Data for this paper came from the British Columbia Linked Health Database. Housed at the University of British Columbia (BC), this database contains all provincial administrative health data collected by the BC Ministry of Health including physician claims, hospital separations, long-term care data, pharmacare data and vital statistics. Each of these components can be analyzed separately, as a stand-alone database, or linked through unique identification numbers and examined in combination or as a whole. The data were originally collected as part of a larger project examining the impact of regionalization in BC from 1990/91 to 1998/99 . However, as the purpose of this particular paper fell outside the objectives of the larger study, a separate request to the BC Ministry of Health for data access was submitted and approved. Ethics approval from the academic institution was also received.
Given that PC patients could not be identified in the data prior to 1992/93, for the purposes of this paper, data from fiscal years 1992/93 to 1998/99 for all databases (with the exception of vital statistics where only 1998/99 figures were available) were included. Results are based on a 100 percent sample of the population aged 50 and over residing in the Capital Regional District, on Vancouver Island, British Columbia, Canada for each of the years. Both univariate and multivariate analyses were used to determine the usefulness of administrative databases for PC research in British Columbia.
The starting point for these analyses was the identification of all patients designated as being in need of PC (it should be noted that not all persons who may require PC are identified as needing it in the databases, nor are all people who use PC services included in the databases). This was made possible through the 'service type' variable found in the direct care services database, one of many databases that make up the continuing care component of the administrative databases. Next, using the unique identification numbers that are consistent across the different databases, it is possible to link these different datasets thereby allowing for the examination of individual service utilization across the various segments of the health care system (specifically: physician claims, hospital separations, and the home support and home nursing care components of continuing care).
The continuing care database is a suite of databases, however for the purpose of this paper only the direct care services and home support databases were accessed. The direct care services databases were used in this paper for 2 purposes. First, and most importantly, it allowed us to identify those designated palliative through an individual variable. It should be noted that those never entering the continuing care system but who receive palliative services in another sector (i.e., entered the hospital, received palliative services and died in hospital) could not be identified as such and thus are not designated palliative for these analyses. Second, the direct care services database also contains information on home nursing care. For this paper, the number of home nursing visits received by each individual in a given year was calculated. Home support utilization is tracked on a monthly basis. From these files it was possible to calculate the number of hours of home support received by each individual in a given year.
All claims made by physicians are tracked in the Medical Services Plan database. It is therefore possible to calculate the number of visits to a physician made by individuals over the course of a year. The hospital separations database is so named because a hospital patient is entered into the system only when they leave the hospital (this includes deaths). Since admission and separation dates are included as variables, it is possible to calculate the number of nights each individual spends in the hospital (no overnight stay is assigned a 0). Total nights in a given year for each individual can then be summed. The vital statistics data that was accessed included underlying cause of death. This variable is comprised of ICD-9 codes that can be classified into disease categories. This procedure allowed for the identification of cancer versus non-cancer deaths used in the multivariate analysis.
Each of these health databases includes a small number of demographic variables such as age and gender, and geographic identifiers such as health authority and census tract. Marital status is included in one of the continuing care databases. This lack of what are typically referred to as 'control variables' is one of the biggest limitations of administrative databases. To compensate, it is possible to link Census variables at an aggregate level to the administrative data. Of course this procedure only works for variables that lend themselves to averages. For example, it is possible to calculate the average household income for an area but it is not possible to calculate the average gender for an area. Using this premise, average household income was linked to the administrative data at the enumeration area level (the smallest geographic unit released by Statistics Canada and the BC Ministry of Health). In brief, the average household income for an enumeration area is assigned to an individual residing anywhere within the enumeration area. Thus, income is an aggregate measure while the remainder of the variables used in this paper are individual measures.
Finally, as mentioned earlier, each of these databases is linkable. The term linkable is used to describe the assignment of the same unique identification number to an individual regardless of database. In other words, Person A will be assigned the same number in the hospital database and the continuing care database. As such, linkable databases allow researchers to examine health service utilization and health status of the same individual across sectors. Without the ability to link, it would be impossible, for example, to examine underlying cause of death (vital statistics database) in relation to the designation of palliative care (continuing care database).
As described in the preceding paragraphs, 5 health service utilization measures were calculated: number of general practitioner visits per year; number of medical specialist visits per year; number of nights spent in hospital per year; number of hours of home support received per month; and number of home nursing care visits received per month. Next, using these variables, three different analyses were conducted:
1. Descriptive statistics (distribution, mean, median) were used to examine PC patient characteristics and health service utilization by gender. All those identified as being in need of PC for each fiscal year were examined by age, gender, income, and the 5 health service utilization measures. The sample size of PC patients ranged from a low of 74 in 1992/93 to a high of 568 in 1997/98.
2. Because service use is likely to differ by diagnosis, we compared patients designated as being in need of PC with patients who were not designated as in need of PC in order to control for diagnosis. Previous research indicates that the majority of persons receiving specialized palliative and end of life care have a cancer diagnosis . While it is acknowledged that different types of cancer place different demands on the health care system, for the purposes of this study, all cancers were examined in combination. Using the underlying cause of death code from vital statistics available for 1998/99, all those who died of cancer in 1998/99 (n = 2,734) were identified. Next, comparisons involving age, gender, income, and the 5 health service utilization measures were then made between cancer patients designated palliative (n = 119) and cancer patients not designated palliative (n = 2,615).
3. To examine the influence of a PC designation on health service utilization, five multiple linear regression models were estimated using the 1998/99 data from the palliative and non-palliative sample (n = 2,734) described above. The following five dependent variables were regressed on age, gender, average household income, and designation of palliative/non-palliative: (1) number of general practitioner visits; (2) number of medical specialist visits; (3) number of nights spent in the hospital; (4) number of hours of home support; and (5) number of home nursing care visits. Prior to conducting the multivariate analyses, assumptions of normality, linearity, and collinearity were tested and adjustments were made where necessary.