New insights into inequalities in life expectancy and mortality: evidence from an Individual-level Socio-Economic Index (ISEI) in Australia

Abstract

Background

Socio-economic inequalities in life expectancy and mortality have received significant academic and policy attention in many high-income countries. Most studies to date have measured area-level inequalities either using a composite socio-economic index or individual-level inequalities using a single or limited variables (e.g. education, income). However, these approaches might underestimate life expectancy inequalities compared to a composite index of individual-level socioeconomic characteristics.

Methods

This study fills a major gap in evidence of life expectancy inequalities by analysing 2016 Population Census data linked at the individual level to death registration data from 2016 to 19 in Australia. We used Multiple Correspondence Analysis to construct an Individual-level Socio-Economic Index (ISEI) comprising variables representing education, income, marital status and whether living alone, housing tenure, migrant status, place of residence and neighbourhood socio-economic index. The ISEI was constructed separately for each sex and 10-year age group. Life expectancy at 25 and 65 years, partial life expectancy from 25 to 65 years and age-standardised and age-specific mortality rates were calculated and compared between ISEI percentiles (approximately 85,000 population each) and percentile groups.

Results

Life expectancy at age 25 years ranged from 43.2 (95% confidence intervals 42.8–43.6) years in the lowest (p1) ISEI percentile to 62.9 (62.5–63.4) years in the highest (p100) for males (gap of 19.7 years) and 51.7 (51.3–52.2) to 64.3 (64.0-64.8) years for females (gap of 12.6 years). The ISEI-life expectancy relationship was strong across all percentiles, with the gradient steepest in the lowest percentiles, flatter in middle deciles and slightly steeper in the highest percentiles. Partial life expectancy from 25 to 65 years had a steep gradient in the lowest ISEI percentiles but had a narrower total gap compared with life expectancy at 65 years. Age-standardised mortality rates (25–64 years) differed between p1 and p100 by a magnitude of 17.7 times for males and 11.8 times for females, with the largest disparities at 35–54 years.

Conclusions

This study shows substantial and concerning inequalities in life expectancy and mortality in Australia. By analysing these indicators with greater granularity than earlier studies by using a multidimensional socio-economic index, it shows wider disparities than previously and the important contribution of a wide range of social determinants.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12963-026-00492-3.

Introduction

Socio-economic inequalities in life expectancy and mortality have long attracted considerable academic and policy attention in many high-income countries (HICs). This interest has intensified amid concerns about the recent slowing, and in some cases stagnation, of improvements in life expectancy observed in several HICs, particularly the USA and UK [1]. It has been argued that these slowdowns have been more concentrated among lower socio-economic groups, with implications for widening inequalities within societies. Both widening absolute and relative inequalities have been recently documented across multiple countries, including the USA (e.g. absolute inequalities in life expectancy by income and education) [2–7], the UK (e.g. absolute inequalities in area-level life expectancy) [8–10], Australia (e.g. relative inequalities in premature mortality by area-level socio-economic index) [11–13], Norway (absolute inequalities in life expectancy by income) [14] and Denmark (e.g. absolute inequalities in pre-retirement mortality by education, income, and occupation) [15]. Policy concern regarding such inequalities is demonstrated by the identification of inclusion, equity and fairness as cross-cutting dimensions in the Australian government’s Measuring What Matters Framework, which includes life expectancy as a key indicator [16].

To date, most studies have measured socio-economic inequalities in mortality either at the area level – ranging from local neighbourhoods to much larger aggregations – using composite indices of socio-economic position, or at the individual level using a single or limited number of indicators such as education, income and occupation [17]. In Australia, for example, socio-economic indices have only been used to study life expectancy and mortality inequalities using one of the Australian Bureau of Statistics (ABS) Socio-Economic Indexes for Areas (SEIFA) measured at the Small Area Level 2 (SA2) of average population approximately 10,000 people [11, 12, 18]. This showed life expectancy for males in the most disadvantaged decile (D1) was 6.8 years lower than the most advantaged decile (D10) in 2017-19 and for females was 4.7 years lower in 2021-22 [12]. Analyses in Australia at the individual-level have only been conducted using linked Census and death registration data for a specific socio-economic variable; to measure a gap in life expectancy at age 25 between the highest and lowest educational groups of 9.1 years for males and 5.5 years for females and to identify higher mortality rates among young adult male labourers and machine operators and drivers [13, 19]. While these studies have been particularly informative at documenting disparities in life expectancy, they might underestimate inequalities either due to their unit of analysis (area level compared with individual) or by using limited socio-economic variables at the individual level (e.g. life expectancy has been found to vary by education group within income percentiles in Norway and by marital status within education groups in the US) [5, 14].

