In the course of 2005 and 2006 a database was collected containing all deaths of the most recent year for seven entire countries (Denmark, The Netherlands, Norway, Sweden, England, Wales, Scotland), two regions in Belgium (Flanders and Brussels, the Walloon region being left out due to a serious lacking behind in death certificate registration), and three regions in Italy (Emilia Romagna, Tuscany, Milan). All these regions have an autonomous public health policy and authority over the death certificate data.

Besides the place of death, we aimed to include a limited number of clinical, socio-demographic, residential and health care system factors, based on factors identified as relevant in the literature. We therefore drafted a typical database (table 1), based on recommendations from all participants to the study. All partners of the study negotiated a dataset maximally resembling this typical database with their national or regional administration of mortality statistics, which was to be integrated in one common European database on deaths. In case variables of the typical database were not available directly via the mortality statistics, partners needed to inquire for possibilities to combine the data register with other registers.

Characteristics of the data collection and the collected data will be described:

- procedure to obtain the data

- most recent year available and total number of deaths in the data file

- place of death information

- other variables, potentially associated with place of death, available on the death certificate data, or included via linkage of the death certificate data with other data files.

There were considerable variations in the procedure to obtain the requested data in the different countries (table 2). Approval of the project, based on a provided project description, by the agencies responsible for the death certificate data was sufficient to get the data files in Italy, Belgium, and Scotland. In the Netherlands this was also the case, but the office's data protection policy to prevent possible identification of individuals implied some restrictions in the use of the data, so that some variables could not be provided (e.g. marital status, place of residence), while aggregations needed to be made for others (e.g. age, cause of death). Data were provided relatively fast in these countries.

In other countries additional approvals were required next to those by the agencies responsible for the death certificate data: in Sweden by the National Board of Health and Welfare; in Denmark by the Danish Data Protection Agency (including an additional approval of access to micro data with a restriction to use the data only within Denmark); in England/Wales by the Micro release panel; and in Norway by the data protection agency and by the Social- and Public Health Department of the Ministry of Health.

In Denmark, Sweden, England/Wales and Norway, the time from the order of the data to the delivery exceeded (sometimes considerably) 6 months. No charges were asked for the datasets in Belgium (Flanders and Brussels), and Italy (all three regions). In the Netherlands, England/Wales and in Scotland the cost price was less than 1,000 euros, in Sweden over 2,500 euros and in Denmark over 3,500 euros.

The most recent year of the available full and error-checked databases -at the time of the initiation of the study (September 2004)- was 2003 for Flanders(BE), Brussels(BE), The Netherlands, Norway, Scotland (UK), and England/Wales (UK); 2002 for Tuscany (IT), Emilia Romagna (IT), the city of Milan (IT), and Sweden, and 2001 for Denmark.

Total number of deaths ranged from 10,108 in Brussels to 505,341 in England (table 3).

The categories of the place of death variable on the death certificate data file corresponded in most countries with the categories that could be marked on the actual death certificates (table 4). In the Netherlands it was most comprehensive, comprising the categories: hospital, psychiatric hospital, nursing home, home for older people, other institute, own home, and other. On the Belgian file, the place of death was divided in hospital, care home (which covers both nursing homes and homes for older people), home and other (subdivided in workplace, public road, or a textual specification by the physician). The Scotland data file comprised hospital (hospital, and joint user), care homes (residential homes, nursing homes, and contracturals), own home, other institution (prison, and homes), and other. The England and the Wales file distinguished hospitals, psychiatric hospitals, care homes (residential homes, and nursing homes), own home, and NHS and private hospices. In Norway hospitals and care homes were grouped in a same category on the death certificate.

However, in three countries the place of death variable on the dataset did not contain all categories that could be marked by certifying physicians on the death certificate [see Additional file 1]. The Italian datasets only made a distinction between 'home' and 'other', while the category 'hospital' from the death certificate was not recorded. The Danish data file only distinguished 'hospitals', 'institutions (but not hospitals)', and 'other', while 'home' could also be marked on the death certificate. In Sweden, place of death, while a certified variable, was not even recorded at all on the death certificate data file. However, 'hospital', 'psychiatric hospital' or 'other' could be deduced from the postcodes of the parish of death, as these institutions have their own postcodes.

In Belgium and in Italy most desired clinical and socio-demographic information was directly available via the countries' death certificate data (table 5). In other countries the clinical and socio-demographic information directly available via the death certificate data was more limited, but in several countries linkages could be made (via unique identifiers) with other population databases. The living environment of the deceased was however not available in Norway, Sweden, and Scotland. In England and Wales the living environment and the civil status of the deceased, recorded in census data, were not linked to the death certificate data due to privacy rules. The level of education could not be retrieved in The Netherlands and in the UK, but in England, Wales and Scotland the social class based on the last occupation (i.e. NS-SEC code) was available for all deaths below 75 years.

