Health Reports
Does geography matter in mortality? An analysis of potentially avoidable mortality by remoteness index in Canada

by Rajendra Subedi, T. Lawson Greenberg and Shirin Roshanafshar

DOI: https://www.doi.org/10.25318/82-003-x201900500001-eng

Despite the tremendous amount of ongoing research, the mechanism of urban–rural health disparities is not fully understood in Canada. Although rural and remote location in itself may not necessarily lead to poor health, it may influence other socioeconomic, environmental and occupational health determinants. There is noticeable heterogeneity within and between rural communities in Canada in terms of socioeconomic and geographic characteristics.^{Note 1} However, in general, people who live in rural communities have limited access to health care services and have worse health outcomes than their urban counterparts.^{Note 2Note 3} This may lead to disproportionate mortality rates between urban and rural communities.

A 2012 Canadian Institute for Health Information (CIHI) report claimed that people who live in rural and disadvantaged areas experience a higher burden of ambulatory-care-sensitive conditions, such as asthma, chronic obstructive pulmonary disease, diabetes, high blood pressure and heart disease.^{Note 4} Studies have established an association between poor health status and the lifestyle, culture, geography and environment in rural areas.^{Note 1Note 5Note 6} Other studies have compared health status and health care needs along the urban–rural continuum.^{Note 1Note 7Note 8} Rural and remote areas are also distinguished by more CSDs with a higher proportion of Aboriginal population than in urban areas.^{Note 9} Avoidable mortality is higher in First Nations adults^{Note 10} and Inuit.^{Note 11} Remoteness can therefore be compounded by the proportion of Aboriginal population in that area.

Urban–rural variation in mortality from chronic diseases and major causes of deaths has been studied in other countries, including the United States,^{Note 12} Australia^{Note 13} and China.^{Note 14} However, due to blurred urban rural and remote classification, no studies in Canada have compared potentially avoidable mortality by relative remoteness.

The potentially avoidable (hereafter avoidable) mortality is a subset of premature mortality that takes place under the age of 75. The avoidable mortality is defined as “untimely deaths that should not occur in the presence of timely and effective health care, including prevention.”^{Note 15} Avoidable mortality represents over 70% of total deaths of individuals younger than age 75 in Canada^{Note 10Note 15} and consists of two subgroups: preventable mortality and treatable mortality. “Preventable mortality” refers to deaths that could be prevented through primary preventative actions in incidence reduction, such as immunization, smoking reduction and seatbelt use. “Treatable mortality” refers to deaths that could be treated through secondary and tertiary prevention efforts, such as effective screening and treatment of an existing disease.^{Note 15} The concepts of preventable and treatable mortality have been adopted widely as performance indicators of broader health systems in countries such as Australia, the United Kingdom, Italy and the United States.^{Note 16Note 17} Previous studies have shown that timely and effective intervention may reduce mortality rates significantly.^{Note 15Note 18}

Conversely, literature also shows that socioeconomic status is the most consistent predictor of health status, and therefore a good predictor of morbidity and mortality.^{Note 15Note 19Note 20} While some authors have argued that level of education is an important social determinant of health,^{Note 21Note 22} others have established an association between income and individual-level health risk factors.^{Note 22Note 23Note 24} There are multiple studies on sex differences in mortality. In general, because of their engagement in more risky behaviour, men are known to have higher mortality rates than women for the majority of the leading causes of death.^{Note 25} However, there is not enough evidence on how avoidable mortality rates vary by sex and by remoteness to draw formal conclusions in this regard.

The majority of previous studies have used the Statistical Area Classification (SAC) and have defined “urban” through the combination of Census Metropolitan Areas (CMA) and Census Agglomerations (CA).^{Note 1} Statistics Canada has defined urban areas as population centres with a population of at least 1,000 and a density of 400 persons per square kilometre.^{Note 26} However, no clear distinction is made between urban, rural and remote areas in Canada. Therefore, researchers and health system planners have identified a need for a pan-Canadian approach to make a clear distinction between urban, rural and remote areas in order to measure health care inequalities in Canada.^{Note 27}

In response to this call, a group of researchers from Statistics Canada recently developed a remoteness index (RI) and assigned a value to each census subdivision (CSD).^{Note 28} This index measures the relative remoteness of Canadian communities (CSDs) on the basis of their size and their proximity to surrounding population centres. Although the RI is a continuous scale, one of its strengths is that it can be easily categorized for the classification of Canadian communities by their relative remoteness. This study uses the new RI classification to distinguish rural and remote areas from urban areas in Canada.

The main objective of this research is to examine major causes of both preventable and treatable mortality by relative remoteness of Canadian communities. Furthermore, it explores the interrelationship between remoteness and avoidable mortality while taking into account three important variables: average household income after-tax, the proportion of postsecondary graduates and the proportion of Aboriginal population by CSD. The central analytical questions in this paper are the following:

• How do avoidable mortality rates vary by relative remoteness of Canadian communities?
• What is the contribution of the remoteness index in determining avoidable mortality rates when the effects of income, education and Aboriginal population are taken into account?

