Specialist palliative care improves patient experience, reduces bed days and saves money:

Less than 1% of people die in high-income countries annually but this group accounts for 8%–10% of healthcare spending.^1,2

Expenditures are primarily driven by acute hospital admissions and often yield poor value with widespread prevalence of potentially modifiable problems including unmanaged symptoms and fragmented care.^3,4

People with serious medical illness largely prefer to be at home if adequate supports are available,⁵ but most in high-income countries visit hospital in the last months of life and hospital is the most common place of death.^3,4,6

Care for serious medical illness is provided by a range of intersecting health and social care providers. Specialist palliative care – defined as care for those with more complex needs which cannot be met by their primary or ‘core’ team alone, requiring a workforce with specialist skills and experience delivering palliative care as their main role⁷ – is one intervention which may improve outcomes.^8,9 In England, less than half of people who die from a non-sudden cause get specialist palliative care.³

Economic evidence to inform improvement efforts is thin.¹⁰ Primary studies are often underpowered for economic evaluation.^10–12 There is potential for decision-modelling, combining evidence from a range of sources, to answer policy questions where primary data are insufficient.^13,14 Systematic reviews of economic evaluations of palliative care specifically identify a lack of modelling studies as a key weakness in the evidence base.^10,15

To estimate the cost-effectiveness of home- and hospital-based specialist palliative care for adults in England.

This was an economic modelling study combining data on mortality, place of death, care utilisation, unit costs, quality of life, specialist palliative care receipt and treatment effect estimates. We chose cohort models as appropriate given study resources and data availability.¹⁶ This study evaluated two models of specialist palliative care: at home, and in acute hospital. We evaluated each model of care separately as insufficient data were available on treatment effects for concurrent receipt of hospital and home specialist palliative care.

Specialist palliative care in England is delivered by multi-disciplinary teams comprising consultants, staff-grade doctors, specialist nurses and allied health professionals.⁷ Teams specifically support patients with specialist palliative care needs, including complex symptom management, psychosocial concerns, advance care planning and training and support of the wider workforce in health and social care. We compared receipt of specialist palliative care to not receiving specialist palliative care, sometimes called usual care. Since treatment effect estimates are drawn from Cochrane reviews, we interpret our main results as reflecting timely, systematic care for an individual.

People with progressive life-limiting illness may benefit from palliative care at different points in their illness.^7,17 Specialist palliative care is predominantly delivered in the last 3–6 months of life.^18,19 We defined our study population as community-dwelling adults in England, where a clinician would not be surprised if they died in the next 3 months.^18–20 We distinguished two analytic cohorts: those living at home (and so candidates for home/community palliative care), and those admitted to hospital (and so candidates for hospital palliative care). We excluded those living permanently in care homes as having different characteristics and patterns of care.²¹ For population estimates on numbers of people and places of death, the most recently identified data came from 2022.²²

Descriptive data on healthcare use and mortality are collected widely in high-income countries. Cochrane reviews found that both home- and hospital-based models of specialist palliative care significantly reduced the odds of dying in hospital.^8,9 Interventions that reduce deaths in hospital are by definition also affecting time in hospital near end of life, but cost-effectiveness implications have not been explored.

If we have good evidence on where people spend time near end of life, their place of death and to what extent specialist palliative care affects this place of death, then we can combine these data to model patient flow between different places near end of life, and estimate the associated costs and outcomes with and without receipt of timely specialist palliative care. Based on meta-analysis of trial data for people with advanced illness,²³ we assumed that specialist palliative care does not affect survival in either direction and fixed survival across treatment counterfactuals so all differences in outcome reflect different time in different places and never differential survival.

We conducted the modelling in seven steps, detailed in the Supplemental Appendix. All but the first step were undertaken separately for each analytic cohort.

First, we conceptualised a cohort Markov model with five states and a 24-h time cycle (Figure 1). The live states were defined as physical locations that adults with serious illness in England spend time: home, hospital, care home and hospice. Each of these locations has different cost and quality of life implications, and approximately 98% of deaths occur in one of these places.²² At the start of each simulation, 100% of the cohort is in the place the intervention occurs: home for home-based care, hospital for hospital-based care. From that point on, the cohort can move between places (and die in any place) as indicated by Figure 1. In the UK context for seriously-ill adults, a hospice is a facility focused on providing specialist palliative care and a care home is a facility providing residential care with expert nursing support.

Second, we identified cost and outcome parameters for each state.

Third, we specified survival curves.

Fourth, we estimated place of death, both for all relevant adults together and stratified by receipt of specialist palliative care.

