Are younger cohorts in the USA and England ageing better?

Abstract

Background

Whether worldwide increases in life expectancy are accompanied by a better health status is still a debate. People age differently, and there is a need to disentangle whether healthy-ageing pathways can be shaped by cohort effects. This study aims to analyse trends in health status in two large nationally representative samples of older adults from England and the USA.

Methods

The sample comprised 55 684 participants from the first seven waves of the English Longitudinal Study of Ageing (ELSA), and the first 11 waves of the Health and Retirement Study (HRS). A common latent health score based on Bayesian multilevel item response theory was used. Two Bayesian mixed-effects multilevel models were used to assess cohort effects on health in ELSA and HRS separately, controlling for the effect of household wealth and educational attainment.

Results

Similar ageing trends were found in ELSA (β = –0.311; p < 0.001) and HRS (β = –0.393; p < 0.001). The level of education moderated the life-course effect on health in both ELSA (β = –0.082; p < 0.05) and HRS (β = –0.084; p < 0.05). A birth-year effect was found for those belonging to the highest quintiles of household wealth in both ELSA (β = 0.125; p < 0.001) and HRS (β = 0.170; p < 0.001).

Conclusions

Health inequalities have increased in recent cohorts, with the wealthiest participants presenting a better health status in both the USA and English populations. Actions to promote health in the ageing population should consider the increasing inequality scenario, not only by applying highly effective interventions, but also by making them accessible to all members of society.

Key Messages

• We assessed whether recent cohorts are ageing in a healthier manner in two large and representative cohorts of the English and US populations.

• A harmonized metric of health based on different domains of functioning was jointly estimated in both countries, allowing the comparison of health trends across cohorts.

• The common metric used allowed us to show that the patterns of ageing and their moderators were similar in both countries.

• We found that recent cohorts are ageing better, but only among the wealthiest groups of the US and English populations.

Introduction

The worldwide population is ageing steadily as a result of increased longevity and reduced fertility rates.¹ A global rise is expected in the world population aged 60 years or older from 841 million in 2013 to more than 2 billion in 2050.² In addition to the increased number of older people, the number of people aged 20–64 years will be reduced to almost half in some countries, such as the UK or USA, by 2030.³ This demographic shift will probably have important socio-economic and health implications, raising concerns about the demands on pension schemes and savings, and the burden of older people on health and long-term-care systems.¹

Whether worldwide increases in life expectancy are accompanied by a better health status is still a debate, and there is a need to disentangle whether healthy-ageing pathways can be shaped by cohort effects. Although the prevalence and incidence of chronic conditions have increased in the older population,^2,4 the evidence on trends in functioning among birth cohorts of older people is mixed and inconclusive. Whereas some studies show a decrease in functioning difficulties and disability for recent cohorts,^2,5–9 others report an increase in the total lifetime days of disability.^6,10–15 Some of these conflicting findings could be explained by country-specific variables that might moderate the relationship between birth cohorts and health status. For example, the positive impact of a higher educational status on health might be different across birth cohorts, as higher education has become more accessible for younger cohorts in some countries.¹⁶ Country wealth and its change over time might also contribute to inconsistencies in results; for instance, one study in the US population showed that recent younger cohorts benefit from a disability reduction in old age but only in the high-income group.¹⁷ Similarly, another study in the UK reported an increasing disability trend in the lower-income group for younger cohorts.¹⁸

Besides potential country-specific moderators, the sparse and conflicting evidence on cohort effects on health might be due to the diverse conceptualizations and measurements of health status employed across studies. Most of the existing studies comparing data across countries are based on severe measures of disability [i.e. limitations in Activities of Daily Living (ADLs) in Instrumental Activities of Daily Living (IADLs) or the presence of chronic health conditions].¹⁹ For example, Bank et al.¹⁹ found that middle-aged people living in the UK had better health status than the US middle-aged population, whereas Cieza et al.²⁰ did not find such differences. These measures of disability limit the scope for assessing health trends in people with mild functioning problems, and do not consider other important domains of health.

