We found death occurred in Switzerland most frequently in a hospital (38.4% of all cases) followed by nursing homes (35.1%) and dying at home (26.6%), which is generally consistent with other published data . Nearly 26% of men died at home versus 20.9% for women. However, 45% of women died in a nursing home compared to 23.6% of men. We suppose that this fact could mainly be attributed to the higher life expectancy of women. Men can often live longer at home due to the presence of a usually younger partner. They tend to die in hospital after an acute deterioration of their health state. Females spend their last phase of life more frequently without a partner in a nursing home. Additionally, we explored the association of place of death and residential region. People in French- or Italian- speaking cantons appeared to pass away to a greater extent in hospitals (43.4%) than the Swiss average (nursing home 32.9% and home 23.8%). Fischer et al.  investigated the place of death in 2001 only in the German speaking part of Switzerland. In comparison with their results, our study found a trend towards dying at home and a shift within institutions from hospitals to nursing homes in the German-speaking part of the country. Similarly, we observed other significant regional differences between age at death among Swiss cantons. This finding is consistent with official data from the SFSO. Further research is needed to determine the extent to which these regional differences are due to patient preferences, supplier-induced demand, differential access to medical services or other factors.
There were substantial and significant differences in the descriptive comparison of health care utilization in the last six months of life between places of death. The mean number of consultations with a general practitioner or specialist as well as the number of different medications and, naturally, the number of days in hospital was consistently highest for the decedents in hospitals. This result can be expected given the generally higher burden of severe illnesses suffered by hospitalized patients. On the other hand, the mean number of consultations with a general practitioner as well as a specialist physician for decedents in nursing homes might not be visible in health insurance data due to aggregated claims data. Therefore, these estimates are possibly biased.
The mean number of different medications consumed by decedents is comparably high in our sample and there is considerable variation, mainly due to people who had no medication at all. Differences in this variable arise when splitting the sample by place of death. In nursing homes we observed an unexpectedly high percentage of people without any medication (24.5%). This number of decedents without any medication might be overestimated as we do not possess any information on medication prescribed in nursing homes.
The majority of people had at least one stay in hospital (60.8%) in the last six months of life and the average length of stay was 16.9 days. However, the differences between the places of death are highly influenced by the number of persons having no stay at all. When restricting the analysis to persons with at least one stay, the mean LOS appeared similar across places of death (27.7 days). The significant duration of hospital stays (25.2 days) for those prior to dying at home was notable. This could point towards the general preference of home as place of death found in prior research [41, 42].
Concomitantly and consistent with high expenditures for hospital stays, last six-month HCE are significantly affected by place of death. The mean HCE for hospital deaths, at CHF 23,193.70, is more than twice the mean amount for those dying at home (CHF 11,194.30) and 40% greater than the mean amount for nursing homes (CHF 16,579.0).
Our further analyses are in line with Kelley et al.  and show that a substantial portion of the previously described variations in place of death are due to patient-level characteristics. Notably, the proportion of variation explained by our model after patient characteristics were controlled for is however larger than the results presented by Kelley et al.. Place of death is significantly associated with age, sex, region and multi-morbidity. Elderly females have a greater probability of dying in a nursing home, whereas a young male would preferably die at home. Additionally, a decedent living in a rural area will presumably die at home and people in urban areas either in a hospital or nursing home. Persons with multiple chronic conditions have a greater chance of dying in an institution than at home. These findings are in line with earlier studies [14, 36, 43].
Strengths and limitations
This study has several strengths. To our knowledge, it is the first empirical investigation, which describes the conditions and health care utilization in the last six months of life with regard to place of death in Switzerland. We use health insurance claims data, which guarantees a uniform data set and presents an ideal basis for the analysis. This study has focused primarily on differences in place of death and health care utilization at the end-of-life, but we cannot comment on the appropriateness of different patterns of care delivered to decedents. In addition, our analysis is based on a period of five consecutive years, which allows us to capture any time effects as well as to reduce standard errors on the interesting variables due to a larger sample.
It is also important to point out to some limitations of our study. From previous unpublished research, we estimate that 2-3% of all claims invoices are paid directly by the patient (e.g. due to high deductibles chosen) and are not reimbursed by the health insurer. This may lead to a possible bias due to a mixture of the different effects in the estimations and missing claims data. Furthermore, we focused on all cases within the mandatory health insurance in Switzerland, which omits people dying from an accident or committing suicide, as these cases might be covered by other insurance policies and the health insurer does not see any referral claims. The claims coming from nursing homes are often set at a flat rate and lack further detailed information. This fact prevented us from observing any medication or medical treatment applied in the nursing home. Therefore, we suppose a slight tendency to underestimate all variables describing health care utilization for people residing in nursing homes. This restriction does not notably affect treatment outside nursing homes (consultations with GPs, outpatient units, pharmacies etc.). To conclude, the place of death is specified by means of the last claim received. The claims contain the date and the duration of the specific treatment, which allows us to compare it with the date of death. The origin of the last claim then defines the place of death. If there are several claims covering the date of death, hospital is taken as place of death. This process makes our data vulnerable to inaccuracies resulting from administrative processes and might lead to a possible overestimation of hospital deaths.