Value of Information

Value of Information in Health Technology Assessment

Any funding decision taken on uncertain evidence carries a chance that the chosen option is not the best one, and value of information puts an expected size on the health or money that such errors would cost. Its measures, from the expected value of perfect information to the expected net benefit of sampling, are computed from the same probabilistic model that produces the cost-effectiveness estimate. HTA bodies and research funders use them to judge whether to adopt a technology now, to tie access to further evidence, or to fund a particular study, and a 2026 methods report from the Health Economics Methods Advisory (HEMA) sets out recommendations on their use in HTA.

Decision uncertainty creates an expected loss

A decision maker chooses the option with the highest expected net benefit under current evidence. The true parameter values are unknown, so that option may not perform best once they are revealed, and the shortfall is an opportunity loss. Averaged over the uncertainty, this expected loss is what further information could recover, which the HEMA report calls the cost of uncertainty.

$$d^*=\arg\max_{d\in\mathcal D}E_{\theta}[NB(d,\theta)]$$

Here $d$ is one of the decision alternatives in the set $\mathcal D$, $\theta$ is the vector of uncertain parameters and $NB(d,\theta)$ is the net benefit of option $d$ when the parameters take the value $\theta$. The opportunity loss of an option in a given state is defined as follows.

$$L(d,\theta)=\max_{d'\in\mathcal D}NB(d',\theta)-NB(d,\theta)$$

where $d'$ ranges over all alternatives, so $L$ is zero whenever $d$ is the best option for that $\theta$. Information has no decision value when the same option is optimal for every plausible $\theta$, so a very uncertain parameter that cannot change the decision adds nothing.

Net benefit puts costs and health on one scale

Value of information is calculated on a net benefit scale because costs and health then combine additively within each draw of the uncertain parameters, which a ratio such as the ICER does not allow. With net monetary benefit the scale is money, and with net health benefit, the measure used throughout the HEMA report, it is health, usually QALYs.

$$NMB(d,\theta)=\lambda E_d(\theta)-C_d(\theta)$$

where $\lambda$ is the cost-effectiveness threshold, $E_d(\theta)$ is the health outcome and $C_d(\theta)$ is the cost of option $d$. Dividing by $\lambda$ gives net health benefit.

Because both scales depend on $\lambda$, the ISPOR task force advises reporting value of information at explicit thresholds or as a function of the threshold. The NICE manual (PMG36) asks for net health benefits at GBP 25,000 and GBP 35,000 per QALY gained. The expected value of perfect information changes with the threshold, since the threshold moves both the preferred option and the size of the loss when it is wrong. When outcomes are measured in DALYs, which count health lost rather than gained, the sign convention has to be fixed before any information value is calculated.

The four measures and the question each answers

The measures run from a ceiling on research value to the net value of one study design. Each has its own page with formulas and computation methods, and the VOI analysis page sets out the applied sequence.

EVPI is the gap between deciding after and before the uncertainty is resolved.

$$EVPI=E_{\theta}\left[\max_{d\in\mathcal D}NB(d,\theta)\right]-\max_{d\in\mathcal D}E_{\theta}[NB(d,\theta)]$$

where the first term is the expected net benefit if the choice could be made after $\theta$ is revealed and the second is the expected net benefit of the choice made now. EVPI also equals $E_{\theta}[L(d^*,\theta)]$, the expected opportunity loss of the current decision.

The measures are nested. EVPPI for any group of parameters cannot exceed EVPI, and the EVSI of a study that informs those parameters cannot exceed their EVPPI, apart from simulation error.

Worked EVPI example with two states of the world

In this illustrative example, two hypothetical strategies, A and B, face two equally likely states of the world. Net monetary benefit is measured per patient in GBP. Current evidence favours B, but perfect information would sometimes select A.

Expected NMB under current information is calculated for each strategy.

$$E[NMB_A]=(0.5\times1{,}000)+(0.5\times200)=600$$

Here 0.5 is the probability of each state and the bracketed figures are the NMB of A in each state.

$$E[NMB_B]=(0.5\times600)+(0.5\times800)=700$$

where the same probabilities weight the NMB of B. Strategy B is selected under current information, with expected NMB of GBP 700. If the state were known before choosing, the better strategy in each state would be chosen.

$$E[\max NMB]=(0.5\times1{,}000)+(0.5\times800)=900$$

where 1,000 and 800 are the best NMB attainable in states 1 and 2.

$$EVPI=900-700=200$$

where the result is in GBP per patient. The loss view gives the same answer, since choosing B loses GBP 400 in state 1 and nothing in state 2, and 0.5 x 400 = 200. Perfect information is worth GBP 200 per affected patient in expectation under these assumptions. That figure is a ceiling, since no feasible study removes all uncertainty, and it must be scaled to the population before any comparison with research cost.

Population value and the timing of research

Per-person values become population values when multiplied across the people whose treatment the information could improve during its useful life. The HEMA report recommends presenting results at population level, since that is where the impact of decisions falls, and comparing the expected population health gain from immediate access with the expected value of future research on the same scale. Its recommendations ask for the population EVPI at a minimum, alongside cost-effectiveness acceptability curves.

$$Population\;VOI=VOI_{person}\sum_{t=t_0}^{T}\frac{N_t}{(1+r)^t}$$

where $N_t$ is the number of people newly facing the decision in period $t$, $r$ is the discount rate per period, $T$ is the horizon over which the information stays useful and $t_0$ is the period in which research results become available, which is 0 for EVPI.

