Lower socioeconomic status and the acceleration of aging: An outcome-wide analysis

The study involved 5,018 participants in the English Longitudinal Study of Aging (ELSA) aged 52 and over (mean 64.44, SD 8.49) assessed in 2004 (baseline) and then 8 y later in 2012. Wealth was divided into quartiles for the purposes of analysis. All analyses included age, age squared (to account for nonlinear age associations), gender, ethnicity, educational attainment, childhood SES (based on paternal or primary caregiver’s occupation when the participant was aged 14), and number of long-term conditions (arthritis, asthma, cancer, coronary heart disease, dementia, diabetes, heart failure, Parkinson’s disease, and stroke) as covariates. As can be seen in SI Appendix, Table S1, lower SES participants were slightly older on average; were more likely to be female and nonwhite (although the overall proportion of ethnic minorities was small); and had fewer educational qualifications, lower childhood SES, and more long-term conditions. All analyses were weighted using inverse probability weights to ensure national representation and to take account of differential nonresponse at follow-up (see SI Appendix for details). The levels of the 19 outcome variables at baseline and follow-up are summarized in SI Appendix, Table S2; the numbers in each analysis varied because of missing data. Bonferroni corrections were applied within each domain to take account of multiple comparisons.

The cross-sectional associations between SES and outcomes are detailed in Table 1. The four measures of physical capability were all related to SES in a graded fashion, with lower hand grip strength, slower gait speed over a standard distance of 8 ft, reduced ability to reach criteria in standing up repeatedly from a chair, and less self-reported physical activity among lower SES participants. Self-reported sensory function was also inversely associated with SES, with lower SES participants more likely to rate their sight only fair or poor, rather than good, very good, or excellent. The association with self-reported hearing was not significant after Bonferroni correction. In the physiological function domain, lower SES was related to higher levels of two markers of inflammation: high-sensitivity plasma C-reactive protein and fibrinogen. There was also a gradient in lung function, with lower forced expiratory volume over 1 s (FEV₁) and forced vital capacity (FVC) in less affluent participants.

Three aspects of cognitive function were assessed: memory (a combination of immediate and delayed recall of word lists), executive function (a verbal fluency task), and processing speed on a letter cancellation task (29). There were significant cross-sectional SES gradients in memory and executive function but not processing speed, with poorer memory and executive function in lower SES categories. We measured two aspects of emotional well-being. Positive affective well-being was assessed with a four-item measure of enjoyment of life that has previously been associated with health outcomes (30), while depressive symptoms were indexed by the eight-item Center for Epidemiologic Studies Depression scale (CESD) (31). Both showed SES gradients, with lower positive well-being and an increased prevalence of depressive symptoms in lower SES groups. In the social domain, we included four indicators of social function and prosocial behavior: participation in organizations such as social clubs, church groups, and resident organizations; number of close friends; volunteering on a monthly basis or more frequently; and cultural engagement (going to museums, concerts, or theater). In all cases, there was a significant social gradient, with lower SES individuals being less likely to participate.

Longitudinal analyses investigated SES differences in deterioration of function over 8 y. Analyses adjusted for baseline levels of each function and the same covariates as in the cross-sectional analyses. Significant results are presented in Figs. 1 and 2, with full statistical details in SI Appendix, Table S3. Because the continuous outcomes were measured on a variety of metrics, we standardized scores into SD changes. In the physical capability domain, there were reductions in average grip strength, gait speed, and self-reported physical activity over time. These decreases were graded by SES, being greater in the less wealthy categories. Differences were substantial; for instance, the reduction in gait speed was 38% greater in the lowest than in the highest wealth category. Although a higher proportion of people in the lower SES groups failed the chair stand test, the linear trend across categories was not significant. In the sensory function domain, incident self-reported sight problems ranged from 10.1% in the highest to 15.6% in the lowest SES category. The SES gradient in incident hearing problems was not significant.

Three of the four markers of physiological function showed social graded impairments over the 8 y follow-up period. The concentrations of both inflammatory markers decreased on average, and the reductions in fibrinogen were greater in higher SES groups. Lung function declined in all groups, with the decreases in FEV₁ and FVC being significantly greater in lower SES participants after controlling for covariates. As can be seen in Fig. 1, the fall in FEV₁ in the lowest SES group was more than double that in the highest group.

