Background

The use of health economics in palliative and end-of-life care has been slow to develop and the cost-effectiveness evidence base remains small. There is a risk that the absence of robust evidence on the cost-effectiveness of palliative care is interpreted as evidence that such services and interventions are poor value for money, and the lack of an evidence base can encourage simplistic analysis based on naive interpretation of the limited data.

There is relatively little record of the use of economic evaluation evidence in decisions on the development and financing of palliative care services. There are several reasons why this is the case. In many countries, palliative care is poorly developed and much of the funding has come from private and voluntary sources (Martin-Moreno et al., 2008). Without a substantial role for public finance in the development of palliative care services there may be no perceived need for formal analysis of costs and benefits, and many voluntary organisations develop in the clear belief that the services they promote are needed. A more practical reason why economic evaluation in this field is poorly developed is that focus on palliative care as a specialist area in medicine and nursing is quite recent; in many countries, services are poorly developed (Martin-Moreno et al., 2008) and the numbers of researchers with an interest in the field remain small.

Part of the explanation is also that there are (and in addition are perceived to be) some particular difficulties in applying economic evaluation techniques in palliative care and in the end-of-life setting. It can be difficult to assess costs where many interventions work through other professionals and may support large inputs from informal and family carers. But the main difficulties are in valuing the benefits (Douglas et al., Reference Douglas, Normand, Higginson and Goodwin2005; Gomes et al., Reference Gomes, Harding, Foley and Higginson2009; Normand, Reference Normand2009). Zimmermann et al. (Reference Zimmermann, Riechelmann, Krzyzanowska, Rodin and Tannock2008) comment that ‘evidence for benefit from specialized palliative care is sparse and limited by methodological shortcomings’. They attribute this in part to the difficulties in carrying out clinical trials. Clinical trials typically assess standardised interventions, whereas in many cases palliative interventions are personalised and may vary greatly between patients. Clinicians may not consider it ethical to enter patients into studies where they have a random chance of getting a standardised treatment rather than getting a package of care designed around their complex needs. Zimmerman et al. (2008) comment that some clinical trials in palliative care evaluate oversimplified versions of palliative interventions. They also point to other shortcomings of the studies reviewed in terms of their assessment of costs of specialist palliative care services.

Some authors argue that it is feasible to conduct methodologically robust studies including randomised controlled trials in palliative care to gather robust evidence on outcomes (Raftery et al., Reference Raftery, Addington-Hall, MacDonald, Anderson, Bland, Chamberlain and Freeling1996; Mitchell et al., Reference Mitchell and Abernethy2005). Randomisation may be hard to achieve where clinicians consider it to be unethical, but even when other suitable studies designs (Black, Reference Black1996; Motheral et al., Reference Motheral, Brooks, Clark, Crown, Davey, Hutchins, Martin and Stang2003) are chosen there can be important challenges. The comparison of new models of care (that are only partly specified) with existing standard care (that is only partly specified) makes it hard to be clear what is being compared with what.

To an extent, this debate has missed the point in that clearly some palliative interventions can reasonably and ethically be evaluated in the context of randomised trials, but some of the interesting difficulties are not related to the overall study design, but come in measurement and in comparing interventions that are patient centred and individual. Although these problems are not unique to palliative care, they are very common in that sector. Teasing out the issues in this context may provide some insight into how they may be distorting the findings of studies in other areas.

Some costing studies in palliative care ignore the costs that fall on families, which can be very important (Zapart et al., Reference Zapart, Kenny, Hall, Servis and Wiley2007) and which might favour care patterns that shift costs from formal to informal providers. There are other interesting issues concerning the inclusion of preferences around processes and more conventional outcomes, and research has shown that including the value of process changes the ranking of options (McNamee et al., Reference McNamee and Seymour2008).