People treated before results arrive cannot benefit, so research delay and gradual uptake shrink the population, while other research, patent expiry and new treatments shorten the useful horizon. The ISPOR task force recommends justifying the horizon and testing alternatives in scenario analyses. Longer follow-up can improve a study's information yet reduce the population able to use it, which is why ENBS can peak at a smaller or shorter design than a power calculation would suggest.

How the HEMA report places value of information in HTA decisions

The HEMA report treats value of information as a way to quantify the consequences of uncertainty and as a framework for choosing among decision options. It recommends that EVPI be reported alongside every probabilistic analysis, including the probabilistic analysis of each scenario, with EVPPI and EVSI used where a research decision is in prospect rather than as a general requirement.

The report stresses that value of information weights a wrong yes the same as a wrong no. Funding a technology that later proves not to be cost-effective displaces health elsewhere, while delaying or denying one that later proves cost-effective loses health, and for some patients that loss cannot be recovered when the decision is revisited. On this logic further research, either only in research or managed access where routine use is compatible with it, is warranted only where ENBS is positive, so that the expected value of resolving the uncertainty exceeds the cost of the research and of the health forgone while waiting. The report also treats a low value of information, even with high uncertainty, as a reason not to ask for more evidence.

Where the population value of research is high compared with the population gain from immediate access, the report suggests considering whether funding can be linked to a requirement for further research. The feasibility and value of that research turn on whether the key uncertainties can be reduced by research, how long the research would take to report, whether its cost would be below the EVPI, and whether immediate funding would incur sunk costs that could not be recovered if the research went the other way. Its recommendations ask for more evidence when it could realistically reduce an important uncertainty and its expected value justifies the cost, delay and burden of obtaining it, including the burden on patients and services. Two judgements sit outside the calculation, namely whether the research would in fact be undertaken and reported, and whether the system can reassess the decision when it reports.

A lower price raises the expected net benefit of immediate access and lowers the cost of uncertainty and the value of feasible research, which can make further research uneconomic while the underlying uncertainty in the evidence remains. The report notes that in England value of information has an established, if intermittent, role in research prioritisation and in guidance on only in research recommendations, whereas other settings have no established route by which a result enters coverage decisions, and it recommends that agencies without one consider establishing it. Where no formal route exists, it describes value of information as a decision principle applied through structured judgement rather than routine calculation.

What the probabilistic model leaves out

Every value of information result is conditional on the probabilistic model behind it. The ISPOR task force asks for probability distributions on all uncertain parameters, with any dependency represented by a joint, correlated distribution. Sampling correlated parameters independently can change which option wins in each draw. Choices such as the survival distribution used for extrapolation are often explored only in scenarios, which the HEMA report calls the main current route to structural uncertainty.

The HEMA report shows why omissions matter. Its Table 5.1 notes that where bias in study estimates has not been parameterised, EVPI understates the cost of uncertainty and the shortfall should be stated; that evidence which does not generalise to the decision setting is not resolved by more evidence of the same kind; and that for extrapolation beyond observed follow-up, time rather than sample size is the missing ingredient. It recommends that assessment reports name the parameters, correlations or other sources of uncertainty left out of the probabilistic analysis, because what is missing bears on which decision option is indicated. Value of information computed from an incomplete model values resolving the uncertainty represented, not all possible information.

Information and implementation

Value of information assumes that practice follows the decision once the evidence is known. In reality uptake is often imperfect, and Fenwick, Claxton and Sculpher set out a single framework in which the value of information and the value of implementation are examined separately but simultaneously, because research is not the only way to change practice.

Common errors in value of information work

Value of information results can look precise while resting on an incomplete model. Several errors recur.

• Swapping expectation and maximisation gives the current decision the benefit of perfect information and drives EVPI to zero.
• Treating EVPI as the value of a feasible study ignores that real studies give imperfect information at a cost.
• Scaling by everyone with the condition indefinitely ignores research delay, gradual uptake and the obsolescence of evidence.
• Calculating value of information from one-way sensitivity analysis fails because, as the HEMA report notes, such analysis gives neither the probability of a different decision nor the expected loss.
• Counting only the study budget in ENBS leaves out the health forgone by participants given the inferior option and by patients waiting for results.

Sources

• Fenwick E, Claxton K, Sculpher M. The value of implementation and the value of information: combined and uneven development. Medical Decision Making. 2008;28(1):21-32. https://doi.org/10.1177/0272989X07308751
• Fenwick E, Steuten L, Knies S, Ghabri S, Basu A, Murray JF, Koffijberg HE, Strong M, Sanders Schmidler GD, Rothery C. Value of information analysis for research decisions: an introduction. Report 1 of the ISPOR Value of Information Analysis Emerging Good Practices Task Force. Value in Health. 2020;23(2):139-150. https://doi.org/10.1016/j.jval.2020.01.001
• 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. Section 4.10.8. https://www.nice.org.uk/process/pmg36
• Rothery C, Strong M, Koffijberg HE, Basu A, Ghabri S, Knies S, Murray JF, Sanders Schmidler GD, Steuten L, Fenwick E. Value of information analytical methods. Report 2 of the ISPOR Value of Information Analysis Emerging Good Practices Task Force. Value in Health. 2020;23(3):277-286. https://doi.org/10.1016/j.jval.2020.01.004
• Steuten L, Thavorn K, Cipriano LE, Guertin JR, McQueen B, Schmidler G, Sullivan S, Wailoo A, Sculpher M. Methods to analyze and present uncertainty in economic evaluation to support HTA decisions. HEMA Report 2. Health Economics Methods Advisory; 2026. https://hemamethods.org/wp-content/uploads/2026/10/HEMA-Report-2_For-Publication_100726.pdf