Findings

View the most recent version.

The dramatic increase in life expectancy in Canada and other economically developed nations during the last century stands as testimony to the success of improvements in public health and advances in medical care.¹   But despite these gains in longevity, inequalities in health outcomes across different subpopulations are still pervasive in Canada and other industrialized countries.^2-6

Irrespective of how socio-economic status is defined and measured (by income, educational attainment or occupational prestige), mortality rates show a gradual but systematic increase at successively lower levels of the socio-economic hierarchy.  Nonetheless, time-series analyses for Canada suggest that, in absolute terms, socio-economic differences in mortality have been diminishing over recent decades, with the specific patterns dependent on sex and the cause of death considered.^7,8

Mortality, however, is only one aspect of population health.  It is now widely recognized that information on morbidity (including disability and reduced health-related quality of life) is crucial for monitoring health trends, setting priorities, and conducting cost-effectiveness analysis of population-level interventions.  Because morbidity varies by socio-economic status,^2,3,9-12 a complete assessment of health disparities requires summary measures reflecting the effects of differences in both mortality and morbidity across socio-economic categories.^13-15 Accordingly, a variety of summary measures, which integrate information on mortality and morbidity into a single numeric index, have been developed.^16-18

A number of studies in Canada have calculated summary measures in order to examine broad national patterns of mortality and morbidity,¹⁹ regional variations in population health,²⁰ the population health impact of specific diseases and risk factors,^21-25 and whether population health has been improving over time.^7,26,27 Few investigations in the Canadian context have focused on socio-economic differences in summary measures of population health, and those that have done so have relied primarily on macro-level indicators (such as neighbourhood income) rather than micro-level indicators (such as household income),^28-30 thereby almost certainly attenuating the association between socio-economic status and health.  One Canadian study used micro-level indicators for analyses of socio-economic disparities in health, but only within a single province.³¹

This article examines socio-economic differences in health for a nationally representative sample of the adult population of Canada, using a summary measure known as health-adjusted life expectancy.³² By weighting years of life according to their quality, health-adjusted life expectancy converts the conventional, purely mortality-driven life expectancy measure into expected equivalent years of full health.  This study combines mortality data from the 1991-2001 Canadian census mortality follow-up study³³ with information about health-related quality of life from the 2000/2001 Canadian Community Health Survey to estimate health-adjusted life expectancy for different income groups. 

Data and methods

Data source

The 1991-2001 Canadian census mortality follow-up study

Death data were obtained from the 1991-2001 Canadian census mortality follow-up study,conducted by Statistics Canada in collaboration with the Canadian Population Health Initiative.  With probabilistic linkage techniques, a 15% sample (n = 2,735,152) of the non-institutional population aged 25 or older who completed the 1991 census long-form questionnaire (the cohort) was matched to 11 years of death records (June 4, 1991 to December 31, 2001) from the Canadian Mortality Data Base.  The linked file contains information on various demographic characteristics, socio-economic status, activity limitations, disability, and cause and date of death.  Additional methodological details on the construction and contents of the linked file have previously been reported.³³

The Canadian Community Health Survey

Data on health-related quality of life were derived from the 2000/2001 Canadian Community Health Survey (cycle 1.1).  The Canadian Community Health Survey is an ongoing cross-sectional survey that collects information on health status, health determinants, and health care utilization.  It is representative of the household population aged 12 or older in all provinces and territories, excluding residents of institutions, of Indian Reserves and Crown Lands, of Canadian Forces bases (military and civilian residents) and of certain remote regions, and full-time members of the Canadian Forces.  The response rate for cycle 1.1 was 84.7%.  For consistency with the Canadian census mortality follow-up  study, the present  investigation was restricted to respondents aged 25 or older (n = 106,283).  More detailed descriptions of the Canadian Community Health Survey design, sample, and interview procedures can be found in other published reports and on the Statistics Canada website.^34,35  

Definitions

Income deciles

For the census mortality linked file, deciles (tenths) of population ranked by income adequacy were created.  First, for each economic family or unattached individual in the non-institutional census population of all ages, total pre-tax, post-transfer income from all sources was pooled across all economic family members and divided by the weighted family size (or “equivalent person unit” scale).  An economic family is a grouping of two or more individuals living in the same household or dwelling unit who are related by blood, marriage or adoption.³⁶ Unattached individuals are economic families of size 1. Under the weighting system, the first person received a weight of 1.0, the second person, 0.4, and all subsequent persons, 0.3.  Next, adjusted family income was partitioned into deciles, using the same cut-points for the entire country.  For the Canadian Community Health Survey, income deciles were constructed in the same manner, except that total household income was used rather than total economic family income. 

