1 Introduction
Cancer has been described as “the emperor of all maladies” [
1], and despite significant improvements in survival rates for many cancers [
2], it is still a ‘dreaded’ disease [
3]. There are instances of health policies assuming that there is a preference for society to fund cancer care, relative to other diseases and conditions. For example, in the documents establishing the original Cancer Drugs Fund (CDF) in England, there is an assumption that the public value health gains to cancer patients up to twice as much as other conditions [
4]. The CDF is unique in providing ring-fenced funds for a named disease, although there are examples of funds to cover specific types of conditions, such as the New Medicines Fund in Scotland, which supports access to drugs for end-of-life or rare conditions [
5,
6]. The end-of-life criteria used by the National Institute for Health and Care Excellence (NICE) also reflect an assumed preference for health gain to patients with limited life expectancy [
7], a feature of many cancers.
Health economic analysis typically assumes that the primary role of publicly funded healthcare is to maximise population health [
8]. This is operationalised in cost-effectiveness analysis by assuming that ‘a quality-adjusted life-year (QALY) is a QALY’, i.e. a given level of health gain is equivalent regardless of the person it accrues to; it does not generally consider aspects outside the specific definition of health used for assessment, such as characteristics of the patient, the intervention or the condition itself [
9,
10]. However, there are some particular circumstances that can be given generous weighting in NICE’s Technology Appraisal Committee deliberations, including severity of disease, end of life, and illnesses in children, and these aim to reflect societal preferences for allocation of healthcare resources [
11].
Giving a preference weighting to cancer, or indeed any specific feature of ill health, requires understanding the trade-off involved: does society value health gains to cancer patients more highly than gains to other patients? More specifically, are we prepared to divert resources to cancer treatment even if it results in lower health gains for the population as a whole? Prioritising one disease type in this way within a fixed budget means that health is foregone by other patients within the population; hence, it has been argued that a strong case must be made to depart from the principle of health maximisation, and that this should reflect society’s views [
12]. Therefore, our study aimed to explore the empirical evidence for a preference among the UK general public for health gain to cancer patients. Preliminary work indicated that limited empirical data exist specifically for cancer in the UK. To make our review more informative, we therefore chose to also look at similar data from other countries, and to consider proxies for cancer, in order to place the UK findings in context and enrich our interpretation; our focus, however, remains on the UK and cancer.
2 Methods
A literature review was undertaken to identify empirical studies examining societal preferences for health gain to cancer patients. A search of the MEDLINE and PubMed electronic databases was conducted during November 2015, and updated in March 2017, using search terms covering both social preferences and cancer. In addition, a search of MEDLINE was conducted for the specific types of studies that would be used to address such preferences, such as discrete-choice experiments (DCEs). EconLit searches were added in March 2017. The search strategies are reported in Online Resource 1, section A. The terms referring to societal views were restricted to the title field to select papers with a direct focus on this topic.
Papers were screened by review of the abstracts, and eligible papers identified by full-text review (by LM). Studies were included if they were empirical studies of preferences for treating cancer patients relative to other conditions, studies of public views (i.e. excluding studies of clinicians and decision makers), and written in English; unpublished papers were not included.Study authors for the UK papers directly addressing health trade-off were contacted for points of clarification. The main data extracted from the studies (by LM, reviewed by SW) were the measure of preference for cancer treatments, its value, and whether a preference was demonstrated, along with key study features. Potential sources of bias for key papers were considered and are described in the Discussion section.
We also explored the literature on preferences for notable features of cancer, specifically severity (i.e. an illness that places patients in a poor health state) and end of life (where a patient’s life expectancy is short as a result of their illness). Both of these are used as prioritising features within health technology assessment, and often as proxies for cancer. While conclusions from studies in severity and end-of-life preferences do not necessarily apply to all cancers, many types of cancer will fall into at least one of these categories, therefore such studies could help support our understanding of cancer preferences. This area has been reviewed by Dolan et al. [
13], Shah [
14] and most recently by Gu et al. with searches run in August 2014 [
15]. These were supplemented with MEDLINE and PubMed searches in February 2016 for additional publications since that work, with an update and Econlit searches conducted in March 2017 (details in Online Resource 1). Inclusion criteria were as above, replacing cancer with severity or end of life.