Fifth, we estimated admissions to hospital, care home and hospice, and bed days in each of these places, incorporating both survival and proximity-to-death effects.

Sixth, we developed the models. We estimated transition probabilities that delivered model outcomes consistent with the place of death and healthcare utilisation estimated in the prior two steps. In estimating counterfactuals, we allowed treatment status to affect only transition probabilities from the place of care, holding all other parts of the model constant. Seventh, we combined model outputs with outcome parameters to derive results.

We sought papers measuring health-related quality of life in older people in the UK with palliative care needs, using EuroQoL EQ5D5L.²⁴ We estimated quality-adjusted life years (QALYs) and quantified treatment effects as incremental cost-effectiveness ratios (ICERs). For resource use in institutions, we combined frequency data with unit costs (UK pounds (£), 2022). To estimate the formal costs associated with being at home, we combined frequency data and unit costs for outpatient care, primary care, and district and community nursing.¹ We estimated informal care costs using the substitution method for a community-based home care worker.² Where costs were identified for a year other than 2022, we adjusted using the UK Consumer Price Index (Health).²⁵

In primary analysis, we estimated formal care costs from the payer perspective. In secondary analysis we combined formal costs with informal care costs. We estimated QALYs from the patient perspective. We combined effects on costs and QALYs to estimate ICERs, interpreting these in the context of a willingness to pay of £25,000 to £35,000 per QALY.²⁸ We imposed a lifetime horizon, with a discount rate of 3.5%.²⁹

In primary analyses, we applied the odds ratios of treatment effect on place of death and their 95% confidence intervals,^8,9 generating estimated time in each place for those receiving specialist palliative care and those not. Combining these estimates of time in place with the relevant sets of unit costs and QALY values generates an average, lower and upper limit of the estimated effect on costs and QALYs, and the associated ICER.

In our secondary analyses, we repeated the primary analyses but added unpaid family care to the cost of being at home. This analysis quantifies the extent to which observed savings for the system are displaced to the household. In sensitivity analyses, we checked robustness of our primary results to purposively identified limitations in our data: the cost of a day in acute hospital, the cost of the intervention itself and uncertain QALY effects.

To quantify the estimated effects at the population level, we combined primary results with population-level need and specialist palliative care receipt. Since our primary analysis estimates treatment effects per clinical trials, that is, care provided in a systematic and timely way from a well-defined baseline, but real-world care provision occurs in a less coherent fashion with some patients receiving timely care and others receiving care very late in the disease trajectory,¹⁸ we incorporated linear dose effects with a range from our primary analysis results (i.e. timely care) and zero (i.e. too late to have any effect).

Main results are presented in Table 3.

We estimated that, in reducing the odds of dying in hospital, formal costs were £7908 lower for home palliative care patients than usual care patients (95% CI: −18,044 to 395), while QALYs were higher by 0.035 (95% CI: 0.033 to 0.037). The corresponding ICER was −£227,781 per QALY (95% CI: −482,524 to 12,151). Total formal costs were £6480 lower for hospital palliative care patients than usual care patients (95% CI: −11,482 to −1671), while QALYs were higher by 0.033 (95% CI: 0.031 to 0.035). The corresponding ICERs were negative across the confidence interval, indicating that palliative care dominated usual care.

For both models, the estimated cost-savings come from reductions in acute hospital bed days and, to a lesser extent, reduced care home bed days. Increased costs associated with being at home are more than offset by the money saved by reduced time in hospital. QALY improvements are driven both by palliative care’s effect on outcome, and on the lower QALYs associated with being in hospital compared to being in other places.

Secondary results, where the cost variable incorporates costs from the caregiver perspective, are presented in Table 4. In both cases, overall estimated effect on costs was smaller than in primary analysis, reflecting how palliative care increases days at home versus hospital, but results were substantively unaffected. At the mean, both models of care dominated the usual care comparator with negative ICERs.

We reran our primary analyses to test assumptions to the cost of a day in acute hospital; intervention costs; and the co-occurring uncertainty of QoL effects alongside place-of-death effects. Results were substantively unaffected at the mean. See Supplemental Appendix for further details.

The population-level implications of our results, combining primary analysis estimates with the number of people estimating to be receiving specialist palliative care, for the calendar year 2022 are presented in Table 5. We estimated that home-based care resulted in 17,098 fewer deaths in hospital (95% CI: −30,681 to −2985), 1.0 million fewer bed days (95% CI: −1.9 to −244,164 million) and £533.1 million lower costs for the NHS (95% CI: −1.2 billion to 31.2 million). We estimated that hospital-based care resulted in 3906 (95% CI: 6269 to −1627) fewer deaths in hospital, 491,601 fewer bed days (95% CI: −789,873 to −205,748) and £283.9 million lower costs for the NHS (95% CI: −503.4 to −73.5 million).