Data and methods

This research used three data sources produced by Statistics Canada. The first data source, the Canadian Vital Statistics – Death Database (CVSD) for the years 2011 to 2015, is an administrative database containing a collection of annual demographics and cause-of-death information in Canada.^{Note 29} Death data are collected from the provincial and territorial vital statistics offices, where deaths that occur in those jurisdictions are registered.^{Note 29} The mortality database was used to create a subset of premature deaths, which in turn was used to create the avoidable mortality database. The International Classification of Disease, 10th edition (ICD-10), codes associated with the CVSD were used to classify avoidable deaths into preventable and treatable causes of death according to the standard classification method developed by the CIHI^{Note 15} (see Appendix A).

The second data source used in this research was the census subdivision RI, which was developed by a team of researchers at Statistics Canada in 2017.^{Note 28} The index value for each CSD ranges from 0 to 1, where “0” represents the most accessible areas and “1” represents the most remote areas. The CSD-level RI scores were classified into five mutually exclusive categories according to the RI values, population size, number of CSDs in each class, and natural break points in the distribution of the RI score (see Appendix B). On the basis of these criteria, the five classes were defined as “easily accessible areas,” “accessible areas,” “less accessible areas,” “remote areas” and “very remote areas.” However, to understand its relationship with avoidable mortality rates, the RI in a scale of 0 to 1 was used as a continuous predictor variable in the multiple regression models.

The third data source, data tables from the 2016 Census of Population, were used to derive the proportion of the population with a postsecondary certificate, diploma or degree, proportion of Aboriginal population and the average annual household income after-tax for each CSD. A common-variable CSD unique identifier (CSDUID) was used to link CSD-level preventable and treatable mortality rates with the RI, along with the education, Aboriginal population and income variables. All the variables were used in four different multiple linear regression models to test the hypothesis that the remoteness index is a good predictor of both preventable and treatable mortality rates in Canada.

Studies have shown that Aboriginal populations in Canada have elevated health risks and higher premature mortality rates.^{Note 30Note 31} Traditionally, the proportion of Aboriginal populations is higher in northern and remote communities in Canada.^{Note 6} Therefore, the proportion of Aboriginal populations could also be a good predictor of higher preventable and treatable mortality rates in the rural and remote communities. However, due to strong correlation of the proportion of Aboriginal population and the RI, it was not possible to use them together in a single regression model. Consequently, CSDs were classified into two groups of low and high Aboriginal identity populations and regression analyses were performed separately for both groups. As applied by previous studies in Canada,^{Note 32Note 33} CSDs where less than 33% of residents reported an Aboriginal identity on the 2016 Census of population were considered as “low-Aboriginal CSDs” and those CSDs where 33% or more of residents reported Aboriginal identity were considered as “high-Aboriginal CSDs”. Multiple linear regression models were used for both groups, and both the preventable and treatable mortality rates were used separately as outcome variables.

Analysis

This research used two types of data analysis strategies to answer the research questions outlined above. First, descriptive data analysis techniques were used to understand the geographic variability of preventable and treatable mortality rates. The preventable and treatable mortality rates per 100,000 population for each remoteness class were used as outcomes of interest. The rates were calculated for groups of mortality causes (e.g., cancers, infections, injuries) and for specific mortality causes (e.g., breast cancer, pneumonia, falls). The direct standardization method was used to age-standardize all rates to the 2011 Census of Population with five-year age groupings. Next, the age-standardized mortality rates per 100,000 population were compared across all remoteness classes, for males and females and for population subgroups, by mortality cause. Since the mortality rates presented in the tables are the average rates for 2011 to 2015, 95% confidence intervals of the mean are also presented. The rates for each geographic region by disease category and sex were tested for statistical significance using one-way ANOVA and Tukey’s HSD test in SAS Enterprise Guide 7.1.

Second, multiple linear regression models were used to test whether the differences in RI, education and income explain the higher preventable and treatable mortality rates for both low-Aboriginal and high-Aboriginal CSDs. The preventable and treatable mortality rates were used as outcome variables whereas the RI, education and income variables were used as predictor variables. The proportion of Aboriginal identity population by CSD was used as a filter variable to classify low-Aboriginal CSDs and high-Aboriginal CSDs. The assumptions of linearity of response and explanatory variables were tested using residual plots, whereas the assumption of non-multicollinearity was tested using the variance inflation factor.

Results

From 2011 to 2015, there were 483,114 premature deaths registered in Canada. Of those, 347,167 were avoidable – about 72% of total premature deaths in Canada. About 65% of total avoidable deaths were preventable, and 35% of those were treatable.

