Data sources, methods, and limitations

Data sources

The data come from two sources: the 2001 Census of Canada and the 2001/2002 Hospital Morbidity Database (HMDB). The census data pertain to the characteristics of the population at the dissemination area (DA) level, the smallest geographic unit for which census data (including family, household and dwelling characteristics and Aboriginal identity) are available.

The HMDB, which is maintained by the Canadian Institute for Health Information, contains annual information about acute-care (as distinguished from long-term care) hospital discharges, including dates of admission and discharge, diagnoses and procedures, date of birth, sex, postal code, and health insurance number. Because Quebec does not submit complete postal code information, which was necessary for this analysis, Quebec data were excluded.

More details about these data sources are provided in Appendix B.

Methods

The term “Aboriginal” refers to people who identified as an Aboriginal person on the 2001 Census. Throughout this report, those who identified as North American Indian are referred to as First Nations. This analysis makes no distinction between Aboriginal people living on or off a reserve. The percentage of Aboriginal residents in a dissemination area (DA) is the number reporting Aboriginal identity divided by the DA’s total population. DAs in which 33% or more of the population was Aboriginal were classified as “high-Aboriginal”; those with less than 33%, “low-Aboriginal.”¹² Census information was also used to further classify each high-Aboriginal DA by predominant identity: First Nations, Métis or Inuit.

Because the Hospital Morbidity Database (HMDB) does not contain information about the Aboriginal identity, the percentage of the DA population that was Aboriginal was assigned to each hospital discharge record based on the patient’s postal code, using the Postal Code Conversion File (PCCF+).¹⁹

Rural postal codes are often not uniquely matched to a single DA and may serve several DAs. The PCCF+ uses an unbiased procedure to assign a single postal code to a DA, after considering the weighted population counts for every DA that could possibly be selected. This could not be done for hospital discharge records for Quebec, which did not include 6-digit postal code information. The exclusion of Quebec meant that a considerable number of DAs with substantial Aboriginal populations were omitted from this analysis (Limitations). Table A in Appendix A displays the count of DAs to which the percentage of the population that was Aboriginal could not be assigned. Table C includes counts of hospitalizations excluded from the analysis because postal code information was incomplete or missing (n= 31,447, 1.5% of all records), or because the DA did not have sufficient census information to calculate the percentage of the population that was Aboriginal (n= 4,750, 0.22% of all included records).

In 2001, there were 52,993 DAs in Canada; 40,840 excluding Quebec. Of these, 2,066 (5.1% of included DAs) were classified as high-Aboriginal. High-/Low-Aboriginal could not be determined for 2,406 DAs (5.9% of all included DAs) because of very small counts for reported identity or missing information. The jurisdictions with the largest percentages of unclassifiable DAs were Newfoundland and Labrador, Yukon, Northwest Territories, and Saskatchewan. Together, the population in these unclassifiable DAs made up about 0.1% of the total population of Canada (Appendix A, Table B). Appendix A Table E lists incompletely enumerated Indian Reserves and Indian Settlements in the 2001 Census and corresponding population estimates.

Information about the characteristics of each dissemination area (DA) came from the 2001 Census. The prevalence of missing data for area characteristics is reported in Appendix A Table D.

DAs that were not located in Census Metropolitan Areas or Census Agglomerations (CMA/CA) and that had weighted populations less than 10,000 were designated rural; otherwise, DAs were classified as urban.

Area average household income quintile is based on summary data for neighbourhood (in the CMA/CA) income per person equivalent (IPPE), adjusted for household size. The PCCF+¹⁹ classified DAs into an income quintile (Appendix C).

Labour force participation rate below jurisdiction rate is a dichotomous variable (yes/no) that indicates whether a DA’s participation rate was below that of the province or territory. The labour force participation rate is the percentage of the population (aged 25 or older) who were employed or actively seeking work during the week before the 2001 Census.

Above-average percentage of population without secondary graduation is a dichotomous variable (yes/no) that indicates whether the percentage of the population aged 20 or older in a DA who did not complete secondary school was greater than the average for the province or territory.

20% or more private dwellings need major repair is a dichotomous variable (yes/no) that indicates whether 20% or more of private dwellings in the DA needed major repair.

10% or more private dwellings overcrowded is a dichotomous variable (yes/no) that indicates whether the ratio of people to rooms in at least 10% the private dwellings in the DA exceeded 1:1.

Limitations

Predominant identity refers to the most prevalent (First Nations, Métis or Inuit) among all reported Aboriginal identities that comprised at least 1% of the total population in a high-Aboriginal DA. Predominant identity does not mean that 33% or more of the DA’s population was, for example, First Nations; the correct interpretation is that 33% or more of the total population of the DA was Aboriginal, among whom the most prevalent identity was First Nations. Moreover, within predominant Aboriginal identity groups, distinct cultures and identities exist.

To designate a DA as “high-Aboriginal,” this analysis selected a threshold of 33% or more of the population reporting Aboriginal identity. However, this threshold does not necessarily apply to “predominant Aboriginal identity.” Because Inuit are concentrated in distinct geographic areas, the 33% threshold yields DAs in which more than 80% of the population was Inuit, and which together accounted for more than 80% of the Inuit population.²⁶ By contrast, the Métis are not geographically concentrated, so high-Aboriginal DAs in which Métis are the predominant identity yield DAs in which a much smaller percentage of the population was Métis and accounted for 50% of all Métis in Canada.²⁶

In 2001, census enumeration was not permitted or was interrupted before it could be completed on 30 Indian reserves and settlements, representing an estimated 34,541 individuals (Appendix A, Table E). Furthermore, the census does not collect Aboriginal identity information from people living in collective dwellings. Therefore, acute-care hospitalizations for high-Aboriginal DAs are likely underestimated.

The results of this study might have been different had it been possible to include Quebec. In 2001, around 79,000 Quebec residents self-identified as Aboriginal, representing about 8% of all people who did so (close to one-tenth of the total Aboriginal population).

This analysis pertains only to acute-care hospitalizations; other hospitalizations are not included. For example, Aboriginal populations have been shown to be disproportionately represented among psychiatric hospitalizations in Ontario.²²

Hospitalization rates for specific conditions were based on the “most responsible diagnosis” (MRDX) on each record. As a result of using the MRDX, hospitalizations of cancer patients were not counted where the most responsible diagnosis was chemotherapy treatment; therefore, rates under-represent total hospitalized cancer cases.

In 2001/2002, implementation of the 10th revision to the International Classification of Diseases (ICD-10) was underway but not complete (Appendix A, Tables F and G; Appendix B). At that time, Canadian hospitals were using one of three classification systems to code their data. For some analyses in this report, codes were converted to the ICD-9 system. This may have produced slight over- or undercounts of some diseases ranked in Tables 2 and 3 for some jurisdictions. Although coding was done with the assistance of a coding expert and in accordance with international standards, acute myocardial infarction rates may be undercounted for hospitals that used the ICD-10 system, because of different coding systems reflecting shorter timing of onset of the condition.

This report used only univariate analytical techniques. Hierarchical multivariate modeling to simultaneously adjust for both patient-level and area-level factors might have produced different results.

Values were assigned to the dichotomous area-level variables in relation to either DA or provincial/territorial benchmarks. This may have strengthened or weakened the association between any given variable and hospitalization rates, compared with what would have been the case if a national benchmark had been used.