Health Services and Delivery Research (Feb 2019)
The role of service factors on variations in place of death: an observational study
Abstract
Background: Previous studies have revealed that there is significant geographical variation in place of death in (PoD) England, with sociodemographic and clinical characteristics explaining ≤ 25% of this variation. Service factors, mostly modifiable, may account for some of the unexplained variation, but their role had never been evaluated systematically. Methods: A national population-based observational study in England, using National Death Registration Database (2014) linked to area-level service data from public domains, categorised by commissioning, type and capacity, location and workforce of the services, and the service use. The relationship between the service variables and PoD was evaluated using beta regression at the area level and using generalised linear mixed models at the patient level. The relative contribution of service factors at the area level was assessed using the per cent of variance explained, measured by R2. The total impact of service factors was evaluated by the area under the receiver operating characteristic curve (AUC). The independent effect of service variables was measured at the individual level by odds ratios (ORs). Results: Among the 431,735 adult deaths, hospitals were the most common PoD (47.3%), followed by care homes (23.1%), homes (22.5%) and hospices (6.1%). One-third (30.3%) of the deaths were due to cancer and two-thirds (69.7%) were due to non-cancer causes. Almost all service categories studied were associated with some of the area-level variation in PoD. Service type and capacity had the strongest link among all service categories, explaining 14.2–73.8% of the variation; service location explained 10.8–34.1% of the variation. The contribution of other service categories to PoD was inconsistent. At the individual level, service variables appeared to be more useful in predicting death in hospice than in hospital or care home, with most AUCs in the fair performance range (0.603–0.691). The independent effect of service variables on PoD was small overall, but consistent. Distance to the nearest care facility was negatively associated with death in that facility. At the Clinical Commissioning Group level, the number of hospices per 10,000 adults was associated with a higher chance of hospice death in non-cancer causes (OR 30.88, 99% confidence interval 3.46 to 275.44), but a lower chance of hospice death in cancer causes. There was evidence for an interaction effect between the service variables and sociodemographic variables on PoD. Limitations: This study was limited by data availability, particularly those specific to palliative and end-of-life care; therefore, the findings should be interpreted with caution. Data limitations were partly due to the lack of attention and investment in this area. Conclusion: A link was found between service factors and PoD. Hospice capacity was associated with hospice death in non-cancer cases. Distance to the nearest care facility was negatively correlated with the probability of a patient dying there. Effect size of the service factors was overall small, but the interactive effect between service factors and sociodemographic variables suggests that high-quality end-of-life care needs to be built on service-level configuration tailored to individuals’ circumstances. Future work: A large data gap was identified and data collection is required nationally on services relevant to palliative and end-of-life care. Future research is needed to verify the identified links between service factors and PoD. Funding: The National Institute for Health Research Health Services and Delivery Research programme.
Keywords