Distributional Cost-Effectiveness

Distributional Cost-Effectiveness Analysis of Who Gains, Who Bears the Opportunity Cost and the Equity Trade-off

Distributional cost-effectiveness analysis (DCEA) extends a standard cost-effectiveness model so that it reports how a health programme changes health in each social group as well as in total. It shows what a single incremental cost-effectiveness ratio cannot, namely whether a programme narrows or widens unfair inequality, and how much total health a fairer option would cost. Asaria, Griffin and Cookson of the University of York set out the framework in two stages, modelling social distributions of health and then evaluating them. The impact plane and public estimates of inequality aversion are treated separately under the equity-efficiency trade-off.

Why a total-health analysis needs a distributional companion

Asaria and colleagues note that standard cost-effectiveness analysis focuses exclusively on maximising total population health, so it is silent on inequality impacts and on trade-offs between total health and unfair inequality. Cookson and colleagues add that standard studies make equity judgements implicitly, such as treating all health-adjusted life years as equally valuable. DCEA makes them explicit, aiming to extract new information about equity as an input to deliberation, not a universal equity formula.

One thing that separates DCEA from a plain subgroup breakdown is opportunity cost. Cookson and colleagues describe the York approach as paying careful attention to the distribution of health opportunity costs from displaced spending within a fixed budget. The tutorial's version, like standard analysis, counts only health benefits and opportunity costs falling on the health sector budget.

Stage one, modelling the social distribution of health

The modelling stage produces, for each option, a distribution of lifetime health across social groups. The tutorial splits it into three steps.

1a. Baseline distribution. The baseline covers the whole general population, not only recipients, because concern about inequality usually relates to the general population and, in a budget-constrained system such as the NHS, extra spending on recipients displaces care anyone could have received. The tutorial measures it as quality-adjusted life expectancy at birth by sex and deprivation quintile.

1b. Changes net of opportunity cost. Prevalence, uptake, costs, other-cause mortality and quality adjustment can all vary by subgroup. In the bowel screening example uptake was the driver, and the adjusted odds ratio for uptake in the most deprived fifth was 0.37 against the least deprived. The tutorial converted costs to health forgone at £20,000 per QALY, so £72 million of costs became 3,600 QALYs, shared equally per head, so a subgroup holding 10% of the population bore 360 QALYs. The authors call equal sharing convenient but not evidence based, and later York work estimated that 26% of the health effect of a change in English NHS spending would accrue to the most deprived fifth and 14% to the least deprived.

1c. Removing variation judged fair. Variation from choice or unavoidable bad luck may be judged not unfair, so the distribution is adjusted to keep only unfair variation. The tutorial treats differences by sex as fair and those by deprivation and ethnicity as unfair, one of eight possible judgements with three variables, and removes the fair part by direct standardisation.

Stage two, evaluating distributions against two objectives

The evaluation stage compares the modelled distributions against the twin objectives of more total health and less unfair inequality. It moves from description to ranking to trade-offs.

2a. Describe. The health cost of a fairer option is its shortfall in total health. Inequality is measured with relative, absolute or shortfall indices, with a range suggested when there is no clear choice; the health inequality and concentration index pages cover the measures.

2b. Rank by dominance. Dominance rules rank options with minimal restriction on the social welfare function. The tutorial adds the Atkinson and Shorrocks rules, built on the Lorenz curve.

2c. Analyse the trade-off. When dominance fails, a social welfare function with an inequality aversion parameter is needed. The tutorial uses the Atkinson index form for relative inequality and the Kolm form for absolute inequality, each summarised as equally distributed equivalent (EDE) health. Written for groups, the Atkinson form is as follows.

$$h_{EDE} = \left[ \sum_{g} s_g \, h_g^{\,1-\varepsilon} \right]^{\frac{1}{1-\varepsilon}}, \quad \varepsilon \neq 1$$

where $h_{EDE}$ is EDE health, $s_g$ is the population share of group $g$, $h_g$ is the health of each member of group $g$ and $\varepsilon$ is the Atkinson inequality aversion parameter. With zero aversion the EDE equals mean health; mean minus EDE is the health per person a decision maker with that aversion would give up for full equality. Cookson and colleagues also list direct weights on health benefits for chosen groups, the approach of equity-weighted analysis.

