Social determinants, health status and 10-year mortality among 10,906 older adults from the English longitudinal study of aging: the ATHLOS project

Socio-demographic and lifestyle measurements

Socio-demographic measurements retrieved from ELSA database and used in this work were: sex (men/women), age (in years), formal household wealth and educational qualification. Specifically, according to the ELSA study’s protocol, wealth status refer to household wealth including financial, physical, and housing wealth, but not pension wealth. Wealth was calculated as net of debt and included the value of any home and other property (fewer mortgages); financial assets covering all types of savings available in England; the value of any business assets and physical wealth, such as artwork and jewellery. For the purposes of the present work, participants were classified into 3 classes of household wealth: Low (1st–2nd quintile, n = 3980, 39%), Moderate (3rd quintile, n = 2002, 20%) and High (4th–5th quintile, n = 4082, 41%). For the educational status, participants were classified into three groups based on the highest qualification achieved, i.e.,: Low status, indicating that the individual left education without any formal qualifications (i.e., 0–12 years of education – compulsory education) or after only completing National Vocational Qualifications (NVQs) at level 1 (n = 5488, 50%); Moderate status, which included participants who had completed high school (i.e., 12–14 years of education) or equivalent qualifications (O-level, A-level, or National Vocational Qualifications [NVQs] at levels 2–3) (n = 2710, 25%); and High status, including those participants with college or university degrees or NVQ at Level 4 or 5 (i.e., 15+ years of education) (n = 2688, 25%). Education and wealth status were assessed at baseline, as well as at each of the six waves.

Participants were also asked how often they took part in vigorous-intensity (e.g., running/jogging, swimming, cycling, aerobics/gym workout, tennis, and digging with a spade), moderate-intensity (gardening, cleaning the car, walking at moderate pace, dancing) and low-intensity (laundry and home repairs) physical activities, using prompt cards with different activities to help them interpret different activity intensities. Response options were: more than once a week, once a week, one to three times a month, and hardly ever/never. At each time point, physical activity was further categorized into 3 categories: Inactive; only light activity at least once a week (but no moderate or vigorous) (n = 920, 8%); Moderate activity at least once a week (but no vigorous) (n = 1155, 11%), and Vigorous activity at least once a week (n = 8829, 81%). These thresholds were chosen based on previous work in ELSA demonstrating robust dose-response associations with mortality [9]. According to their smoking habits, participants were classified as never smokers (n = 3890, 36%), former smokers (i.e., those who had quitted smoking before their enrollment in the study, n = 5017, 46%) and current smokers (i.e, those who were still smoking during their enrollment in the study, n = 1998, 18%). Moreover, the frequency of any alcohol consumed in the past 12 months was recorded and responses varied from 1 “almost every day” to 8 “not at all in the past 12 months” and were classified into 3 groups: “Never” (n = 1278, 12%), “Twice a week or less” (n = 6537, 60%) and “More than twice a week” (n = 3087, 28%).

A metric of health status across all ELSA waves

To answer the research question of the present work, i.e., whether the effect of education and wealth status on all-cause mortality is mediated by aging factors, a health metric, as a proxy of healthy aging, has been used. In a previous publication of the ATHLOS project a health metric that incorporated factors associated with aging process, has already been introduced and validated [4]. Briefly, a set of 45 items were identified, comprising: 39 self-reported health questions related to impairments in body functions, limitations in Activities of Daily Living (ADL), and limitations in Instrumental Activities of Daily Living (IADL), and the other six variables were a set of tests covering cognitive functioning and walking speed [4]. This health metric has been calculated for each one of the six ELSA waves. The theoretical range of the health metric was from 0 to 100; higher values in the health metric score are indicative of better health status. The Health metric score descriptives (i.e., mean ± sd and median (range)) of the ELSA project participants across the six ELSA waves are presented in Additional file 1: Appendix 1.

10-year follow-up (2002–2012)

All-cause deaths were recorded. The time of death was available only by year, thus, binary time-specific event indicators were created for each study period. However, for n = 358 participants although the status of death was available, time of death was unknown, thus, information was retrieved looking whether respondents took part in the following surveys; if they were interviewed in later waves, they were assumed to be alive at least until the year of the last survey they responded, otherwise they were considered lost to follow-up. Thus, the participants were characterized as: (1) survivors or censored: if they did not experience a fatal event and were followed-up for all time-periods; (2) dead: if they had a fatal event during the study period; or (3) lost to follow-up: drop-outs. Mortality status was updated at February 2012. using the National Health Service Central Register, based on the informed consent provided by the participants at baseline.