Observed cost-effects were realised primarily by reduced days in acute hospital. QALY improvements arose both from palliative care improving this outcome and by higher QALYs associated with places other than hospital. Including informal care costs reduced the estimated cost-effectiveness ratio as caregiver hours increase with time spent at home, but key conclusions were substantively unaffected. Combining our primary analysis with population-level estimates of palliative care need and receipt suggests that each year in England these two models of palliative care support over 20,000 people to die outside of hospital, save approximately 1.5 million NHS bed days and reduce system expenditures by £817 million.

Causal evidence of effects on place of death and quality of life is long established,^8,9 but – to the best of our knowledge – this study is the first to examine the economic implications.¹⁰

This paper therefore represents the best available evidence of the cost-effectiveness of specialist palliative care for adults in England, and quantifies for the first time the improvements in care and associated cost-savings.

At a time when there is widespread unmet need,³⁷ and growing needs due to population ageing,³⁸ this evidence can inform planning to meet needs, curb cost growth and promote good-value care.

Since a minority of people who might benefit currently receive specialist palliative care, it is likely that in the short term expanding access would yield further cost-savings and quality improvements. However, given well-established inequities in access and outcomes,³⁹ meeting current needs also requires new approaches to identifying, engaging and meeting the needs of underserved groups.⁴⁰ In the long term, as the population ages, meeting needs will require significant expansion of the specialist workforce. It is important to consider delivery of palliative and end-of-life care by primary or ‘core’ teams, and how the range of intersecting primary and specialist providers can work most effectively together.

The challenges of providing good end-of-life care during a century of demographic ageing are not unique to England,^41,42 and other countries can adapt and extend our approach to their own contexts and data. To this end, all modelling materials are provided in the Supplemental Appendix.

The strength of this study is that it combines high-quality data from multiple sources to derive credible evidence of cost-effectiveness on interventions that previously have not been subject to widespread economic evaluation. This method harnesses the best available data on specialist palliative care through Cochrane reviews while eliding limitations of other approaches to economic evaluation such as within a clinical trial (when sample size and time horizon are typically insufficient¹¹), or observational data (which face challenges of bias⁴³ and prospective identification of a sample^44,45). Hospital palliative care dominated the usual care comparator in primary analysis. In the home-based evaluation, in the most pessimistic scenario from the formal cost perspective, expenditures increased slightly but the corresponding ICER (£12,151 per QALY) was well inside the threshold for funding.

The most important limitation is that we used aggregate, population-level data in a cohort model.¹⁶ There is good evidence that patient quality of life,⁴⁶ care costs,⁴⁷ intervention costs,²⁶ patient preferences,⁵ and treatment effects on outcomes^18,48 all vary by myriad factors including diagnosis, phase of illness, beliefs, household conditions, palliative care dose and supply-side factors. In particular, the timing of palliative care and related dose effects are central to cost-effectiveness and clinical efficacy,^49,50 and we lack a clear relationship between timing and treatment effect estimates to model that more precisely. A second key assumption is that Cochrane evidence on place-of-death effects is ‘true’ for our population; trial evidence is noted for internal validity but not always generalisability. We modelled effects on NHS costs with a sensitivity analysis incorporating informal costs, but some costs in the model are borne by third sector providers – particularly hospice and care home days.

It follows that this study should be the start and not the end of decision models that identify how and for whom specialist palliative care makes the biggest difference, addressing ex ante specific decision problems for policy, both so that services can optimise allocation of current limited capacity and plan for meeting future needs. Such analyses will optimally use individual-level data and dynamic modelling that allow for more thorough exploration of the demand- and supply-side heterogeneity. While we tested results to key hypothesised parameter uncertainties, we identified other potential uncertainties beyond the scope of our exercise that could be addressed with individual-level data (e.g. survival curves by age and sex; differences in quality-of-life trajectories by location, independent of treatment; unit costs associated with proximity to death or death in an institution; diminishing cost-effectiveness of the intervention as some people prefer not to leave hospital for home; relationship between proximity to death and treatment effect on place of death). Incorporating these uncertainties into future work is essential to appropriately meeting the needs of the more-than-half of people in England who might benefit from specialist palliative care but currently do not receive it.