There is hence the potential to improve measurement of inequalities in mortality by combining the strengths of these two approaches using a composite index of socio-economic position at the individual-level. To date, there are only limited examples of the use an individual-level socio-economic index to measure mortality and health outcomes; these include an affluence Index in Denmark based on individual wealth and income [20], individual-level social vulnerability, material and social deprivation indices in 10 European countries from survey data [21, 22], an individual-level social deprivation index in France from a survey of attendees at health centres [23, 24], a socio-economic vulnerability index in South Korea from longitudinal data [25], and a deprivation index in New Zealand from survey data [26]. While not strictly measured at the individual level, the ABS have also developed an Index of Household Advantage and Disadvantage, but it has not been used to measure health outcomes [27].

In Australia, person-level linkage of the Population Census with death registration data can be used to construct an individual-level socio-economic index to measure inequalities in life expectancy. Unlike the individual-level indices above, the linked Census and death registration data have both full population coverage and the availability of several socio-economic variables to facilitate highly granular analysis of socio-economic inequalities. The objective of this study is therefore to develop an Individual-level Socio-Economic Index (ISEI) using 2016 Population Census data linked to 2016-19 death registration data in Australia to measure individual-level inequalities in life expectancy and mortality. This period pre-dates the COVID-19 pandemic to reflect patterns of inequality in years where normal mortality patterns prevailed. Compared with previous studies in Australia, the ISEI is expected to show wider disparities in life expectancy and mortality, and more generally to significantly enhance our understanding of the extent and various dimensions of socio-economic inequalities in mortality. The Methods section describes the ISEI in detail and the Results section presents the findings from its application to both life expectancy and mortality rates.

Methods

Study design

This retrospective cohort study used 2016 Population Census data linked to subsequent death registration data from the ABS Person Level Integrated Data Asset (PLIDA) to calculate death rates and life expectancy estimates for the period from 9 August 2016 to 31 December 2019 [28]. The study population comprised individuals residing in Australia on Census night (9 August 2016), who were followed for mortality outcomes until 31 December 2019. There were 17,024,726 people in the study cohort, contributing 54,969,490 person-years of follow-up during the study period.

Census and death registration data

The 2016 Census was conducted on 9 August 2016 and collected a range of demographic, social, economic, living arrangement and other information about the Australian population [29]. The 2016 Census had a net undercount of 1%, so we adjusted the population counts by using ABS published undercount adjustment factors specific to sex, five-year age group and state/territory [30]. We also redistributed missing values of variables to non-missing categories (see Additional File A.1 and Table A.1). Registration data comprising all deaths registered in Australia that occurred after the Census, i.e. 10 August 2016 to 31 December 2019, were used to enable measurement of deaths of people who were counted in the Census [31].

Linkage of Census and death registration data

Census and death registration data were linked in PLIDA using both deterministic and probabilistic methods via the Person Linkage Spine, which identifies individuals residing in Australia in any time period using data from Services Australia, Department of Social Services and the Australian Taxation Office [32, 33]. The Person Linkage Spine was linked to 89.2% of 2016 Census cases and 98.8% of 2016-19 death registration cases, resulting in a combined linkage rate of 88.1% (0.892 × 0.988).

