Decision

Decision: How Economic Evidence Becomes a Choice About Health Care Resources

Health economics exists to inform choices about how limited health care budgets are spent, and the decision is the point at which the analysis meets that choice. Here a decision is taken by a body with authority over resources, such as a health technology assessment (HTA) agency, a payer or a commissioner, and it selects one of several mutually exclusive courses of action: fund a technology or not, fund it for everyone or for a subgroup, fund it now or wait for better evidence. The page uses the term in this resource allocation sense, not for decision theory in general or for the psychology of individual choice. It covers the options an HTA decision offers, why the choice rests on expected values while uncertainty bears mainly on a separate choice about research and, in some cases, on the category of recommendation, how NICE decisions become funded care in England, and how the concept differs from the narrower pages on coverage decisions, decision criteria and decision rules.

The choice an economic evaluation is built to inform

Standard texts, such as that of Drummond and colleagues, define economic evaluation as the comparative analysis of alternative courses of action in terms of both their costs and their consequences. That definition puts the decision first: without alternatives and someone who must choose between them, there is nothing to evaluate. The joint definition of HTA agreed by INAHTA, HTAi and partner organisations in 2020 makes the same link, stating that the purpose of HTA is to inform decision-making in order to promote an equitable, efficient and high-quality health system.

A decision of this kind has a decision-maker with a budget and a perspective, a set of options that includes current practice as the comparator, an objective such as obtaining the most health from a fixed budget, and evidence on the costs and effects of each option. The structured statement of these parts is the decision problem, which in a NICE appraisal corresponds to the scope, which the NICE manual (PMG36) describes as the framework for the evaluation. A health economic model then links the evidence to each option.

The options open to an HTA body

Section 6.4 of the NICE manual (PMG36) lists the recommendation types a committee normally makes: recommended as an option; recommended only in specific circumstances, known as an optimised recommendation; recommended with managed access, for medicines only; recommended for use during an evidence generation period, for HealthTech only; recommended only in a research context; and not recommended. The manual also states that the committee cannot make recommendations on price, although it can consider a commercial arrangement or a managed access proposal, a form of managed entry agreement.

Claxton and colleagues framed the same choice as approve, approve with research, only in research, or reject. They found that the right category depends only partly on whether a technology is expected to be cost-effective. For a technology expected to be cost-effective, only in research may be preferable to approval when the research cannot be done once the technology is approved, and only in research or even rejection may be preferable to approval, with or without research, when adoption carries significant irrecoverable costs.

Deciding on expected values, and deciding whether to wait

A decision made under uncertainty still has to pick one option. Claxton argued that the choice should rest only on expected net benefit, whether or not differences are statistically significant, and that the distribution of net benefit matters only for the separate question of whether more information should be acquired. In terms of net monetary benefit, the adoption choice is:

$$ j^{*} = \arg\max_{j} ; \mathbb{E}_{\theta}\left[ \lambda Q_j(\theta) - C_j(\theta) \right] $$

where $j^{*}$ is the option chosen, $j$ indexes the mutually exclusive options, $\theta$ is the set of uncertain model parameters, $\mathbb{E}_{\theta}$ is the expectation over their joint distribution, $\lambda$ is the value placed on one unit of health effect (the cost-effectiveness threshold), $Q_j(\theta)$ is the health effect of option $j$, for example in QALYs, and $C_j(\theta)$ is its cost.

The second choice, whether to wait for or commission more evidence, uses the same distribution. The expected value of perfect information is the gain from being able to pick the best option for every possible value of the uncertain parameters:

$$ EVPI = \mathbb{E}{\theta}\left[ \max{j} NMB_j(\theta) \right] - \max_{j} \mathbb{E}_{\theta}\left[ NMB_j(\theta) \right] $$

where $EVPI$ is the expected value of perfect information per person affected by the decision, $NMB_j(\theta)$ is the net monetary benefit of option $j$ for parameter values $\theta$, and the other terms are as defined above. This quantity is the starting point of value of information analysis.

What the expected-value rule assumes

The rule of choosing the option with the highest expected net benefit rests on three assumptions. The first is that the decision-maker is risk neutral, so the spread of possible results does not count against an option once its expected value is known. The second is that the threshold reflects the health forgone elsewhere; the NICE manual (PMG36, section 6.3.1) states that, given the fixed budget of the NHS, the appropriate maximum acceptable ICER is that of the opportunity cost of programmes displaced by new, more costly technologies. The third is that adoption can be reversed without loss if later evidence points the other way. The 2012 framework relaxes this last assumption through irrecoverable costs, and section 6.2.32 of the NICE manual asks the committee to take into account the likelihood of decision error and its consequences for patients and the NHS, which looks beyond the expected value alone.

