Net Benefit

Net Benefit: Valuing Health and Costs on One Scale in Economic Evaluation

An economic evaluation usually produces two results for each option, an expected cost and an expected health effect, and a decision needs a way to weigh one against the other. Net benefit does this by stating what a unit of health is worth, through a cost-effectiveness threshold or a direct money valuation, and combining health and cost into a single number that can be compared across options. Because that number is linear in costs and effects, it can be averaged across simulations, regressed on patient characteristics and ranked across many options, which a ratio cannot do as easily.

This page is the overview of the net benefit family in health economics. It sets out the monetary and health forms, the incremental form and its link to the ICER, the net benefit of cost-benefit analysis, a worked example with three options at the two values named in the NICE manual (PMG36), and the part net benefit plays in uncertainty analysis. Each form has its own page for detail. The term also has an unrelated meaning in the evaluation of clinical prediction models, covered briefly near the end so that readers looking for decision curve analysis can find the right page.

Why a threshold lets costs and health be combined

Costs are measured in money and health effects in units such as quality-adjusted life years (QALYs), so the two cannot be added directly. The cost-effectiveness threshold, written $\lambda$, supplies the rate of exchange. It states the value of one unit of health in money or, read the other way, how much health a given sum could produce elsewhere in the health system.

The interpretation of $\lambda$ shapes what a net benefit result means. On a supply-side reading, $\lambda$ reflects the health displaced when other care is not funded, so a negative incremental net health benefit signals a net loss of population health. On a demand-side reading, $\lambda$ reflects a value that individuals or society place on health. The arithmetic is identical under both readings, but the claim that a positive result supports is different, which is why the threshold and its basis are reported with every net benefit figure.

Net monetary benefit and net health benefit

Net benefit can be expressed on either scale, and the two forms carry the same information. The monetary form values health at $\lambda$ and subtracts cost, while the health form converts cost into the health it could have bought and subtracts that from the health gained. For option $j$:

$$\text{NMB}_j = \lambda E_j - C_j, \qquad \text{NHB}_j = E_j - \frac{C_j}{\lambda}$$

where $\text{NMB}_j$ is the net monetary benefit of option $j$ in money, $\text{NHB}_j$ is its net health benefit in health units, $E_j$ is its expected health effect, $C_j$ is its expected cost and $\lambda$ is the threshold in money per unit of health.

Because $\text{NHB}_j$ equals $\text{NMB}_j$ divided by $\lambda$, the two forms rank options in the same order at any positive threshold, and the choice between them is one of presentation. Net health benefit keeps the result in QALYs, which suits a health system that reads the threshold as health forgone. Net monetary benefit is convenient for regression and for audiences used to money values. Stinnett and Mullahy set out the health form in 1998 as a framework for analysing uncertainty in cost-effectiveness analysis without relying on ratios. In the same year Tambour, Zethraeus and Johannesson proposed multiplying effects by a price per unit of effectiveness, so that net benefit in money could be analysed with standard statistical methods.

In cost-effectiveness analysis, the level of net benefit for a single option has no meaning on its own, since it depends on which costs and effects are counted. Only differences in net benefit between options inform a decision.

Incremental net benefit and the ICER

Decisions turn on differences between options, so net benefit is often reported in incremental form. The incremental quantity compares a new option with a named comparator and has both a monetary and a health version. For a new option against its comparator:

$$\text{INMB} = \lambda \Delta E - \Delta C, \qquad \text{INHB} = \Delta E - \frac{\Delta C}{\lambda}$$

where $\text{INMB}$ is incremental net monetary benefit, $\text{INHB}$ is incremental net health benefit, $\Delta E$ is the health effect of the new option minus that of the comparator and $\Delta C$ is the corresponding difference in cost. INMB equals the difference between the two options' NMB values and INHB equals the difference in their NHB values. The general term incremental net benefit covers both, and incremental net monetary benefit has its own page.

The link to the ICER follows from rearranging the threshold rule. When $\Delta E > 0$, the rule that the ratio $\Delta C / \Delta E$ should fall below $\lambda$ is the same condition as $\lambda \Delta E - \Delta C > 0$, so INMB is positive exactly when the ICER is below the threshold and zero when the ICER equals it. When $\Delta E < 0$, multiplying through by a negative number reverses the inequality, and a new option that saves money but loses health is favoured only if its ICER is above $\lambda$.

Net benefit needs no such case analysis. A positive value favours the new option in every quadrant of the cost-effectiveness plane, and the measure stays finite when $\Delta E$ is close to zero, where the ratio becomes unstable. The ICER still has a use in showing what an extra unit of health costs at the margin, and the decision rule page sets out how the two rules are applied together.

