Elsevier

Social Science & Medicine

Volume 57, Issue 9, November 2003, Pages 1621-1629
Social Science & Medicine

Does inequality in self-assessed health predict inequality in survival by income? Evidence from Swedish data

https://doi.org/10.1016/S0277-9536(02)00559-2Get rights and content

Abstract

This paper empirically addresses two questions using a large, individual-level Swedish data set which links mortality data to health survey data. The first question is whether there is an effect of an individual's self-assessed health (SAH) on his subsequent survival probability and if this effect differs by socioeconomic factors. Our results indicate that the effect of SAH on mortality risk declines with age—probably because of adjustment towards ‘milder’ overall health evaluations at higher ages—but does not seem to differ by indicators of socioeconomic status (SES) like income or education. This finding suggests that there is no systematic adjustment of SAH by SES and therefore that any measured income-related inequality in SAH is unlikely to be biased by reporting error. The second question is: how much of the income-related inequality in mortality can be explained by income-related inequality in SAH? Using a decomposition method, we find that inequality in SAH accounts for only about 10% of mortality inequality if interactions are not allowed for, but its contribution is increased to about 28% if account is taken of the reporting tendencies by age. In other words, omitting the interaction between age and SAH leads to a substantial underestimation of the partial contribution of SAH inequality by income. These results suggest that the often observed inequalities in SAH by income do have predictive power for the—less often observed—inequalities in survival by income.

Introduction

There is a large and fast growing literature comparing the existence of socioeconomic inequality in morbidity and mortality across a wide range of countries and populations.1 The literature on socioeconomic inequality in measures combining health status and length of life like quality adjusted life years (QALYs) is much thinner.2 One of the reasons for this is that mortality registrations often do not include indicators of socioeconomic status (SES) such as income, education or occupation. Another reason is probably that for reliable estimates of socioeconomic inequality in QALYs, large data sets with large spans of longitudinal follow up of both health status and mortality experience are required and the collection of such data sets is very costly and therefore relatively rarely undertaken (cf. Wolfson & Rowe, 2001).3 This explains why much of the literature on socioeconomic inequalities in morbidity and in mortality has developed relatively independently. Interestingly, the results obtained—at least in terms of cross-country comparisons—do not always match. For instance, Mackenbach et al. (1997) report socioeconomic inequalities in France and Finland to be relatively high in mortality, but rather low in morbidity, and the reverse appears to be true for Denmark. However, the results based on mortality and morbidity need not necessarily coincide because they partly reflect different dimensions of health, but are strongly related in view of the consistently reported high predictive power of self-assessed morbidity for subsequent survival chances. Idler and Benyamini (1997), for example, quote evidence from no less than 27 studies documenting that a respondent's global health rating, based on the simple question ‘How is your health in general?’ with response categories ranging from ‘very good’ or ‘excellent’ to ‘poor’ or ‘very poor’ is an independent and powerful predictor of subsequent individual mortality. Of course, self-assessed health (SAH) will include the evaluation of non-life threatening conditions on the one hand, will not reflect unknown future mortality risks (like, e.g. accidents) on the other hand, but the relationship nevertheless appears to be sufficiently strong to persist even after controlling for other mortality risk factors. Moreover, recent findings for Sweden (Burström & Fredlund, 2001) suggest that the predictive power of SAH for mortality is unaffected by social class: SAH predicts survival equally well for high and low social classes.

This suggests that there is little reason to expect that the comparability of SAH across populations groups is problematic because of a problem which has been termed ‘state-dependent reporting bias’ (e.g. Kerkhofs & Lindeboom, 1995), ‘scale of reference bias’ (e.g. Groot, 2000) or response category cut-point shift (e.g. Sadana, Mathers, Lopez, Murray, & Iburg, 2000; Murray, Tandon, Salomon, & Mathers, 2001). Basically, it occurs if subgroups of a population use systematically different threshold levels for their SAH evaluation, despite having the same level of ‘true’ health. These differences may be influenced by, among other things, age, sex, education, language and personal experience of illness. Basically, it means that different groups appear to ‘speak different languages’ when they are responding to the same question. Evidence of significant cut-point shift by income or education would have important implications for the measurement and explanation of, for example, inequalities in SAH by income or education. If, given the same level of true health, the assessment of reported health differs systematically by SES, this will bias the measured degree of socioeconomic inequality in health. If, for instance, those with lower SES are more inclined to report poor health at the same true health, e.g. because of general feelings of dissatisfaction, then true socioeconomic health inequality will be overstated by using SAH, If the reverse is true, and lower SES persons are less inclined to report poor health given true health, e.g. because of lower health expectations, then true SES-related health inequality will be underestimated.

