Article Text

Download PDFPDF

Trends and inequalities in late-life health and functioning in England
  1. Linda G Martin1,
  2. Robert F Schoeni2,
  3. Patricia M Andreski2,
  4. Carol Jagger3
  1. 1RAND Corporation, Arlington, Virginia, USA
  2. 2University of Michigan, Ann Arbor, Michigan, USA
  3. 3Newcastle University, Newcastle upon Tyne, UK
  1. Correspondence to Dr Linda G Martin, RAND Corporation, 1200 South Hayes Street, Arlington, VA 22202, USA; lmartin{at}


Background Recently, late-life disability rates have declined in several countries of the Organisation for Economic Co-operation, but no national-level trend analysis for England has been available. The authors provide such analysis, including measures both early and late in the disablement process, and the authors investigate the extent to which temporal trends are associated with population changes in socioeconomic position (SEP).

Methods The authors fit logistic models of trends in self-reports and nurse measures of 16 health indicators, based on cross-sectional data from those aged 65 years and older from the 1992 to 2007 Health Survey for England.

Results Overall, prevalence rates of limitations in seeing, hearing and usual activities declined (p<0.05); ever smoking, measured high blood pressure, high cholesterol, and high C reactive protein decreased (p<0.05); and the proportion with limitations in self-care activities remained stable. But obesity and limitations in walking 200 yards and climbing stairs increased (p<0.05). Increases over time in education and non-manual social class membership were associated with declines in smoking, C reactive protein and problems with usual activities. Had the changes in SEP not occurred, the increases in problems walking and climbing would have been greater. People with less education or of manual social classes experienced relatively worse trends for hearing, mobility functions and usual activities. The opposite was true for seeing.

Conclusions Recent trends in late-life health and functioning in England have been mixed. A better understanding of which specific activities pose challenges, how the environment in which activities are conducted influences functioning and the causes of relatively worse trends for some SEP groups is needed.

  • Ageing
  • disability
  • elderly
  • health status
  • socioeconomic
  • epidemiology
  • social inequalities
  • eastern Europe
  • dementia
  • longitudinal studies
  • diabetes
  • hypertension
  • social epidemiology

Statistics from


Life expectancy at age 65 in England and Wales increased from 14.0 to 19.2 years between 1960 and 2007, with over half of that improvement occurring since 1990.1 As people live longer, questions arise about whether extra years of life are spent in good or bad health and whether people can continue to function and live independently.2–4 The answers have important implications for demand for healthcare, quality of life and late-life participation in society, including in the workforce. The soundest evidence regarding questions of temporal trends in health and functioning is derived from analysis of individual-level data collected in three or more repeated nationally representative cross-sectional surveys or censuses (or waves of a longitudinal survey with limited attrition) that span 6 years or more and use the same survey design.5

On the basis of this standard, the declines in the prevalence of limitations in physical and sensory functioning (eg, climbing, seeing) and limitations in activities (eg, self-care and household tasks) among older Americans in recent decades are well documented.5–7 However, even as overall disability has declined, the prevalence of chronic conditions in the USA has increased, suggesting that there has been a change in the debilitating nature of particular conditions whose diagnosis and management may have improved over time.8 ,9 Moreover, there is evidence that the improvements in functioning have not occurred across all socioeconomic groups and that in some cases, there has actually been deterioration in functioning among the least advantaged.10 Although there are numerous studies of trends in late-life disability in other developed countries, most do not meet the above criteria for strong evidence.11 ,12 The exceptions include studies from Finland, Japan, the Netherlands and Sweden, all of which found improvements in functioning.13–16 We were not able to identify for England as a whole any comparable trends studies meeting our standard for sound evidence. However, there is a rich research literature documenting disparities in health by socioeconomic position (SEP) in England.17–22

In this paper, we use individual-level data from the annual Health Survey for England (HSE) from 1992 to 2007 to examine trends in functioning for the 65 and older English population, as well as trends in health behaviours and markers of physical dysregulation that may be associated with functioning. We also investigate if changes in educational and social class distributions are associated with the overall trends, and we estimate trends separately for education and social class groups.