Because of the high proportion of missing income data on the Canadian Community Health Survey (21.5%, compared with  1.4% for the census), the extent of possible bias was addressed by performing two sets of analyses.  The first set completely excluded cases with missing income deciles, and the second set was conducted after the missing decile information was filled in using hot deck imputation.³⁷ In the latter procedure, groups of Canadian Community Health Survey respondents matched on age, sex and educational attainment were created.  For the cases with missing income deciles in each of these groups, a decile value was randomly selected from the pool of complete cases (the “donor” cases). Because the hot deck imputation did not appreciably alter the results (data not shown), all of the analyses presented here are based on cases with originally complete income data from the Canadian Community Health Survey.

Analytical techniques

Health-adjusted life expectancy

Health-adjusted life expectancy was estimated using a modified version of the Sullivan³⁸ method.   Chiang’s³⁹ method was used to calculate abridged (five-year age group) period life tables, corresponding standard errors, and 95% confidence intervals for each population subgroup of interest (by age, sex and income decile).  Before computing the life tables, it was necessary to:  (1) transform baseline age to age at the beginning of each year of follow-up; (2) calculate deaths and person-years at risk separately for each year (or partial year) of follow-up; and (3) pool deaths and person-years at risk at the beginning of each year of follow-up.³³

Health-related quality of life weights were derived from the Health Utilities Index Mark 3 instrument (HUI3),^40,41 administered to all respondents to the 2000/2001 Canadian Community Health Survey.  The health utilities index measures eight basic domains or attributes of health status:  vision, hearing, speech, ambulation, dexterity, emotion, cognition, and pain.  Each attribute has five or six levels, ranging from normal to severely limited functioning.  For example, the levels of ambulation range from 1 (“able to walk around the neighbourhood without difficulty, and without walking equipment”) to 6 (“unable to walk at all”).  Respondents were asked a standard set of questions on usual functional ability or capacity, which can be mapped to the levels of the eight attributes.  A respondent’s health status was thus represented by an eight-element vector listing each of the attribute levels, which were then summarized by a weighted scoring function into a single value representing overall health-related quality of life.  The global score has a theoretical range of -0.36 (the worst possible health state) to 1.00 (the best possible health state), where 0.00 represents a health state equivalent to death.

Mean global health utilities index scores were computed from the Canadian Community Health Survey for each population subgroup (by age, sex and income decile).  Ten-year rather than five-year age groups were used with the Canadian Community Health Survey data to ensure stability of the mean estimates.  Therefore, for computing health-adjusted life expectancy at ten-year intervals (for ages 25 through 75), the appropriate life table elements for five-year age groups from the census-mortality linked data were collapsed (for example, combining ages 25 to 29 with ages 30 to 34).  Survey sampling weights were applied to correct the point estimates of the health utilities index means for unequal selection probabilities, post-stratification adjustments, and unit non-response.  The Rao-Wu bootstrap technique was used to adjust the standard errors and 95% confidence intervals for the effects of stratification and clustering.^42,43

After the life table values and corresponding mean global health utilities index scores were assembled, health-adjusted life expectancy was computed using the following formula, separately by sex and within each income decile:

where:
HALE is health-adjusted life expectancy;
x is the exact age for which HALE is estimated (25 to 75, by ten-year intervals);
i is an indicator representing the lower limit (x) of the age interval (x, x + a);
L_i is the number of life years lived by the age group (x, x + a);
l_x   is the number of survivors at age x;
H_i  is the mean global health utilities index score for the age group (x, x + a), with H_i = 1.00 indicating full health; and
w is the total number of age groups in the life table.
Thus, the higher the average level of health-related quality of life for a given age group on a scale with an upper limit of 1.00 (full health), the closer health-adjusted life expectancy will be to conventional life expectancy.