Worked example: adopt now, then ask whether more evidence is worth it

The figures below are illustrative. A new technology B is compared with current care A at a given threshold $\lambda$, and uncertainty about B's effect is summarised in three scenarios for its incremental net monetary benefit per patient.

1. Make the adoption choice on the expected value. The expected incremental net monetary benefit of B is:

$$ 0.25 \times 8000 + 0.50 \times 2000 + 0.25 \times (-6000) = 2000 + 1000 - 1500 = 1500 $$

where each term is a scenario probability multiplied by the incremental net monetary benefit of B in that scenario, in pounds per patient. The expected value is positive, so B is chosen on current evidence, even though there is a 0.25 probability that it is the wrong choice.

2. Value the removal of uncertainty. With perfect information, B would be chosen in the favourable and central scenarios and A in the unfavourable one, where the gain over A is zero:

$$ 0.25 \times 8000 + 0.50 \times 2000 + 0.25 \times 0 = 2000 + 1000 + 0 = 3000 $$

where the terms are as in step 1. The expected value of perfect information is £3,000 minus £1,500, or £1,500 per patient. Measuring net monetary benefit relative to A leaves this result unchanged, because the net monetary benefit of A is subtracted from both terms of the EVPI equation.

3. Interpret the two choices. The adoption choice on current evidence is B. The £1,500 per patient is an upper bound on the value of further research for each patient affected; multiplied by the number of patients expected to be affected over the time horizon of the decision, with discounting, it gives the population EVPI. Research is only potentially worthwhile if the population EVPI exceeds its cost, which is a necessary condition but not a sufficient one: whether a particular study is worth doing is judged with the expected value of sample information and the expected net benefit of sampling. The category of recommendation (including rejection) then also turns on whether that study could still be done once B is in routine use, on whether adopting B carries irrecoverable costs, and on the expected net benefit forgone by patients outside the study if B is recommended only in research, here £1,500 per patient.

From recommendation to funded access in England

In many health systems an HTA body advises and a separate payer makes the coverage decision, so the recommendation and the funding decision sit with different organisations. In England the two are tied by law. Under regulation 7 of the regulations setting out NICE's constitution and functions (SI 2013/259), relevant health bodies must comply with a technology appraisal recommendation to provide funding, normally within three months of publication unless NICE specifies a longer period in defined circumstances.

The NICE committee does not use a precise maximum acceptable ICER, and its manual (PMG36) describes cost-effectiveness as necessary but not the only basis for decision making, so the decision criteria it weighs go beyond the cost per QALY.

Decision, coverage decision, decision criteria and decision rule

Several pages on the wiki deal with parts of the same process, and the general concept on this page sits above them. The table separates what each term refers to.

Usage differs between sources. NICE calls its outputs recommendations but describes the committee's task as the decision to recommend a technology.

Where the term is misread

A common error is to let statistical significance stand in for the decision. A technology with positive expected incremental net benefit that is not significantly different from zero is still the better choice on current evidence; the uncertainty bears on whether to gather more evidence and, where costs are irrecoverable or approval would block research, on the category of recommendation. A second error is to treat the decision as binary when the real options include restricted use, use with evidence generation and use only in research.

A third error is to judge a decision by its outcome. A decision taken on expected values can turn out badly once new evidence arrives without having been wrong on what was known at the time. Finally, a decision is only as sound as its set of options, since leaving out a relevant comparator can change which option has the highest expected net benefit.

Sources

• Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford: Oxford University Press; 2006.
• Claxton K. The irrelevance of inference: a decision-making approach to the stochastic evaluation of health care technologies. Journal of Health Economics. 1999;18(3):341-364. doi:10.1016/S0167-6296(98)00039-3.
• Claxton K, Palmer S, Longworth L, Bojke L, Griffin S, McKenna C, Soares M, Spackman E, Youn J. Informing a decision framework for when NICE should recommend the use of health technologies only in the context of an appropriately designed programme of evidence development. Health Technology Assessment. 2012;16(46):1-323. doi:10.3310/hta16460.
• 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.
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). London: NICE; 2022, updated March 2026. Chapters 2 and 6.
• O'Rourke B, Oortwijn W, Schuller T; International Joint Task Group. The new definition of health technology assessment: a milestone in international collaboration. International Journal of Technology Assessment in Health Care. 2020;36(3):187-190. doi:10.1017/S0266462320000215.
• The National Institute for Health and Care Excellence (Constitution and Functions) and NHS England (Information Functions) Regulations 2013 (SI 2013/259), regulation 7. https://www.legislation.gov.uk/uksi/2013/259/regulation/7