This report is consistent with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines.
4 Discussion
Respondents in these studies view cancer as ‘special’ and deserving prioritisation, and, in some contexts, showed increased WTP (tax or personally) for cancer care or prevention. However, when presented with the opportunity cost of that choice, the results are inconsistent. As such, the literature reviewed does not provide a strong body of evidence supporting preference for health gains in cancer
per se, and gives no clear indication for a weighting factor. This finding is consistent with other authors’ reviews (e.g. Linley and Hughes [
20], Chamberlain [
55] and Shah [
58]); this paper updates and systematically extends that work. Evaporation of a preference for treating cancer when faced with its opportunity cost was also demonstrated by Gold et al. in a qualitative study [
59]. The impact assessment for the establishment of the CDF in 2010 also found little support for its assumption of a cancer preference, and the absence of a specific cancer preference in the Linley and Hughes paper is commonly cited in critiques of the CDF, including the Scottish and Welsh governments’ decisions not to implement similar funds [
60,
61].
The evidence on severity suggests that the public show a preference for health gains to patients with severe disease; however, the support for a preference for health gains at the end of life is equivocal. The variability in these results illustrates the challenges of designing experiments to determine definitive weights for these parameters for use in Health Technology Assessments (HTAs), with the results being sensitive to framing effects (preferences shifting with different descriptions of the problem). Interpretation is further complicated by evidence of heterogeneity in attitudes within the population surveyed, raising questions of how to represent an overall societal view [
56]. Examples of the use of severity in HTA include Sweden and The Netherlands (variable threshold), as well as France (as a dimension of clinical benefit assessment) [
62]. End-of-life criteria are used in HTA by NICE and the Scottish Medicines Consortium, allowing more flexibility in the cost per QALY under specific criteria [
5,
7].
As the two UK studies requiring a trade-off disagree on the role of cancer in public preferences, it is important to explore potential sources of bias. The studies differ on several design elements, which might contribute to the contradictory results, and we suggest there is an overall tendency for the study by Erdem and Thompson [
18] to overestimate, and the study by Linley and Hughes [
20] to underestimate, preference for cancer. First, complexity; these are difficult choices and it is possible that in the multiattribute DCE, respondents resort to simple decision heuristics [
63], such as prioritising the cancer patient, to make the decision more manageable. Indeed the authors’ own further analysis indicated use of selection by aspects [
64] and attribute non-attendance [
65] in the responses, with cancer consistently attracting respondents’ attention. This suggests that respondents were not fully considering the trade-offs in the scenarios, which resulted in overestimation of coefficients and WTP for cancer. The Linley and Hughes study [
20] describes the choice in a simple health gain scenario. It may be that this provides respondents with a stark picture of the implications of favouring a particular group, in which the opportunity cost implications are not acceptable to the majority of respondents.
Second, the choices permitted: the DCE requires an all-or-nothing choice between two options (and a ‘none’ option), whereas the single parameter choice experiment in effect offers an 11-point scale. A forced choice can overestimate the degree of preference, particularly if there is a uniform direction of preference among respondents, but only a minimal perceived difference between the options. The results reflect the number of respondents with a given preference, but not the extent of that preference. In contrast, the scale may underestimate preference because of central tendency bias (respondents under-using the extremes of a scale). A similar hypothesis was tested by Skedgel et al., who compared the results of a binary-choice DCE with constant-sum paired comparison (budget sharing) and found that allowing respondents to distribute a budget across two alternatives provided richer preference data [
47].
Third, the DCE may be sensitive to the choice of comparator conditions, which include conditions that might be considered behaviour-related (such as obesity), and where cancer is the only one that is typically considered immediately life-threatening. Any bias is likely to be in favour of cancer in this context.
Finally, both studies are limited in some aspect of generalisability. Erdem and Thomson report a relatively small and localised study (250 respondents in West Yorkshire), which may not be broadly generalizable for the UK, although comparison with the 2011 census indicated the sample was similar to the West Yorkshire population. The Linley and Hughes study provides a simple choice scenario; however, in reality, healthcare prioritisation is complex and based on multiple criteria, as outlined by Erdem and Thompson, therefore the single-parameter approach may not generalise to complex decisions. The direction of potential bias here is non-obvious.