The variance of health-adjusted life expectancy was estimated by adapting methods proposed by Mathers⁴⁴ (see  Health Canada, 2004²⁸ ), which take into account stochastic fluctuations in the observed death rates and the mean global health utilities index scores.  All analyses were performed using a combination of SAS Version 9.1 for Windows (SAS Institute, Cary North Carolina) and Microsoft Excel 2002 for Windows.

Health-adjusted life expectancy by educational attainment

To verify the robustness of the association between health-adjusted life expectancy and socio-economic status, differences were also estimated for varying levels of education.  For both the census mortality linked file and the Canadian Community Health Survey, self-reported information on highest level of education was grouped into four categories:  less than secondary graduation, secondary graduation (or trades certificate), postsecondary certificate or diploma (short of a university bachelor’s degree), and university degree (bachelor’s or higher).

Results

Sample characteristics

For the most part, the distribution by age, sex and income decile was similar in the census mortality follow-up study and the Canadian Community Health Survey (Table 1).  However, the Canadian Community Health Survey had considerably more missing data on the income decile (21.5%) than did the census mortality data (1.4%).  As well, 1,565 respondents (about 1.5% of those aged 25 or older) to the Canadian Community Health Survey were missing data on the health utilities index, and were excluded from the analyses.

Table 1
Sample sizes for 1991-2001 Canadian census mortality follow-up study and corresponding weighted population estimates for 2000/2001 Canadian Community Health Survey

Disparities in mortality

For both men and women, a clear socio-economic gradient emerged for remaining life expectancy at age 25, as well as for the percentage expected to survive to age 75 (Table 2).  From the lowest to the highest income deciles, a gradual yet steady increase in remaining life expectancy at age 25 was evident.  For men, the difference in life expectancy between the extreme deciles was 7.4 years, and for women, 4.5 years.  The proportion expected to survive to age 75 also climbed steadily, with 51% of men in the lowest income decile expected to reach age 75, compared with 75% of those in the highest income decile.  The corresponding figures for women were 69% versus 84%, a smaller inter-decile gap.

Table 2
Remaining life expectancy at age 25 and percent expected to survive to age 75, by income decile and sex, Canada, 1991-2001

Disparities in health-related quality of life

Average health-related quality of life scores also exhibited appreciable socio-economic gradients. Table 3 shows mean health utilities index scores by income adequacy decile for people aged 25 to 34.  For both sexes, mean scores tended to rise with income.

Table 3
Mean health utilities index scores for Canadian Community Health Survey respondents aged 25 to 34, by income decile and sex, 2000/2001

It is useful to evaluate these disparities in terms of a criterion called the minimal clinically important difference, whichis the smallest difference that reflects a meaningful impact on (or change in) health-related quality of life.⁴⁵ For global scores on the health utilities index—in particular, the Mark 3 version used here—a minimal clinically important difference of 0.03 has been established through a combination of empirical research and expert opinion.⁴⁶ In the present study, the difference in mean global health utilities index scores between men in the highest income group (decile 10) and those in the lowest (decile 1) was 0.087, almost three times the minimal clinically important difference.  The corresponding difference for women was 0.078.

Disparities in health-adjusted life expectancy

Remaining health-adjusted life expectancy at age 25 is shown in Table 4, and disparities across the income deciles in both conventional life expectancy and health-adjusted life expectancy are displayed in Figures 1 and 2.  As with conventional life expectancy, for both sexes, the results for health-adjusted life expectancy demonstrated a nearly linear gradient with respect to income. Disparities in health-adjusted life expectancy between the highest and lowest deciles were 14.1 years for men and 9.5 years for women, whereas the corresponding disparities in conventional life expectancy were only 7.4 and 4.5 years, respectively.