An issue with measuring life expectancy and death rates using the linked Census-death registration data is dealing with the 11.9% of cases that are not linked. There are two sources of bias of unlinked cases. Firstly, unlinked Census cases are more likely from younger adults and lower socio-economic groups and so are expected to have higher death rates relative to their peers than linked cases [33]. This means that final estimates of life expectancy using linked cases will be higher (and mortality rates lower) than for the whole population (i.e. both linked and unlinked cases). Secondly, the relative differences in mortality according to socio-economic status in unlinked cases may also be different to linked cases. To overcome this issue, we redistributed unlinked death cases to unlinked Census cases using negative binomial regression models, to have final estimates for the whole population (linked and unlinked cases). This resulted in three different sets of data for our study population: (1) linked data only, (2) linked and unlinked data, with the unlinked death cases redistributed to unlinked Census cases using a detailed model (including all variables in the ISEI), and 3) linked and unlinked data with the unlinked death cases redistributed to unlinked Census cases using a basic model (including only age and ABS Index of Relative Socio-economic Advantage and Disadvantage or IRSAD). The full process is described in Additional File A.2.

Individual-level Socio-Economic Index (ISEI)

The Individual-level Socio-Economic Index (ISEI) was developed using variables from the 2016 Census [29]. Variables considered for the ISEI were those representing characteristics with existing evidence of a relationship with mortality and health outcomes, including income, education, social relationships, housing tenure, migrant status, neighbourhood socio-economic status and urban-rural residence [4, 11, 34–38]. We excluded variables that measure or are clearly a consequence of poor health, i.e. where there is reverse causality: labour force status, because a primary reason for not being in the labour force is poor health, and having a long–term health condition or disability and need assistance with core activities. We also restricted the ISEI to people aged 25 years and above, the age when an individual would have likely reached the highest education category of Bachelor’s degree or higher.

We calculated the ISEI separately for different age groups, therefore measuring a person’s socio-economic status against people of a similar age. This was done because many of the variables may represent socio-economic status differently at different ages and so may bias an individual’s socio-economic position; for example, attaining Year 11 education or lower is relatively less common among younger than older generations and so indicates relatively poorer socio-economic status for younger people.

The method used to develop the ISEI was Multiple Correspondence Analysis (MCA) [39]. MCA is a dimensionality reduction method used to create an index from a set of categorical variables. It is similar to Principal Components Analysis (PCA), which is instead used on continuous variables and is the basis for IRSAD. In MCA, dimensions are extracted which represent the underlying structure of the data and explain the maximum amount of inertia (variance) present. The first dimension (Dimension 1) explains the greatest amount of the inertia of the variables. For each dimension, each variable category is assigned a coordinate which explains how well associated that category is with the overall direction of a dimension. A summary normalised score is calculated for each observation based on the average of the Dimension 1 coordinates for that individual’s categories, which are then normalised to a mean of 0 and a standard deviation of 1.

Additional File A.3 shows how the MCA was used to develop the ISEI. The combination of variables selected for the ISEI was that with the highest percentage of inertia explained by Dimension 1, averaged over all 10-year age groups (25–34, 35–44 … 85 + years) (results for each MCA shown in Table A.3); the final percentage of inertia explained by Dimension 1 was 55.8%. The final variables used were (full categories for each variable in Table A.4):

• highest level of education

• equivalised household income (i.e. adjusted according to the size and composition of households)

• marital status and whether live alone

• housing tenure

• whether migrated to Australia since 2006

• place of residence according to state/territory and capital city classification [40], and

• ABS IRSAD measured for Small Area Level 1 (average population approximately 400) [18]. Further detail of the IRSAD is provided in Additional File A.3.

Table A.5 shows the MCA coordinates from Dimension 1 of each variable category at each age group and each sex, with an accompanying description of the results. Table A.6 presents the contribution of each variable to the ISEI, with the IRSAD making the largest contribution, followed by equivalised household income and educational attainment. We also assessed the correlation of the variables in Table A.7, which was not high enough to warrant for any to be removed from the ISEI. The summary normalised scores computed from the coordinates were used to calculate ISEI percentiles of equal population size for each sex and 10-year age group, with percentiles ranging from 1 (lowest socio-economic status) to 100 (highest socio-economic status). The age-specific ISEI percentiles were used for analyses across broader age ranges; for each sex, each percentile has a population aged 25 years and above of approximately 85,000. Figure A.2 shows how the distribution of each variable’s categories varies across each ISEI percentile or percentile group. In summary, individuals disproportionately in the lowest ISEI percentiles are those who have not completed Year 12, have the lowest household income, rent from a state/housing authority, are not married and live alone, are not migrants, live outside a major city and live in a lower socio-economic neighbourhood.