Statistical analysis

The unadjusted associations were assessed by applying t-test, ANOVA and Pearson r correlation where appropriate. In order to assess the over time age/gender-adjusted effect of educational level and household wealth on the health metric calculated at each one of the 6 ELSA waves, a mixed-effects, multilevel regression was conducted using data from all of waves (specified as levels and included in the model as random effects). Then, smoking habits, physical activity and alcohol consumption, were introduced into the model, to assess their potential mediating effect on the education-health and wealth-health associations. Sobel’s test was applied to test the significance of the mediation hypothesis [10]. Furthermore, Cox proportional hazard models assessed the associations of health metric, household wealth and educational qualification on the 10-year mortality rate. The proportionality assumption was assessed graphically and the estimated mortality hazard ratios (HRs) per 10 units increase in the health metric, adjusted for age, sex, education and household wealth for each of the ELSA waves were presented in Fig. 1. Moreover, the corresponding 10-year survival curves were constructed (i.e., according to the educational and household wealth level recorded at baseline) and compared by applying the log-rank test (Fig. 2). All reported p-values were based on two-sided tests. STATA 14 (Stata Corp LLC, Texas, USA) software was used for the statistical calculations.

Education, household wealth, lifestyle behaviors and the health metric of functioning and cognition, across study’s course

Indicators of lifestyle behaviors, such as smoking, physical activity and alcohol consumption were also associated with the health metric of functioning and cognition. In particular, the metric was inversely associated with smoking status (p < 0.001) (i.e., smokers had lower functioning and cognition abilities across aging) and positively associated with education (i.e., better education level, higher functioning and cognition across aging), household wealth (i.e., better wealth status higher functioning and cognition), physical activity level (i.e., engaging in physical activities higher functioning and cognition during aging) and alcohol consumption pattern (moderate alcohol drinking better functioning and cognition level) (all p-values < 0.001). The aforementioned lifestyle factors also acted as mediators in the education and wealth associations with the health metric (p < 0.001). All the three, reduced significantly (i.e., mediated) the effect size of both the educational and economic level on the health metric, which nevertheless remained significant. A linear, positive correlation was observed for both physical activity and alcohol consumption with the health metric, while smoking (currently or formerly) was associated with lower health scores (p < 0.001); current smokers did not differ from the former in terms of the health metric (p = 0.921). No significant interaction was observed between sex of the participants and educational level (p = 0.339) or household wealth (p = 0.116) in their relationship with the health metric.

Education and household wealth as predictors of 10-year all-cause mortality: The mediating effect of the health metric of functioning and cognition

The health metric was inversely and progressively associated with 10-year all-cause mortality risk across the six waves (Fig. 1). Moreover, the age-gender adjusted all-cause mortality risk was 22% lower for the moderately educated (p < 0.001) and 16% lower for the highly educated (p = 0.011) as compared to the lowest education class; there was no difference between moderate and high education classes (p = 0.384). After further adjusting for the metric that evaluated healthy aging status of the participants, only the moderate educated class differed significantly compared to the lowest class as regards the 10-year mortality rate (p = 0.018). Sobel test yielded a significant mediating impact of the health metric on the observed education-mortality association (p = 0.02). Concerning household wealth, both the middle and highest classes had 26 and 35% reduced mortality risk as compared to the lowest class, respectively (p < 0.001), while the middle class compared to highest, had also increased mortality risk (14%, p = 0.057). As in the case of education level, the introduction of the health metric in the aforementioned model mediated the wealth-mortality association (p < 0.001) by reducing the effect size of wealth status on mortality (Sobel’s p = 0.01). However, the association of household wealth on mortality remained significant, even after adjusting for the health metric of functioning and cognition, with the highest and middle classes having lower risk as compared to the lowest class (p = 0.001); no differences between the middle and highest class were observed as regards their effect on mortality (p = 0.279). Moreover, no significant interactions were observed between gender of the participants and educational level (p = 0.09) or household wealth (p = 0.188) as regards 10-year mortality risk. The corresponding estimated survival curves are provided in Fig. 2. Similar were the results when analyzing education and wealth status from the rest of the ELSA waves (data not presented here).

Strengths and limitations

The fact that the reasons for the abstinence from alcohol and/or smoking were not recorded could be seen as a limitation of the present work, as a pure health state is frequently the cause and thus, it was not possible to distinguish between the abstainers of free choice and those who abstained after medical advice while the only available data on alcohol consumption was the 3-level distinction as follows: i) not drinking, ii) twice a week or less and iii) more than twice a week. Also, the exact years of educational attainment were not available in the ELSA survey and thus the corresponding educational levels were used. Additionally, there was not available information on the existence of several co-morbidities or drugs used that could probably have affected the results. Despite the aforementioned limitations, the use of a large and nationally representative sample of the English population coupled with the long follow-up with adequate reassessments (6 waves) makes it possible to draw safe and generalizable conclusions. Additionally, the employment of a latent score on health, which can be compared across longitudinal studies opens and facilitates the way for further research at multinational level.