Premature mortality due to social and material deprivation in Nova Scotia, Canada

Area-based deprivation indices were developed in the UK [4],[27],[28] as a tool for investigating socio-economic variations in health and as a surrogate indicator of individual-level socioeconomic status. They have since been modified to reflect the local reality and data availability of various populations around the world [29]-[37]. For this study, two indices of deprivation were constructed following the methodology detailed by Pampalon and colleagues [38] developed to measure socioeconomic deprivation within a Canadian context. The indices were composed of six variables from the 2001 Canadian census known to have utility as geographic proxies of socioeconomic conditions [21],[33],[39],[40]. For people age 15years and over, these variables were: the proportion of people with no high school diploma, the individual average income, the employment rate, the proportion of separated, divorced or widowed, the proportion of single-parent families (lone parent), and the proportion of persons living alone. The first three indicators reflect the material dimension of deprivation; the others reflect its social aspect. All variables, with the exception of the proportion of single-parent families, were adjusted to the age and sex structure of the 2001 NS population aged 15years and older, using indirect standardization [41]. Transformations (log" for continuous variables, arcsin of square root" for proportional indicators) were applied to normalize the indicators. Variables were combined using a Principal Component Analysis (PCA), a standard factorial approach that recognizes the interlinked nature of variables by accounting for their correlation and co-variation [42]. Following a varimax rotation, two independent components with eigenvalues exceeding 1.0 were retained for interpretation. These components were defined as `material index´ and `social index´ of deprivation, respectively.

The indices were constructed at the smallest unit of census geography, the dissemination area (DA), which comprises generally a population of 400-700 persons but which can be as low as 40 persons in rural NS and as high as 3,600 in urban NS. DAs were defined as urban when in proximity to a census metropolitan area with a population density of 400 or more people per square kilometer as outlined in Du Plessis et al. [43]. In the 2001 census, NS was covered by 515 urban and 771 rural DAs (excluding First Nations reserves, for which details of population and census variables were incomplete). PCA produced factor scores for all 1,286 DAs. The DAs were ranked according to their factor scores and grouped into weighted population quintiles, one distinct set of quintiles for each level of geography (i.e. urban, rural, NS as a whole). This was done to account for differences in the range of factor scores by level of geography. In all instances, quintile 1 (Q1) represented the most privileged segment of the population and quintile 5 (Q5), the least. This process was carried out separately for each of the deprivation indices.

Mortality data coded ICD-9 (1995 - 1999) or ICD-10 (2000-2005) for NS residents who died between 1995-2005 were obtained from NS Vital Statistics. Deaths were grouped into five categories: cancer (ICD9-140-208; ICD10-C00-C97), circulatory system (ICD9-390-459; ICD10-I00-I99), external causes (ICD9-800-999; ICD10-V01-Y98), other causes and all causes. PM was defined as deaths occurring prior to the median age at death (75 for men, 81 for women) observed in this period. Age 75 is often used as a fixed upper threshold age for the calculation of PM, however, an older cut-off was used for females as to reflect their longer life expectancy. Residential postal code at death was used to assign each death to a DA using the Statistics Canada Postal Code Conversion program (PCCF+, version 5G). There were 87,484 deaths over the 11-year period. Of these, 74,610 deaths had postal code information and 73,088 (98%) were successfully geo-referenced to a DA. Two percent of deaths occurred before age 15 and these were excluded. PM rates were based on a total of 35,266 premature deaths and calculated using the 2001 NS population aged 15years plus, obtained from Statistics Canada. An aggregated dataset of premature death counts was used to estimate PM rates for each quintile of material and social deprivation, from the most (Q1) to the least privileged (Q5), and for groups experiencing extreme socioeconomic conditions, including those materially and socially most privileged (Q1material-Q1social; Q1 & Q1) which accounted for 7.0% of the NS population aged 15years and older, and those materially and socially least privileged (Q5material-Q5social; Q5 & Q5) which accounted 4.3%.