Life expectancy and death rate estimates

Additional File A.4 describes the calculation of life expectancy and mortality rates. We analysed differences in the following metrics for each ISEI percentile or group of percentiles:

• life expectancy at age 25 years,

• partial life expectancy from 25 to 65 years; that is, the average number of years expected to be lived between age 25 and age 65 years (i.e., out of a maximum of 40 years),

• remaining life expectancy at 65 years,

• age-standardised death rates at ages 25–64 using the undercount-adjusted Census population as the standard population, and,

• age-specific death rates for 10-year age groups (using 20 quantiles, i.e. percentile 1–5, 6–10 … 96–100); these were calculated based on five-year age groups and age-standardised to the 10-year age groups.

Age-specific death rates at ages 85 years and above were adjusted using the Kannisto logistic model [41]. We calculated 95% confidence intervals using bootstrapping (1,000 samples) assuming a Poisson distribution of deaths. Both age-standardised and age-specific death rates were analysed using rate ratios calculated with the highest ISEI percentile or quantile (percentile 96–100) as the denominator and 95% confidence intervals also assuming a Poisson distribution of deaths. We also calculated the Relative Index of Inequality (RII) of the age-standardised death rates at ages 25–64 years. The RII measures the ratio of the regression-modelled death rate of the lowest ISEI percentile to the highest ISEI percentile, importantly accounting for inequalities across the whole population distribution. We also present a concentration curve of age-standardised death rates at ages 25–64 to show cumulative death rates from p1 onwards for each percentile as a percentage of the sum of death rates across all percentiles.

Sensitivity analysis comparing linked and unlinked (detailed model) data, linked and unlinked (basic model) data and linked data only was conducted using absolute difference in life expectancy at 25 years and the RII of the age-standardised death rates at ages 25–64 years. Our final results present those from the linked and unlinked data using the detailed redistribution model, because this model uses all ISEI variables (compared with the basic model using just IRSAD) and shows wider inequalities than the basic model in this sensitivity analysis. We also conducted analysis to assess the impact of removing each of the household income and IRSAD variables on our results. Further detail is provided in Additional File A.5.

Results

Life expectancy at age 25 (LE₂₅) had a strong relationship with the ISEI (Fig. 1). For males, it ranged from 43.2 (95% confidence intervals 42.8–43.6) years in the lowest percentile (p1) to 62.9 (62.5–63.4) in the highest percentile (p100), a difference of 19.7 years. For females, LE₂₅ ranged from 51.7 (51.3–52.2) years in p1 to 64.3 (64.0–64.8) in p100, a gap of 12.6 years. There is a noticeable shape in life expectancy across percentiles, with the relationship steepest in the lowest ISEI percentiles, flatter to about p80 and then slightly steeper to p100. LE₂₅ was higher than in the preceding percentile for each male percentile p2 to p12 and each female percentile p2 to p9. The 9.9-year life expectancy gap between p12 and p1 for males and 6.7-year gap between p11 and p1 for females was greater than the gap to p100 for each sex. For females, there was a noticeably sharper slowdown of increase in LE₂₅ above p15 than for men for whom the slowdown was more gradual. For females, the rate of increase in life expectancy above p80 was more than double that from p20 to p80. Given the skewed shape of LE₂₅, the national level was equivalent to approximately p40 for each sex.

Fig. 1.

Life expectancy at age 25 by ISEI percentile and sex, Australia, 2016-19 . Note: Linked & unlinked (detailed model) data. Figure A.7 and Table A.11 show 95% confidence intervals for each percentile. Dashed line is the national life expectancy at age 25. Source: ABS [29, 31]

Figure 2 shows a very clear relationship between partial life expectancy at ages 25–65 years (PLE₂₅₋₆₅) and ISEI percentile, with a particularly large decline from p20 to the lowest percentiles for males. From p1 to p100, PLE₂₅₋₆₅ ranged from 34.4 (34.1–34.6) years to 39.6 (39.6–39.7) for males (gap of 5.3 years), and from 36.9 (36.7–37.1) to 39.7 (39.7–39.8) years for females (gap of 2.9 years), with about three-quarters of this gap being between p1 and p20 and a much flatter gradient from p20 to p100. Life expectancy at age 65 (LE₆₅) was lower than for PLE₂₅₋₆₅ due to the average number of years expected to be lived being lower at older compared with the younger ages. LE₆₅ also had a wider range from 15.2 (14.9–15.5) years in p1 to 24.0 (23.7–24.4) in p100 for males (difference of 8.9 years) and 19.7 (19.3–20.0) to 25.2 (24.9–25.6) for females (gap of 5.6 years). The relationship of LE₆₅ with the ISEI is slightly different to PLE₂₅₋₆₅, with a steeper gradient above p80 than from p20 to p80, similar to LE₂₅.