Conceptual framework

In designing our analysis, we relied on a conceptual model of the disablement process that has commonly formed the basis for investigations of the health of older people.23 In this framework, a biochemical or physiological abnormality (eg, inflammation) may be manifested as an impairment (eg, joint stiffness), which in turn may lead to difficulty in cognitive, sensory or physical functioning (eg, stooping) and limitation in carrying out activities (eg, bathing in a bathtub). Activity disability does not simply reflect the health of the individual but rather represents a gap between an individual's capacity and the demands of a task being carried out in a particular environment. The WHO has developed an alternative framework, the International Classification of Functioning, Disability, and Health (ICF), which also characterises disability as such a gap.24 The model of the disablement process also has been extended in recent years to include indicators of health behaviours (eg, smoking) that may be engaged in prior to evidence of pathology.25


We use a time series of cross-sectional data from the 1992 to 2007 HSE, an annual survey of private households in England that is conducted through initial in-person interviews and subsequent nurse visits. The survey is based on a multistage stratified probability sampling design that is representative of the general household population.26 Weights to account for non-response to the initial in-person interview were not provided for waves prior to 2003. For consistency over our long study period, interview non-response weights for the later years are not used in our analyses. Analyses (not shown) find that the unweighted prevalence estimates of the demographic and socioeconomic indicators on which we focus and of the health outcomes from the interviews do not differ substantially from those estimated using interview non-response weights for 2003–2007. Sensitivity analyses (not shown) indicate that the basic trends are substantially the same if weights are used for the years that they are available.

We use two measures of SEP: (1) social class, which may change over the course of a lifetime, and (2) education, which reflects early-life circumstances and is one of the mechanisms of intergenerational transmission of social class. For both, combination of HSE response categories into broader groupings minimises any discontinuities resulting from changes in specific response categories in the HSE.

Table 1 presents for 1992 and 2007, as examples, a profile of the respondents aged 65 years and older, by sex, education and social class. During the 16-year study period, the population became older, better educated and more likely to be a member of a non-manual social class.

Table 1

Descriptive statistics and definitions of socioeconomic position indicators for people aged 65 years and older, Health Survey for England, 1992 and 2007 (per cent)

There were no missing responses for age and sex, but roughly 0.5% of education responses and 0.3% of social class responses were missing. The missing responses for these variables are included in the last category for each variable in the table.

Weights that adjust for non-response to the subsequent nurse visits also were not provided until 2003. However, preliminary analysis (not shown) indicated that rates of response to the nurse visit, given response to the initial interview, decreased significantly from 85% in 1992 to 74% in 2007, but women and individuals in households of semiskilled/unskilled manual workers were over time increasingly more likely to have a nurse visit. We fit logistic models of non-response to the nurse visit separately for each survey year as a function of age, sex, education and social class, using the categorisations shown in table 1. The parameter estimates from these models were used to calculate individual probabilities of responding to the nurse visit by year, and their inverses were used as nurse visit weights.

Proxy reports were not allowed in most survey years. For consistency over time, data obtained via proxy reports are not used in our analysis. The survey did not cover people living in institutions, but the likelihood of entering and staying in long-term residential care declined over the study period, so the estimated trends in rates of some of the health problems in the community may be slightly biased upward.27 ,28

Indicators of health and functioning

The health and functioning measures for which information is collected by the HSE vary by year. To estimate trends in a particular measure, we use data from as many years as possible from 1992 through 2007, always ensuring that each outcome is assessed consistently over the observations we use.

The 16 measures of health and functioning whose trends we analyse are listed in table 2. The indicators, all of which are defined dichotomously (yes or no), are organised into domains suggested by our conceptual framework and reflect both early and late stages of the disablement process: two indicators of health behaviour, seven of physical dysregulation (so-called biomarkers), five of sensory and physical functional limitations and two of activity limitations. The HSE also collected self-reports of chronic conditions, which are related to the early stages of the disablement process. Preliminary analyses (not shown) found upward trends in reports of most: heart conditions, stroke, high blood pressure, diabetes, cancer, nervous system disorders and musculoskeletal conditions. These upward trends may indicate increased health problems but also may reflect change in diagnostic thresholds, greater popular awareness of specific conditions, increased access to medical care, improved management of conditions and longer survival with conditions. Accordingly, for our analysis of indicators of the early stages of the disablement process, we decided to focus solely on health behaviours and biomarkers of pathology, the interpretation of whose trends is less ambiguous.

Table 2

Health outcomes, definitions, years data available, sample sizes and prevalences for the first and last years, Health Survey for England, 65+

For each indicator, we provide the source of information (initial interview or nurse visit), the definition, the years for which consistent information is available, the number of valid responses, the unadjusted prevalence in the first and last years of available data and the number and percentage of responses missing for the relevant years.