The influence of deprivation on PM was modeled with Poisson regression. In Model 1, quintiles of material and social deprivation were used as categorical variables and so accounted for the main effects of the two indices. In Model 2, for every combination of material and social deprivation quintiles, mean material and social deprivation scores were calculated and modeled with their interaction with population location (urban/rural). PM rate ratios and absolute excess mortality were also examined. Rate ratios (rate for the least privileged (Q5material-Q5social) divided by the rate for the most privileged (Q1material-Q1social), and corresponding 95% confidence intervals were derived from a Poisson regression model. The excess mortality measure estimated the absolute number of premature deaths for any subgroup that could be potentially avoided if the whole population had the same PM rate as that of the most privileged group. Data analyses were performed using SAS 9.1 and R 2.13.0. The study received ethics approval from Capital Health and IWK Health Centre Research Ethics Boards.

The PCA identified two main components, together accounting for 67% of the variation associated with the six indicators. The first component reflected material deprivation, with high loadings for education (0.89), income (-0.84) and employment (-0.62); the second component reflected social aspects, with high loadings of the proportion of separated, divorced or widowed (0.89), the proportion of persons living alone (0.78) and of single-parent families (0.64). The population profile by quintile of material and social deprivation is presented in Table11. Of particular interest is the comparison between the least and most privileged groups (Q5material-Q5social vs. Q1material-Q1social, respectively) which shows that the former had 4.1 times higher proportion of people without a high school diploma (e.g. 47.6% vs. 11.5%); 1.7 times lower employment rates; 3.1 times higher number of people living alone; 2.4 times higher number of people identified as separated, divorced or widowed; and 6.1 times higher number of single-parent families. In addition, the least materially and socially privileged people earned less than half the income of the most privileged ($16.7K vs. $40.5K). These differences between the least and most materially and socially privileged groups were observed in both rural and urban NS, but were generally greater in urban populations (Table11). The exception was for employment rate for which the gap between the most and least privileged group was greater in rural NS (Table 1).

Of the 35,266 premature deaths included in the study, 14,054 (40%) were attributed to cancer, 9,793 (28%) to disease of the circulatory system, 2,646 (8%) to external causes and 8,773 (25%) to other causes (Table22). The total number of premature deaths was greater in rural than urban NS (20,506 vs 14,752) but crude PM rates did not differ significantly between urban and rural areas, with the exception of other causes mortality for which the rate was higher in urban populations (Table22). Both crude (Table22) and adjusted PM rates (Model 1, Figure11) increased monotonically with increasing levels of material and social deprivation. For social deprivation, these rates showed higher mortality in Q4 for cancer and all causes mortality.

Crude rate ratios (RRs) in PM for those in the most and least privileged population groups are presented in Table33. For all causes, the PM rate for NS was 2.5 times higher in people experiencing a combination of the most extreme conditions of material and social deprivation relative to the most privileged (Table33). PM due to cancer, diseases of the circulatory system, external causes and other causes was 1.9, 2.9, 4.1 and 3.1 times higher respectively in the least compared to the most privileged groups (Table33).Non-significant differences in RR were observed between urban and rural populations, with RR in urban being slightly higher. Figure22 shows the relationship between the mean material and social deprivation scores and PM, adjusting for interacting material, social and urban/rural effects (Model 2). Table44 shows the predicted percentage change in PM corresponding to a change of one quintile level. Again, a significant increase in PM for both major and all causes of death was observed with increasing material and social deprivation (Figure22; Table44). Material deprivation had a similar influence upon PM rates in urban and rural populations (Figure22; Table44). The exception was PM due to external causes for which an increase in material deprivation scores equivalent to one quintile was associated with a 17% increase in PM among those living in rural areas (Table44) compared to a 7.7% increase in PM rate for those living in urban areas. The influence of social deprivation upon PM rates was also significant for major and all causes mortality and was generally of larger magnitude for rural populations (Figure22; Table44). An increase in social deprivation of one quintile was associated with an increase of 14%, 21%, 25% and 19% in PM, due respectively to cancer, circulatory system, other causes, and all causes of death, among rural populations. In contrast, an increase in social deprivation equivalent to one quintile was associated with significant, but lower comparative increases of 7.8%, 11%, 15% and 8.8% in PM among those living in urban areas. The exception to this pattern was external causes, for which an increase in social deprivation of one quintile resulted in a comparable increase in PM rate (20%; Table44) in both rural and urban populations. However, irrespective of social deprivation, rural populations had a 16% higher risk of dying prematurely due to external causes than urban populations.

