://www.ajph.org).

TABLE 2—

Association Between Socioeconomic Position and Total Mortality, by Age at Death: The Australian Diabetes Obesity and Lifestyle Study, 1999–2012

Index of Socioeconomic Position	Mortality Risk Before Age 70 Years, HR (95% CI)	P (Trend)	Mortality Risk at Any Age, HR (95% CI)	P (Trend)
IRSD quintile		.013		.002
1 (most disadvantaged)	1.48 (1.08, 2.01)		1.35 (1.12, 1.63)
2	1.19 (0.86, 1.63)		1.08 (0.88, 1.32)
3	1.00 (0.72, 1.39)		1.01 (0.82, 1.24)
4	1.09 (0.78, 1.51)		1.06 (0.85, 1.31)
5 (least disadvantaged; Ref)	1		1
ARIA		.11		.08
Inner regions	1.36 (1.07, 1.73)		1.18 (1.02, 1.37)
Outer regions	1.27 (0.99, 1.62)		1.18 (1.01, 1.39)
Remote regions	0.84 (0.41, 1.70)		1.02 (0.75, 1.38)
Major cities (Ref)	1		1
Educational level		.001		.002
Secondary school education (lowest)	1.64 (1.17, 2.30)		1.39 (1.08, 1.79)
Trade or technical certificate	1.28 (0.90, 1.81)		1.31 (1.01, 1.70)
Associate degree and othersa	1.10 (0.71, 1.71)		1.05 (0.77, 1.44)
Bachelor’s degree and postgraduate qualifications (Ref)	1		1

Note. ARIA = Accessibility/Remoteness Index of Australia; CI = confidence interval; HR = hazard ratio; IRSD = Index of Relative Social Disadvantage. Data are adjusted for age and gender.

^aThis educational level category includes associate degree, undergraduate diplomas, and nursing or teaching qualifications.

As determined by ARIA score, living in inner regions carried a higher risk of premature mortality than living in major cities, but there was no gradient of higher mortality as people lived farther from major cities. As age of death decreased, there was a trend of increasing hazard ratios, from 1.18 (95% CI = 1.02, 1.37) for mortality at any age to 1.44 (95% CI = 1.08, 1.94) for mortality before 65 years, but confidence intervals were overlapping (Table A).

For education, those in the lowest level of education had a significantly higher risk of premature mortality than those in the highest level. There was a significant gradient of increased risk as educational level decreased. The magnitude of the relationship increased as the mortality age threshold decreased, with hazard ratios from 1.39 (95% CI = 1.08, 1.79) for mortality at any age to 1.66 (95% CI = 1.11, 2.48) for mortality before 65 years (Table A).

Mediating Factors

The overall contribution of the modifiable risk factors (behavioral and CVD risk factors) varied according to the index explored, with a contribution of approximately 77%, less than 50%, or less than 25% in the relationship of premature mortality to IRSD, educational level, and ARIA index (inner regions compared with major cities), respectively (Table 3; Table B, available as a supplement to the online version of this article at http://www.ajph.org).

TABLE 3—

Mediating Factors of the Relationship Between Socioeconomic Position Indicators and Mortality: The Australian Diabetes Obesity and Lifestyle Study, 1999–2012