Despite the inconsistency of results in this specific resource allocation context, there is a strong response to cancer in many of the studies reviewed. This is consistent with dual-processing theories of cognition as outlined by Kahneman [
66], who described one processing system as fast and intuitive (System 1), while the other is slower and more deliberate (System 2). Using this model, we can describe the immediate responses to questions on cancer as triggering System 1, responding based on fear and dread, with trade-off questions that require further consideration invoking System 2. Simple heuristics in complex scenarios may be dominated by System 1, and not checked by System 2 so long as the resultant choices are coherent and acceptable. A similar explanation was proposed by Robb et al. [
3] for qualitative observations of respondents responding initially to cancer questions with dread, while also acknowledging significant improvements in outcomes. Shah et al. [
67] also observed participants in qualitative research struggling to reconcile logical resource allocation decisions with their intuitive response. Although there have been recent critiques of Kahneman’s work [
68], it provides a useful structure for exploring the implications of the study’s findings. For example, it may be that the application of accountability for reasonableness principles [
69] by HTA agencies (e.g. NICE [
70]) makes these processes inherently reliant on System 2 judgements, and this will contribute to decisions being seen as unacceptable by the public if they are responding based on System 1.There are also implications for preference research, including the need for researchers to be clear on which type of response they are aiming to measure, and careful consideration of language in the materials and questions presented to respondents, with prior qualitative work to identify ‘trigger’ words that could prompt use of a simple heuristic. Parallel qualitative work with survey respondents may also help to understand the basis for their choices, as described by Shah [
56].
4.1 Limitations and Further Research
At the review level, we suggest that the main risk of bias across studies is likely to be publication bias. It may be that negative results (i.e. showing no evidence of a cancer preference) have a lower probability of publication, hence this review would overestimate any preferences.
The review is limited by the small number of studies that address the question of the value of health gains in cancer directly. This probably stems from the focus in economic evaluation on generic measures of health to allow comparison across disease areas; standard textbooks recommend that studies valuing health states are not labelled with specific conditions, and the value tariffs in common use were generated using unlabelled health state descriptions [
71]. The impact of named conditions on how the public values health states has been reviewed by Brazier et al. [
71] and Rowen et al. [
72], showing mixed results. The study by Rowen et al. finds a reduction in the values assigned to severe health states labelled as cancer compared with irritable bowel syndrome (IBS) or no label [
72]. In a study not covered in these reviews, Mason et al. [
73] used named conditions in a prioritising exercise and did not observe cancer behaving noticeably differently from the other conditions; however, they do comment that familiar conditions such as cancer tended to receive more extreme scores, with less familiar illnesses more likely to be ranked in the middle. Despite the inconsistencies between studies, it appears that the name of the condition can be important in how the public value a given state, which, by extension, could affect the value of any health gain. Focusing on disease-blind health state valuations may be obscuring such public preferences.
Our review is limited by its focus specifically on societal attitudes to cancer in the context of resource allocation and health maximisation; however, we cannot ignore the reaction to cancer seen in other types of studies. Our conclusions are rooted in a health maximisation paradigm, which does not explicitly consider broader aspects beyond health gain. An alternative view that maximised something other than health (for example, utility as in welfare economics) would use measures such as individual WTP, which can be considered as a holistic reflection of what is important to the individual. In our review, such studies showed a cancer preference, suggesting there are features associated with cancer that are not reflected in standard measures of health. This potentially leads to undervaluation of interventions in health technology assessment, not only in cancer but also in other diseases, if aspects that are valued by the public are not captured. We hypothesise that the apparent uniqueness of cancer may be in the particular combination of such features, including, for example, the impact on the family, or the value of hope. Further research is needed to explore characteristic features of cancer in comparison with other significant health conditions such as cardiovascular disease and dementia. Our future work aims to identify broader aspects of health outcome that contribute to the value that patients and the public attach to therapeutic interventions.