The conceptualization of health used in the present article is within the framework of the World Health Organization²¹ and the International Classification of Functioning²² (ICF), based on an individual’s intrinsic capacity to function in several domains (e.g. mobility, seeing, cognition, etc.) and its interactions with the environment. Considering the sparse and mixed evidence, it is necessary to clarify whether healthy-ageing pathways can be shaped by cohort effects and whether these cohort effects are different across countries. Thus, the present study aims to compare trends in health status in two large nationally representative samples of older adults from England and the USA, assessing whether younger cohorts of older adults are ageing in a healthier manner in these two populations.

Methods

Sample and study design

The sample comprises participants from the first seven waves of the English Longitudinal Study of Ageing (ELSA) (2002–14)²³ and the first 11 waves from the Health and Retirement Study (HRS) (1992–2012).²⁴ The ELSA and HRS are biannual longitudinal studies conducted on nationally representative samples of people aged 50 years and over from the English and US populations, respectively. All participants in the ELSA provided written informed consent and the National Research Ethics Service granted ethical approval for all the ELSA waves (MREC/01/2/91). Further details of the ELSA sample, study design and data collection are available at the ELSA project website (https://www.elsa-project.ac.uk/). In HRS, participants provided written verbal and an informed consent document to participate. The requirements from the University of Michigan's Institutional Review Board were used for the collection of the HRS data. Additional details of the study design and sampling procedure are available on the HRS website (hrsonline.isr.umich.edu).

Measures

A common latent health score based on self-reported health items related to impairments in body functions, limitations in ADLs and IADLs, and measured tests of cognitive performance and walking speed was created from the original datasets in both samples to assess potential cohort effects on health in both countries. This metric of health was estimated simultaneously for ELSA and HRS using a Bayesian multilevel item response theory approach,²⁵ with scores ranging between 0 and 100, and higher scores indicating a better health status. This overall measure of health allows comparisons between both countries.^2,26

Further details on the metric and the statistical model can be found elsewhere.^26,27 The level of educational attainment was harmonized in both studies. In HRS, 0, 1–12 and 13+ years of education were considered as ‘no education’, ‘medium education’ and ‘high education’, respectively. In ELSA, the highest level of educational attainment was used, considering no qualification, up to National Vocational Qualification (NVQ) Level 2/General Certificate of Education O level and NVQ level 4 and over as ‘no education’, ‘medium education’ and ‘high education’, respectively. Quintiles of household wealth were also obtained for both ELSA and HRS, considering savings, investments, value of any property or business assets, net of debt and excluding pension assets.

Statistical analysis

Descriptive statistics for the socio-economic variables were computed for both the ELSA and HRS samples. Two Bayesian mixed-effects multilevel models based on the hierarchical–multilevel framework proposed by Bell and Jones²⁸ were used to assess cohort effects on health in ELSA and HRS separately. According to Bell and Jones,²⁷ the best way to discern age–period–cohort effects is to constrain one of these effects to zero, thus allowing appropriate inference about the other two effects. As, in this study, we were particularly interested in modelling birth-cohort effects on life-course trajectories of health, period effect was constrained to zero. The fixed part of the model included the continuous linear and quadratic effects of age and year of birth, as well as the quintile of household wealth and the level of education. In addition, the interactions among all the fixed effects were also included in the fixed part of the model. Multilevel variance inflation factor (MVIF) was used to assess collinearity among predictors, with variables presenting values over 10 indicating potential collinearity problems.²⁹ The random part of the model comprised three levels: the measurement occasion, the person and the cohort, with measurements across waves nested within individuals and individuals within cohorts. Thus, in the random part, the cohort effect is modelled using a categorical variable defined by 5-year birth cohorts, since cohorts are conceptualized as contexts in which individuals are nested. Bayesian estimation using Markov Chain Monte Carlo was used to estimate the parameters of the mixed-effects multilevel models. Gibbs sampling and diffuse prior distributions for the model parameters were specified to approximate maximum-likelihood estimation. For the random effects included in the model, the variance partition coefficient (VPC) was obtained to indicate the percentage of variance of health accounted for by each level in the model. As a sensitivity analysis, the Bayesian mixed-effect multilevel models were conducted separately by gender in both ELSA and HRS studies. STATA was used for descriptive analyses and Bayesian multilevel models were conducted using MLwiN.^30,31

