Our results confirm that it is possible to use the existing literature to estimate lifetime QALYs and quantitatively compare groups with different conditions to identify who are worse off, and to compare the lifetime account with the proportional shortfall about prospective severity. For example, we found that lifetime QALYs are lower for childhood deafness and rheumatoid arthritis than for acute stroke and hip osteoarthritis. A proportional shortfall approach yields a different ranking, the main difference being that acute stroke victims become worse off, second only to rheumatoid arthritis. This information is highly relevant for transparent and evidence-based political discussions on how to assign higher weight to health gains for those who are worse off.
Resources will be allocated to different patients groups, depending on whether extra weights are given to those who can expect to attain fewer QALYs over their lifetime versus to those who can expect greater relative future loss of QALYs. To our knowledge, this is the first study based on data from published economic studies to consider the worse off as being those with fewer lifetime QALYs, and which compares this lifetime health approach with the prevailing view of defining the worse off in terms of relative losses in current and future health.
Most proposed definitions of the worse off have a future-oriented perspective . The measure of lifetime QALYs is persuasive since, in principle, absolute differences in past and future health is taken into account. The length and quality of life lived before time of intervention may be relevant in judgements about who are worse off, and it is not obvious that these concerns should be disregarded. For example, early onset of disease is the main reason why deaf children achieve fewer lifetime QALYs than deaf adults (annual quality of life losses are assumed to be similar for both groups, an additional table provide quality-adjustment weights) [see Additional file 2]. Children who are born deaf will suffer from lack of language skills throughout their entire life if left untreated. Despite substantial prospective impairment in quality of life among adults who develop profound deafness, they have developed normal language skills earlier in life, and are less prone to all of the negative lifelong effects of early deafness . Consequently there are good reasons to give higher priority to quality-improving interventions, such as the cochlear implant, to the deaf child than to the deaf adult. Lifetime QALYs identifies the child as worse off. Proportional shortfall of QALYs does not discriminate between the two groups. Past health would have to be treated as a separate concern if this was to be taken into account in the proportional shortfall approach (Figures 2 and 3).
Even if deaf children will suffer from lack of language skills their entire life, they will also, however, adapt to their disability and learn skills required for functioning without hearing at a younger age. The lifetime QALY approach therefore entails the same adaptation problems as health state valuations in standard QALY applications . Discrete choice techniques may adjust for adaptation over time by obtaining health state valuations from the general public rather than directly from patients . Nevertheless, it is critical what respondents are asked in preference elicitation studies and there is need for additional empirical and normative work to fully understand the complexities of adaptation over a lifetime .
Absolute shortfall of future health (QALY loss) is a measure suggested by the UK Department of Health and National Institute for Health and Care Excellence to be used in weighting the societal wider impact of conditions, while using proportional shortfall in weighting the burden of illness, in value-based assessment of interventions . We define absolute shortfall of QALYs as the expected future loss of QALYs from a condition, i.e. the absolute difference between the QALY expectancy in absence of illness (QALEN) and the remaining QALYs with standard care at the time of intervention. The lifetime QALY approach treats lifetime QALY attainments and lifetime QALY loss as providing the same information. Values of these two lifetime measures are inversely correlated, and rank orders will not change if QALEN were the same for all interventions.
Arrow has argued that equivalent formulations of a choice problem should yield the same preference order . In principle, a lifetime QALY approach would therefore yield the same preference order for both QALY attainment and QALY shortfall if QALEN were identical for all age groups. Nevertheless, people are typically more averse to losses than equivalent gains . Whether people actually are less averse to QALY attainments than equivalent QALY shortfalls over a lifetime remains unclear. More empirical work is needed to test preferences of the general public on these matters.