	Mortality Risk Before Age 70 Years		Mortality Risk at Any Age
Index of Socioeconomic Position	HR (95% CI)	% Risk Reduction (95% CI)	HR (95% CI)	% Risk Reduction (95% CI)
IRSD (most disadvantaged compared with least disadvantaged)
Model 1: age, gender	1.48 (1.08, 2.01)	(Ref)	1.35 (1.12, 1.63)	(Ref)
Model 2: age, gender, and education	1.31 (0.96, 1.80)	35 (12, > 200)	1.27 (1.05, 1.55)	23 (3, 112)
Model 3: age, gender, and behavioral variables	1.31 (0.96, 1.79)	35 (0, 87)	1.28 (1.06, 1.54)	20 (9, 54)
Model 4: age, gender, behavioral variables, and stress	1.29 (0.94, 1.77)	40 (14, > 200)	1.28 (1.06, 1.55)	20 (5, 57)
Model 5: age, gender, and all cardiometabolic variables	1.36 (0.99, 1.87)	25 (0, 150)	1.30 (1.07, 1.57)	14 (0, 71)
Model 6: age, gender, education, behavioral variables, cardiometabolic variables, and stress	1.11 (0.80, 1.53)	77 (27, > 200)	1.21 (1.00, 1.47)	40 (12, 175)
ARIA (inner regions compared with major cities)
Model 1: age, gender	1.36 (1.07, 1.73)	(Ref)	1.18 (1.02, 1.37)	(Ref)
Model 2: age, gender, and education	1.33 (1.05, 1.69)	8 (0, 59)	1.18 (1.01, 1.37)	0 (0, 48)
Model 3: age, gender, and behavioral variables	1.33 (1.05, 1.68)	8 (5, 59)	1.17 (1.01, 1.37)	6 (0, 33)
Model 4: age, gender, behavioral variables, and stress	1.30 (1.02, 1.65)	17 (1, 60)	1.14 (0.98, 1.33)	22 (0-156)
Model 5: age, gender, and all cardiometabolic variables	1.36 (1.07, 1.73)	0 (0, 20)	1.20 (1.03, 1.39)	−11 (0, 7)
Model 6: age, gender, education, behavioral variables, cardiometabolic variables, and stress	1.28 (1.00, 1.63)	22 (0, 114)	1.16 (1.00, 1.36)	11 (0, 53)
Educational level (lowest level compared with highest level)
Model 1: age, gender	1.64 (1.17, 2.30)	(Ref)	1.39 (1.08, 1.79)	(Ref)
Model 2: age, gender, and IRSD	1.53 (1.08, 2.17)	17 (0, 54)	1.30 (1.00, 1.68)	23 (0, 118)
Model 3: age, gender, and behavioral variables	1.47 (1.05, 2.07)	27 (13, 91)	1.28 (0.99, 1.65)	28 (3, 66)
Model 4: age, gender, behavioral variables, and stress	1.45 (1.03, 2.05)	30 (10, 90)	1.26 (0.98, 1.63)	33 (10, 146)
Model 5: age, gender, and all cardiometabolic variables	1.51 (1.07, 2.12)	20 (5, 81)	1.32 (1.02, 1.70)	18 (5, 145)
Model 6: age, gender, IRSD, behavioral variables, cardiometabolic variables, and stress	1.34 (0.95, 1.90)	47 (14, 158)	1.15 (0.89, 1.50)	62 (16, > 200)

Note. ARIA = Accessibility/Remoteness Index of Australia; CI = confidence interval; HR = hazard ratio; IRSD = Index of Relative Social Disadvantage. Data are hazard ratios from different Cox proportional hazards models, with adjustment for individual risk factors and calculation of the percentage of risk reduction attributed to each factor. CIs around the risk reduction were derived from bootstrapping techniques. The fully adjusted model included adjustment for age, gender, education or IRDS, smoking status, physical activity, diet quality index, diabetes, hypertension, waist circumference, high-density lipoprotein cholesterol, triglycerides, and stress.

IRSD and premature mortality.

In the analysis of the relationship between living in the most disadvantaged areas and mortality before age 70, there was a 35% risk reduction after adjustment for the 3 behavioral variables combined (smoking status, physical activity, and diet quality) and a 25% risk reduction after adjustment for the CVD risk factors (diabetes, hypertension, waist circumference, HDL cholesterol, and triglycerides). The contribution of smoking was 35%, 33%, and 95% for mortality at any age, before age 70 years, and before age 65 years, respectively (data not shown). Physical activity level and diet quality contributed approximately 10% of the risk reduction (data not shown). With the addition of education to model 1, the relative risk of premature mortality associated with IRSD went from 1.48 to 1.31 (35% risk reduction). The risk reduction was similar if education was added in a model including all modifiable risk factors (data not shown).

ARIA and premature mortality.

In the relationship between ARIA and premature mortality, the age- and gender-adjusted hazard ratio of 1.36 (95% CI = 1.07, 1.73) fell to 1.28 (95% CI = 1.00, 1.63) for the fully adjusted model, representing a 22% risk reduction. Each CVD risk factor accounted for less than 7% of the excess risk (data not shown).

Education and premature mortality.

In the relationship between educational level and premature mortality, there was a 27% risk reduction with adjustment for the 3 behavioral variables and a 20% reduction with adjustment for CVD risk factors. The relationship remained significant after adjustment for smoking status, physical activity and diet quality, and stress. We observed a small increase in the contribution of smoking status as age of death decreased.

DISCUSSION

Using a population-based Australian sample and focusing on death before age 70 years, we describe a relationship between 3 indices of SEP and premature mortality. We found significantly higher mortality risk for people living in the most disadvantaged areas compared with those in the least disadvantaged areas. These results were no longer significant after we adjusted for all behavioral and CVD risk factors and stress. Using individual-based SEP, we found a relationship between education class and mortality, which was not significant after adjustment for risk factors. Using the ARIA index, we found a relationship between living in inner regions versus living in major cities, which was poorly explained by the modifiable risk factors. We also showed that the strength of the relationship between IRSD or education and premature mortality explained by smoking and CVD risk factors increased with falling age at death.

