For many, the notion that ‘a QALY is a QALY’ is a fundamental principle of resource allocation decision making and cost-utility analysis [
7,
14]. While this principle is widely assumed in the practice of economic evaluation, it is not universally accepted. It has been suggested by some that there are occasions when differential consideration should be given to health gains (or other benefits of treatment) based on the characteristics of those receiving care. It has been argued, for example, that the principle of equal value for all QALY gains is disadvantageous to people at the end of life [
15,
16], that it is ageist [
17,
18] and that it is based on assumptions about the characteristics of individuals that do not hold [
19‐
21]. As a result, a literature has developed on how it might be possible to give greater weight, and therefore greater access to scarce resources, to some populations over others. Such weights are variously referred to as distributional weights or equity weights, referring to the idea that, by weighting health gain, a more equitable distribution of health can be achieved relative to the dominant health maximisation approach implied in the acceptance of the idea that ‘a QALY is a QALY’.
An overview of the current literature
Much of the literature on equity weighting is concerned with the identification of attributes for weighting or estimating weights for individual attributes. A review from Paulden et al. [
22] identified 19 individual candidate attributes for weighting from the existing literature. And in recent years, a number of empirical studies on equity weights and distributional concerns have been published. A recent systematic review [
2] identified 64 such studies, published between 1989 and 2014. All studies included in that review reported on the identification of attributes deemed to be important in weighting, from a range of different populations (most commonly the UK, US and Australia). Studies included in the review were mixed as to whether they focused on a single attribute (28 studies) or multiple attributes (36 studies). A minority (22 studies) also attempted to estimate distributional weights for the identified attributes. Of these, 19 studies estimated weights for single attributes, and three studies [
23‐
25] that estimated weights for multiple attributes (and of these, two are based on the same research into the social value of the QALY [
23,
24]). Research that focuses on the existence of preferences between characteristics is also ongoing [
26].
The Social Value of a QALY (SVQ) research project [
23,
24] estimated weights using two different methods—a discrete choice experiment (DCE) and a person trade-off exercise (PTO). This research found that irrespective of the method by which weights were derived, people expressed a preference for health gains that are accrued by young people or the elderly. Only one attribute apart from age was considered to be important in health distribution; using the PTO method (but not in the DCE), it was found that illness severity should be given consideration.
The other study identified by Gu et al. [
2] to estimate multi-attribute weights is from Dolan and Tsuchiya [
25], who examined the trade-off between maximising total health gain against reducing inequalities in health. In this study, weights were estimated through trying to estimate a social-welfare function, incorporating equity considerations. They showed that a sample of the general public gave greater weight to health gain when a person was expected to die at the age of 60 compared to someone expected to die at the age of 70. They also found that the less time a person spent in full health, the greater the weight the respondents gave to gains in health. The overall results from this study suggest that age at death, and the length of time lived in good health, are both important priorities for equity weighting among the general public.
The most recent study to consider equity weighting is from Rowen and colleagues [
3], and was undertaken as part of a programme of work looking at value-based pricing in decision making [
27]. This study focused on whether greater value should be placed on populations according to the burden of illness they experience (which might otherwise be described as severity) and/or whether they are end-of-life. The results of this work found that both the ‘severity’ and ‘end-of-life’ were considered important considerations for weighting in decision making. This study was limited by design to considering burden of illness and end-of-life as potential equity weights.
The SVQ work represents the most comprehensive effort to date to understand the effects of multiple equity attributes on public preferences for health gain. While this work did identify a small number of instances where health gains would be differently valued by the public, most of the estimated weights were small [
23]. By contrast, Dolan and Tsuchiya found relatively large weights applied to differences in the age of the beneficiary at onset and at death [
25]. One possible reason for the discrepancy between the two studies is the number of attributes that respondents were asked to consider. In the SVQ project, respondents were asked to consider attributes relating to age at death, age at onset of ill-health, the severity of health lost and the potential health gain from treatment. Dolan and Tsuchiya [
25] only required respondents to consider the age at death and age of onset of ill-health. This supports the idea that there are important interactions to be considered in developing equity weights.
