Because of the problems with prognosticating who is dying, it is difficult, if not impossible, to obtain information from a representative sample of dying patients in prospective end-of-life care research [34, 35]. Therefore, a retrospective nationwide mortality follow-back survey, with data collection shortly after the patient had died, was chosen for this study.
Observational unit
General practice is highly accessible in Belgium: almost all of the population (95%), including care home residents, have a regular general practitioner (GP) who they consult regularly [36]. Hence, GPs can provide a good public health perspective on end-of-life care and dying in the country and it would be possible to generate a population-based sample of deaths via GPs. Therefore, we decided to work with a representative sample of GPs in Belgium.
A Sentinel Network of General Practitioners is a network of practices or community based physicians who monitor one, several or an exhaustive list of health problems on a regular or continuing basis. The information from these practices is used to monitor the health of the entire population [32, 37]. The Belgian Sentinel Network of GPs is a nationwide network, operational since 1979, and has proved to be a reliable surveillance system for a wide variety of health-related epidemiological data e.g. on diabetes, stroke, cancer, accidents [33, 38–43].
The GPs in the Network collect data via continuous and weekly paper-and-pencil registrations using a standardized registration form. Each registration programme lasts minimally one year. Each year, about 8 different themes are surveyed. The general objectives of a Sentinel Network of GPs are: to evaluate public health problems and their importance within the general population; to continuously observe certain health problems over time for example to study the impact of prevention campaigns; and to study the management and follow-up of health problems by the general practitioners. Recorded data must concern an important health problem not subject to surveillance of another system, unless the Sentinel Network provides complementary information to this end. The participation of the GPs is voluntary. Feedback is regularly distributed to the participants, concerned authorities, the medical press, scientific associations and interested individuals. The turnover of the GPs, from year to year, is low, which contributes to the collection of data of high scientific quality. Also, only regularly participating GPs (i.e. who register at least 26 weeks per year) are included for data analyses. The Sentinel Network is funded by the regional governments of Belgium and co-ordinated by the Scientific Institute of Public Health [44].
In 2005, the first year of this study, the Network consisted of 181 regularly participating GP practices (205 GPs), who were representative of all 10,578 GPs in Belgium in terms of age, gender and geographical distribution, and also of the GPs in the Northern (Dutch-speaking) and Southern (French-speaking) regions of Belgium separately. The Network covered 1.75% of the total Belgian population [40, 44].
Study population
The unit of measurement in this study was the death case. Primary inclusion criteria were:
In order to focus this study on care delivered at the end of life or on dying patients (i.e. patients who were theoretically able to receive care in the terminal phase of life) we additionally excluded all deaths that had occurred "suddenly and totally unexpectedly [28, 45]."
Retrospective data collection procedure
For the purpose of this study, the GPs registered deaths on a weekly standardized registration form, during 2 consecutive years (2005–2006) from January 1st until December 31st. To shorten the time between death and registration – hence preventing recall bias as much as possible – the physicians were instructed to register all deaths, immediately after being informed about the patient's death. GPs use patient records and information coming from hospital physicians as much as possible when filling in the forms. To control for possible non-response (e.g. GPs who forgot to report one of their deaths), all GPs received a summary of all reported deaths after each year of registration. GPs were asked to verify all these reported deaths and indicate if there were deaths that were not yet reported. If this was the case, we asked about basic demographic information (age, sex, place and cause of death) of the unreported deaths.
Definition of concepts
In this study, the terminal phase under consideration was defined as the last three months of life of patients that died non-suddenly [34, 46–48].
Transitions between end-of-life care settings were defined as moves or changes in location of care, during the last three months of life. Home (or with relatives, in service flats), care home (elderly/nursing home), hospital and (inpatient) palliative care unit, were differentiated. Care setting trajectories are a combination of moves between different settings.
Caregivers involved in patient care were subdivided into informal caregivers (e.g. partner, child, sibling, friends) and formal caregivers (specialized multidisciplinary palliative care services, GP, clinical specialist, nurse, caretaker/home carer, spiritual caregiver, physio-, occupational-, speech therapist).
End-of-life care was measured in terms of end-of-life care goals i.e. main treatment aim (cure, prolonging life, or palliative/comfort/supportive care), specific potentially life-prolonging and palliative treatments received, and physical, psychosocial and/or spiritual content of care [1, 2]. Other end-of-life practices measured were the absence/presence of continuous deep sedation until death of the patient, and of medical end-of-life decisions with possible life-shortening effect i.e. (1) non-treatment decisions; (2) alleviation of pain or other symptoms, in dosages which are large enough to include the hastening of death as a possible or certain side-effect; and (3) physician-assisted dying (including euthanasia, physician-assisted suicide, and life-ending without the patient's explicit request). This categorisation was based on a pre-existing internationally validated framework [27, 28, 49].
Measurement instrument
To develop the registration form for this study, we selected instruments used in other (Dutch) retrospective and quantitative studies whenever possible [17, 27, 28, 50–54]. In case a specific concept could not be measured with an existing instrument, questions were developed on the basis of relevant literature and in dialogue with the Belgian Scientific Institute of Public Health (coordinating the Sentinel Network of GPs) and with a counselling project Advisory Board consisting of 5 MDs (of which 4 GPs), a palliative care physician, 2 psychologists, a medical sociologist, a health scientist, an anthropologist and a representative from a home care nursing organisation. We also copied questions from the official death certificates in Belgium, in order to optimize comparison afterwards.