Fig. 2.

a) Partial life expectancy at ages 25–65 and b) life expectancy at age 65 by ISEI percentile and sex, Australia, 2016-19. Note: Linked & unlinked (detailed model) data. Figure A.8 and Tables A.12-A.13 show 95% confidence intervals for each percentile. Black dashed line at 40 years indicates the maximum possible partial life expectancy between 25 and 65 years. Blue and red dashed lines show sex-specific national levels of partial life expectancy at ages 25–65 and life expectancy at age 65. Source: ABS [29, 31]

Figure 3a demonstrates the large disparities in age-standardised death rates at age 25–64 years. The male death rate at p1 was 17.7 (15.7–20.0) times higher than in p100, while for females it was 11.8 (10.1–13.7) times higher. As with life expectancy, in the low percentiles there was a particularly sharp decline in death rates for each successive percentile, with the death rate at p1 being approximately three times higher than p16 for each sex. The ratios of death rates to p100 were quite similar for males and females from p75 and higher but were stronger for males in the lower percentiles. The national age-standardised death rate was equivalent to about p30-p35 for each sex (Fig. 3b). The RII for males was 11.0 (9.8–12.4) and females was 7.4 (6.6–8.3) (Table A.9). The concentration curve shows that 25% of age-standardised death rates (as a percentage of the sum of death rates across all percentiles) occurred in the population in the lowest eight male and nine female ISEI percentiles, 50% occurred in the lowest 24 male and 28 female percentiles, and 75% occurred in the lowest 51 male and 56 female percentiles (Fig. 3c).

Fig. 3.

a) Age-standardised death rate (per 100,000, normal scale), b) Ratio of age-standardised death rate to percentile 100 (log scale) and c) Concentration curve of age-standardised death rates by ISEI percentile and sex, ages 25–64, Australia, 2016-19. Note: Linked & unlinked (detailed model) data. Figure A.9 and Table A.14 show 95% confidence intervals of age-standardised death rate for each percentile. Table A.15 shows 95% confidence intervals of ratio of age-standardised death rate for each percentile. Dashed lines refer to the following: a) sex-specific national age-standardised death rate, b) ratio of 1, c) Cumulative death rates at each percentile (starting at p1) as a percentage of the sum of death rates across all percentiles: 25%, 50% and 75%. Equality line indicates a hypothetical scenario of equal death rates across each percentile. Source: ABS [29, 31]

Death rates had a particularly strong relationship with the ISEI at all age groups from 25–34 to 55–64 years (Fig. 4). For males, the ratio of the lowest quantile (p1-p5) to the highest quantile (p96-p100) exceeded 10 for each of these age groups, reaching 14.5 (12.7–16.6) at 45–54 years and 13.9 (11.5–16.7) at 35–44 years. For females, this ratio was 9 or higher for the 25–34 to 55–64 age groups, reaching 12.1 (8.3–17.4) in the 25–34 age group. The ratios declined to the 65–74 age group but remained at 5.0 (4.7–5.4) for males and 3.9 (3.6–4.3) for females. Male and female ratios were very similar across all quantiles at the 75–84 and 85 + age groups, where the ratios were consistently much lower than at younger age groups.

Fig. 4.

Ratio (log scale) of age-standardised death rate for ISEI percentile to ISEI percentile 96–100 by age group and sex, Australia, 2016-19. Note: Linked & unlinked (detailed model) data. Table A.16 shows 95% confidence intervals of the ratio of age-standardised death rate rates by age group. Source: ABS [29, 31]

Removal of household income and IRSAD from the ISEI moderated the strength of the relationship of life expectancy but still showed large differences, as shown in Table A.10 and Figures A.5-A.6. The biggest decline in difference was for females when IRSAD was removed, where the relationship was flatter through the middle percentiles.