The problem with testing for cut-point shift is that true health is not directly observable and therefore the benchmark is difficult to measure in a valid and reliable way. Typically, researchers have relied on other indicators of SAH (Kerkhofs & Lindeboom, 1995) or chronic conditions (Groot, 2000) to test for cut-point shift. While the test of an income-SAH interaction effect in survival prediction cannot be seen as a definitive test of such cut-point shifting because morbidity and mortality are related but inherently different dimensions of health status, it seems nevertheless worthwhile to explore whether their covariance is affected by such an interaction effect between SAH and SES indicators in a mortality prediction model. The longitudinal mortality follow up of the Swedish Level of Living Conditions Surveys allows for such a test of interaction.

The above observations lead to the following two research questions which we address in this paper: (1) given that mortality is obviously a very objective indicator of health risk, is there a significant interaction between SAH and other mortality predictors, in particular income and education? In the absence of such an interaction, the evidence on income-related inequalities in SAH carries more weight, since it indicates that income-related inequalities in mortality, even if unobserved directly, are also likely to follow similar patterns. (2) If SAH predicts mortality, to what extent can observed socioeconomic inequalities in current SAH (e.g. by income) predict inequalities in future survival?

Section snippets

Modeling mortality risk

Because the observation of the duration until death is censored by the length of the follow-up period (cf. data description), we analyse the effect of SAH and other covariates on mortality risk during follow up using a Cox's (1972) proportional hazard model. This is a flexible semi-parametric model which makes no distributional assumptions about the functional form of the baseline hazard, as is required for most other hazard functions (cf. Greene, 1993). Besides depending on time, the hazard

Data and variable definitions

The data used in this paper are taken from Statistics Sweden's Survey of Living Conditions (the ULF survey), which was linked to all-cause mortality data from the National Causes of Death Statistics and to income data taken from the National Income Tax Statistics. Since 1975, annually a random sample of adults aged 16–84 is interviewed about living conditions. We have used pooled data from the annual surveys conducted in 1980–1986 for adults aged 20–84. After deletion of missing values, the

Effect of SAH on mortality risk

The Cox regression results of three model specifications with and without interaction terms with the SAH dummies are presented in Table 2. We can see that—as expected—being male, being older, and having functional limitations or high blood pressure increases mortality risk, whereas higher (self-assessed) health, income and education all show protective effects on survival of Swedish adults. When SAH interactions are included, some of these coefficients become non-significant, in particular for

Conclusion

In this paper we have sought to provide an empirical answer to two questions using linked morbidity and mortality data for Sweden. The first question was: what is the effect of an individual's SAH reporting on his subsequent survival probability and does this effect differ by other characteristics, in particular income or education? We find that SAH does have a substantial predictive power for survival and that its effect differs by age, sex and hypertensive status, but not by income, education

Acknowledgements

This paper derives from the project ‘Economic determinants of the distribution of health and health care in Europe’ (known as the ECuity II Project), which is funded in part by the European Community's Biomed II programme (contract BMH4-CT98-3352). We are grateful to the EC for financial support, and to Andrew Jones, Emmanuela Gakidou, Brian Nolan and other participants at the Rome workshop of the ECuity II Project for useful suggestions and comments on an earlier version of the paper.

References (26)

  • U.-G Gerdtham et al.

    Do life-saving regulations save lives?

    Journal of Risk and Uncertainty

    (2002)
  • Gerdtham, U.-G., & Johanneson, M. (2003). Absolute income, relative income, income inequality and mortality, Journal of...
  • W.H Greene

    Econometric analysis

    (1993)
  • Cited by (195)

    • The evolution of health over the life cycle

      2022, Review of Economic Dynamics
    • Health, wealth, and informality over the life cycle

      2021, Journal of Economic Dynamics and Control
    View all citing articles on Scopus
    View full text