Preventable mortality rate

There was a clear geographic gradient of preventable mortality rates by remoteness for both sexes, and the rates were statistically significant for most geographic regions. The rates for males were almost double those for females, regardless of remoteness (Figure 1).

The age-standardized preventable mortality rate was calculated for eight broad disease groups, as defined by the CIHI (see Appendix A). In the easily accessible and accessible areas, the highest rate of preventable deaths was associated with cancer-related causes for both sexes. However, injury, including both intentional and unintentional injuries, was the leading cause of death for males in less accessible, remote and very remote areas, and for females in very remote areas (Table 1). Diseases of the circulatory system were the third major cause of preventable deaths for all remoteness categories and for both sexes. Most of the differences between remoteness categories and between males and females were statistically significant at the 95% confidence level. The top three causes of death alone accounted for almost 80% of preventable mortality in each remoteness category.

A further analysis was done to understand the top 10 disease-specific preventable causes of death by remoteness categories. For all remoteness categories and for both sexes, lung cancer was the leading cause of death, followed by ischaemic heart disease. Death by suicide and self-inflicted injuries, and alcohol-related diseases were more common in remote and very remote areas than in easily accessible areas. Suicide and self-inflicted injuries were the second major causes of preventable deaths in very remote areas, and the rate was more than three times higher in very remote areas than in easily accessible areas. Transport accidents increased significantly along with remoteness, and were the fourth major cause of preventable death for males in all remoteness categories except easily accessible areas.

Treatable mortality rate

Data analysis shows that about 25% of all premature deaths and about 35% of all avoidable deaths from 2011 to 2015 in Canada were treatable. Treatable mortality rates varied significantly by relative remoteness and were higher for remote areas than for more accessible areas, regardless of sex. Differences were noticeable for all remoteness categories; treatable mortality rates were higher for males than for females, but the gaps were slimmer than those observed in regard to preventable mortality rates (Figure 2).

Treatable mortality rates were calculated for grouped causes of death to understand the geographic variability. Treatable mortality rates increased as remoteness increased for both males and females, and the differences were statistically significant. For males, diseases of the circulatory system were the leading treatable cause of death, followed by cancer. In contrast, cancer was the leading treatable cause of death for females, followed by diseases of the circulatory system (Table 3). This finding was consistent across all remoteness categories.

The findings in Table 3 were further investigated by calculating the disease-specific treatable mortality rates for the top 10 major causes of death (Table 4). There was a clear geographic variation for most of the treatable causes of death. For all geographies, ischaemic heart disease was the leading cause of treatable mortality among males whereas breast cancer was the leading cause of treatable mortality among females. Colorectal cancer was the second major cause of treatable mortality for both sexes.

Multivariate analyses

The findings of four separate linear regression models are summarized together in Table 5. The first model represents the preventable mortality rates for low-Aboriginal CSDs (n = 3,573; RI range 0 to 0.82) and the second model represents preventable mortality rates for high-Aboriginal CSDs (n = 769; RI range 0.09 to 1). The first model shows that the RI and income were statistically significant predictors of preventable mortality rates at 95% confidence level but not the education. However, the second model shows that, education and income are significant predictors of preventable mortality rates at 95% confidence level (Table 5). These results indicate that remoteness is an important factor of preventable mortality for low-Aboriginal CSDs, whereas education is an important factor of preventable mortality for high-Aboriginal CSDs. The income variable is a statistically significant predictor of preventable mortality for both low-Aboriginal and high-Aboriginal CSDs.

The positive parameter estimate for the RI (89.18) in the first model indicates that the preventable mortality rate increases along with the increased remoteness, whereas the negative parameter estimate for the income in the first model (-0.98), and for both income and education in the second model (-2.66; -3.44) indicate that preventable mortality rate decreases with increased average household income after-tax and increased proportion of postsecondary graduates (Table 5).

The findings for treatable mortality rates in the third and fourth models are similar to that of preventable mortality. Similar to the previous models, only education and income are statistically significant predictors of treatable mortality rate at 95% confidence level for high-Aboriginal CSDs (n = 3,412; RI range 0.09 to 1). However, all three variables (RI, education and income) are statistically significant predictors of treatable mortality rate at 95% confidence level for low-Aboriginal CSDs (n = 672; RI range 0 to 0.82). These findings also indicate that relative remoteness is a determinant of treatable mortality for low-Aboriginal CSDs but not for the high-Aboriginal CSDs. The positive value of the parameter estimate for the RI (63.85) in the third model indicates that the treatable mortality rate increases along with remoteness, whereas the negative parameter estimates for education and income in the third (-0.98; -0.18), and also in the fourth model (-1.95; -3.33) indicate that the treatable mortality rate decreases along with the increased average household income and the proportion of postsecondary graduates.