Discussion

The Individual-level Socio-Economic Index constructed for this study demonstrates substantial and unacceptably large inequalities in life expectancy and mortality in Australia. There is a gap of almost 20 years in male life expectancy at age 25 and almost 13 years for females between the highest and lowest percentile. Partial life expectancy from 25 to 65 years has lower absolute disparities than life expectancy at 65 years but a particularly sharp decline in the lowest ISEI percentiles; this narrower range reflects the much lower absolute death rates in the younger compared with older age range. In fact, there are very large disparities in death rates at ages 25–64 years, being almost 18 times higher for males and almost 12 times higher for females in the lowest compared to highest percentile, with 50% of age-standardised death rates in this age group occurring in approximately the lowest 25 ISEI percentiles. Notably, for each metric there is a substantial decline in life expectancy and rise in mortality in the lowest approximately one-seventh of percentiles, with the absolute difference in life expectancy from the 11th −12th percentile to the lowest percentile being similar to the difference to the highest percentile. This shows the importance of understanding the overall gradient of inequalities across all percentiles and not just measuring differences between the highest and lowest percentiles. The wide working-age (25–64 years) mortality inequalities reflect that a higher proportion of deaths at these ages are preventable than older ages, but is also concerning given Australia’s relatively high mortality levels at these ages compared with non-English speaking countries [42].

The availability of linked Census and death registration data facilitates the development of a individual-level multidimensional index such as the ISEI which reflects the reality that life expectancy and mortality outcomes are determined by the interaction of a wide range of social determinants of health, including characteristics of individuals, their households and neighbourhoods. The ISEI enables a granularity of analysis not previously possible by measurements conducted at the area-level or for a single or limited number of individual characteristics. Furthermore, the ISEI also measures socio-economic status relative to a person’s age group, thus overcoming issues with measurement of variables such as education which can reflect socio-economic status differently depending on the cohort. The gap in life expectancy at 25 years far exceeds that found for life expectancy at birth in Australia at the SA2 level (6.8 years for males and 4.7 years for females) between the most advantaged and disadvantaged IRSAD deciles, differences between the highest and lowest life expectancy at birth of the 107 Small Areas Level 4 (SA4; average population almost 250,000) (13.2 years for males and 11.2 years for females) and the gap in life expectancy at age 25 years between the highest and lowest educational groups (9.1 years for males and 5.5 years for females) [12, 13, 43].

The life expectancy gap in this study also is larger than found in the USA by various studies: a 16.1 years gap in partial life expectancy between 30 and 90 years between white married men with a university degree and white never married men with a high school diploma or less, a 14.6 years male and a 10.1 years female life expectancy gap at age 40 (adjusted for race and ethnicity) between the richest 1% and poorest 1% income of individuals, and a 10.7 years gap in life expectancy at birth between the 1 st and 99th percentile of counties ordered by life expectancy; only when comparing the highest and lowest life expectancy of over 3000 counties was there a 20.1 year gap [2, 5, 7]. The gap revealed in our analysis is also higher than in Norway, where life expectancy at age 40 years differed between the richest and poorest 1% according to income by 13.8 years for males and 8.4 years for females [14]. In the UK, analysis of local areas of average approximate population of 8000 showed a gap in life expectancy at birth between the 1 st and 99th percentiles of about 13.6 years for men and 14.2 years for women [10]. That UK study however did find a 27.0 year male and 20.6 year female gap between the highest and lowest life expectancies of their almost 7000 small areas [10]. It is noticeable that the above studies that use percentiles show a somewhat similar shape of life expectancy to our analysis; that is, a sharp decline in life expectancy in the bottom percentiles, although in Australia and Norway this decline is relatively larger compared with the increase from percentiles 20 onwards while in UK local areas there is a sharper rise that in their highest life expectancies [2, 10, 14]. Nevertheless, comparison of our results with these studies should be interpreted carefully given differences in the life expectancy metric used (life expectancy at 25 years compared with life expectancy at birth or 40 years) and that none of those studies used an individual-level socio-economic index.