Results

The overall sample consisted of 55 684 people participating in at least one wave of either ELSA (n = 18 396; 54.5% of women) or HRS (n = 37 288; 56.2% of women). Table 1 displays the characteristics of the ELSA and HRS samples at baseline. Supplementary Table 1, available as Supplementary data at IJE online, shows the age range across the different birth cohorts considered in both studies.

Table 1.

Baseline characteristics of the ELSA and HRS samples

	ELSA (n = 11 906)	HRS (n = 12 648)
Age at baseline
M (SD)	64.97 (10.20)	56.45 (4.49)
Median	64	56
IQR*	17	6
Year of birth at baseline
M (SD)	1936.66 (10.22)	1935.28 (4.48)
Median	1938	1936
IQR*	17	7
Female, N (%)	6663 (55.96)	6783 (53.63)
Level of education, N (%)
No qualification/0 years of education	4911 (41.32)	83 (0.66)
Medium education/up to 12 years of education	3531 (29.71)	8034 (63.52)
High education/13+ years of education	3444 (28.98)	4531 (35.82)
Quintile of household wealth, N (%)
1st quintile	2126 (19.29)	2230 (17.63)
2nd quintile	2193 (19.90)	2224 (17.74)
3rd quintile	2210 (20.05)	2804 (22.17)
4th quintile	2214 (20.09)	3331 (26.34)
5th quintile	2277 (20.66)	2039 (16.12)

^*IQR, interquartile range.

Results from the mixed-effects multilevel models in the ELSA and HRS samples are presented in Table 2. All predictors presented MVIF values below 10 in both ELSA (MVIF for birth year = 6.43; MVIF for age = 6.27; MVIF for household wealth = 1.21; MVIF for education: 1.31) and HRS (MVIF for birth year = 3.91; MVIF for age = 3.83; MVIF for household wealth = 1.41; MVIF for education = 1.30), suggesting the absence of potential collinearity problems with these results. As can be seen in the fixed part of the ELSA model, there is a negative age effect on health with a linear shape (β = –0.311; p < 0.001). Figure 1 displays the longitudinal course of health status in each study. There was also a small but statistically significant quadratic effect of age on health status (β = –0.014; p < 0.001). The level of education and quintile of household wealth were also positively related to health (p < 0.001). No significant effects were found for year of birth or the interaction between year of birth and linear age effects. Nonetheless, some interactions presented significant effects on health. The age–high education interaction presented a negative coefficient (β = –0.082; p < 0.05). Figure 2 shows health trajectories by education level in each study. On the other hand, there were significant interactions between year of birth and belonging to the 3rd (β =  0.112; p < 0.05) and 5th (β =  0.125; p < 0.001) quintiles of household wealth. Figure 3 presents birth-cohort effects on health across the five quintiles of household wealth. Regarding the random part of the model, 51.78% of the variation in health was explained by inter-individual differences (Person level VPC = 0.518), whereas the birth-cohort level explained less than 1% of the variability in health (Cohort level VPC = 0.002).

Table 2.