The bowel cancer screening case study

The tutorial's case study compared four options for the NHS Bowel Cancer Screening Programme, namely no screening, standard screening, a reminder targeted at the two most deprived quintiles and areas with the highest share of people from the Indian subcontinent, and a universal reminder. The universal reminder maximised population health and the targeted one minimised unfair variation; dominance removed the first two options, but the generalised Lorenz curves of the reminders crossed.

At zero aversion the universal reminder gave 1,000 more population QALYs; the targeted reminder was preferred above an Atkinson parameter of 8 or a Kolm parameter of 0.12, and the tutorial cites English public elicitation work giving about 10.95. When deprivation-related variation was treated as fair, the universal reminder won at every aversion level shown, and when the least healthy subgroup bore all opportunity costs, no screening had the highest Atkinson EDE at $\varepsilon = 30$.

How NICE uses DCEA in technology appraisal

The NICE manual for technology appraisal and highly specialised technologies (PMG36, section 4.12) says a company includes a DCEA only with clear evidence of significant health inequalities in the eligible population, and only as supporting evidence. Subgroup cost-effectiveness based solely on social characteristics stays out, and costs and benefits are not weighted by social characteristics. Under section 4.12.4, external assessment groups do not do DCEAs in HealthTech appraisals or multiple technology appraisals of medicines, where companies can provide the evidence. Under sections 6.2.35 to 6.2.38, substantial inequality impacts let the committee apply flexibility to the range normally considered cost-effective, after considering displacement and opportunity cost.

The PMG36 support document on health inequalities, last updated 13 May 2025, groups people by Index of Multiple Deprivation quintile, assumes equal uptake unless evidence shows otherwise and applies no inequality aversion weights to QALYs. Opportunity costs are shared equally in the base case, with slight (22% to 18%) and moderate (24% to 16%) gradients towards the most deprived quintile as scenarios, and net health benefit is reported by quintile.

The document asks for results at several values within the ranges in sections 6.3.4 to 6.3.8 of the manual and, where that range is £20,000 to £30,000 per QALY, at those two values. The manual, last updated 31 March 2026, now uses £25,000 to £35,000 in those sections, so the two named figures no longer correspond to the manual's range.

Worked example with one technology, two groups and two opportunity cost assumptions

The figures are illustrative. An area of 100,000 people has two groups of 50,000 people each, a more deprived group with quality-adjusted life expectancy of 64 years and a less deprived group with 72, so mean health is 68. A new technology gains 400 QALYs in the more deprived group, where uptake is lower, and 1,400 in the other, and adds £30 million to costs.

1. Health opportunity cost. At £25,000 per QALY, the lower end of NICE's current range, £30 million displaces 1,200 QALYs.

2. Net health by group. The table compares equal sharing with an illustrative gradient in which the more deprived group bears 60%.

3. Describe. Total health rises by 600 QALYs, 0.006 years per person, either way. With equal sharing the groups move to 63.996 and 72.016 years, widening the gap from 8 to 8.020; under the gradient they move to 63.9936 and 72.0184, a gap of 8.0248. Both cases are win-lose. Because the more deprived group loses while mean health rises, the generalised Lorenz curves cross and dominance cannot rank funding against not funding.

4. Apply the Atkinson EDE. At $\varepsilon = 2$ the EDE is the harmonic mean, $h_{EDE} = 2 / (1/64 + 1/72) = 67.765$ at baseline, 0.235 years below the mean. With funding and equal sharing it is 67.770, a gain of 0.0048 years per person or 484 EDE QALYs across the area. The table sums the change in EDE health from funding over the population.

5. Find the switching values. For a small change, funding raises EDE health while the relative weight on a gain to the more deprived group, about $(72/64)^{\varepsilon}$, is below the ratio of net gain to net loss, which gives the approximation below.

$$\varepsilon^{*} \approx \frac{\ln (800/200)}{\ln (72/64)} = \frac{1.386}{0.1178} \approx 11.77$$

where $\varepsilon^{*}$ is the switching value of inequality aversion and $\ln$ is the natural logarithm. Under the gradient, $\ln(920/320) / \ln(72/64) = 1.056 / 0.1178 \approx 8.97$. Solving the EDE formula directly gives about 11.76 and 8.95.

