Background The relation between income inequality and mortality in Canada is unclear, and modifying effects of characteristics such as immigration have not been examined.
Methods Using a cohort of 2 million Canadians followed for mortality from 1991–2001, we calculated HRs and 95% CIs for income inequality of 140 urban areas (Gini coefficient, Atkinson index, coefficient of variation; expressed as continuous variables) and working age (25–64 y) or post-working age (≥65 y) mortality in men and women according to immigration status, accounting for individual and neighbourhood income, and sociodemographic characteristics. Major causes of mortality were examined.
Results Relative to low income inequality, high inequality was associated with greater working age mortality in male (HRGini 1.08, 95% CI 1.04 to 1.13) and female (HRGini 1.12, 95% CI 1.06 to 1.18) non-immigrants for all income inequality indictors. Results were similar for female post-working age mortality. There was no relation between income inequality and mortality in immigrants. Among Canadian-born individuals, associations were greater for alcohol-related mortality (both sexes) and smoking-related causes/transport injuries (women).
Conclusion Income inequality is associated with mortality in Canadian-born individuals but not immigrants.
Statistics from Altmetric.com
- Effect modifiers (epidemiology)
- income distribution
- socioeconomic factors
- social conditions
- migration health
- mortality SI
- social inequalities
Individual income is an established risk factor for mortality,1 2 but inequality in the distribution of income within countries may also influence health. Countries with high income inequality are those in which extremes of wealth and poverty are more pronounced. Understanding the relation between income inequality and health has been a priority in Europe and the USA since the 1990s.3 Income inequality is associated with mortality in several Western countries,4 but no such relation has been established for Canada.5–8 However, Canadian studies are limited by cross-sectional ecological designs, and multilevel prospective designs are preferred for evaluating income inequality and health.2 4 In fact, it was through prospective studies of populations from the USA9–12 and Europe13–15 that income inequality was more strongly established as a risk factor for mortality.
Nonetheless, the relation between income inequality and mortality has been debated despite this evidence;9–15 in part because no associations were found in some studies accounting for ethnicity and individual income.1 16–20 However, other underlying structural or demographic characteristics of populations may need to be addressed to accurately understand the relation between income inequality and mortality. Immigration patterns in particular may be important in places such as Canada where nearly 20% of the population is foreign-born.21 Immigrants may be generally healthier than native-born populations,22 23 and may inadvertently mask associations between income inequality and mortality. Immigration patterns have not been accounted for in previous Canadian cross-sectional analyses of income inequality and mortality.5–8 To clarify these issues, we evaluated whether the relation between income inequality and mortality in a large Canadian cohort was modified by immigration.
Data and variables
Data were drawn from the Canadian Census Mortality Follow-up Study (CCMFS)—a cohort containing a 15% sample of the Canadian population aged 25 years and over living in Canada on the day of the census (4 June 1991) who were not residents of a long-term institution and were followed for mortality for 10.6 years until 31 December 2001.24 The cohort contains individuals representative of the Canadian population who responded to a detailed census questionnaire. Individuals residing in all urban areas of Canada (n=140 cities) who were part of the CCMFS cohort were included (N=2 077 000). Urban areas consist of 25 census metropolitan areas (N=1 637 900) and 115 smaller urban centres with populations over 10 000 known as census agglomerations (N=439 000). Rural areas were not included as they were too small to allow meaningful calculations of income inequality.