Table 4
Remaining health-adjusted life expectancy (years) at age 25, by income decile and sex, Canada, 1991-2001

Figure 1
Remaining life expectancy and health-adjusted life expectancy at age 25, by income decile, men, Canada, 1991-2001

Figure 2
Remaining life expectancy and health-adjusted life expectancy at age 25, by income decile, women, Canada, 1991-2001

Health disparities across education levels

The results of the supplemental analyses of health disparities across educational attainment categories are shown in Appendix Tables A, B and C.  Without exception, all of the health indicators (life expectancy, survival probabilities, mean health utilities index scores, and health-adjusted life expectancy) showed steady improvement with increasing levels of education.

Table A
Remaining life expectancy at age 25 and percent expected to survive to age 75, by highest level of educational attainment and sex, Canada, 1991-2001

Table B
Mean health utilities index scores for Canadian Community Health Survey respondents aged 25 to 34, by highest level of educational attainment and sex, 2000/2001

Table C
Remaining health-adjusted life expectancy at age 25, by highest level of educational attainment and sex, Canada, 1991-2001

Discussion

This study indicates strong and consistent evidence of socio-economic disparities in health. Income-related disparities in health-adjusted life expectancy were found to be considerably larger than those for the conventional life expectancy indicator. For both men and women at age 25, the difference in remaining health-adjusted life expectancy between the highest and lowest income groups was much larger than the corresponding difference in overall life expectancy: 6.8 years more for men, and 5.0 years more for women. These results highlight the generally worse morbidity that lower-income groups experience in addition to their higher mortality.  The findings are consistent with those of other Canadian investigations using area-level^28-30 and individual-level³¹ indicators of health and socio-economic status, and with numerous studies of socio-economic differences in health expectancy in the United States and Europe.^13-15  This information suggests that the remediation of both fatal and non-fatal health outcomes may have the potential to substantially reduce health disparities related to socio-economic circumstances in Canada.

Some additional perspective on the magnitude of socio-economic disparities in health-adjusted life expectancy can be provided through comparisons with the impact of specific health conditions.  For example, Manuel and colleagues²³ found that all cancers, which represented the greatest burden of disease in the population, reduced health-adjusted life expectancy at birth by 2.8 years for men and by 2.5 years for women. By contrast, in this analysis, which examines health-adjusted life expectancy at age 25, the difference between the highest income decile and the overall average was estimated at 5.8 years for men and 3.1 years for women. For men, this was around twice the impact of all cancers combined, while for women, it was about the same as the impact for all cancers combined. Because of differences in methodology and data sources, these findings and those of Manuel and colleagues²³ are not completely comparable.  But methodological differences alone are unlikely to account for such large discrepancies between the impact of a major disease which has been the object of tremendous research and clinical effort (such as the “war on cancer” in the United States, declared in 1971),⁴⁷ and the impact of socio-economic factors, which remain, by comparison, relatively poorly understood.  Further research examining pathways by which socio-economic status affects health is warranted to understand how such large differences are generated and to point to potential areas for intervention.

Limitations

The results of this study pertain to the non-institutionalized population aged 25 or older (except that the mortality data included people who were institutionalized after the 1991 Census). Future work should investigate ways of including the morbidity and mortality experience of institutional residents—the most disabled segment of the population—as well as that of people younger than 25, to get a more comprehensive picture of morbidity and mortality in relation to socio-economic status.

The Canadian Community Health Survey data may be affected by a certain element of self-selection, since not everyone contacted agreed to participate in the survey.  The health utilities index data from the survey may be subject to self-report error, and may not perfectly reflect the health-related quality of life of the 1991-2001 Canadian census mortality cohort.  However, a more representative and temporally consistent source of data on health-related quality of life was not available.

Data on income were missing for a large share of the Canadian Community Health Survey sample. However, imputing the missing income deciles and then recalculating the estimates with complete decile information for the survey component did not substantially change the results (data not shown).   

