Marginal Analysis

Marginal Analysis of the Next Step in Health Care Spending

Marginal analysis in health care looks at the next step of spending on a service, setting what that step adds in cost against what it adds in health, instead of judging the service by its totals. The marginal benefit of expanding a service is the extra health produced by that next step, such as the quality-adjusted life years gained by adding clinic sessions or by reaching one more group of patients, and it usually falls as a service extends to people at lower risk, while the cost of each step often rises as it reaches people who are harder to reach. Most budget decisions in a health system concern changes of this kind; few concern whether to provide a service at all.

The two principles of opportunity cost and the margin

Mitton and Donaldson, in their guide to health care priority setting, describe two economic principles that underlie the task. The first is opportunity cost, meaning that investing resources in one way loses the benefit those resources could have produced elsewhere. The second is the margin, which concerns changes to the resource mix. If the budget rises, the question is how best to spend the extra money; if it falls, resources should come from the areas producing the least benefit; and if it stays flat, some areas may be cut back so that others can expand and benefit to the population improves.

Both principles follow from scarcity, since within a fixed budget the value of any extra spending is judged against the health it displaces. This places marginal analysis inside welfare economics and the theory of allocative efficiency, in which the best mix of services is reached when moving money between them can no longer raise total benefit.

Why the average figure misleads

The average cost-effectiveness of a service divides its total cost by its total health gain. That figure says whether the service as a whole was good value, but it says little about the next step, because early steps usually reach the people who gain most. The marginal cost per QALY of the next step can therefore sit far above the average even when the average looks very good.

A well-known teaching case is a protocol of six sequential stool guaiac tests for colonic cancer, for which Neuhauser and Lewicki reported that the marginal cost of the sixth test might be 20,000 times the average cost. Brown and Burrows, noting that the original analysis gave an incremental cost per case detected of more than USD 47 million, later argued that it contained two fundamental errors and that, once corrected, the marginal cost was quite modest; the average cost-effectiveness ratio page covers the case. Whatever the figure, marginal estimates need the same scrutiny as any other model output.

Measuring a marginal change and judging it

A marginal value is a change in cost or benefit divided by the change in activity that caused it. The activity unit must be stated, for example patients treated, sessions run or coverage reached, and costs and benefits must refer to the same change over the same period.

$$MC = \frac{\Delta C}{\Delta Q} \qquad MB = \frac{\Delta B}{\Delta Q}$$

where $MC$ is marginal cost, $MB$ is marginal benefit, $\Delta C$ is the change in total cost, $\Delta B$ is the change in total benefit and $\Delta Q$ is the change in the quantity of activity. When benefit is valued in money, expansion improves welfare while $MB$ exceeds $MC$. In health care the benefit is usually measured in health units, so each step is judged against the health that the same money would produce elsewhere.

$$NHB_j = \Delta E_j - \frac{\Delta C_j}{\lambda}$$

where $NHB_j$ is the net health benefit of expansion step $j$, $\Delta E_j$ is the extra health from that step (in QALYs), $\Delta C_j$ is its extra cost and $\lambda$ is the cost of producing one QALY with the services that would be displaced. A step is worth taking when $NHB_j$ is positive, which is the same as its cost per QALY, $\Delta C_j / \Delta E_j$, lying below $\lambda$. Multiplying through by $\lambda$ gives the equivalent money form, the incremental net monetary benefit.

Costing at the margin also matters inside formal appraisal. The NICE technology appraisal manual (PMG36) says that when a technology is already recommended for another use and enough spare capacity exists, an analysis using marginal costs may be supplied in addition to the analysis based on average costs.