Another possible source of difference in the two sets of results is the approach used to estimate weights. Dolan and Tsuchiya [
25] used an approach based on trying to identify a latent social welfare function based on the stated preferences of individuals. The SVQ project team undertook a discrete choice experiment [
23] and person-trade-off exercise [
24]. Given differences in the underlying conceptual basis of each method, it is possible that this would lead to different weights being obtained. If the difference in results is down to the empirical method selected, then the wide variation in estimated weights should raise concerns about the validity of any results obtained to date. Further research into the most appropriate methods of deriving weights should be considered a priority.
Equity weighting in current practice
As highlighted above, although much research has been done to derive individual or joint weights for equity attributes, there is no accepted formal system for applying weights in economic evaluations. Yet, despite this, weighting is routinely done, both implicitly [
28] and explicitly [
12], in the practice of decision making. This creates obvious problems. Such an approach lacks transparency, leads to a simplistic consideration of equity and is likely to result in a series of inconsistent, ad hoc decisions. The end-of-life criteria as applied by NICE are an ideal example of the flawed current process.
Box 1. NICE end-of-life criteria
The treatment is indicated for patients with a short life expectancy, normally less than 24 months and; |
There is sufficient evidence to indicate that the treatment offers an extension to life, normally of at least an additional 3 months, compared to current NHS treatment, and; |
The treatment is licensed or otherwise indicated, for small patient populations | |
In 2009, NICE introduced a series of criteria that would apply to cost-effectiveness decision making processes for a small subset of treatments, in a subset of patients (see Box
1). The rationale for this was that this patient group was somehow disadvantaged by the existing systems and did not receive an equitable allocation of resources. The practical effect of this policy has been to introduce a distributional weight for allocating resources when considering a particular population. Under the revised scheme, treatments meeting the new end-of-life criteria could be approved for use in the NHS under a less stringent threshold for cost-effectiveness than required for other treatments. Now, instead of being required to provide an additional QALY at a cost of £20,000 to £30,000, qualifying treatments must only generate each additional QALY at an incremental cost estimated to be in the region of £50,000 [
29].
In effect this policy has created the first explicit equity weight to be used in practice by NICE—if the end-of-life threshold is in fact £50,000/QALY, then it is 2.5 times that of the lower limit of the standard threshold. If the threshold is taken to represent the opportunity cost of health displaced, this gives some sense of how much more weight decision makers place on the QALYs that accrue to the beneficiaries of the policy.
The approach chosen by NICE in relation to end-of-life treatments cannot be justified by equity concerns. Paulden and colleagues [
30] have shown that simply applying a differential threshold to treatments or populations fails to identify who bears the opportunity cost of the additional weight given to the beneficiaries’ health. Where the patients who bear the opportunity cost of the decision are similar to those who gain, this can lead to differential weights being applied to similar patients, a violation of the principle of horizontal equity. The argument of Paulden et al. [
30] is summarised as follows.
Consider a new treatment for a specific illness, where the opportunity cost is borne solely by other patients who also have that illness. Suppose that, for every £20,000 spent on the new treatment, one QALY is displaced among those patients who bear the opportunity cost. A healthcare payer wishes to assign 2.5× the value to QALYs for patients with the illness in question, and so decides to assign an acceptable threshold of £50,000 per QALY (2.5× the standard threshold of £20,000 per QALY). However, treatments approved at this threshold would displace 2.5 QALYs among patients who bear the opportunity cost—who in this example all have the same illness as the beneficiaries of the new treatment. The payer would be choosing to displace 2.5 QALYs in order to fund a treatment that generates just 1 QALY among patients with an identical illness. In this case, it would be more equitable to retain the original threshold. The revised threshold is only suitable where those who bear the opportunity cost have no overlap with the beneficiaries of treatment. Of course, in practice it might be unlikely that all patients who bear the opportunity cost have an identical illness to the beneficiaries—more likely, we would expect that the group of patients that bears the opportunity cost contains some with the illness in question (who should receive the same special consideration as the beneficiaries of the new treatment) and other patients who do not. This implies that the correct threshold should be somewhere between £20,000 and £50,000, depending upon the prevalence of the illness in question among the patients who bear the opportunity cost. This prevalence would only be revealed if the make-up of the opportunity cost group is known. For a full discussion see Paulden et al. [
30].