Our analysis highlights that those in the lowest ISEI percentiles are characterised by people with low levels of education, low income, who rent from a state/territory housing authority, who – demonstrating the importance of marriage and social connectedness – are not married and in particular live alone or are divorced or separated, are not recent migrants and who live in low socio-economic neighbourhoods, all consistent with past literature [4, 11, 34–38, 44, 45]. Although these populations are also highly likely to reside outside of major cities, almost two-fifths of the bottom five percentiles reside in major cities, demonstrating that high mortality levels in these locations too. For context, the life expectancy of the lowest percentiles is lower than for the global level, which was 50.1 years for males and 54.8 years for females in 2018 [46]. The ISEI did not include Indigenous status, consistent with the ABS SEIFA indices, with separate analyses showing Indigenous Australians have overall lower socio-economic status compared with non-Indigenous Australians. The ABS estimated life expectancy at age 25 of 47.9 years for Indigenous males and 51.7 years for Indigenous females, which is consistent with that of the lowest percentiles of the ISEI and which demonstrates Indigenous disadvantage is experienced across many of the dimensions included in the ISEI, not only for a point in time but over the entire life course [47].

Among the high ISEI percentile groups, having high education, high income, being married, owning your own home (with or without a mortgage), being a recent migrant and living in a high socio-economic area are the most common characteristics. Over 99% of people in the top 5 percentiles live in a major city or the ACT, with 46% living in Sydney, 22% in Melbourne, and 11% in the ACT. Our study’s findings of a life expectancy at age 25 years of 64.3 years in the highest percentile of females is among the highest recorded that we are aware of in a sizeable population, which in our study was approximately 85,000; no subnational areas in the Global Burden of Disease has a life expectancy at age 25 this high (the highest in 2019 was 63.7 years in Toyama, Japan), which demonstrates the survival advantage of the highest socio-economic groups in Australia [46]. Life expectancy is only clearly higher in some small areas in the UK study referred to above with average population of about 4000 of each sex; life expectancy at birth reaches 95 years for both males and females in parts of London [10].

This analysis focused on inequalities in life expectancy and all-cause mortality in Australia in the pre-COVID period. Previous research can reveal the possible contribution of causes of death to inequalities according to the ISEI. A detailed analysis of contributions to inequalities in life expectancy using IRSAD at the SA2 level showed that lung cancer, ischaemic heart disease, chronic obstructive pulmonary disease, diabetes, substance-related deaths and, for men, suicides and transport accidents, made the largest contribution to absolute inequalities of all specific causes of death [12]. More recent analysis has also shown particularly large SA2-level inequalities in overweight- and obesity-related cardiovascular disease mortality at 35–74 years [48]. Risk factors such as smoking and obesity have a strong socio-economic gradient in Australia and so are likely to contribute to these differences [49, 50]. Although Australia has a relatively strong public health system, increasing health insurance premiums and out-of-pocket specialist fees make treatment of health conditions more challenging for lower socio-economic populations. It is likely that chronic multimorbidity, exposure to multiple risk factors and relatively poor access to treatment combine to contribute to the sharp decline in life expectancy at lower percentiles. The ISEI has much potential for application to better understand not only causes of death but, by linking with health survey data, morbidity, risk factors and health service utilisation across different socio-economic groups. It also has potential for use by policymakers to identify those populations that are of higher risk of mortality (e.g. not finished high school, low income, not married and live alone).

While the ISEI measures socio-economic status with significant granularity, it has limitations. For the measurement of health outcomes, reverse causality because of poor health or disability may affect some ISEI variables (i.e. being of poor health or having a disability, for which mortality rates are higher, may increase the likelihood of a person being of low socio-economic status) [51]. To address this issue, we did not include any variable in the ISEI that would be highly correlated with poor health, such as employment status and any measure of health status, which are included in the IRSAD. We also assessed removal of household income from the ISEI for this reason, however it only made a small difference to the results by lowering the life expectancy gap for men by less than one year and slightly increasing the gap for women. Another possible issue is high correlation between variables, however our analysis revealed that this was not a problem. Removal of the IRSAD variables did reduce inequalities, especially in the middle percentiles for women, which is to be expected given neighbourhood socio-economic status is a recognised social determinant of health.