Parameter estimates of the mixed-effects multilevel models in the ELSA and HRS studies

	ELSA			HRS
Fixed part coefficient estimates	Mean estimate	95% CI*		Mean estimate	95% CI*
Constant	55.096***	52.993	56.180	51.269***	48.646	54.670
Age	–0.311***	–0.377	–0.196	–0.393***	–0.482	–0.200
Age2	–0.014**	–0.019	–0.004	–0.027***	–0.028	–0.021
Birth year	–0.043	–0.109	0.012	0.022	–0.082	0.124
Birth year2	0.004	–0.003	0.008	–0.002**	–0.006	–0.001
Age * birth year	0.004	–0.005	0.015	–0.014***	–0.016	–0.011
Education (ref. no.)
Medium	2.502***	2.059	3.024	4.412***	3.100	5.680
High	4.066***	3.630	4.479	9.532***	6.942	10.935
Quintile of household wealth (ref. 1st quintile)
2nd quintile	3.582***	2.901	4.778	1.514***	1.216	4.255
3rd quintile	5.454***	4.291	7.814	2.383***	1.971	6.966
4th quintile	6.692***	5.297	9.636	2.918***	2.424	8.708
5th quintile	7.868***	6.092	11.665	3.577***	2.984	10.885
Age * medium education	–0.018	–0.065	0.044	–0.088**	–0.166	–0.007
Age * high education	–0.082**	–0.143	–0.032	–0.084**	–0.163	–0.002
Age * 2nd quintile of household wealth	–0.031	–0.103	0.029	–0.039**	–0.130	–0.013
Age * 3rd quintile of household wealth	–0.009	–0.079	0.051	0.015	–0.055	0.041
Age * 4th quintile of household wealth	–0.063*	–0.123	0.006	0.044*	–0.074	0.073
Age * 5th quintile of household wealth	0.012	–0.058	0.073	0.070*	–0.062	0.101
Birth year * medium education	0.051*	–0.002	0.103	–0.059	–0.153	0.035
Birth year * high education	0.017	–0.049	0.075	–0.047	–0.144	0.047
Birth year * 2nd quintile of household wealth	0.065	–0.010	0.176	–0.017	–0.101	0.008
Birth year * 3rd quintile of household wealth	0.112**	0.031	0.239	0.055**	–0.011	0.080
Birth year * 4th quintile of household wealth	0.062	–0.027	0.212	0.120**	0.023	0.148
Birth year * 5th quintile of household wealth	0.125***	0.044	0.251	0.170***	0.098	0.199
Random part variance estimates
	Variance	95% CI*		Variance	95% CI*
Level 3: Cohort	0.245	0.001	1.345	11.749	2.331	39.768
Level 2: Person	81.277	0.010	115.982	101.503	0.059	108.998
Level 1: Occasion	75.433	45.024	152.826	68.553	63.299	155.984

^* p < 0.05;

^** p < 0.01;

^*** p < 0.001 (Bayesian p-values).

^*CI, Bayesian Credible Intervals.

Figure 1.

Health trajectories in the HRS and ELSA studies.

Figure 2.

Interaction between age and level of educational attainment in the HRS and the ELSA studies.

Figure 3.

Interaction between year of birth and quintiles of household wealth in the HRS and the ELSA studies.

In the fixed part of the mixed-effect multilevel model conducted in the HRS sample (Table 2), there was a significant and negative linear effect of age on health (β = –0.393; p < 0.001) (Figure 1). The negative quadratic effects of age (β = –0.027; p < 0.001) and year of birth (β = –0.002; p < 0.05) were associated with health. Higher levels of education and household wealth were positively associated with a better health status (p < 0.001). Several interactions were found to be significant in the HRS sample. Regarding the interactions, the age effect on health was found to be significantly smaller in: (i) younger cohorts (age * birth year: β = –0.014; p < 0.05); (ii) participants with medium (age * medium education: β = –0.088; p < 0.001) and high levels of education (age * high education: β = –0.084; p < 0.05) (Figure 2); and (iii) participants belonging to the 2nd quintile of household wealth (age * 2nd quintile of household wealth: β = –0.039; p < 0.05). In addition, there were some significant interactions between the year of birth and the level of household wealth (Figure 3). In that regard, earlier cohorts belonging to the 3rd (Birth year * 3rd quintile of household wealth: β =  0.055; p < 0.05), 4th (birth year * 4th quintile of household wealth: β =  0.120; p < 0.05) and 5th (birth year * 5th quintile of household wealth: β =  0.170; p < 0.001) quintiles of household wealth presented a better health status than those belonging to the 1st quintile of household wealth. According to the results from the random part of the HRS multilevel model, 55.83% of the variation in health was explained by the person level (VPC = 0.558), whereas the cohort level only accounted for 6.46% of its variance (VPC = 0.064). Results from the sensitivity analysis conducted separately by gender in each study revealed practically identical results for men and women. These results are available upon request.