In addition, the introduction of this equity weighting system has been undertaken with little regard to the methodological literature on applying equity criteria to economic evaluations. The Gu et al. review [
2] identified seven studies that considered whether people value end-of-life differently in terms of distributional effects. Five of these studies found little to no evidence that this was the case. And among those studies that did find end-of-life to be an important criterion, only one attempted to estimate a distributional weight [
31]. Pinto-Prades et al. estimate that the societal value of a QALY for a person at the end of life is 1.41 times greater than for others [
31]. In the UK context and using a baseline threshold of £20,000/QALY, this would imply a threshold of £28,200 per QALY, lower even than the upper limit of the standard threshold, suggesting no need for specific criteria.
Multi-attribute equity considerations
As illustrated in the Gu et al. review [
2], it is feasible and relatively straightforward to estimate equity weights for single attributes. However, it is unrealistic to believe that those concerned with equity in provision of treatments are concerned only with single-attribute populations. It is far more likely that decision makers will be routinely concerned with multi-attribute populations in allocation decisions. For example, it is common for public discourse to suggest that both children and those at the end of life deserve exceptional consideration [
12,
28], a clear example of a multi-attribute scenario.
It is also straightforward to apply single attribute weights in practice (assuming that, unlike the end-of-life criteria, they are evidence based). One possibility is that the weight could simply be applied directly to the QALY estimates used in the calculation of the cost-effectiveness ratio, thereby changing the ICER. Another approach, as taken by NICE, would be to apply a differential threshold criterion [
10]—though it should be noted that these approaches may lead to different decisions and so should not be considered equivalent. For example, an intervention that is ‘dominated’ (i.e. more expensive and less effective than a comparator) without a weight applied may no longer appear dominated (and may even appear cost-effective) when the weight is applied. If, instead, the threshold is altered as an alternative to applying weights, then the dominated intervention will always appear dominated and so cannot appear cost-effective, irrespective of where the threshold is set [
30]. It should also be borne in mind that while the end-of-life criteria may be weighted positively, weights may also be negative, reflecting attributes disfavoured by the public.
It is more difficult to deal with the multi-attribute scenario, and there are important methodological issues raised in a situation where there is a conjoint distributional problem. The most immediate solution to the problem would be to apply each weight individually in sequence. This approach works if it is assumed that weights are independent from one another and can be combined multiplicatively. Distributional weights estimated for individual attributes could then be applied, with no limit on the number of attributes considered. For the end-of-life child, the calculation is simply:
$$\frac{{\text{Incremental cost}}}{{\text{ (Incremental QALY } \times \text{ Weight}_{{\text{End of Life}}} \text{ } \times \text{ Weight}_{{\text{Child}}} \text{)}}}$$
It is unlikely, however, that equity weights associated with individual attributes are independent of one another. What is more likely is that the jointly estimated distributional weight (hereafter the joint-weight) is different from the product of the independently estimated weights. There is no conceptual basis on which to predict whether the joint-weight applied to the end-of-life child is greater than, less than, or equal to the product of the independently estimated weights. This can only be determined empirically. The Gu et al. review [
2] highlighted the need for additional research on dealing with this joint distributional problem. In the UK, NICE proposed a maximum threshold of £50,000 per QALY when considering joint weights, although, as Paulden et al. show, this could lead to logically inconsistent decision making [
30].