There is a difficulty in specifying exactly what should be included in the analysis, as the role and scope of palliative care services can be difficult to define. Palliative care professionals often work partly through other professionals and in co-ordination and integration of other services (Higginson et al., Reference Higginson, Hearn, Myers, Naysmith, Baxter, Douglas, Eaders, Francis, Goodwin, Harvey, Jones, Kennet, Lonstaffe, Luchterhand, Normand, Tookman and Wilson1999; Douglas et al., Reference Douglas, Higginson, Myers and Normand2000; Skilbeck et al., Reference Skilbeck, Corner, Bath, Beech, Clark, Hughes, Douglas, Halliday, Haviland, Marples, Normand, Seymore and Webb2002). It can be quite difficult to identify exactly what the palliative care professionals actually do (Higginson et al., Reference Higginson, Hearn, Myers, Naysmith, Baxter, Douglas, Eaders, Francis, Goodwin, Harvey, Jones, Kennet, Lonstaffe, Luchterhand, Normand, Tookman and Wilson1999; Douglas et al., Reference Douglas, Higginson, Myers and Normand2000), and to be sure if their inputs are linked to any observed differences in outcomes (Higginson et al., Reference Higginson, Finlay, Goodwin, Hood, Edwards, Cook, Douglas and Normand2003). This is important if the study intervention is to be replicated and introduced in normal clinical settings.

There is a growing literature on the costs of palliative care interventions and in particular on how such interventions can reduce other costs (Johnson et al., Reference Johnson, Abernathy, Howell, Brazil and Scott2009). Some studies also include limited measures of patient outcome or experience (Fassbender et al., Reference Fassbender, Fainsinger, Brenneis, Brown, Braun and Jacobs2005; Brumley et al., Reference Brumley, Enguidanos, Jamison, Seitz, Morgenstern, Saito, McIlwane, Hillary and Gonzalez2007; Tamir et al., Reference Tamir, Singer and Shvartzman2007; Taylor et al., Reference Taylor, Ostermann, Van Houtven, Tulsky and Steinhauser2007; Hanson et al., Reference Hanson, Usher, Spragens and Bernard2008). There are no published studies that have reported broader estimated costs (including costs to families and informal carers) or outcomes and experiences in detail. Results generally show that outcomes are better or no worse, and that costs are lower for the palliative interventions studied. It could be argued that in the absence of evidence that palliative care produces worse outcomes, evidence of lower costs provides a strong prima facie case for choosing such services. However, there is a need for more robust evidence to demonstrate that outcomes are no worse.

Measurement and evaluation in palliative care

A key function of economic evaluation is to make the different possible uses of resources comparable within a common framework. Standard ways of carrying out and presenting economic evaluation in health care have emerged, with the typical use of incremental cost-effectiveness ratios and cost-effectiveness acceptability curves (Briggs et al., Reference Briggs, Sculpher and Claxton2006; Gray et al., Reference Gray, Clarke, Wolstenholme and Wordsworth2011). Comparability requires that the metrics used are as similar as possible (Normand, Reference Normand2009). In the case of palliative interventions, this causes a serious problem, as the most commonly used metric in economic evaluation is the quality-adjusted life year (QALY). This is derived by measuring the expected additional years, weighted for health-related quality of life. It is common for HTA agencies to use a threshold for how high a cost is acceptable to justify an additional QALY – normally in the range from $30,000 to $60,000.

A problem in palliative interventions is that there is evidence that most people would consider that some interventions are cost-effective despite clearly failing on this type of rule (Normand, Reference Normand2009), a version of what has been described as the QALY Problem (Hughes, Reference Hughes2005). The choices for policy makers and analysts are to reject this evidence as simply the consequences of limited understanding and rationality on the part of the public or to look with more care at what may be wrong with the currently used measurement tools when we move into the domains of complex interventions. Additionally, applying the tools in difficult circumstances may reveal more widespread problems with the tools. They may not be fully fit for purpose even when their application produces plausible results and does not raise questions about their fitness. This article focuses on two issues – the problem of context and the problem of assuming that time is additive. It also looks briefly at the problem that preferences in palliative care seem to be nearly lexicographic and do not behave as economists would wish. These issues are probably related, especially in the sense that context includes difficulty of specifying when the interventions take place (or at least when they are perceived to take place), how this shapes preferences and how these appear not to follow the axioms of economics.