The person-equivalence scale used to compute adjusted household income was only one of many possible approaches.  For example, other Canadian studies have used the ratio of total income to the Statistics Canada low income cut-off established for the applicable family and community size group.³³ In addition, a variety of alternative indicators of socio-economic status (such as education or occupational prestige) could be used to examine gradients in health.  Nonetheless, it is well known that the socio-economic gradient in health is robust to the choice of measurement method.  Recomputation of the same set of estimates (for life expectancy, survival probabilities, mean health utilities index score, and health-adjusted life expectancy) for another indicator of socio-economic status (educational attainment) yielded essentially the same pattern of results (Appendix Tables A, B and C).

While the measurement properties of the health utilities index have been supported in numerous studies,^40,46,48 other reliable and valid measures of health-related quality of life can be used in the computation of health-adjusted life expectancy, such as the EuroQol Five Dimensions Index⁴⁹ and the Short-Form Six Dimensions Index.⁵⁰  However, the health utilities index was the only such measure available from the Canadian Community Health Survey, which, in turn, was the only nationally representative source of health-related quality of life data with a sample of sufficient size.  Future studies in the Canadian context could use other instruments if they are included on national surveys.  Alternatively, weights could be assigned to particular levels of disability based on expert opinion.^29,30,51   Comparison of the current results and those obtained with such alternative methods would show the sensitivity of health-adjusted life expectancy and of the corresponding socio-economic gradient to changes in the way health-related quality of life is measured.

The 1991-2001 Canadian census mortality follow-up study dataset contained only baseline (1991) information on family income, and the Canadian Community Health Survey provided only cross-sectional estimates of morbidity.  Since these characteristics are expected to change over time, it would have been preferable to have income and morbidity information for each year of follow-up, thus providing the input for a more dynamic, multistate life table or microsimulation-based approach to computing health-adjusted life expectancy.⁵²

Mathers’ method⁴⁴ for calculating the variance of health-adjusted life expectancy assumes that the rates of mortality and morbidity are uncorrelated, but violations of this assumption could result in underestimation of the variance using his method.⁵³   Because individual risk functions for the two variables were not known, it was not possible to estimate the correlation between morbidity and mortality from the present data.  Nevertheless, the bias introduced in the variance estimate may have been reduced because all the calculations were done by age, sex and income decile, which themselves explain a substantial amount of the variance in health-related quality of life and mortality.

Although life expectancy, survival probabilities, the health utilities index, and health-adjusted life expectancy were each found to be associated with income in the current study, causality cannot be inferred.  In a study of the effects of poverty and material hardship on mortality in Finland, Martikainen and colleages⁵⁴ concluded that a large part of the association observed between income and mortality was not due to a direct causal impact, but rather to the mutual dependence of mortality and income on other background factors such as educational attainment and occupational prestige.  Identifying the true causal mechanisms underlying the socio-economic gradient in health is important, since the effectiveness of interventions may rest on correctly understanding the forces involved. The real drivers of health inequalities could be differences in education and occupation.  It has also been suggested that poorer health status may be an antecedent to both lower incomes and earlier mortality (“reverse causality”).⁵⁵   Although health status affects socio-economic status to some degree (for example, chronic illness can result in foregone income and reduced opportunities), several prospective studies have demonstrated that forward causality (socio-economic status influencing health) is more important in generating the observed socio-economic gradient in health.^56-59 Nevertheless, the current descriptive analysis makes no claims about the causal mechanisms underlying the observed health disparities.

Conclusion

The 1991-2001 census mortality follow-up study has opened a wide range of opportunities for analyzing socio-economic correlates of health in Canada.  The health disparities estimates presented here are novel in that they use Canadian microdata on socio-economic status, mortality and morbidity for a large, representative sample of the household population aged 25 or older.  The results provide comprehensive, robust, and policy-relevant information on the distribution of health outcomes by socio-economic status.  This study may be useful to identify areas in which interventions for reducing health disparities would yield the greatest benefits.

Acknowledgements

The authors gratefully acknowledge the input of Geoff Rowe on the study methodology, as well as comments by Michelle Rotermann on the exposition. All errors are the responsibility of the authors.

Funding sources

Major funding for the creation of the Canadian census mortality follow-up study was provided by the Canadian Population Health Initiative, part of the Canadian Institute for Health Information.