Health Reports
Trends in mortality inequalities among the adult household population

by Emma Marshall-Catlin, Tracey Bushnik and Michael Tjepkema

DOI: https://www.doi.org/10.25318/82-003-x201901200002-eng

Health inequalities can be defined as avoidable differences in health status between population groups.^{Note 1} These differences have been attributed to personal, social, economic and environmental factors. For example, income has been tied to health outcomes in that people with low income may have less access to health services and higher-quality material resources (such as shelter and food), and the knowledge and skills obtained through formal education can influence the understanding of health messaging and the ability to effectively access health services.^{Note 2} The routine measurement of population health status indicators like mortality is important to assess progress in the reduction of inequalities and is a priority in Canada.^{Note 3} Though mortality rates overall have been decreasing over time,^{Note 4} inequalities remain. In the past, people in higher socioeconomic groups have experienced lower mortality rates than those with less income or lower educational attainment.^{Note 5} Despite increasing attention and actions taken,^{Note 3} regional studies suggest that mortality inequalities across socioeconomic groups may be increasing in Canada, especially for women.^{Note 6Note 7Note 8}

Many previous estimates of health inequalities have used area-based measures of socioeconomic indicators.^{Note 9Note 10} However, area-based measures detect only a portion of social inequalities in health, and individual-level data are necessary to measure the association more precisely.^{Note 11} The Canadian Census Health Environment Cohorts (CanCHECs) now make this possible. The CanCHECs combine mortality data with sociodemographic information at an individual level for millions of records for each cohort year available (1991 to 2011), allowing for a robust analysis of the relationship between socioeconomic status and mortality over time.

This study estimates mortality rates in 1991, 1996, 2001, 2006 and 2011 according to individual-level measures of income and education. The main objectives of the study are to quantify mortality inequalities across these measures among the adult household population in Canada, and to examine whether these inequalities have changed over time.

Data and methods

The CanCHECs are population-based linked datasets that follow the non-institutional (household) population at the time of the census for different health outcomes such as mortality, cancer and hospitalizations. The 2006 and 2011 CanCHECs were constructed using Statistics Canada’s Social Data Linkage Environment (SDLE).^{Note 12} The SDLE helps create linked population data files for social analysis through linkage to the Derived Record Depository (DRD), a dynamic relational database containing only basic personal identifiers. Survey and administrative data are linked to the DRD using generalized record linkage software that supports deterministic and probabilistic linkage. The 1991, 1996 and 2001 CanCHECs were constructed before the creation of the DRD using the same standard generalized record linkage software.^{Note 13Note 14Note 15} In recent months, the 1991, 1996 and 2001 CanCHECs were deterministically linked to the DRD using social insurance numbers in order to update and attach different health outcomes (mortality, cancer and hospitalizations) in a manner consistent with the 2006 and 2011 CanCHECs.

For the 1991, 1996, 2001 and 2006 censuses, a mandatory long-form census questionnaire was administered to the non-institutional population (about one in five households) including those in collectives (such as rooming houses and hotels) who usually live in Canada on Census Day. In 2011, the long-form census was replaced with the voluntary 2011 National Household Survey, which was administered to the non-institutional population living in private dwellings (about one in three households, excluding non-private dwellings such as rooming houses and hotels).

A cohort weight that adjusts for the census sampling design and non-linkage and bootstrap weights that allow for appropriate variance estimation were created for each CanCHEC. Applying the weights helps ensure that estimates are representative of the household population for the corresponding census year.

This study includes CanCHEC members who were aged 25 or older on Census Day and who did not live in a collective dwelling at the time of the census.

Mortality

Mortality data were based on the Canadian Vital Statistics Death Database, which was linked to the DRD. The linkage rate of deaths to the DRD exceeded 99% between 1991 and 2016.

Definitions

Income quintiles were derived by summing total pre-tax income from all sources for all economic family members or unattached individuals for the year prior to the census, and then calculating the ratio of this total income to the Statistics Canada low income cut-off for the applicable family size, community size group and year.^{Note 16} Weighted quintiles were derived based on this ratio for each census metropolitan area, census agglomeration or provincial residual for each cohort.

Educational attainment was the highest level of education completed as of Census Day and was grouped into four different categories: less than secondary graduation, secondary graduation (or trades certificate), postsecondary certificate or diploma (excluding university degree), and university degree or equivalent.

Statistical analysis

The mortality follow-up period for each CanCHEC was restricted to five years after Census Day. A five-year follow-up period was chosen to ensure enough deaths to provide robust estimates of mortality and to minimize mortality overlap in follow-up periods across the different CanCHECs.