There is an important sense in which end-of-life care is not an option. We have the choice to operate or not to operate to help those with cataract or hip problems, and there is genuinely an option to do nothing. In end-of-life care (assuming we are not totally callous), the issue is more whether we do the right things and do them in the right ways rather than whether we do them at all (Douglas et al., Reference Douglas, Normand, Higginson and Goodwin2005). In the United States, in many cases there is a formal point at which palliative care begins, given the rules of the Medicare Hospice Benefit, but most people would consider that the dividing line between curative and palliative intent should be blurred, and in some senses it should not be possible to define a time when we enter the end-of-life phase. In addition, there is not a clear date when the benefits of accessing palliative care begin, as a part of the benefit is knowing it will begin when needed. Anticipation of well-managed support and symptom control at the end of life may be enjoyed over a very long period before that. Again this may be important in other areas of care where the timing of availability of services is critical, such as access to emergency services for heart disease or following accidents.

It is also more difficult to assess the benefits of palliative care when some interventions inevitably have both palliative and curative intent – for example, the use of bisphosphonates may be justified for their management of hypercalcaemia and its related symptoms, but has the associated effect of extending life by a small amount (Ross et al., Reference Ross, Saunders, Edmonds, Patel, Wonderling, Normand and Broadley2004). Overall, these issues mean that the services we are interested in evaluating may be complex, interwoven with other services, enjoyed in anticipation (or in retrospect), as well as at the time of delivery and may be ones where priorities are very strong for certain aspects such as timely and effective access to symptom control (Douglas et al., Reference Douglas, Normand, Higginson and Goodwin2005).

Another important issue in measurement in end-of-life care evaluation is the duration of benefits. Although it is argued above that it is often not useful to define the starting point of palliative care, it is typical for the interventions to be offered over a short period, and for the direct effect to be enjoyed over a short time. Most studies in epidemiology and HTA assume that time is effectively cardinal and additive. Years of life or QALYs gained can be added. Even when the costs of palliative care are quite modest, the consequence will often be that palliative interventions appear not to be cost-effective as there is not enough time for them to generate QALYs (Normand, Reference Normand2009). Although it is obviously useful to assign cardinal properties to outcome measures that have time as at least one dimension, there is no good evidence that this is justified. In the same way that the theoretical difficulties raised by Arrow's General Possibility Theorem (Arrow, Reference Arrow1950) are typically noted by economists and then ignored, it is common in economic evaluation to impose both interpersonal comparability of utilities and cardinal properties of time without explicit questioning the validity of doing so. There is an attractive apparent lack of discrimination in making these assumptions, but this may be undermined if such assumptions are imposed but lead to systematic undervaluing of some benefits and overvaluing of others.

There are ideas from psychology that may offer some insight into the ways in which people behave, which suggest that time is not cardinal and not additive when they assess the impact of events. The purpose of this discussion is not to replace in all circumstances the assumptions of cardinal and additive time with specific alternative assumptions, but rather to demonstrate that there are circumstances where additivity seems not to apply. Therefore, the focus is on one well-established framework where empirical studies provide evidence for rejecting cardinality and additivity.

The idea of the Peak End Rule (Kahneman, Reference Kahneman1999) is that the value gained from an experience depends crucially on its best (or worst) level and its level at the end of the experience. For example, the enjoyment of a holiday may depend on the best days and the final days, but has been shown not to be necessarily associated with the overall length of the holiday or day-to-day enjoyment (Kemp et al., Reference Kemp, Burt and Furneaux2008). In end-of-life care, there is a strong focus of users of services on avoiding bad experiences. Douglas et al. (Reference Douglas, Normand, Higginson and Goodwin2005) found near lexicographic preferences in choice of service mix. In the case of palliative day care people were unwilling to make comparisons between bundles of services with and those without adequate access to symptom control regardless of the volume and quality of other services offered, but once this access to symptom control was available they demonstrated strong interest in the additional services on offer. Such patterns of preferences are difficult to reconcile with the utility maximisation objective that underlies economic evaluations.