The equi-marginal principle and where reallocation stops paying

Marginal analysis across several services leads to a single condition for the best allocation. Mitton and Donaldson state that, for a funder focused on health gain, the goal is reached when no further reallocation from one service to another would increase total benefit, which in theory is where the ratio of marginal costs to marginal benefits is equal across services. Written for two services, the condition is as follows.

$$\frac{MB_A}{MC_A} = \frac{MB_B}{MC_B}$$

where $MB_A$ and $MB_B$ are the marginal benefits of the last unit of services A and B, and $MC_A$ and $MC_B$ are their marginal costs. If the last pound spent on A buys more health than the last pound spent on B, moving money from B to A raises total health until the two ratios meet, usually because returns fall in A as it grows and rise in B as it shrinks.

The NICE manual applies the same logic to national decisions. It states that decisions about a new technology must consider the health care programmes for other patients that may be displaced, and that, given the fixed NHS budget, the appropriate maximum acceptable ICER is the opportunity cost of programmes displaced by new, more costly technologies. Because NICE lacks complete information on the costs and QALYs of all competing programmes, it uses a cost-effectiveness threshold range. In the current manual (last updated 31 March 2026), when the most plausible ICER is below GBP 25,000 per QALY gained, the decision to recommend normally rests on the cost-effectiveness estimate and the acceptability of the technology as an effective use of NHS resources, and between GBP 25,000 and GBP 35,000 the committee makes explicit reference to further factors.

Worked example of expanding a community service in four steps

The figures below are illustrative. A commissioner is considering expanding a community service in four steps of 1,000 patients each. The first steps reach the patients at highest risk, so later steps cost the same or more (more outreach) and gain less. The commissioner's displaced services are assumed to produce one QALY for every GBP 20,000, so $\lambda$ is GBP 20,000.

Marginal ratios. Each step's extra cost is divided by its extra QALYs, giving 200,000 / 40 = 5,000; 200,000 / 25 = 8,000; 250,000 / 12 = 20,833; 300,000 / 6 = 50,000.

Averages. Cumulative cost and QALYs are 200,000 and 40, then 400,000 and 65, then 650,000 and 77, then 950,000 and 83. The averages are 5,000, then 400,000 / 65 = 6,154, then 650,000 / 77 = 8,442, then 950,000 / 83 = 11,446.

Net health benefit. Each step's cost is turned into QALYs forgone by dividing it by 20,000 to give $NHB_1 = 40 - 10 = 30$, $NHB_2 = 25 - 10 = 15$, $NHB_3 = 12 - 12.5 = -0.5$ and $NHB_4 = 6 - 15 = -9$.

The decision. Expansion should stop after step 2, at 2,000 patients, where the total net health benefit is 30 + 15 = 45 QALYs. Expanding to 4,000 patients gives a total of 45 minus 0.5 minus 9, or 35.5 QALYs, a loss of 9.5 QALYs to the population as a whole. The stopping point depends on $\lambda$. At GBP 25,000 per QALY, step 3 has a net health benefit of $NHB_3 = 12 - 10 = 2$ QALYs and expansion would stop at 3,000 patients.

The average for the full expansion, GBP 11,446 per QALY, sits well below GBP 20,000 and would wrongly suggest that all four steps are good value; the early steps carry the later ones. Read in reverse, the table is also a disinvestment case, because if the service were already at 4,000 patients, releasing the GBP 300,000 spent on step 4 would buy 15 QALYs elsewhere for a loss of 6, a gain of 9 QALYs.

Programme budgeting and marginal analysis (PBMA)

PBMA turns the two principles into a process that health organisations can run within a budget cycle. Mitton and Donaldson describe it as based on the same principles as economic evaluation, note that it has been argued to be more pragmatic, and describe it as usable within a single programme of care, across programmes or for funding decisions on new technologies and drug formularies.

Their guide sets out seven stages, which are to decide the aim and scope; compile a programme budget, which maps current activity and expenditure; form a marginal analysis advisory panel; set locally relevant decision criteria; identify options for service growth, for resource release through efficiency gains and for resource release through scaling back or stopping services; make recommendations on moving resources between these lists; and check validity with wider stakeholders before final decisions. Growth and reduction options are rated against criteria set in advance, weighted where possible, and placed in one combined ranking, and resources are moved from the lowest ranked funded item to the highest ranked growth option until no further gain is expected from switching.