The number of deaths by sex, age group and population group during the five-year follow-up period was calculated for each CanCHEC. In addition to the number of deaths, the number of people living during the follow-up periods (i.e., the at-risk population) by sex, age and population group was also calculated. Person-years at risk were calculated based on the date of the census and the date of death or the end of the follow-up period, as applicable, and then assigned to the appropriate age group. For example, since most individuals did not remain the same age for an entire follow-up year, someone who turned 65 exactly halfway through would contribute 0.5 person-years at risk to the 60-to-64 age group and 0.5 person-years at risk to the 65-to-69 age group for that year of follow-up.

Sex-specific all-cause age-standardized mortality rates (ASMRs) per 100,000 person-years at risk were calculated for each education and income category. The ASMRs were age standardized based on the 2011 Canadian standard population^{Note 17} in five-year age groups. The age groups started at age 25 and ended at 85 or older, for a total of 13 age groups. The ASMR calculation was based on the number of deaths and person-years at risk in each age group. Trends over time in ASMRs were analyzed by calculating the average percentage change (APC) over the five time points.

Absolute rate differences (RDs) and relative rate ratios (RRs) comparing the lowest (first income quintile; less than secondary graduation) with the highest (fifth income quintile; university degree or equivalent) were calculated based on the ASMRs, with confidence limits computed using the ASMR standard errors.

All mortality estimates were weighted using the cohort weight and calculated using PROC RATIO in SUDAAN in SAS Enterprise Guide 5.1. Standard errors were estimated using the 500 bootstrap weights provided with each CanCHEC. APC over time was estimated using Joinpoint version 4.6.0.0. An APC was calculated for each income and education category by fitting a linear regression model assuming a constant rate of change in the logarithm of the weighted age-standardized rates from one cohort year to the next. The models incorporated the appropriate standard errors, the tests of significance used a Monte Carlo permutation method, and the estimated slope from each model was then transformed back to represent an APC.

Results

Table 1 presents the distribution of the cohort population of men and women according to income quintile and educational attainment. The proportion of the adult household population that did not graduate from secondary school decreased for both men and women between 1991 and 2011. The weighted proportion of women over the age of 25 with a university degree increased from 11% in 1991 to 24% in 2011. For men, it increased from 15% to 23%.

Mortality inequality by income quintile

ASMRs were highest among those in the lowest income quintile, for both men and women, across all time points (Table 2). Mortality rates decreased significantly for men in all income quintiles between 1991 and 2011 (the average change from one cohort year to the next ranged from -1.9 for Q1 to -2.7 for Q5, p < = 0.05), and for women in all income quintiles except the lowest quintile (the average change ranged from -1.1 for Q2 to -2.0 for Q5, p < = 0.05) (Figure 1). A gradient in the average percentage decrease over time was observed, with a larger average decline in ASMRs for men and women in the higher income quintiles compared with those in the lower income quintiles.

The difference in ASMRs in absolute terms between the highest and lowest income quintiles decreased significantly over time for men, from a rate difference (RD) of 827 deaths (95% confidence interval [CI] 719; 935) per 100,000 person-years at risk in 1991 to an RD of 672 deaths (95% CI 636; 707) in 2011 (Table 3). For women, the trend was less clear given the wide confidence intervals around the estimate for 1991, though the results suggest an increase in absolute terms between 1996 and 2011. For both women and men, the difference in mortality rates in relative terms increased significantly between 1991 and 2011 (from a rate ratio [RR] of 1.4; 95% CI 1.3, 1.5 to an RR of 1.7; 95% CI 1.6, 1.7 for women, and from an RR of 1.7; 95% CI 1.6, 1.8 to an RR of 1.9; 95% CI 1.9, 2.0 for men) (Table 3).

Mortality inequality by educational attainment category

Women and men who had not graduated from secondary school had the highest ASMRs compared with those with higher educational attainment, and this difference persisted over time (Table 2). ASMRs for men at all education levels decreased over time (the average change between each time point ranged from -1.5 for no HS [less than secondary] to -2.4 for university, p < = 0.05), whereas only women with a postsecondary diploma or higher had a statistically significant average decrease in their ASMRs between 1991 and 2011 (Figure 2).

The absolute difference in ASMRs between those who had not graduated from secondary school and those with a university degree or equivalent increased over time for women, from an RD of 261 deaths (95% CI 133; 389) per 100,000 person-years at risk in 1991 to an RD of 379 deaths (95% CI 354; 404) in 2011 (Table 3). For men, there was little change over time in absolute terms, but there was a significant increase in the relative difference, from an RR of 1.5 (95% CI 1.3; 1.7) in 1991 to an RR of 1.9 (95% CI 1.8; 1.9) in 2011. Women also saw an increase in their RR over time from 1.4 (95% CI 1.2; 1.7) to 1.8 (95% CI 1.7; 1.9).