Discussion

The results of the descriptive analysis indicate that there is significant variation in the preventable and treatable mortality rates by relative remoteness of Canadian communities. Although the health care system in Canada is publicly funded, there seems to be geographic variation in terms of health care policy, access and utilization. This variation has led to disproportionate mortality rates in remote communities. Higher preventable and treatable mortality rates in the relatively remote areas could be attributed to geographic barriers, limited health care services, unmet health care needs, and historical and environmental factors that shape the socioeconomic status of Aboriginal rural communities in Canada.^{Note 2}

The proportion of Aboriginal populations is higher in remote CSDs in Canada. However, the model did not show a significant relationship between mortality and the RI in CSDs with a high proportion of Aboriginal people. This could be a result of a smaller sample size and a distribution bias toward higher values of RI in the CSDs with a high proportion of Aboriginal people. For CSDs with a low proportion of Aboriginal people, the majority with 84% of all CSDs, the RI was a significant predictor of both preventable and treatable mortality rates.

The higher mortality rates reported in remote communities could be a combination of both geographic remoteness and higher mortality rates observed in the Aboriginal population. However, the findings on Aboriginal population should be interpreted cautiously taking into account the broader historical context of colonization and racial discrimination that created present day health inequalities between Aboriginal and non-Aboriginal population.^{Note 34} In addition, intergenerational effects of residential school system, social isolation and inadequate health services, adversely affected physical, social, emotional and mental health and wellbeing of Aboriginal population, which are directly or indirectly related to higher mortality rates.^{Note 34Note 35}

The preventable mortality rate for males was almost double than that for females. This indicates that females are more likely than their male counterparts to take preventive measures and utilize health care facilities. This finding is consistent with the literature, which shows differences between men and women in health care utilization.^{Note 36Note 37}

Injury-related preventable deaths were higher than cancer-related deaths in remote and very remote areas. This finding is consistent with other literature that shows a higher risk of injuries in rural areas.^{Note 31Note 38Note 39} The higher preventable mortality rates associated with injuries could also be associated with increased exposure through longer commutes and occupation for example and reduced access to health care facilities.^{Note 38}

Diseases of the circulatory system were the leading cause of treatable mortality among males, and cancer was the leading cause of treatable mortality among females. A comparative study of Canada and Australia by Pong et al.^{Note 7} found higher mortality rates caused by circulatory diseases among men living in rural areas of Canada. A study by Shields and Wilkins^{Note 40} also found that, in Canada, men were more likely than women to die from heart disease. Both of those findings are consistent with this study’s findings. Other studies have shown that breast cancer mortality rates have been declining in recent years due to of increased screening and better treatment.^{Note 41} However, breast cancer is still a leading cause of treatable mortality among women in Canada. The treatable mortality rates associated with diabetes were high for relatively remote communities among both males and females. Previous studies have shown higher mortality rates associated with diabetes for males than for females.^{Note 42} However, there is no previous study allowing comparison of rates by relative remoteness.

This research established that remoteness classes can be used as new dimensions to measure geographic variability of avoidable mortality in Canada. In addition, this research contributes to the literature by establishing a relationship between remoteness and avoidable mortality. This is a novel research in Canada that has flagged possible issues and identified areas for more detailed investigation. Despite the interesting findings, this research has a few limitations. First, the RI classification is not a standard classification, but the authors’ own classification of the continuous RI values. Second, the average household income is not adjusted for the cost of living; therefore, income is not a strong predictor of avoidable mortality compared to RI and education. Finally, the third limitation is related to the mortality database. The CVSD does not identify the individuals in the mortality database as Aboriginal or non-Aboriginal, which does not allow to make a separate analysis for Aboriginal and non-Aboriginal population by remoteness. Having the actual number of deaths by Aboriginal population would have helped to identify whether or not the higher rate of mortalities reported in very remote communities were the results of remoteness.

Conclusion

Avoidable mortality is a key indicator of health behaviour, health policy and health care utilization. However, avoidable mortality is greatly affected by the geographic remoteness of a community, along with other socioeconomic characteristics of the population. This study used the newly developed RI, the Census of Population, and Canadian Vital Statistics data to explore the geographic variation of avoidable (preventable and treatable) mortality rates by relative remoteness of Canadian communities.

The results indicate that there is a clear gradient of both preventable and treatable mortality rates by remoteness for both sexes. Regardless of the declining overall avoidable mortality rates in Canada, preventable and treatable mortality rates were substantially higher in more remote areas than in easily accessible areas. Preventable and treatable mortality rates were significantly higher for males than for females, regardless of remoteness. However, the gender gap was smaller for treatable mortality rates. The RI was positively correlated with Aboriginal population. Therefore, despite being a statistically significant predictor of both preventable and treatable mortality rates for low-Aboriginal CSDs, RI was not a significant predictor of mortalities for high-Aboriginal CSDs.

Appendix