Changes in memory were assessed over 8 y, but the changes in executive function and processing speed were measured over a 6-y period because these tests were not administered in wave 6 but in wave 5 (2010) of ELSA. As can be seen in Fig. 2, performance of all three cognitive tests deteriorated over time, with fewer words recalled in the memory test, reduced success in the verbal fluency test, and slower speed on the letter cancellation task. In all cases, the decline was socially graded, with little change in the highest SES group progressing to larger reductions in the lowest SES group. Changes in the two measures of emotional well-being were also socially graded, with greater reductions in enjoyment of life and more incident depressive symptoms in the lower SES categories. There were also significant SES differences in changes over time in three of the four social function measures. Membership of organizations tended to decline but was maintained at baseline levels in the highest SES group. The likelihood of volunteering and of engaging in cultural activities on a regular basis also varied across SES groups, being greater in more affluent categories.

We carried out a number of sensitivity analyses to explore different explanations of results. First, we repeated analyses after omitting baseline levels of the outcomes as covariates. The results (SI Appendix, Table S4) show no changes in the pattern of results for incident or binary outcomes but reduced statistical significance for the continuously distributed outcomes. Only the differences in changes in memory remained significant, indicating that the SES variations in decline were dependent on cross-sectional differences. Second, analyses were limited to people aged 75 and younger. The results in SI Appendix, Table S5, show a similar pattern as in the full analysis, with 12 of the 15 significant associations being unchanged. However, the SES differences in grip strength, sight impairment, verbal fluency, and processing speed were no longer robust, whereas the gradient in C-reactive protein concentration was significant. Third, we tested the possibility that differences in smoking rates underlie the pattern of results. Smoking is socially graded and is associated with the acceleration of cognitive decline, inflammation, and several other outcomes. Smoking rates were 8.3% in the highest wealth quartile, increasing to 10.3, 14.7, and 24.9% across the remaining quartiles. However, adding smoking as a covariate had a limited influence on longitudinal results, with three associations no longer being significant (grip strength, sight impairment, and verbal fluency) (SI Appendix, Table S6). Next, we considered whether differences in marital status were influential since marriage/partnership is more common among higher SES individuals and is related to several of the outcomes assessed here (32, 33). Repeating the analyses with baseline marital status as a covariate resulted in only one change (grip strength) to the significance of longitudinal associations. Finally, we conjectured that although the associations in Figs. 1 and 2 were independent of the presence of long-term health conditions, there might be health effects that were not detected by these measures. We therefore repeated the analyses adding self-rated health as a covariate. The associations between SES and grip strength and incident sight impairment were no longer significant, but there were no other differences from the main analyses.

Lower SES is associated with increased risk of many of the long-term health problems of aging, and declines in function and phenotypes could be consequences of SES gradients in illness. Our results indicate that lower SES is related to acceleration of a broad range of age-related impairments independently of diagnosed health conditions or self-rated health. It is striking that these associations were observed across six domains of physical, mental, and social function that typically deteriorate with age. Associations were independent of age, gender, ethnicity, educational attainment, and childhood SES as well as reported illness status. Although the overall magnitude of decline varied across outcomes, SES differences in change were similar, averaging 0.1 to 0.2 SDs between the highest and lowest SES groups. Sensitivity analyses indicated that the pattern of results could not be explained by the influence of smoking status, marital status, or additional indicators of health and were present in younger participants as well as the full sample.

The cross-sectional results summarized in Table 1 largely corroborate previous findings (11–15, 18) but provide little insight into the temporal relationships. Processes such as sensory impairment, inflammation, and low psychological well-being could potentially contribute to reduced SES as well as being exacerbated by lower SES. Most existing longitudinal studies of SES and age-related decline have focused on individual domains such as cognitive function or physical capability and have shown mixed relationships (11, 14, 21, 22). Work combining different outcomes has been carried out in the successful aging research literature, but findings have varied across domains of function and measures of SES (16, 34, 35). Our approach used a single indicator of SES (wealth), relating it to 19 outcomes in a single dataset, applying a uniform approach to analysis. This outcome-wide approach has been advocated for population studies because it provides information about the broad impact of specific exposures and avoids many of the methodological difficulties of exposure-wide studies (26). We selected wealth as a more precise indicator of contemporary socioeconomic resources at older ages than measures such as education and occupational status that are typically acquired in early life or midlife. Income can also be misleading since some older people with financial resources seek to limit their incomes in order to preserve their capital (28). All analyses also took childhood SES and educational attainment into account, indicating that the differences with wealth were not proxies for earlier life SES but represented accumulated socioeconomic resources at older ages.