Because Belgium consists of two large regions in which Dutch or French is spoken, the instrument was first developed in Dutch and then translated into French via forward-backward procedure.
The standardized weekly registration form is shown in Additional File 1 in English (forward translation from the Dutch version). The first part A concerned several patient- and residence characteristics and was filled in for all patients who met the primary inclusion criteria. Questions from Part A based on official death certificates were Q2, Q3, Q4, Q5, Q8, Q10, Q11 and Q12. Question 17 was taken from previously designed questionnaires in Belgium and the Netherlands [27, 28, 45]. Questions designed in dialogue with the Scientific Institute of Public Health and the counselling Advisory Board were Q6, Q7, Q9, Q13, Q14, Q15 and Q16.
For all patients who had died non-suddenly, the GPs registered part B consisting of: places of care and wishes for place of death, the caregivers involved during the last three months of life, the (goals of) end-of-life care and medical decision-making. Several aspects of the medical care provided were measured using three separate time frames: last week, second to fourth week and second to third month before death. Questions from Part B based on previously designed questionnaires in Belgium and the Netherlands [17, 27, 28, 50–54] were Q3, Q4, Q5 (partly), Q6 and Q9 until Q16. Literature concerning life-sustaining therapies and indicators of aggressive and palliative care [11, 21, 55–58] additionally guided the development of Q5. Questions designed in the counselling Advisory Board were Q1, Q2, Q7 and Q8.
The quantitative registration instrument implied that all items had closed-ended response options. Also, in order to avoid the confounding effect that positive or negative connotations of several terms may or may not have (for example euthanasia or terminal sedation), we only used wording that describes the actual medical practices.
Finally, GPs were sent accompanying instructions at the beginning of each year, clearly stating the inclusion criteria of the study, and clarifying the manner in which the questions concerning cause of death and care setting trajectory needed to be filled in.
Ethical considerations
The Scientific Institute of Public Health, co-ordinating the Belgian Sentinel Network of GPs, asks the GPs to give written informed consent at the beginning of each year, after being fully informed about the objectives and method of the research themes. The registration of deaths is part of this procedure.
Furthermore, strict procedures regarding patient anonymity are employed [59]. Every patient that is registered within the network receives an anonymous reference from the GP him/herself. An independent "Trust Service Provider" is responsible for supplementary anonymisation of patient information i.e. the patient's date of birth is registered by the GP but replaced with the patient's age before data-entry, and postal code of habitual residence is transformed into more aggregate indicators such as province and region of care. Concerning the GPs' identity, all his/her identification codes are replaced in the data files with anonymous codes during data cleaning.
The protocol of the present study was approved by the Ethical Review Board of Brussels University Hospital of the Vrije Universiteit Brussel (2004).
Pilot study
The feasibility of the study design and the measurement instruments have been tested in an extensive pilot study of 3 months in 2004 [60]. The pilot study was also intended to estimate the number of non-sudden deaths that GPs would register per year, and the statistical power for the analyses of the main study in 2005 and 2006. The physicians were instructed to register all deaths of patients, aged one year or older, on a continuous basis during the months of inclusion (13 consecutive weeks), as well as the last death before this period.
In 2004, 176 GP practices reported one or more deaths. In total 502 deaths were reported, of which 466 were part of the GPs' practices. Patients who were not part of the GPs' practices (whose death was certified by the GP) were excluded. Sixty-eight percent (n = 318) of this group died during the inclusion period, and 62% (n = 214) of the latter group had died non-suddenly.
On the basis of this pilot study, the registration instrument was slightly adapted. It was estimated that 1310 patients who were part of the GPs' practices would be registered in 2005 by the 181 participating practices, and that 890 of these patients would have died non-suddenly.
Data management and plan of statistical analysis
Several control measures are used to ensure data quality and to limit missing data. Data are entered weekly using a dbase-based programme. Range and skip checks prevent key-punching errors. Automatic follow-up of missing data for key variables and a consistency check during data-entry with the possibility of contacting GPs by phone, improve data quality. Key-punch errors are corrected via double data-entry.
Data cleaning and data analyses for this study are performed using SPSS14.0 (SPPS Inc). During data cleaning, no manual changes are made. Instead, all operations are stored via SPPS syntax-files to create a working data file. Databases are stored with a central data manager at the Scientific Institute of Public Health. To answer the first two descriptive objectives of the study concerning the places of care, and medical care provided, we plan to use mainly descriptive statistics i.e. valid percentages and 95% Confidence Intervals. To study the characteristics associated with these outcome measures (objective 3), we will use multivariate logistic regression analyses.
To verify if a representative population-based sample of deaths can be obtained via this representative sample of GPs, we will evaluate whether the deaths registered by the Sentinel Network of GPs are comparable in terms of age, sex, place of death and cause of death, to the deaths occurring within the general population (using bivariate and multivariate statistics). Since there are no recent mortality statistics available for the Southern Region of Belgium, the comparison will be made only for the data from the Northern and Brussels Capital Region (66% of the total population).