Discussion

This study aimed to shed light on the ongoing debate about whether recent birth cohorts are healthier than those that preceded them. To do so, we analysed the patterns of health status across birth cohorts over the past two decades in two large and representative samples of older adults from the English and US populations.

In this study, we have found variables that moderate the relationship between health, age and birth cohort. The existence of these moderators might explain the inconsistent results found in previous studies regarding cohort effects on health status. Our study particularly underlined that household wealth is an important variable that moderates the relationship between birth cohort and health, so that an increasing health gap was observed between the wealthiest and the poorest for the recent cohorts in both the US and English populations. The increasing inequality of successful ageing in favour of higher-income groups has been previously reported.^18,32 However, our study has additionally revealed that, whereas rich recent cohorts are healthier than previous birth cohorts, the poorest seem to be in a similar health status than their past cohort counterparts. The positive-gradient health income has been previously explained by the easier access that high-income people have to material resources (wealth and high-quality health-care services) and also to non-material factors (education, healthy habits and coping resources).³³

On the other hand, a similar decrease in health status across the lifespan was found in the ELSA and HRS studies. However, the life-course effect on health seemed to be slightly smaller for people with higher levels of education. This moderating effect is consistent with previous research highlighting the role of education as an important determinant of healthy ageing.^26,34 In that regard, it is likely that people with high levels of educational attainment have a better understanding of their own health and the importance of following a healthy lifestyle. Additionally, higher education enables access to qualified occupations that take place in healthier environments, and possibly better access and adherence to preventive and curative health-care interventions.

The present work has several strengths. This study relies on a robust harmonized metric of health to assess and compare trends in health status across the lifespan of birth cohorts from two large representative samples of the US and English populations. The metric is grounded on a common conceptual framework, comprising several indicators of functioning, disability and performance on diverse domains of functioning. In addition, the measurement approach used in this study allows comparison of the health status over time and across studies with distinctive features (e.g. study-specific items or different numbers of waves). Regarding the limitations of this study, it should be considered that the harmonized health metric is mostly based on self-reported items that could be influenced by response-style biases.³⁵ In addition, the absence of a main cohort effect on health might be affected by the range of the cohorts considered. In that regard, further research should be carried out to analyse health trends in more recent cohorts.

Conclusions

Health inequalities in terms of wealth have increased in recent cohorts, with the wealthiest participants presenting better health status in both the US and English populations. Thus, household wealth moderates birth-cohort effects on health status. On the other hand, higher levels of education are associated with a better health trajectory across the life course. The common metric used allowed us to show that the patterns of ageing and their moderators were similar in both the US and UK populations. The results of our study suggest that actions to promote health in the ageing population should consider the increasing inequality scenario. In that regard, the application of effective interventions should be not only guaranteed, but also facilitated to all members of society. Such interventions could comprise actions focused on promoting and assessing healthy lifestyles, a reasonable use of medications and encouraging social participation or training in cognitive health.³⁶

Funding

This work was supported by the Ageing Trajectories of Health: Longitudinal Opportunities and Synergies (ATHLOS) project. The ATHLOS project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 635316. The first seven ELSA waves have been funded jointly by UK government departments and the National Institute on Aging, in the USA. The HRS is funded by the National Institute on Aging (U01 AG009740) and the Social Security Administration and is performed at the Institute for Social Research, University of Michigan. Javier de la Fuente work is supported by the FPU pre-doctoral grant (FPU16/03276) from the Spanish Ministry of Education, Culture and Sport.