Discussion

Mortality rates have, in general, fallen over time in Canada.^{Note 4} However, in the present study, this decline has not been shared equally across all income and education levels, nor have the changes over time been the same for men and women. Although all-cause age-standardized mortality rates (ASMRs) for both sexes fell in absolute terms between 1991 and 2011 regardless of income or education, men had a greater reduction compared with women. Regarding income-related mortality inequality, in absolute terms the difference in ASMRs for men in the lowest versus the highest income quintile decreased significantly between 1991 and 2011, while it increased for women from 1996 to 2011. In relative terms, the ratio of the ASMR in the lowest versus the highest income quintile increased for both sexes over time. Mortality inequality according to education category for women followed the same pattern as income, though the absolute rate difference between men with less than secondary graduation versus men with a university degree remained relatively unchanged over the period.

Historically, men have had higher mortality rates than women, which has been explained by more risky behaviours, biology and health system use, among others.^{Note 6} The present study found that the mortality gender gap narrowed because of a steeper decline in mortality rates for men than women, regardless of income or education. These results are consistent with recent literature that has suggested that improvements in treatment for cardiovascular disease (CVD) coupled with an increased prevalence of CVD among women are contributing factors.^{Note 4}

The findings regarding income-related mortality inequality showed different patterns for men and women. Among men, larger absolute but smaller relative declines over time in the ASMR in the lowest income quintile compared with the highest income quintile led to a narrowing of the absolute—but a widening of relative—mortality inequality. For women, those in the lowest income quintile had a smaller absolute and smaller relative decline in their ASMR compared with those in the highest income quintile, which led to a widening of absolute and relative inequalities in mortality over time. This trend of increasing inequality among women has been reported in Ontario using area-based measures^{Note 6Note 8} and elsewhere in the world^{Note 18} and may reflect various aspects of disadvantage. These could include the increased uptake among women of risky behaviours such as smoking^{Note 4} coupled with the higher prevalence of this behaviour among lower income women in Canada,^{Note 19} as well as growing inequalities in access to health care.^{Note 20}

For women, mortality inequality according to education categories was similar to that of income in that there was a widening over time in both absolute and relative inequalities between those who did not graduate from secondary school and those with a university degree or equivalent. Among men, there was little change over time in the absolute difference between the ASMRs of those with less than secondary graduation compared with those who had a university degree or equivalent, though the relative difference between the two did increase over the period. As with low income, the mortality decline disadvantage of those with less education has been partly attributed to behavioural risk factors such as smoking and alcohol.^{Note 21} Furthermore, the population size of the low education category has declined significantly, and it may be increasingly composed of individuals who are more homogeneous in terms of characteristics that compound the risk of ill health.^{Note 22} For example, it has been shown that, between 1998 and 2011, education-related inequalities in health status among women in Canada increased significantly.^{Note 23}

Strengths and limitations

Absolute and relative measures of mortality inequality each have their uses in monitoring progress towards the reduction or elimination of health inequalities,^{Note 16Note 24} and a strength of this study is that it presents both. Another strength is the use of the CanCHECs—large, nationally representative census cohorts with individual-level income and education linked to administratively collected mortality data. Although area-based measures contribute to the understanding of neighbourhood effects on health inequalities,^{Note 25} this study’s use of individual-level data analysis allows for less misclassification of sociodemographic characteristics. Five cohorts permitted an analysis of trends over a 20-year period; however, two decades of data have raised concerns about comparability. Changes to question wording and collection mode from one census to the next over this period^{Note 26Note 27Note 28Note 29} should be kept in mind when interpreting the results of this study. Notably, in 2006, there were substantive changes to questions on educational attainment, and, for the first time, census respondents could choose to allow linkage to their tax records rather than self-report their income data. The former change corrected the underreporting of secondary school completion in previous census years^{Note 30}, while the latter reduced clustering around round dollar amounts (such as $30,000), thereby causing income distributions to be more variable than in previous censuses.^{Note 31} The use of income quintiles for this analysis, however, helped to minimize the impact of increased variability in the income distribution from 2006 onward.

These results are limited to the household population aged 25 or older and therefore do not apply to the institutionalized population in Canada. Another limitation relates to potential inclusion bias associated with data linkage. Although the cohort weights were designed to help mitigate this bias, unknown bias might exist if people missing from the cohorts differed systematically from those who were included.

Conclusion

Although mortality rates have fallen significantly over the past 20 years, income- and education-related mortality inequalities persist for the adult household population in Canada, and have increased for women. Additional research to determine the reasons underlying these trends could help address the complex challenge of reducing health inequalities in Canada.