Mortality was evaluated for two separate age groups at the time of cohort inception, men and women of working age (25–64 y) and post-working age (≥65 y) as previous research suggests relations depend on age and sex.10 Principal cause of death was assessed using the 9th and 10th revisions of the International Classification of Diseases (ICD)25 for deaths in 1991–1999 and 2000–2001, respectively. We considered general causes of death (cancer, cardiovascular, injury-related), specific causes (cancer of the lung, prostate/breast and colorectum, ischaemic heart disease, cerebrovascular disease, transport injuries, suicide) and behavioural or preventable causes (alcohol-related or smoking-related) used in previous research of socioeconomic inequalities in mortality.25 ICD codes for these causes are available elsewhere.25
Urban-level income inequality was measured using the Gini coefficient, Atkinson index and coefficient of variation, since associations with health may vary with the indicator, and use of different indicators may help capture associations that depend on sensitivity to various parts of the income spectrum.26 Indicators were calculated for each of 140 urban areas using family income (before-tax, after-transfers), equivalised for family size and relationships between members according to the method of the Organisation for Economic Co-operation and Development (after-tax income was not available).27 For individuals not in a family (ie, those living alone), personal income was used instead. Income in dollars was obtained directly from the CCMFS cohort.28 The Gini coefficient measures deviation of the income distribution from perfect equality and ranges from 0 (perfect equality) to 1 (perfect inequality).7 26 The Atkinson Index also ranges from 0 to 1 and is interpreted similarly.26 The Gini and Atkinson indices were computed using a SAS macro from the Luxembourg Income Study (http://www.lisproject.org/key-figures/sasprograms/allkf_sas.txt) with a ε value of 1 for the Atkinson Index. The coefficient of variation was obtained from the SD of the income distribution divided by its mean (we verified that income was normally distributed).26
Census respondents were asked whether they had immigrated and, if so, the year of immigration.28 In analyses, immigration status was assessed for the categories non-immigrant, long-term immigrant (>10 y) and recent immigrant (≤10 y). A 10-year cut-point was used to differentiate recent from long-term immigrants as immigrant health in Canada tends to converge with non-immigrants after a decade.29 Finer categorisations based on source country or for narrower time periods since immigration were not considered as the relation between immigration and mortality was not the main focus of the study. Income was measured relative to Statistics Canada's low income cut-off for families and communities grouped as quintiles (low, low-moderate, moderate, high-moderate, high).24
Covariates included individual age (25–34, 35–44, 45–54, 55–64 y for working age; and 65–74, 75–84, 85+ y for post-working age), education (no high-school graduation, high-school diploma and/or trade certificate, post-secondary non-university, university degree), employment (employed, not employed, not in labour force), marital status (legally married, common law, not married), visible minority (no, yes), provincial region (British Columbia and Territories, Prairies, Ontario, Québec, Atlantic) and mean neighbourhood household income (low, moderate-low, moderate-high, high). Neighbourhoods were specified as enumeration areas—the smallest unit for which Statistics Canada disseminated 1991 socioeconomic data (containing 750 inhabitants on average).
Directly standardised mortality rates were calculated using the 1991 population structure as the reference. We used Cox proportional hazards regression to calculate age-adjusted HRs and 95% CIs for the relation between income inequality and working age or post-working age mortality in men and women. The census day was specified as the start time. The proportional hazards assumption was verified with log(-log Survival) curves for all variables plotted against time. Models first evaluated income inequality alone and were subsequently adjusted for individual income, immigration and other covariates (employment, education, marital status, visible minority, mean neighbourhood household income and region). Effect modification was tested with income inequality-by-immigration interaction terms, and fully adjusted models were run for non-immigrants, long term immigrants and recent immigrants, separately. Specific causes of mortality were evaluated separately in models with working age and post-working age groups combined, and with long-term and recent immigrants combined, to increase statistical power.
For regression analyses, indicators of income inequality were expressed as continuous cumulative rank scores ranging from 0 to 1 to account for the population distribution across levels of income inequality30 (individuals in the same category were assigned the average rank).31 The cumulative rank score was used as a continuous variable in regressions after verifying that quartiles of income inequality were linear with the log-hazard of mortality. HRs obtained from regression of the cumulative rank score on mortality are interpreted as the relative difference in mortality between the hypothetical most and least unequal areas of the cumulative income inequality distribution.30 These HRs are equivalent (or equal) to the Relative Index of Inequality—a summary measure of inequality increasingly used in the literature.30
The data were hierarchically structured with individuals nested in 140 urban areas. Clustering in urban areas was accounted for with the robust sandwich estimator.32 We verified that results accounting for clustering in neighbourhoods were similar. We also verified that models run without small urban areas yielded similar results.
Statistical analyses were undertaken using the TPHREG procedure of SAS 9.1 (SAS Institute Inc). The CCMFS was approved by the research ethics committee of the University of Toronto and the Statistics Canada Policy Committee.
Long term and recent immigrants constituted approximately 20% and 6% of the population, respectively (table 1). There were 202 354 deaths (9.7%) over the follow-up period. Age-adjusted mortality rates per 100 000 person-years were 530 and 310 for working age men and women versus 6600 and 3900 for post-working age men and women, respectively. Death was more frequent among low-income individuals (table 2). Among post-working age individuals, deaths were proportionately higher for non-immigrants than for immigrants. However, among working age individuals, higher death proportions were present for long-term immigrants relative to non-immigrants.