Given imperfect linkage rates between the Census and death registration, we had to redistribute deaths according to Census cases using different models. However, this showed that the detailed model did not exacerbate inequalities compared with linked data only. The Census also only measured socio-economic status at a point in time when it is not only the cumulative result of processes across the life course but also potentially changes over time. Research from Denmark reveals that transitions between household types between ages 50 and 65, for example, are not significant enough to affect results of inequality analysis [35]. While the ISEI is measured specifically for different 10-year age groups to provide a more stable measure than for smaller age groups, a trade-off is that there can be considerable socio-economic change that can occur for individuals within certain age ranges, especially 25 to 34 years; this is minimised to an extent by the household income indicator being equivalised to the size and composition of households. Further, the Census also does not measure wealth which is an important determinant of life expectancy in Australia, however income would represent much of its impact [52]. Finally, the most recent Population Census in 2021 was not included in this analysis because the subsequent years for which it could be linked to death registration data (2021-24) include the impact of the COVID-19 pandemic, when normal mortality patterns were disrupted. Socio-economic inequalities in mortality could have been distorted during the COVID-19 pandemic due to disparities in COVID-19 infection and survival rates (and other causes experiencing excess mortality), inequalities in the beneficial impact of social restrictions introduced to prevent spread of the virus, as well as pre-existing mortality disparities. However future analysis of life expectancy during the COVID-19 pandemic using the ISEI would be particularly insightful.

Conclusion

The significant inequalities in life expectancy and mortality shown by the ISEI are concerning for Australia, with segments of the population that share many of the health concerns found in lower socio-economic groups in several other high-income countries [1]. Given that Australia has one of the highest life expectancies in the world, these findings demonstrate that the concept of it being “the land of the fair go” clearly does not apply to mortality outcomes [46, 53]. Future research can use the ISEI to compare trends in life expectancy inequalities over time, commencing with the 2011 Census which was the first to be linked with death registration data; the ISEI can be used to measure relative socio-economic status at a given point in time (i.e. the Census) and so overcomes challenges in assessment of trends in variables whose composition changes over time, such as education [54]. The ISEI can also assess the contribution of causes of death to these inequalities to help inform interventions to reduce disparities in life expectancy. On an international scale, it could also facilitate comparisons with other countries where Census or national registry and death registration data have been linked, such as Japan and Denmark, to compare the socio-economic gradient in life expectancy [15, 20, 55].

Abbreviations

ACT

Australian Capital Territory

COVID-19

Coronavirus disease

D

Decile

IRSAD

Index of Relative Socio-Economic Advantage and Disadvantage

ISEI

Individual-level Socio-Economic Index

LE₂₅

Life expectancy at age 25 years

LE₆₅

Life expectancy at age 65 years

MCA

Multiple Correspondence Analysis

p

Percentile

PCA

Principal Components Analysis

PLE₂₅₋₆₅

Partial life expectancy between ages 25 and 65 years

PLIDA

Person Level Integrated Data Asset

RII

Relative Index of Inequality

SA2

Small Area Level 2

SA4

Small Area Level 4

USA

United States of America

UK

United Kingdom

Author contributions

TA conceptualised the paper, acquired the data, conducted the data analysis, and led the authorship of each draft and the final version of the manuscript. HL, ST, AC and VC-R reviewed and revised each draft of the manuscript and provided intellectual content. All authors read and approved the final manuscript.

Funding

TA acknowledges support from the Australian Research Council (FT240100506). ST acknowledges support from the Medical Research Future Fund (Million Minds 1179461, Cardiovascular Mission 2034149). VC-R acknowledges support from the Australian Research Council (DP210100401).

Data availability

The data underlying this article cannot be shared publicly because they can only be accessed by approved users of PLIDA via the ABS DataLab.

Declarations

Ethics approval and consent to participate

Ethics approval for the study was granted by the University of Melbourne Medicine and Dentistry Human Ethics Sub-Committee (Reference 2023-25166-43934).

Consent for publication

Not applicable.

Competing interests

Tim Adair is Co-Editor-in-Chief of Population Health Metrics. He was not involved in any decisions of this journal regarding this manuscript. Otherwise, the authors declare they have no competing interests.