The proportion of missing responses varies from 0.1% to 26.2%. Preliminary analysis (not shown) indicated significant trends over time in item non-response for some outcomes. For 10 indicators—obesity, the seven biomarkers and the two activity limitations—there were differential trends in item non-response by some of the demographic and socioeconomic variables shown in table 1. Accordingly, for these 10 measures, we developed annual item non-response weights based on annual logistic models of item non-response for each outcome that controlled for age, sex, education and social class. We use these weights for all estimates and models of these outcomes. For the seven biomarkers, which are ascertained during the nurse visit, we multiply the item non-response weights by the nurse visit weights described previously to obtain the full weights.

Statistical analyses

For each separate measure of health and functioning, we fit a logistic model that combines all the data for the population aged 65 years and older for all the survey years for which the information is available and includes controls for sex, age group and survey year. Our focus is the coefficient on the survey year or so-called trend variable, where trend equals 0 in 1992, 1 in 1993, 2 in 1994 and so on. We examined plots of the data and concluded that the patterns over time are fairly well represented by such a linear trend for most outcomes. Taking the exponential of the coefficient on the trend variable (yielding the OR), subtracting one and then multiplying by 100 produces the estimated annual per cent change in the outcome, which we report in our tables. In the few cases in which change is great and non-linear (most notably, having high cholesterol and using cholesterol medication), repeatedly applying the multivariate-model-based estimated rate of change to the base unadjusted prevalence may result in unrealistic estimates of prevalence at the end of the period in comparison to the actual unadjusted prevalence. Our interest is less in the precise estimation of rates of change and more in calculating a summary indicator that reflects direction and statistical significance of change for a large number of model specifications.

To assess the extent to which changes in SEP are associated with these basic trends, we separately add education and social class (as defined in table 1) to our base models and report the resulting estimates of annual per cent change. ORs for the SEP variables themselves also allow us to examine the extent to which there are SEP gradients in outcomes.

Finally, we estimate annual per cent change in each of the 16 outcomes for the three education and the first four social class groups separately by fitting logistic models, controlling for trend, sex and age, on subsamples representing each of these SEP groups.

For all models in which weights are used as detailed earlier, we use robust SEs to assess significance of coefficients.


In recent decades, there have been statistically significant changes in all but two of the 16 outcomes, as shown in table 3 for the base model. The estimated annual decline controlling for age and sex is especially large for measured hypertension, high total cholesterol, high C reactive protein (CRP; an indicator of inflammation associated with arthritis, cerebrovascular disease and poor cognitive function) and problems seeing. There were relatively large estimated increases for obesity, use of medication for hypertension or high cholesterol and problems walking 200 yards.

Table 3

Estimates of annual per cent change in outcomes from logistic models, Health Survey for England, 65+*

For two outcomes (ever smoking and having measured hypertension or taking antihypertensives), adding either education or social class to the base model reduces the small negative trends to statistical insignificance, suggesting that changes in SEP could account for the trends initially found.

For high-risk CRP and problems with usual activity, the base model trends were significantly negative, but adding education reduces by 10% or more the estimate of annual per cent decline (eg, for high-risk CRP ((3.8−3.3)/3.8)×100=13.2%), implying that had educational attainment not increased, the improvements in these outcomes would not have been so great. When social class is added to the base model, only the estimate for usual activities is similarly affected.

For problems walking 200 yards and climbing, the base model trends were significantly positive and adding education increases the estimates of annual growth by 10% or more, implying that had educational attainment of the older population not increased, the worsening of these outcomes would have been greater. Adding social class to the base model similarly increases the estimate of growth for problems climbing.

We also fit models including both education and social class indicators (not shown), and the estimates were similar to those for the base plus education models.

Results from these base plus SEP models also allow us to examine the extent to which there are educational or social class gradients. The estimates are too numerous to present, but for most outcomes, more socioeconomic advantage was associated with lower odds of having a problem, although the gradients were not necessarily monotonic, especially for the four categories of social class. Exceptions in the case of education were high cholesterol or taking medication and problems speaking, for which the odds of having the problem for neither the low or medium group were significantly different from those of the high group. For high cholesterol, odds were lower for the low-education group than for the high group. For social class, there were reversed gradients for high cholesterol and high cholesterol or taking medication. There were no significant differences by social class for problems speaking.