Considering the distribution of mean material and social deprivation scores for urban and rural populations, there is a greater distribution gap between the most and least privileged in urban NS. With regards to material wealth, those most privileged in urban areas (i.e. u1 [Q1 urban]; Figure22) were comparatively better off than their rural counterpart (r1 [Q1 rural]; Figure22). With regards to social wealth, those least privileged in urban areas (i.e. u5 [Q5 urban; Figure22) were comparatively worse off than their rural counterparts (r5 [Q5 rural]; Figure22). These differences observed between comparable quintiles in urban and rural populations, were not associated with a significant health advantage in those most materially privileged; nor with a significant health disadvantage in those most socially deprived.

Mortality attributable to variability in death rates across quintiles of material and social deprivation for urban and rural NS is presented in Table55. Material deprivation alone may have accounted for 7,245 premature deaths over an 11year-period in NS (3,825 in urban; 3,420 in rural). The independent effect of social deprivation was even more pronounced, accounting for 9,993 premature deaths (5,032 in urban; 4,961 in rural). Over an 11year-period, 14,693 premature deaths (6,878 in urban; 7,815 in rural) could have been avoided if material and social disparity did not exist (i.e. if all population quintiles had the same mortality rate as Q1, the most privileged). Overall, the combined effect of material and social deprivation accounted for nearly half of all premature deaths recorded in urban populations; and over a third of those recorded in rural populations (Tables22 and 5). These proportions varied by cause of death, ranging from 31% (cancer), 43% (circulatory system), 44% (external causes) and 42% (other causes) in rural areas; and from 34% (cancer), 51% (circulatory disease), 54% (external causes) and 60% (other causes) in urban areas (Tables22 and 5). For NS as a whole, the combined effect of material and social deprivation accounted for 42% of all premature deaths.

The study revealed substantial inequalities in socioeconomic conditions in NS, Canada. Similar disparities, although of varying magnitude, have been observed for other regions of the country [14]. For example, in British Columbia the ratios of the least to the most privileged persons were 2.5, 1.4, and 5.3 for the proportion of people without high school diploma, lower employment rate and number of single-parent families; compared to 4.1, 1.7, and 6.1 for Nova Scotia, respectively [14]. While these data indicate greater discrepancies between the `rich and the poor´ in NS, the gap in average income between the least and most privileged groups was greater in the larger metropolitan areas of both Toronto ($32.8K) and Vancouver ($28.9K) compared to Nova Scotia as a whole ($23.8K).

Inequalities in mortality were not confined to differences between the `rich´ and the `poor´ or between the most and least socially or materially deprived, but rather, were observed over the entire socioeconomic spectrum, thus affecting every segment of the population. PM rates decreased monotonically from the most to the least disadvantaged quintile for both major and all causes mortality, with the exception of cancer and all causes mortality for which social deprivation resulted in higher PM rates in Q4. Socioeconomic inequalities were associated with more than a doubling in PM rates (i.e. 2.5 time higher) for the approximate 32,000 Nova Scotians (i.e. 4.3% of the population) experiencing a combination of the most extreme levels of material and social deprivation. Inequalities of similar magnitude have also been reported for Canada as a whole [14]) and Scotland [17]. However, in Scotland, the ratio in PM rate between the most and least privileged, increased from 2.2 in 1981 to 4.3 in 2001 for all causes premature deaths. This widening gap was attributed to a sharp decline in PM in those most privileged at a time of increased PM in those most deprived.

The impact of material deprivation on PM was similar for urban and rural populations, whereas the impact of social deprivation on PM rates was significantly higher for those living in rural areas. The exception was PM due to external causes, which was higher in the most materially deprived rural populations and for which the impact of social deprivation on PM rate were similar in urban and rural populations. The mechanisms contributing to these overall differences are not well understood. With regards to external causes, some studies have reported increased mortality due to external causes with increasing material and social deprivation [44]; others have reported higher mortality due to external causes in rural populations [45],[46]; but few have examined the impact of the interaction between urban and rural status, socioeconomic indices and external causes of PM.