Choosing among several options by maximum net benefit

With more than two mutually exclusive options, net benefit gives a direct rule. Each option's net benefit is calculated at the same threshold and the option with the highest value is preferred on cost-effectiveness grounds. In symbols:

$$j^{*} = \arg\max_{j} \left( \lambda E_j - C_j \right)$$

where $j^{*}$ is the preferred option and the maximum is taken over every option in the decision problem. This rule selects the same option as a correctly conducted fully incremental analysis, in which dominated and extendedly dominated options are removed and sequential ICERs are compared with $\lambda$. It also avoids a common error, comparing every option with one baseline, which can favour an option that is extendedly dominated or whose ICER against the next less costly option on the frontier exceeds the threshold. The worked example below shows the second case.

Worked example: three screening intervals at £25,000 and £35,000 per QALY

The figures in this example are illustrative and hypothetical and do not describe a real programme. Three mutually exclusive policies are compared per person over a lifetime horizon: no screening (A), screening every three years (B) and annual screening (C). The two thresholds are the values at which the NICE manual (PMG36) asks for net health benefits to be presented.

Step 1: net monetary benefit of each option

Each NMB in the table is the threshold multiplied by expected QALYs, minus expected cost. The same formula applies to every cell, so one calculation is shown in full. For option B at £35,000 per QALY it is:

$$\text{NMB}_B = 35{,}000 \times 8.030 - 2{,}000 = 279{,}050$$

where 35,000 is $\lambda$ in pounds per QALY, 8.030 is the expected QALYs of B and 2,000 is its expected cost in pounds. At £25,000 per QALY, option A has the highest NMB, so no screening is preferred. At £35,000 per QALY, option B has the highest NMB, £250 per person more than A and £100 more than C, so three-yearly screening is preferred.

Step 2: the same result in health units

Dividing by the threshold gives net health benefit, which leads to the same ranking. The results are then in QALYs per person instead of pounds. At £35,000 per QALY the two leading options are:

$$\text{NHB}_B = 8.030 - \frac{2{,}000}{35{,}000} \approx 7.9729$$

where 8.030 is the expected QALYs of B, 2,000 is its expected cost in pounds and 35,000 is $\lambda$, giving net health benefit in QALYs per person. For no screening:

$$\text{NHB}_A = 8.000 - \frac{1{,}200}{35{,}000} \approx 7.9657$$

where 8.000 is the expected QALYs of A and 1,200 is its expected cost in pounds. The incremental form gives the gain from B directly:

$$\text{INHB}_{B,A} = 0.030 - \frac{800}{35{,}000} \approx 0.0071$$

where 0.030 is $\Delta E$ in QALYs and 800 is $\Delta C$ in pounds for B against A. The gain of about 0.0071 QALYs per person equals the £250 NMB difference divided by £35,000. At £25,000 per QALY the sign changes:

$$\text{INMB}_{B,A} = 25{,}000 \times 0.030 - 800 = -50$$

where 25,000 is the lower value of $\lambda$ in pounds per QALY, so B loses £50 of net monetary benefit per person against A at that threshold.

Step 3: agreement with the ICER frontier

The ICER frontier reaches the same answer. Neither B nor C is dominated, and the sequential ICERs rise along the frontier, so no option is extendedly dominated.

$$\text{ICER}_{B,A} = \frac{800}{0.030} \approx 26{,}667$$

where 800 is the extra cost in pounds and 0.030 the extra QALYs of B over A. Moving from B to C:

$$\text{ICER}_{C,B} = \frac{1{,}500}{0.040} = 37{,}500$$

where 1,500 is the extra cost in pounds and 0.040 the extra QALYs of C over B. No screening is therefore preferred below about £26,667 per QALY, three-yearly screening between that value and £37,500, and annual screening above £37,500, which matches the net benefit ranking at both thresholds.

Step 4: the error of a common baseline

Comparing annual screening only with no screening gives a misleading result. Against A, option C costs £2,300 more and adds 0.070 QALYs.

$$\text{ICER}_{C,A} = \frac{2{,}300}{0.070} \approx 32{,}857$$

where 2,300 is the extra cost in pounds and 0.070 the extra QALYs of C over A. This ICER is below £35,000, and the INMB of C against A at that threshold is £150, so C looks worth adopting when judged against A alone. Yet B has £100 more NMB than C at £35,000, because the step from B to C costs £37,500 per QALY. The maximum net benefit rule avoids the error without any sequential steps. The margin between the best options is small, about 0.0071 QALYs per person, which is the kind of case in which the uncertainty around net benefit deserves attention.

Net benefit in cost-benefit analysis

In cost-benefit analysis the benefits themselves are valued in money, so no threshold is needed to convert health. Net benefit is then the present value of benefits minus the present value of costs, both measured as changes against the status quo or another comparator, and a positive value indicates that the option produces more value than it uses from the perspective adopted. When flows over time are discounted to a common date, this is the net present value of the option.