As shown in table 4, for nine of the 16 outcomes, trends are in the same direction (up, down or flat) across all education groups. Of the seven exceptions, six show relatively worse trends for those with less education (ie, trends for the less educated are increasing when those for the more educated are flat or declining; alternatively, trends for the less educated are flat when those for the more educated are declining). In contrast, only those with the least education showed a significant decline in problems seeing.

Table 4

Estimates of annual per cent change in outcomes from logistic models by education group, Health Survey for England, 65+*

In table 5, for 10 of the 16 outcomes, trends are in the same direction across all social class groups. Of the six exceptions, four show relatively worse trends for those of manual social classes. The trends in problems seeing and climbing stairs do not show a clear gradient across social classes.

Table 5

Estimates of annual per cent change in outcomes from logistic models by social class group, Health Survey for England 65+*


In recent decades, there was a significant decline in the proportion of older people in England who have problems hearing, seeing and carrying out their usual activities and no change in the proportion with problems with self-care. The indicators analysed here are not precisely comparable in their wording to measures used in studies from other countries, but the downward trends in activity limitations and problems seeing are similar.5–7 13–16 In contrast, in England, there were increases in reports of inability to walk 200 yards or climb a flight of stairs. Also preliminary analysis of self-reports of diseases from the HSE found upward trends for most. US studies have found a similar pattern of declines in activity limitations accompanied by increases in reports of chronic conditions.8 ,9 Many factors may be associated with increased reports of such conditions, including increased access to medical care, improved management of conditions and longer survival with conditions. For example, the prevalence of coronary heart disease (CHD) among the older population in the UK increased between 1996 and 2005 as a result of decreasing CHD mortality, even as CHD incidence fell.29

More indicative of trends in underlying health status are changes in measured biomarkers of pathology. Downward trends in measured hypertension and high cholesterol and increases in use of medications for these conditions suggest that increased medical attention to these problems has been important. The enormous increase in use of lipid-lowering medication (probably overestimated by our multivariate models with a linear trend but undoubtedly very large) likely also influenced the decline in CRP since statins reduce inflammation as well as cholesterol.30 The general pattern of these biomarker trends is similar to that found for the older US population from 1999 to 2006.31 Our results also are consistent with those of studies of trends in quality of CHD management associated with the institution of pay-for-performance incentive schemes in the UK in April 2004.32 ,33

As in many other countries, trends in health behaviours in England were mixed: reports of ever smoking declined but obesity increased substantially. The consequences of these trends for late-life functioning may be expressed for decades to come.

Increases in education and shifts towards non-manual social classes among the older English population were associated with downward trends in ever smoking, hypertension or taking medication, high-risk CRP and problems with usual activity, but only for ever smoking and hypertension/medication did changes in SEP completely account for the trends. Were it not for the change in SEP, there would have been even larger upward trends in problems walking 200 yards and climbing. These findings are consistent with those of US trend studies showing increases in educational attainment strongly associated with improved health and functioning at older ages.34 ,35

For six of the 16 outcomes in the analysis by education group and four of the 16 outcomes in the analysis by social class, trends among those with less socioeconomic advantage were relatively worse than trends among those with more. Most troubling were the SEP disparities in trends in ever smoking and problems with hearing, walking, climbing and usual activities. These increasing disparities are consistent with trends in the USA.10 Only for problems seeing did those with less education in England show relatively greater improvement than those with more. This pattern is consistent with the recent large increase in cataract surgery and the association of such surgery with residence in areas with lower income, employment and health.36

Our study would have been enhanced by information on trends in specific types of activities that pose challenges to the older population, beyond the broad categories of self-care and usual activities. Also, it would have improved our understanding of how needs for assistance are changing and potentially informed the design of interventions if we had had information on trends in the use of environmental modifications and assistive technologies to maximise functioning. Our results paint a mixed picture of trends in late-life health and functioning in England, particularly among disadvantaged groups. Efforts to brighten this picture in the future are sorely needed.

What is already known on this subject

  • Disability rates among older people have generally declined in recent decades in Finland, Japan, the Netherlands, Sweden and the USA.

  • In the USA, the trends have been more favourable among advantaged than disadvantaged socioeconomic groups.

  • A comparable high-quality study of temporal trends in disability for England as a whole is not available.

What this study adds

  • In England, the trends are mixed—problems with walking and climbing have increased, problems with self-care have not changed but problems with seeing, hearing and carrying out usual activities have declined.

  • Those with the lowest education have relatively worse trends for over a third of the outcomes considered.



  • Competing interests None.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement The data from the Health Survey for England are publicly available.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.