Context
Investigators in palliative care rely heavily on routinely collected data, which carry risk of unobserved confounding and selection bias. ‘Natural experiments’ offer opportunities to generate credible causal treatment effect estimates from observational data.
Objectives
We aimed first to review studies that employed ‘natural experiments’ to evaluate palliative care, and second to consider implications for expanding use of these methods.
Methods
We searched systematically seven databases to identify studies using ‘natural experiments’ to evaluate palliative care’s effect on outcomes and costs. We searched three grey literature repositories, and hand-searched journals and prior systematic reviews. We assessed reporting using the Strengthening the Reporting of Observational Studies in Epidemiology checklist and a bespoke methodological quality tool, using two reviewers at each stage. We combined results in a narrative synthesis.
Results
We included 17 studies, which evaluated a wide range of interventions and populations. Seven studies employed a difference-in-differences design; five each used instrumental variables and interrupted time series analysis. Outcomes of interest related mostly to healthcare use. Reporting quality was variable. Most studies reported lower costs and improved outcomes associated with palliative care, but a third of utilisation and place of death evaluations found no effect.
Conclusion
Among the large number of observational studies in palliative care, a small minority have employed causal mechanisms. High-volume routine data collection, the expansion of palliative care services worldwide and recent methodological advances offer potential for increased use of ‘natural experiments’. Such studies would improve the quality of the evidence base.