There is also a strong tradition in palliative and end of life care of wanting a good end. A study of patient preferences in end of life care (Douglas et al., Reference Douglas, Normand, Higginson and Goodwin2005) suggests that the Peak End Rule may fit better than the additivity of time rule that would usually be applied in HTA. The Peak End Rule also helps to explain why people do not seem interested in trading off between different elements of care at least until the key issues of managing the peak and the end have been addressed. There may of course be other frameworks that work better in end of life care than the Peak End Rule, but at this stage the argument is simply that we can do better than to assume time is additive.

There are some practical implications of accepting the importance of avoiding crises in symptom management and focusing on ensuring a good end to life for how we might approach the evaluation of palliative care. If we cannot simply add up cardinal units of time of varying quality to assess outcomes or benefits, it may be better to compare trajectories of experience as a whole, seeking evidence of preferences between these trajectories. Reluctance simply to trade-off units of quality-adjusted time (e.g. if preferences are lexicographic) is less of a problem when the comparison is between pathways to death. Longer trajectories may or may not be preferred to shorter ones (ceteris paribus they probably are), but comparison of different pathways to death does not require that these are disaggregated into time segments. Options within end of life care can be quite feasibly compared in this sort of framework, with preferences for being assessed for different costs and different trajectories. There may be a need to define the pathways as including not only use of services but also availability of services, as this has been shown to be valued.

More difficult problems arise in comparing end of life trajectories with other treatment and care, as it is typical to produce economic evaluation statistics that are based on additive time, even when the utility measures used to generate QALYs take account of condition-specific outcome measures and allow the potential for utility scores to be informed by context (Petrillo and Cairns, Reference Petrillo and Cairns2008). However, in principle it should be possible to set up experiments that would allow comparison of alternative trajectories in palliative care and other treatment comparisons. It may also emerge from such exercises that the assumption of additive time in estimating benefits should be questioned in other circumstances.

Some concluding comments

As the programmes of evaluation and priority setting in end of life care develop, it is becoming more important to have tools and approaches to measurement that are fit for purpose. It may be more feasible in the short run to develop outcome tools that are useful in setting priorities within palliative care, and then to try to calibrate these against the metrics commonly used in other contexts. If it can be shown that many conventional metrics used in economic evaluation are not well suited to end of life care settings, then it is not helpful to ignore end of life care evidence because it fails to show effects that are unlikely ever to be shown using unsuitable tools. Living with some heterogeneity in the evidence used for policy choices may be less serious than attempting to fit all evaluation activity into frameworks that are systematically flawed in some contexts. In practice, priority setting is seldom carried out without some elements of judgment being used, and in this case the judgments may need to be made in making comparisons of evidence that uses different measurements and metrics. This can be helped using approaches and ideas discussed by Petrillo and Cairns (Reference Petrillo and Cairns2008), where the approach is to make comparison between outcomes of different metrics rather than to use a single set of metrics in evaluation.

It is likely that failure to take account of the particular complexities in outcome measurement in palliative care will have two undesirable effects – first, it will discourage efforts to develop cost-effectiveness evidence in this domain, and second it will perpetuate the risk that failure to demonstrate cost-effectiveness will be interpreted as evidence that such interventions are not cost-effective. There is no reason to be less critical in evaluating end of life care than care in other settings, and no reason a priori for giving higher priority to care in this setting, but equally it is not reasonable to discriminate against a care setting because it raises difficulties in using conventional evaluation tools and metrics.