6. Interpret. At an aversion of 10.95, the public estimate cited in the case study, funding raises EDE health if displaced care is shared equally and lowers it if the more deprived group bears 60%. NICE would not apply the EDE step, but its gradient scenarios expose the same sensitivity.

Full, aggregate and extended analyses

Aggregate DCEA, proposed by Love-Koh and colleagues, spreads the QALY gains of a standard appraisal using the socioeconomic pattern of hospital use for the disease. Of 27 technologies recommended in NICE single technology appraisals from 2012 to 2014, it found 14 that increased health and reduced inequality, 8 that did the reverse and 5 that increased both. It needs fewer data than a full DCEA but ignores technology-specific differences in use and benefit. Extended cost-effectiveness analysis, from the Disease Control Priorities project, adds financial risk protection; Cookson and colleagues noted in 2017 that DCEA did not examine it.

The handbook and the published evidence

Cookson, Griffin, Norheim and Culyer edited the handbook Distributional Cost-Effectiveness Analysis, subtitled Quantifying Health Equity Impacts and Trade-Offs (Oxford University Press, 2020). A 2021 Value in Health editorial reported a systematic review of 54 studies, in which 78% found a favourable equity impact and 6% an unfavourable one, which the authors suggest may reflect publication bias and a focus on benefits over who bears opportunity costs.

Limitations and common misreadings

DCEA is demanding in data; the tutorial had to assume some characteristics independent. Equity cannot be read from benefits alone. In the worked example the more deprived group gains 400 QALYs yet ends up worse off once displaced health is counted. The fairness adjustment, the inequality index and the level of aversion are value judgements that DCEA reports rather than settles. Results by social group also differ from subgroup cost-effectiveness, which NICE excludes; the wider goal is covered under health equity.

Sources

• Asaria M, Griffin S, Cookson R. Distributional cost-effectiveness analysis: a tutorial. Medical Decision Making. 2016;36(1):8-19. https://pmc.ncbi.nlm.nih.gov/articles/PMC4853814/
• Asaria M, Griffin S, Cookson R, Whyte S, Tappenden P. Distributional cost-effectiveness analysis of health care programmes: a methodological case study of the UK Bowel Cancer Screening Programme. Health Economics. 2015;24(6):742-754. https://pubmed.ncbi.nlm.nih.gov/24798212/
• Cookson R, Griffin S, Norheim OF, Culyer AJ, editors. Distributional Cost-Effectiveness Analysis: Quantifying Health Equity Impacts and Trade-Offs. Oxford: Oxford University Press; 2020. https://doi.org/10.1093/med/9780198838197.001.0001
• Cookson R, Griffin S, Norheim OF, Culyer AJ, Chalkidou K. Distributional cost-effectiveness analysis comes of age. Value in Health. 2021;24(1):118-120. https://pmc.ncbi.nlm.nih.gov/articles/PMC7813213/
• Cookson R, Mirelman AJ, Griffin S, Asaria M, Dawkins B, Norheim OF, Verguet S, Culyer AJ. Using cost-effectiveness analysis to address health equity concerns. Value in Health. 2017;20(2):206-212. https://pmc.ncbi.nlm.nih.gov/articles/PMC5340318/
• Love-Koh J, Cookson R, Gutacker N, Patton T, Griffin S. Aggregate distributional cost-effectiveness analysis of health technologies. Value in Health. 2019;22(5):518-526. https://pubmed.ncbi.nlm.nih.gov/31104729/
• Love-Koh J, Cookson R, Claxton K, Griffin S. Estimating social variation in the health effects of changes in health care expenditure. Medical Decision Making. 2020;40(2):170-182. https://pmc.ncbi.nlm.nih.gov/articles/PMC7430104/
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 13 October 2026. Sections 4.12, 6.2.35 to 6.2.38 and 6.3.4 to 6.3.8. https://www.nice.org.uk/process/pmg36
• National Institute for Health and Care Excellence. Support document: health inequalities. Technology evaluation methods support document for PMG36, last updated 13 May 2025. https://www.nice.org.uk/process/pmg36/resources/support-document-health-inequalities-15313210669/chapter/distributional-cost-effectiveness-analysis-methods