In all countries, the cause of death variable was provided as an ICD-10 (3 digits) coded variable, except in The Netherlands and in Italy, where the data protection policies called for certain aggregation. In these countries we negotiated to have 27 pre-determined aggregated cause of death categories, for which we in broad outlines followed the instruction manual by the U.S. Department of Health and Human Services 43.

Besides socio-demographic and clinical variables we also aimed to include a number of residence and healthcare system characteristics. As the municipality (or the parish, council, or local authority) of residence was available on the data files, the variables urbanisation, contextual SES-measures, and number of hospital beds per 1,000 inhabitants were operationalized by linking existing statistics to this place of residence of the deceased.

The Dutch data protection policy, however, did not allow providing us a database containing the municipality of residence of the deceased. At our request the inclusion of the residence characteristics was therefore done in advance by the Dutch Central Bureau of Statistics.

Death certificates have a long tradition as health indicator and as a monitoring tool for public health policy. A major strength is that of completeness: death certificates allow describing patterns within a whole population and not just for a sample. The issue of place of death can be studied across patient populations and across settings, which has been indicated as one of the limitations in many of the previous place of death studies26. Our study in nine countries covers more than 1.1 million deaths. This provides more statistical power, potentially leading to more reliable results, and making it is possible to use multivariable statistical models with many associated factors, or to generate meaningful results for specific subpopulations (e.g. lung-cancer or HIV patients, low educated people, specific regions) 42. As our study demonstrates, most variables on the death certificate data are available for researchers. Especially the Italian and Belgian death certificate data provided many variables besides the place of death. Fewer variables were available directly via the death certificates in other countries, but linkage with other databases made it possible to include several variables, indicated in the literature as relevant to studying place of death 151924263334364647484950. Linkage could be made with unique identification numbers to include a number of important socio-demographic variables, or via the residence of the deceased to include variables such as health care system statistics or contextual socio-economic status. In principle there is an even larger potential, via linkages with other databases, to include additional relevant information (e.g. hospitalisation and individual healthcare use). This could eventually move us further beyond place of death towards researching care at the end-of-life, and might shape good opportunities for health policy to monitor on how health care resources are allocated within each country and between countries within Europe. However, this possibility and the implications this will have on the procedure to use the data for research, needs further investigation.

Because of the comparability of death certificate data as a study method throughout time and across different countries, they facilitate reliable comparison of results. Comparison of temporal trends of place of death across nations, and placed against other mortality trends (e.g. cause of death, age, living conditions) can for example potentially allow us to monitor the impact of public health policy (e.g. implementation of palliative care services, reorganization of home care).

Finally, an advantage is that the data are relatively easy to obtain and at a relatively low cost price (range of 0 to +/-3500 € in our study).

Our study shows that there is considerable variation between countries in the type and comprehensiveness of information on the death certificate data. The place of death variable, while certified in all countries, was in several countries not or only in a limited way coded on the death certificate data files. Therefore several countries only allowed making rough distinctions between dying at home vs. other, or dying in a hospital vs. other. In The Netherlands the variable did not appear on the database before 2003; in Italy only since 2003 in more than 2 categories.

Additionally, procedures to obtain the data and the rules of data-protection were sometimes a barrier to easily get and use the data. The required permissions from different instances often complicated getting the data and integrating them into one data file, resulting in long waiting periods. The privacy and data-protection rules also limited the use of the data.

Finally, next to well known weaknesses of death certificates concerning incorrect cause of death certification, misclassification 5152 and possible country and time variation in (mis)classifications (which might be a potential impediment for comparisons between countries and over time), another weakness for studying place of death is that death certificate data do not contain all variables regarded as relevant in predicting the place of death 16465053, such as information about patients' preferred place of death, or qualitative information about the dying process (e.g. characteristics of the course of the disease, the predictability of death, the use or need of a specific therapy). This can perhaps be overcome by linking death certificates to other information (e.g. on the quality of end of life care). In itself, however, death certificate data, primarily having an administrative purpose and only secondary a research purpose, remain not well designed for non-etiological purposes, like monitoring 'good death' or quality of dying 29. The use of death certificate data reveals statistical patterns, but does not allow us to draw conclusions on the choices, behaviours, attitudes, processes, or feelings that underlie or precede these patterns 30.