This study showed that about 40% (14,696 deaths) of premature deaths over an 11year-period were attributable to socioeconomic inequalities and thus, potentially avoidable. Of these, more than half were associated with social deprivation alone, a factor seldom accounted for in estimates of health risk in Canada. Due to varying study methodologies and limited research reporting on social disparities in premature mortality, it is difficult to compare these results to other studies. Nonetheless, a recent study indicates that up to 30% of excess deaths (all deaths) reported in sixteen European cities could be attributable to socioeconomic disparities [18]. This figure is somewhat lower than that reported here, but may reflect a greater impact of socioeconomic inequalities on premature mortality in comparison to its impact on all deaths. Thus, the magnitude of the burden of PM due to social and material inequalities has far-reaching implications worldwide. Inequalities are undesirable; they affect everyone in terms of loss of potentially productive members of society, and represent added costs for the health care system and public sector [13],[47].

This study was based on 11years of provincial vital statistics data of which 98% was successfully geo-referenced, enabling deaths to be linked to census-derived deprivation scores. Other strengths include the use of validated composite measures of deprivation [26], which provide a more complete representation of the variability in deprivation relevant to health than do single indicator variables such as income [42],[48],[49]. In addition, the weight assigned to each variable included in the construction of the material and social indices of deprivation is determined based on the correlation structure that exists among the variables at the geographic level of interest, rather than being determined a priori[38].

A limitation of this study is the lower population densities of rural areas which can result in unstable modeled results [50]-[52]. Also, DAs can cover larger areas in rural NS, possibly resulting in more heterogeneous population profiles. In addition, as demonstrated earlier, the distribution in material and social wealth varied between urban and rural populations. Each of these factors could have reduced the estimated inequalities in PM rates due to the social and material deprivation in rural populations. A second limitation is that area-based indices can be prone to ecological fallacy when inferences are generalized to the individual level [53]. They are also affected by the modifiable areas unit problem (MAUP), which affects the inference of the results from one scale of observation to another [53]. Third, this study did not account for spatial dependency between DAs. Spatially correlated random effect terms are often used to account for this dependency; however, data provided for the study was aggregated by quintile of social and material deprivation and urban/rural regions and so did not permit such an analysis. Failure to account for spatial dependency may have artificially narrowed the confidence intervals for the β coefficients and resulted in an underestimation of the type I error rate. Finally, when calculating a population attributable fraction one assumes a causal relationship between the risk factors and health outcome of interest and independence of the considered risk factors from other factors that influence risk [54]. However, it is unlikely that factors contributing to social and material deprivations are completely independent of other factors linked to PM, thus resulting in a possible overestimation in the overall attributable fraction.

Overall findings of a pervasive impact of socioeconomic deprivation on PM rates in NS are consistent with findings reported in other regions of Canada as well as in the United Kingdom, United States, Australia and elsewhere [1],[15],[20],[37],[38],[47],[55],[56]. Poor health outcome was not confined to the most disadvantaged. Socioeconomic inequity affected everyone; a pattern highlighted by the World Health Organization (WHO) not only for the most disadvantaged countries, but for countries of all income levels [24]. The estimated twofold difference in PM rate between the least and most privileged population segments of NS is comparable to the 2.3 fold difference in PM rates seen between lower and higher income countries [57].

Canada acknowledges that raising the health status of people with the greatest need would have a major impact on overall health and could also improve the nation´s productivity, as suggested by the WHO Commission´s report on health equity. Using a recently validated index of deprivation, our study demonstrates the feasibility of identifying and quantifying, at a small area-level, social and material factors that contribute to PM and health inequity. It is likely that the overall impact of social and material inequalities on health will continue to increase as the difference in wealth between the rich and the poor continues to grow [58]. Provincial and Federal governments in Canada and elsewhere have a responsibility to acknowledge and address these serious and growing issues that impact on health equity. Part of the effective delivery and evaluation of such policy changes must be the compilation of small area-level measures of health inequity and their determinants.