The results mentioned in this section were obtained via queries with the two research tools, mainly the causal probabilistic model. Various scenarios of daily clinical practice were simulated by testing the impact of the variables and their different states on a target variable. The output of the model was calculated in the form of probability intervals with lower and upper values. Therefore, some of the results are expressed ranges. The remainder of the results are expressed as percentages, which were calculated based on differences in the lower values of the probability interval since a higher prevalence is attributed to variations in the lower value. The results from these queries are detailed in Table 4 and Table 5 in Additional file 5.
Results from the causal probabilistic model
Dependencies of disease and treatment-related variables
Among illness-related variables, the following dependencies were observed. As long as effective anticancer treatment resources are available, these resources are most likely to be used to treat metastatic cancer until near the EOL, with a 64-80% likelihood of use for this purpose, with the goal of palliating symptoms and improving survival. Ongoing cancer treatment prevents or delays communication about EOL issues between oncologists, patients and families. Ongoing cancer treatment decreases the probability of open communication by 40% compared with that of treatment discontinuation. In the investigated study population, the last chemotherapy was administered at a median of 34 days before death. Of the patients, 60 to 80% were probably only partially informed about the proximity of death when undergoing anticancer treatments. Thus, patients and their relatives are more likely to remain in a closed rather than open state of awareness of dying. When patients are not aware of the terminal stage of the disease, they probably do not communicate about their preferences for the place of care and POD. Accordingly, assessing preferences for the EOL period is difficult when patients and families are not ready to discuss the subject. In the current study, these preferences were unknown among approximately two-thirds of the 116 patients at a median of 38 days prior to death. Eighty-seven percent of these patients died in a hospital.
Communication between health care professionals, patients and families about the imminence of death shows a positive impact on the awareness context of 40%. Most patients and their relatives (60-80%) shift from a closed to an open state of awareness of dying following open communication.
In addition to the influence of ongoing anticancer treatment on communication and the awareness context, other aspects must be taken into account. As long as palliative chemotherapy is continued, the treatment of toxicities and their consequences require additional supportive care, which leads to higher rates of hospital admissions and consequently to more days spent in a hospital. The more days patients spend in a hospital, the more likely they are to die in that setting. The number of hospital days is also influenced by the symptom burden. Severe symptoms increase the probability of prolonged hospital stays. A large proportion of cancer patients suffer from a high symptom burden or unstable clinical conditions during the last weeks of life, requiring inpatient care as long as the treatment approach is active.
The interplay of these disease and treatment-related variables shows an influence on the home death rate of 13%.
Dependencies of patient and family preferences
For patients living at home, preferences for home as the place of care (37-63%) are slightly higher than preferences for the hospital (28-53%). Patients resident in a nursing home prefer to be cared for in that setting (57-67%), followed by home (23-31%) and then the hospital (8-14%). Family caregivers’ preferences contrast the wishes of patients, especially in urban areas and agglomerations, where family caregivers clearly prefer hospitals (48-82%) versus homes (16-49%). In rural areas, there is almost no difference in preferences for home (35-60%) versus the hospital (37-62%). The family’s preference shows the highest overall impact on POD at 51%, which is much higher than the impact of the patient’s preference, at 14%. The family’s preference is strongly dominant over the patient’s preferences. A significant congruence between the family’s preference and POD can be observed, independent of the patient’s preference: when the family’s preference is home, the probability of a home death is high (53-70%); when the family’s preference is the hospital, the patient is more likely to die in a hospital (59-87%); and when the family’s preference is a nursing home, the patient is more likely to die in a nursing home (50-79). The best probability for home death is subject to a preference for home care that is expressed by both patients and family caregivers (64-76%).
Dependencies of family-related variables
The family’s preference for the last place of care is mostly influenced by the family system’s conditions for home care. The probability of suitable conditions is best (80-100%) when at least one person from the family is available to assist with home care, when the family members have somewhat solid emotional relationships, when the family members are fully informed about the patient’s approaching death, when the family members have an open state of awareness, and when the patient has a low degree of dependence. Among these variables, the relevance of the state of awareness seems to be important. In a state of closed awareness, the suitability of the family’s conditions for home care drops from 80-100% to 60%. An even more obvious trend becomes apparent if no family member is available for home care. Then, the probability of suitable conditions decreases from 80-100% to 20-40%. An analogous decline from 80-100% to 40% is observable in the family's availability for home care when the patient suffers from severe symptoms, has a high degree of dependence and is in need of continuous help and support.
Economic resources seem to play a role only for families without their own resources in place for home care since the families then must pay for expensive assistance from professional carers. In this event, economic resources show an impact of 20% on the family’s preference. These circumstances might be country specific. In the current study population, continuous assistance provided by health professionals or care workers with migrant backgrounds is not provided for by law and is therefore not covered by health insurance. Only three of the 116 families in the study had chosen this solution.
The interplay of all family-related variables, excluding the family’s preference, demonstrates a final impact on POD of 13%. However, a clear prediction for home death cannot be made as long as the family’s preference is unknown.
Dependencies of care network and policy-related variables
To support the design of policies, the impacts of the following potentially modifiable variables were tested. The use of specialist palliative home care services, access to home visits by GPs, access to home care nurses, access to volunteer hospice services and coverage of home care costs by health insurance reflect the effect of an interdisciplinary home care network and health care policy. The interplay of this group of modifiable variables demonstrates an impact on the home death rate of only 6% when patient and family preferences are unknown. In contrast, in the subgroup of patients and families with a congruent preference for home care, an interdisciplinary home care network can consistently increase the probability of home death by at least 58%. These patients actually have the highest probability of dying at home overall, at 76-83%. Without the use of specialist palliative home care services, the chance of dying at home decreases by 24%, but the likelihood of a home death is still higher than that of a hospital death, at 51-59% versus 41-50%, respectively. No difference is observed in the impact on the likelihood of a home death between routine and continuous professional home care. Continuous professional assistance does not seem to be more effective than routine home care.
As shown before, if the family’s preference for the POD is the hospital, despite full access to an interdisciplinary home care network, the probability of dying at home drops significantly by 57%, from 76-83% to 19-40%. This finding confirms the strong impact of the family’s preference on POD and leads to the hypothesis that if interventions by the interdisciplinary home care network could influence the family’s choice for the place of care, the home death rate would increase significantly. However, these variables reveal a low impact on the family’s preference of only 4%. As a result, the family decision-making process depends neither on the home care network nor on the health care policy but rather mainly on the family conditions for home care.
In the search for hypothetical interventions to better support family conditions, open EOL communication that improves the state of awareness and good symptom control can be identified as an option. The impact of such an intervention demonstrates an increased probability of suitable conditions for home care by 35%, yielding a final influence on the home death rate of 17%.
Results from the classifier
The results from the classifier mainly confirm the results obtained from the causal model. A high probability of home death (97%) is observed for patients who live in a rural environment, have a low symptom burden, have spent few days in a hospital, have an open awareness of dying, have a suitable family system for home care, have congruent preferences for home care with their families, receive home care assistance provided by a home care service, have access to GP home visits and have access to specialist palliative home care. In urban areas, the probability of death at home drops slightly to 82%. Without the involvement of a specialist palliative home care service, the probability decreases significantly to 46%. When severe symptoms set in, even though all other conditions appear favourable, the probability of home death drops from 82% to 59%. In the case of a long hospital stay, the chance of death at home is only 23%. This tendency becomes even more apparent when the family’s awareness of dying is closed. Then, the probability of home death decreases to just 13%.