Median (IQR) values for income inequality were 0.32 (0.31 to 0.34) for the Gini index, 0.21 (0.19 to 0.23) for the Atkinson index and 0.68 (0.63 to 0.75) for the coefficient of variation. Although death proportions varied slightly across levels of income inequality, a directional trend was not apparent. Median (IQR) proportions for the immigrant population per urban area were 11.5% (3.5% to 18.4%).
Table 3 shows HRs for the relation between income inequality and mortality. Contrary to what may be expected, models adjusted only for age suggested that high income inequality was associated with a lower hazard of mortality relative to low inequality. Adjustment for individual income did not change these associations. However, further adjustment for immigration tended to nullify or reverse associations to the expected direction in both men and women, especially for working age mortality. Most indicators of income inequality in fully adjusted models were associated with a higher risk of working age mortality, but not with post-working age mortality.
Tests for interaction between income inequality and immigration were statistically significant in both sexes and age groups except for the coefficient of variation in men. When models were run for non-immigrants alone, a statistically significant relation was present between income inequality and mortality among both working age and post-working age men and women except for the coefficient of variation in post-working age men (table 4). The magnitude of the hazard tended to be greater for working age than post-working age men and women. Among long-term immigrants, associations were either null (for working age mortality) or protective (for post-working age mortality). HRs tended to be even lower among recent immigrants, although not statistically significant with the exception of male post-working age mortality.
Tables 3 and 4 demonstrate that all three indicators of income inequality tended to yield similar results and, therefore, that associations were not dependent on one particular part of the income spectrum.
Among non-immigrant men, colorectal cancer, alcohol-related and ‘other’ causes were most strongly related to the Gini coefficient (table 5). The relation between income inequality and colorectal mortality was also present for immigrant men. Patterns were different among non-immigrant women for whom cancer (lung in particular), smoking-related causes and transport injuries were important causes associated with income inequality. Like men, income inequality among non-immigrant women was associated with alcohol-related and ‘other’ causes of mortality. Among immigrant women, income inequality was associated with a lower hazard of cardiovascular mortality.
This study of a large prospective cohort is the first to demonstrate that income inequality is associated with mortality in Canada independent of individual income. Furthermore, the relation between income inequality and mortality is restricted to non-immigrant Canadians. Income inequality was most strongly associated with mortality among non-immigrant working age men and women. Elevated risks were also observed for post-working age women, but associations were borderline for post-working age men. In contrast, associations among immigrants were either absent or protective. Unadjusted models suggesting a protective effect of income inequality on mortality were, therefore, in part explained by the modifying effect of immigration. Though precision was not high, the causes of death primarily associated with income inequality among non-immigrant Canadians were colorectal cancer in men, lung cancer/smoking-related/transport injuries in women and alcohol-related in both sexes. Among immigrants, protective associations were observed for prostate cancer (men) and cardiovascular mortality (men and women).
The literature cites two general pathways through which income inequality can influence health. The first involves direct effects though individual factors such as poverty that are characteristic of unequal societies.3 33 Under this hypothesis, income redistribution would be sufficient to elevate the health status of the poorest up to the standard of the wealthiest. However, the extent to which income inequality might operate through individual factors is unclear, given that individual income did not confound the associations we observed between income inequality and mortality. Our results are in fact in line with previous research refuting statistical artefact as an explanation of the relation between income inequality and mortality.34 Since neighbourhood income did not influence the associations in our data, it is unlikely that urban-level income inequality operates through neighbourhood (or contextual) material pathways in Canada. The other pathway identified in the literature involves the potential for inequality to generate divisive conditions in society impacting social status, friendships, sense of control and social capital.3 35 These routes are possible, although our data could not be used to explore their role.
Pathways may be different among immigrants for whom elevated HRs were generally absent. Immigrants to Canada come from diverse countries, including Europe in early decades of the last century and, increasingly, Asian countries in recent decades (with consequent challenges related to knowledge of official languages and visible minority status).21 The majority settle in urban areas, especially large metropolitan centres like Toronto and Vancouver that have even higher rates of foreign-born populations than do Miami and Los Angeles.21 Indictors of income inequality may potentially be proxies for other factors influencing where immigrants live rather than income inequality per se. Despite being highly skilled, immigrants face an income gap that improves with length of time in Canada;36 sense of community may also change over time. These factors may, in part, be related to why associations between income inequality and mortality for long-term immigrants tended to approach those of non-immigrants.