Evaluation of premature death—defined as death before 55 years, before 75 years, or avoidable deaths^5,20,34—may be a more informative way to evaluate the overall health of a population, to monitor progress in its health, and to assess differences in health between states or countries.³⁵ Our study suggests that the impact of modifiable risk factors on health inequalities may be greater among those dying younger than those who die at an older age. This is important because the burden of a risk factor that affects people in their economically active lives will have a significant impact on society and should be a strong imperative to improve health inequalities.

In the IRSD mediation analysis, the percentage of area-based SEP risk explained by behavioral and CVD risk factors and by stress was nearly 80%. The main modifiable mediator of this relationship was smoking, which is the biggest cause of adult death in developed countries. Smoking is strongly related to low educational level, low social class,³⁶ and living in the poorest areas.³⁷ Moreover, our results suggest that it is an even larger contributor to socioeconomic inequalities in premature mortality than in overall mortality. We would have expected a greater impact of physical activity and diet quality in the relationship, but those variables are less easily measured, thus leading to a possible underestimation of their role.^38,39 Educational level had a confounding effect. This highlights the complexity of the relationship between area-based disadvantage and individual SEP, and the difficulty of determining how to intervene.

For ARIA and mortality, the mediating effect of risk factors was much lower. Because ARIA is a geographical index, such results could reflect the existence of other mediating factors, such as distance to health care centers.

In most previous studies using area-based indices of SEP, no adjustment was made for individual SEP risk factors.^40–42 In the few studies that have adjusted for individual factors, the relationship between area-based SEP and higher mortality was partially attenuated.^15,43–45 Waitzman et al. described a robust effect of poverty-area residence on all-cause mortality in people younger than 55 years.¹¹ Others found similar results in the general population⁴⁶ or in subgroups (White men,⁴⁷ older men³⁷). Haan et al. reported less than 6% change of the risk association after adjustment for health behavior variables, but this study occurred more than 40 years ago (1965–1974).¹⁵ The authors assumed that the exposure to environmental factors in poorer areas (higher psychological stress, higher crime rates, poorer housing, lack of transportation, environmental contaminants) could explain the relationship with higher mortality rate. One study included dietary factors as mediating factors, but they were analyzed in combination with many other variables.⁴⁴ In our study, diet quality had only a small impact on the relationship between SEP and mortality.

Other studies assessed the impact of educational level on mortality and the contribution of mediators.^1–3 In the National Health and Nutrition Examination Survey III (NHANES III) study, the relationship between individual measures of SEP and mortality remained significant after controlling for known biomedical factors, smoking status, and self-reported global health.³ One other study showed that the contribution of mediating factors (biomechanical exposure, job insecurity, physical exposure, social support) was more pronounced for premature mortality.²¹ Interestingly, our study showed that the relationship between low educational level and premature mortality was also partly confounded by living in a poorer area.

Strengths and Limitations

A major strength of our study is that we had 3 SEP measures (2 area-based and 1 individual-based) and a broad array of covariates to adjust for individuals’ characteristics. Moreover, AusDiab is a national population-based study.

Several limitations must be acknowledged. First, the AusDiab survey has a low mortality rate and for some analyses there were few deaths in each SEP category, potentially leading to lack of significance in the findings and wide confidence intervals. Another limitation of this data set is that we do not have data on the full spectrum of personality-based traits that are key factors in the pathway from socioeconomic position to adverse health, nor do we have data to understand the broader social cultural context—for example, the way an individual interacts with his or her environment. We acknowledge that we cannot fully understand how to intervene to reduce health disparities without an understanding of the broader social context. When stress was taken into account, there was a slight reduction in the magnitude of the relationship between IRSD and premature mortality.

Moreover, confidence intervals obtained by bootstrapping techniques for each risk reduction were large. Finally, mediation analyses could be influenced by imperfect measurement of the included risk factors or the importance of unmeasured risk factors.

Implications

Our study demonstrates that the relationship between living in a disadvantaged area and mortality is partly explained by mediating factors (which on the whole are modifiable), the impact of which seems greater in younger adults. First, it is important to recognize the unexplained part of the relationship and pursue research to explore how socio-environmental factors in poorer areas may influence this relationship.^15,48,49 Second, we suggest that different policies could be implemented in different target groups, such as an area-based approach to target low socioeconomic groups, which is a strategy already developed in some countries with regard to smoking policies.⁵⁰

Conclusions

Our study demonstrates a higher premature mortality for people living in more disadvantaged areas, which was partially mediated by known modifiable risk factors. To decrease health inequalities, prevention policies should target such areas and especially focus on smoking status. Mediating factors showed a similar pattern explaining the relationships between area-based or individual SEP markers and mortality, with a greater impact observed as the age at death decreased.