The two traditions share the idea of a single scale but differ in where the value of health comes from. In a cost-effectiveness framework, $\lambda$ is set by the decision maker or estimated from the health system's opportunity cost. In cost-benefit analysis the money values are drawn from individual preferences, for example through willingness to pay. The net benefit approach page covers the framework that expresses cost-effectiveness results in this money form, and the cost-benefit analysis page covers valuation, discounting and the comparison of net present value with benefit-cost ratios.

Net benefit under uncertainty

Costs and effects are uncertain, and the linearity of net benefit is what makes it the standard currency of uncertainty analysis. The mean of net benefit across simulations equals net benefit calculated from mean costs and mean effects, which does not hold for a ratio. Claxton argued in 1999 that adoption decisions should rest on mean net benefit alone, whatever the statistical significance of the difference, with the distribution of net benefit relevant only to whether more information should be acquired.

Several standard outputs are built from the net benefit of each option in each simulation of a probabilistic sensitivity analysis:

• expected net benefit, the mean for each option, which identifies the option to adopt at the chosen threshold;
• the probability that each option has the highest net benefit, plotted across thresholds as a cost-effectiveness acceptability curve; and
• value of information measures, which compare expected net benefit with perfect or better information against expected net benefit on current evidence.

With patient-level data from a trial, net benefit at a chosen threshold can be calculated for each patient and regressed on a treatment indicator and covariates. Hoch, Briggs and Willan set out this net-benefit regression framework in 2002, noting that it allows established econometric techniques, adjustment for imperfect randomisation and the identification of subgroups. The regression is repeated across a range of thresholds when $\lambda$ is uncertain.

Choices that change a net benefit result

A net benefit figure is conditional on the analytic choices behind it and cannot be compared with another figure unless those choices match. The threshold is the most visible, because it scales the health term and can switch the preferred option, as the worked example shows between £25,000 and £35,000 per QALY.

The NICE manual (PMG36, section 4.2.16) asks for expected net health benefits to be presented as well as ICERs, using values placed on a QALY gain of £25,000 and £35,000. It notes that they may be particularly informative when decision-making modifiers are applied, when there are several technologies or comparators, when differences in costs or QALYs are small, or when a technology provides less health benefit at lower cost, and it adds that net monetary benefits can also be shown.

Other choices matter as much as the value of $\lambda$:

• the comparator set, since the maximum is taken only over the options included;
• the perspective, time horizon, discount rate and price year, which determine the costs and effects that enter $E_j$ and $C_j$;
• the health outcome unit, since a threshold per QALY cannot be applied to effects measured in life years or events avoided; and
• in cost-benefit analysis, the method used to place money values on health and other benefits.

Net benefit in decision curve analysis is a different measure

In the evaluation of clinical prediction models, net benefit has a separate meaning, used in the decision curve analysis that Vickers and Elkin introduced in 2006. There it is the proportion of patients who are true positives minus the proportion who are false positives, with false positives weighted by the odds of the threshold probability at which a patient would opt for treatment, and it is expressed in net true positives per patient, not in money or QALYs. That measure involves no cost-effectiveness threshold and, in its standard form, no costs, and readers looking for it will find it under decision curve analysis.

Common misreadings of net benefit

Most errors come from forgetting that net benefit is conditional on the threshold or from mixing its forms. The following misreadings follow from the distinctions drawn above.

• Treating a positive NMB for a single option as evidence that the option is worth adopting, when only differences between options carry meaning.
• Reporting net benefit without the threshold, or comparing figures calculated at different thresholds.
• Treating the monetary and health forms as rival results, when they always rank options the same way at a positive threshold.
• Applying the rule that the ICER must fall below the threshold when $\Delta E < 0$, where the inequality reverses and only the net benefit rule reads the same way.
• Choosing the option with the highest probability of being cost-effective when a different option has the highest expected net benefit.
• Reading positive incremental net benefit as proof of affordability, which is a question for budget impact analysis.
• Confusing the cost-effectiveness quantity with the decision curve quantity, which has different units and ingredients.

Sources

• Stinnett AA, Mullahy J. Net health benefits: a new framework for the analysis of uncertainty in cost-effectiveness analysis. Medical Decision Making. 1998;18(2 Suppl):S68-S80. doi:10.1177/0272989X98018002S09
• Tambour M, Zethraeus N, Johannesson M. A note on confidence intervals in cost-effectiveness analysis. International Journal of Technology Assessment in Health Care. 1998;14(3):467-471. doi:10.1017/S0266462300011442
• 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
• Hoch JS, Briggs AH, Willan AR. Something old, something new, something borrowed, something blue: a framework for the marriage of health econometrics and cost-effectiveness analysis. Health Economics. 2002;11(5):415-430. doi:10.1002/hec.678
• Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Medical Decision Making. 2006;26(6):565-574. doi:10.1177/0272989X06295361
• Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford: Oxford University Press; 2006.
• 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; published 31 January 2022, last updated 13 October 2026. Section 4.2.16.