This study represents a further investigation of transitions experienced by all patients registered in a comprehensive palliative care program (PCP) at the Queen Elizabeth II Health Services Center (QEII) in Halifax, Nova Scotia, Canada between January 1, 1998 and December 31, 2002 with a recorded date of death on or prior to December 31, 2002. Details of the initial investigation describing the distribution of transitions in care among these patients have been previously published [6].
The research ethics board of the Nova Scotia Capital District Health Authority, Halifax, Nova Scotia provided ethical approval for this research. Personal identifying information was not included on the records used for this project. All patient information available was provided anonymously to preserve confidentiality.
Data
Data for this study were obtained from the QEII PCP database, augmented with aggregate Statistics Canada Census (2001) information. During the five year study period, 4434 patients were admitted to the program; 90% with a cancer diagnosis. The number of patients admitted to the program each year remained relatively stable (mean, 887; standard deviation [SD] 38). Information contained within the PCP database is considered to be valid and reliable. Data are entered by a trained entry clerk who is dedicated to ensure the completeness and accuracy of all information. The data entry program contains a number of checks to reduce the incidence of missing and incorrect information. For instance, where feasible, new information is automatically validated against existing data, fields are made mandatory to force data entry and formats forced for particular fields. In addition, a series of reports are generated at regular intervals identifying what information is missing. Collection of this missing information is immediately sought to complete the record; hence few records include missing data.
For each subject, demographic information (gender, date of birth, date of death, postal code), diagnoses, caregiver relationship (for example, spouse, daughter, son, friend), primary reason for referral (pain, other symptom management, patient or family support, staff support, home consultation), location of death and program transition data were extracted from the PCP database. For each transition, the date, location of care (for example, home, acute care, long-term care) and an indicator of the services provided were included. These services could involve inpatient hospital care in the Palliative Care Unit, medical or surgical services from other acute inpatient care units, outpatient care provided by the Nova Scotia Cancer Center (NSCC) PCP clinic, PCP Home Support Service, care in a long-term care facility or care by a family physician.
Because individual level income information was not available from the PCP database, an aggregate measure of income derived from census information and grouped by enumeration area or 'neighbourhood' was employed. These 'neighbourhood income quintiles' were linked to the PCP database information via the postal code. Similarly, a geographical indicator of urban or rural residency, based on population density, was also generated and linked to the file. Prior to data linkage and release of the data for analysis, all personal identifying information was removed.
Measures
We define a transition as either: 1) a change in location of where the patient was cared for by the PCP; or 2) a change in which service provided care. For example, a transition may be a move to or from the home, a specific acute care unit or a long-term care facility. A transition would also occur if the patient stayed in a single location, for example at home, but the care being provided is transferred from PCP staff (active care by PCP staff) to NSCC staff doctors (without PCP involvement), or to their family physician or home care nurse (non-active, no longer cared for by PCP staff) or vice versa. This change in service provider or 'transfer of care' transition scenario is illustrated below:
Home → Home
(Active: care by PCP staff) → (Non-active: care by family physician)
Each patient's diagnosis was categorized within one of six cancer groupings or as having problems other than cancer such as circulatory or respiratory disease. Site specific cancer groupings included the lung, colorectal, female breast, prostate, lymphatic and hematopietic tissue and all other cancers. Four locations of death were defined: in the hospital (excluding the inpatient PCP unit), within the inpatient PCP unit, at home and death occurring within a long-term care facility. Because the total time a patient was enrolled in the PCP varied, a 'survival' variable was created which was defined as the number of days between date of the initial admission to the PCP and death.
Analysis
The total number and location of transitions experienced by patients from their initial admission to the PCP and during the final two and four weeks of life period were counted and summary statistics provided. Next, patient characteristics were examined, categorized and descriptive statistics presented.
Due to the highly skewed distribution of the total number of transitions experienced by patients admitted to the PCP, negative binomial regression was used to examine the association between total transitions and patient characteristics. To account for the variation in length of survival time, the time between each patient's date of enrollment and death, the log of survival was incorporated as an 'offset' in each regression model. Univariate or unadjusted regression analysis was followed by multivariate regression where patient characteristics found to be significantly associated with the total number of transitions, at the 0.05 level of significance in the unadjusted analysis were included in the initial model. Manual backwards elimination methods were then used to develop the most parsimonious model of total number of transitions and patient characteristics. All patient characteristics retained in the final multivariate model were significant at the 0.05 level. Regression coefficients are exponentiated and reported as rate ratios (RR). Similarly, the association between patient characteristics and the number of transitions occurring within the four and two week periods prior to death was examined. All analyses were conducted using SAS software [7].