The longitudinal analyses of continuously distributed variables were conducted with baseline levels of the outcomes as covariates in order to take account of the observed cross-sectional associations. Although it has been argued that adjustment for baseline may not be appropriate (36), this is a controversial issue because simple analyses of change scores may not provide unbiased estimates (37, 38), and the inclusion of baseline covariates has been advocated on statistical grounds, including improving efficiency, precision, and power and to avoid the pitfall of regression to the mean (39, 40). To explore and evaluate the differences in results we conducted sensitivity analyses excluding baseline covariates; findings for the six incident or binary outcomes were unchanged, but for the continuous outcomes the magnitude of several of the coefficients was substantially diminished, consequently affecting statistical significance. Previous studies have shown that changes in many capability outcomes are highly influenced by baseline values (14, 23, 41); therefore, we believe that a more appropriate approach in this setting is the use of baseline adjustment.

No single factor is likely to drive these associations between the extent of age-related decline and SES. Rather, multiple processes associated with SES disparities may contribute to different domains of function. For instance, greater wealth may provide a pathway toward more mentally stimulating environments and cultural resources that will impact on cognitive and social function; better access to green spaces and exercise facilities may enhance physical activity and help maintain physical capability; wealth may be directly related to eye care; and greater chronic life stress in less affluent groups may influence mental well-being and inflammation, while exposure to environmental pollution may promote faster decline in lung function among lower SES groups.

This study involved a large representative sample of older people in England and included a combination of self-reported and objective outcomes. The multifactorial nature of data in ELSA provided the opportunity to evaluate a broad range of outcomes and to explore several potential explanations of the findings. However, as with most longitudinal studies, there was marked attrition over the 8-y follow-up period. Part of this was because of deaths which were more common among lower SES participants. Of the original sample, 18.7% died before the 8 y follow-up, with death rates ranging from 29.9% in the lowest to 15.7% in the highest SES group. This survivorship bias might be expected to act against the hypotheses tested because the participants lost to the study would have been more vulnerable to functional decline, leaving a more robust lower SES group. We addressed this issue by inverse probability weighting to account for several factors associated with attrition. Unfortunately, there were some facets of age-related decline that could not be assessed, including changes in sleep patterns and nutrition. Other variables such as body mass index could have been analyzed but were excluded because the influence of aging on changes in adiposity independently of ill health is not clear.

Ferrucci et al. (10) have argued there is a hierarchical relationship between the molecular/cellular, phenotypic and functional levels in aging, with alterations at the basic biological level being buffered by compensatory resilience mechanisms that delay their impact on phenotypic processes, that in turn drive functional impairment. Whether lower SES impacts on biological and cellular aging processes is poorly understood; research to date has involved cross-sectional studies using education as a marker of early life SES, and inconsistent associations with telomere length and epigenetic aging have been reported (42–44). This study provides evidence for pervasive associations between lower SES and the acceleration of decline across a wide range of age-related processes at the functional and phenotypic levels. The observation that relationships were independent of reported health conditions and self-rated health suggests that SES impacts on important aging metrics directly. Such differences might persist even if social inequalities in long-term health conditions were reduced.

ELSA is a nationally representative longitudinal panel study of English adults aged 50 and older (45). The study was approved through the National Research Ethics Service, and all participants provided informed consent. The baseline for these analyses was wave 2 (2004/2005) of data collection, and follow-up was in wave 6 (2012/2013). There were 8,402 participants with complete data on wealth (SES) and covariates at baseline, of whom 1,588 passed away, and 1,796 did not participate in wave 6, leaving 5,018 in the analytic sample. Wealth was measured with a detailed assessment of the participant’s financial, housing, and physical wealth (such as land, business wealth, and jewelry) but excluded pension wealth. The mean wealth in the four quartiles was £665,064, £247,413, £143,252, and £25,783, for the highest to lowest, respectively. The covariates in all analyses were age, gender, ethnicity (white or nonwhite), education divided into three levels of attainment, childhood SES, and the number of long-term health conditions present. The data used in these analyses are available from the UK Data Service (access GN 33368) at ukdataservice.ac.uk/.

Full details of the measures are provided in SI Appendix.

The English Longitudinal Study of Ageing is administered by a team of researchers based at the University College London, NatCen Social Research, the Institute for Fiscal Studies, and the University of Manchester. Funding is provided by National Institute on Aging (Grant R01AG017644) and by a consortium of UK government departments coordinated by the Economic and Social Research Council.