Causes of death associated with income inequality were mainly behaviour related. However, results should be interpreted in light of low precision due to small counts for some causes. Alcohol-related causes stood out the most for both men and women. Alcohol consumption has been targeted as a major avoidable risk factor worldwide.37 Though a study of working age Danish men found no association between income inequality and alcohol-related mortality,18 our results suggest the opposite may be the case for Canadian-born individuals. The association between income inequality and colorectal cancer mortality in Canadian-born men suggests that income inequality may influence nutrition given the potential relation between food quality and colorectal cancer.38 In Canadian-born women, lung cancer and other smoking-related causes of mortality were associated with income inequality, which is concerning given the absence of an association among men. Smoking was historically adopted by Canadian women much later than men39 40 and our results suggest that female smoking may have increased more quickly in areas with high-income inequality. Interestingly, mortality from transport injuries was also associated with income inequality in Canadian-born women (but not men), which suggests gender-related inequalities in transport safety may be present in more unequal parts of Canada, although results should be interpreted with caution as mortality from transport injuries among women is rare in urban areas of Canada.
Our findings shed light on how the relation between income inequality and mortality varies across age and sex. In the USA, income inequality has been independently associated with working age (<65 y) but not post-working age (≥65 y) mortality, and the relation tended to be stronger among men compared to women (however, modifying effects of immigration were not considered).10 Similar patterns were observed in Norway.13 In comparison, we found that income inequality was associated with working age mortality for Canadian-born individuals and that associations were just as strong for women as men. In contrast to the USA,10 we found positive associations between income inequality and post-working age mortality, especially for Canadian-born women. Because the association between any risk factor and mortality in older individuals is expected to be attenuated relative to younger ones,41 the fact that we found an association among older women reinforces the importance of income inequality as a determinant of mortality in Canada whatever the age.
Few prospective studies examined whether individual factors modify the influence of income inequality on mortality. An analysis of Swedish individuals aged 40–64 years evaluating occupation as a modifier found that income inequality was associated with a higher risk of mortality for unskilled manual workers only.15 A US study examining individual income and ethnicity as modifiers found effects on mortality were largest in poor white people.9 Unfortunately, no prospective study of income inequality and mortality examined immigration as a modifier despite evidence that immigrants may be healthier than native-born populations.22 29
This study was subject to limitations. We could not account for lagged effects of income inequality42 or changes in income inequality over time;10 41 however, effects are unlikely to be differential between immigrants and non-immigrants. We used census data rather than tax returns and associations may be overestimated with the former.42 We did not have data on second generation immigrants, although we accounted for multiple individual-level variables that may be intermediate factors, which may have unnecessarily attenuated associations.10 We cannot account for the ‘salmon effect’—a bias related to the potential for immigrants to return to their country of origin when seriously ill,43 which may be problematic if the practice was relatively more common in areas with higher (or lower) income inequality. However, this bias would only affect results for immigrants and not those for non-immigrants for whom we demonstrate a positive association with income inequality. We did not verify HRs with age as the underlying time (rather than time-on-study)44 because research points to no practical difference between either time scale as long as age is adjusted for, which we have done.45 46 Last, we could not account for time-varying covariates, including residential mobility.
This study found that income inequality was associated with mortality among non-immigrants in Canada (but not immigrants). Canada, so far considered insulated from such effects, is indeed influenced by extremes of wealth and poverty. Future research on income inequality and health in other countries should evaluate modifying effects of cultural or other population characteristics, including immigration patterns. Health policies for reducing mortality inequalities in Canada may need to address income inequality, especially for prevention of colorectal cancer and smoking/alcohol-related mortality.
What is already known on this subject
Income inequality is associated with mortality in some countries but not all. Modifying effects of immigration have not been examined.
What this study adds
Income inequality is associated with mortality in Canada among native-born individuals but not immigrants. The relation between income inequality and mortality is masked when immigration is not accounted for.
The authors acknowledge the Health Information and Research Division branch of Statistics Canada for facilitating access to the data. NAR acknowledges the generous support of an FRSQ Chercheur Boursier Jr II career award.
Funding The Canadian Census Mortality Follow-up Study was supported by the Canadian Population Health Initiative. All investigators are independent of the funders.
Competing interests None.
Ethics approval The Canadian Census Mortality Follow-up Study was conducted with the approval of the research ethics committee of the University of Toronto and the Statistics Canada Policy Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
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.