Proportional shortfall of QALYs discriminates between conditions with respect to relative differences in prospective health. The measure identifies the worse off as those who stand to lose the largest fraction of their health potential . The rationale appears to be that people of all ages is entitled to fulfil the health potential they had reason to expect in absence of illness. When we compare the rankings of conditions according to lifetime QALYs and proportional shortfall, we find that the proportional shortfall measure hardly discriminates between childhood deafness, adult deafness, and acute stroke (Figure 3), despite the fact that the number of healthy life years for these patient groups differs considerably: the lifetime QALYs were 38.5, 64.0, and 76.4, respectively (Figure 2). The measure of lifetime QALYs distinguishes between conditions with respect to the individual burden of disease over the total lifetime.
We argue that underlying concerns about fair distribution of health are better captured by directing resources towards those with fewer lifetime QALYs rather than to those with a higher proportional shortfall of QALYs, because the former would seek to reduce inequalities in lifetime health, while the latter would seek to reduce inequalities in the future health potential only [15, 39].
Our results elucidate the balance between the worst condition and the most effective intervention. Usually, both concerns point in the same direction. For example, among the eight cases, our results show that cochlear implantation in children is the most effective intervention and that the deaf child is also among the worst off. However, sometimes, one intervention will maximise health and increase inequality. In the comparison between the two condition-intervention pairs (A) rheumatoid arthritis/TNF inhibitor and (B) morbid obesity/gastric bypass (Figure 3), a decision to offer gastric bypass first and decline to offer the TNF inhibitors would maximise average individual health outcomes, but it would also increase the inequality in lifetime health, and the inequality in future health potential, between patients with rheumatoid arthritis and obese patients. In cases where there is a conflict between concerns for the worse off and the effectiveness of the interventions, decision makers must balance competing concerns or rely on fair procedures . Decision makers could give greater weight to the TNF inhibitors because of fairness considerations: patients with rheumatoid arthritis represent the worse off group of the two.
Our study has some limitations. First, only the time aspect of past health is captured by the method we used to calculate lifetime QALYs. Past differences in quality is not taken into consideration since we lack data on past quality of life for the various conditions. Our sources, conventional CEAs, start calculating QALYs at the time of intervention. Among the eight conditions, morbid obesity is likely to be ranked relatively higher if past suffering were taken into account. Morbidly obese patients who are eligible for surgery at the average age of 48 would probably have suffered due to obesity and obesity-related comorbidities over many years, often since adolescence or childhood . The time elapsed between disease onset and the time of intervention would be much shorter for rheumatoid arthritis, atrial fibrillation, and hip osteoarthritis. Knowledge about quality of life losses in this period would result in fewer lifetime QALYs, but it is not obvious that this would change their rankings. The inclusion of past quality of life losses is practically challenging, and theoretically perhaps the most controversial implication of the lifetime health account .
Second, few studies have reported lifetime QALYs, so there is a shortage of available data. Third, the comparability of QALY data across various studies and analytical decision models is impeded by the variations in model structure complexity, varying time horizons, and varying discount rates. Additional file 2 shows key model assumptions in the source studies. Our results are based on undiscounted data. The discount rates used in the base cases of the source studies differ. The reasons to discount health outcomes at a specific rate are usually independent of the conditions and interventions that are being assessed. Therefore, using discounted QALY profiles may affect their relative size, and consequently our rankings, without good reason. There are good fairness reasons to argue that a life year in the past, present, or future should have the same value, but there is no agreement about the role of discounting . We decided to include only studies that estimated effects over a lifetime horizon. However, investigators seem to be reluctant to pursue a lifetime analysis [see Additional file 1]. The extrapolation of effects over a lifetime involves uncertainty, but textbooks and guidance on health economic evaluations generally recommend that a lifetime horizon be applied . Fourth, the validity of the results is subject to limitations in terms of the QALY methodology. Standard QALY calculations involve rather strong underlying assumptions . Some uncertainty is attributed to the quality-adjustment weights . The results are probably sensitive to the quality-adjustment weights used in the source studies [see Additional file 2]. Our approach requires that QALYs be generated in as consistent a manner as possible. The QALY consensus group has pointed out the need to develop a reference method for estimating QALYs .