The criteria need not be health gain alone; Mitton and Donaldson list access, innovation and sustainability among examples. A survey by Mitton and Donaldson identified 78 uses of PBMA in 59 regional health authorities between 1974 and 1999, with a positive impact, defined as setting priorities or shifting resources, in 59% of the exercises where longer-term impact was known. A later review by Tsourapas and Frew found that reported success depended on its definition, ranging from 22% of studies for adopting the framework for future use to 65% for implementing some or all of the panel's recommendations.

Marginal analysis, incremental analysis and the ICER

Marginal and incremental analysis share one rule, which is that the extra cost is set against the extra benefit, never the totals. They differ in what is being changed. Marginal analysis changes the scale of one activity, while incremental analysis compares distinct alternatives for the same decision, such as a new drug against current care. The two coincide when the alternatives are successive scales of the same programme, as in the four steps above.

The ICER applies the same reasoning to a menu of options. PMG36 defines it as the ratio of expected additional total cost to expected additional QALYs and asks for a fully incremental analysis with dominated options and those subject to extended dominance removed, so each option is compared with the next non-dominated one, not with doing nothing.

Assumptions, limits and common misreadings

Marginal analysis rests on assumptions that rarely hold exactly. The theory assumes divisible activities and smooth changes in cost and benefit, but many services come in lumps. A new scanner, ward or team carries a large step cost, and the margin is then a whole unit, not a small change. Results also depend on where the service currently sits on its production function, since capacity limits and minimum safe service levels can stop allocation from reaching the equi-marginal point.

Data are a practical constraint. Costs and outcomes are usually reported as totals or averages, and Mitton and Donaldson note that priority setting needs data and time, that measuring the benefit of services is hard across broad service areas, and that obtaining resource releases can be difficult, especially without fiscal pressure. A marginal estimate built from averages hides the problem it is meant to solve.

Several misreadings recur. In economics "marginal" refers to the next unit of activity and carries no sense of "minor". A low average cost per QALY does not show that further expansion is worthwhile. The ICER is not automatically a marginal cost, since it compares two specified alternatives and becomes a marginal ratio only when those alternatives differ in scale. The $MB = MC$ rule applies only when benefits are valued in money; with health outcomes the benchmark is the health forgone elsewhere. Marginal analysis also leaves equity open, because it shows the efficiency effect of a change, while choices about which groups to serve sit with the decision criteria, as the equity-efficiency trade-off page discusses.

Sources

• Mitton C, Donaldson C. Health care priority setting: principles, practice and challenges. Cost Effectiveness and Resource Allocation. 2004;2:3. https://doi.org/10.1186/1478-7547-2-3
• Mitton C, Donaldson C. Twenty-five years of programme budgeting and marginal analysis in the health sector, 1974-1999. Journal of Health Services Research and Policy. 2001;6(4):239-248. https://doi.org/10.1258/1355819011927558
• Tsourapas A, Frew E. Evaluating 'success' in programme budgeting and marginal analysis: a literature review. Journal of Health Services Research and Policy. 2011;16(3):177-183. https://doi.org/10.1258/jhsrp.2010.009053
• Neuhauser D, Lewicki AM. What do we gain from the sixth stool guaiac? New England Journal of Medicine. 1975;293(5):226-228. doi:10.1056/NEJM197507312930504
• Brown K, Burrows C. The sixth stool guaiac test: $47 million that never was. Journal of Health Economics. 1990;9(4):429-445. doi:10.1016/0167-6296(90)90004-m
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). London: NICE; 2022, last updated 13 October 2026. Sections 4.2.16, 4.4.20, 4.10.8, 6.2.25, 6.3.1, 6.3.4 and 6.3.7. https://www.nice.org.uk/process/pmg36
• Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. 4th ed. Oxford: Oxford University Press; 2015.