• concept of health
• quality/value of life/personhood
• decision-making
• distributive justice
• health economics

In their insightful article, ‘The Disvalue of Death in the Global Burden of Disease’, Solberg et al argue that there is a potential incoherence in the way disability-adjusted life years (DALYs) are calculated. Morbidity is measured in years lived with disability (YLDs) in a way quite unlike the way mortality is measured in years of life lost (YLLs). This potentially renders them incommensurable, like apples and oranges, and makes their aggregate—DALYs—conceptually unsound. The authors say that it is ‘vital’ to address this problem, that ‘[n]eglecting [it] is not an option’, and that ‘one cannot add YLLs and YLDs together in [their] current form’.1 

Though one might object to their argument in various ways, let us assume the authors are correct that there is a potential inconsistency here. I want to ask why (or whether) we should be troubled by that. Now, this question may scarcely seem worth asking. Consistency and coherence are regarded as non-negotiable in so many domains that of course we should demand them in a measure such as the DALY. And this desire for coherence is clearly shared by both the architects of the DALY and their critics.2 3

I think, though, that things are more complicated. In 2012, I wrote an article discussing the epidemiological perspective from which DALYs should be calculated.4 Simplifying greatly, here is the problem: both YLDs and YLLs can be calculated either from prevalence data or from incidence data. The most natural way to calculate YLDs bases them on prevalences, because when we think about morbidity our initial thought is of people who are suffering. The most natural way to calculate YLLs bases them on an incidence measure, because when we think about mortality we focus on events (ie, deaths). As Murray noted long ago, it does not seem to make sense to add these together. The original GBD therefore chose to calculate YLDs based on incidences, despite its awkwardness, because doing so was ‘more consistent with the method for calculating [YLLs]’.2

I suggested this was unnecessary. Consider the California Academic Performance Index (API), an educational assessment tool. As originally implemented, the API evaluated school performance on the basis of academic measures such as standardised test scores. That, however, gave schools an incentive to allow poorly performing students to drop out, thus increasing the school’s average scores. The solution was straightforward: add a component to the API reflecting a school’s retention rate.5 Depending on how this is done, it can turn out that a school can never increase its API by having a student drop out.

Should we be concerned that the revised API seems to involve an incoherence, since it makes no theoretical sense to add standardised test scores to retention rate? I think not. The API’s job is to help us identify good versus poorly performing schools. Most of us think that schools should aim both to increase their students’ academic performance and to keep their students from dropping out. If the API captures these dual goals and weights them in a way that reflects the relative importance we place on them, it serves its purpose and is an effective tool.i

I think something similar is happening with the DALY.ii With respect to any individual, it seems natural to describe us as having two health-related goals: to help her avoid undesirable morbidities and to keep her alive (or, to keep her from ‘dropping out’ of the population). If YLDs do a good job of capturing the way we care about the former and YLLs do a good job of capturing the way we care about the latter, and if their relative weightings reflect the importance we place on each, then the DALY serves its purpose. If that is right, then just as we should not be bothered by any inconsistency in adding incidence-based YLLs to prevalence-based YLDs, perhaps we similarly need not be worried about the potential inconsistency identified by Solberg et al.

This suggests thinking of the DALYs as an index, rather than a measure. (Indeed, DALYs are often described as an index.6) Unlike a measure, which aims to faithfully represent a feature of the natural world and can thus be accurate or inaccurate, an index is simply a tool, which is valuable to the extent that it helps us identify things we care about. Body mass index is a good index to the extent that it helps us to identify obesity-related health risks. Governance indices, such as the Ibraham Index of African Governance, are valuable to the extent that they help us to identify states which are doing a good versus bad job of justly responding to the needs of their citizens.

I believe we have independent reason to think that our concern for health may be better captured through an index, rather than a measure.iii I of course can’t offer a full argument in this short commentary. But briefly, it seems that there is no way to quantify health itself; there is no objective fact about whether a person with pneumonia literally has more health than someone with a broken thumb.7 When it comes to valuing health, I think that many people (dis)value mortality in a very different way than they (dis)value morbidity. From an ethical perspective, death seems qualitatively different than reduced quality of life, which is perhaps why trade-offs between the categories can seem harder, or at least more controversial, than trade-offs within either category.8 9 That we must sometimes make such trade-offs need not mean that we see any robust common denominator between them, beyond the importance we place in each.

It may seem that I have now pushed the pendulum too far in the other direction. If DALYs are best thought of as an index, and adding apples to oranges is not problematic in an index, then does that mean anything goes? Are consistency and coherence irrelevant? No. Coherence in an index can be valuable instrumentally. It may, for example, make an index easier to understand or to explain to others. (I think the DALY’s use of time as a common denominator is beneficial in this way.) Internal consistency may also guard against bias, by preventing the architects of an index from exercising their discretion in too many discrete places, thus allowing bias or interest group influence to creep in undetected. And so forth.

What’s the upshot? Contra Solberg et al, it can be perfectly reasonable for DALYs to add apples to oranges. To show that any incommensurability between YLDs and YLLs is problematic, the authors would need to show that it causes some further problem—for example, by making DALYs harder to understand than they would be under an alternative interpretation. (I take no position on whether that is the case here.) Contra Murray and the architects of the DALY, noting that a methodological change would improve coherence is not sufficient justification for making it. It must also be shown that increased coherence would yield some further benefit, such as preventing the measure from being gerrymandered by interest groups.iv Consistency and coherence may often or even usually be beneficial in the DALY, but we shouldn’t expect them to always be.