Population EVPI

Population EVPI for Research and Funding Decisions

Population EVPI is the figure that lets the per-patient expected value of perfect information be compared with the cost of a research programme. A study, once reported, informs the treatment choice for every later patient facing the same decision, so the value of removing uncertainty grows with the number of those patients and with how long the decision stays relevant. In health technology assessment it acts as the first screen in a value of information analysis, because a population EVPI below the expected cost of research shows that no study of that decision can be worthwhile. Most of the judgement lies in the size of the beneficiary population and the length of the time horizon, and modest changes to either can move the result across a research budget.

Why the per-patient value has to be scaled up

A probabilistic cost-effectiveness model works on a representative patient, so EVPI comes out of it as a per-person quantity, in net health benefit or net monetary benefit, while research costs are paid once for a whole jurisdiction. Report 2 of the ISPOR Value of Information Task Force notes that EVPI, EVPPI and EVSI are usually calculated per person, but that information generated by research informs decisions for future individuals, so the more useful measure is the population value of research. Report 1 adds that, once gathered, information is of value every time a choice is made between the options, so EVPI should be scaled up for the beneficiary population.

The scaling needs no new uncertainty analysis, since the per-person EVPI comes from the probabilistic sensitivity analysis already run. What it adds are two judgements about the future, the number of patients whose treatment will be informed and the period over which the information stays useful.

The population EVPI formula

The ISPOR methods report gives the population value of research as the per-person value multiplied by the discounted size of the beneficiary population over the time horizon. The same expression scales EVPPI and EVSI.

$$EVPI_{pop} = EVPI \times \sum_{t=0}^{T} \frac{I_t}{(1+r)^{t}}$$

where $EVPI_{pop}$ is population EVPI, $EVPI$ is the per-person expected value of perfect information, $I_t$ is the number of patients whose treatment the information would inform in period $t$ (the incidence in the task force notation), $T$ is the last period in which the information is still useful and $r$ is the discount rate for one period.

The first period carries no discount. The methods report explains that the value of research should be discounted over the horizon so that more weight goes to decisions informed in the near term. When the same number of patients enters every period, the sum has a closed form.

$$\sum_{t=0}^{T} \frac{I}{(1+r)^{t}} = I \times \frac{1 - (1+r)^{-(T+1)}}{1 - (1+r)^{-1}}$$

where $I$ is the constant number of new patients per period and $T$ and $r$ are as before. At 3.5% a year over ten years, periods 0 to 9, the ratio on the right is about 8.6077, against 10 with no discounting. The 3.5% rate is the reference case rate for costs and health effects in the NICE manual (PMG36, section 4.5.1).

Because the per-person EVPI depends on the cost-effectiveness threshold, population EVPI does too. For a NICE appraisal a natural pair is GBP 25,000 and GBP 35,000 per QALY, the two values at which PMG36 (sections 4.2.16 and 6.3.2) asks for net health benefits to be presented.

Choosing the beneficiary population

Report 1 recommends that the beneficiary population be calculated from the prevalent cohort, the incident cohorts or both, as the decision problem requires, and the methods report notes that the estimate usually comes from epidemiological data. A prevalent group with the condition enters once, at the start, if its treatment choice is still open, while incident patients, counted from incidence data, enter in every period. Counting the same people in both the prevalence figure and the incident cohorts inflates the result.

Four adjustments then change who actually benefits.

• Time for research to report. Report 1 notes that the benefits of a study are not realised until it is completed, so the population is usually adjusted for the time the study takes, and patients enrolled in a future study while the decision is delayed will generally not benefit.
• Participants in the better arm. Those allocated to the arm that proves optimal benefit while the study runs, which Report 1 says matters most in rare diseases, where the sample is large relative to the population.
• Uptake and implementation. The methods report says gradual uptake or implementation of findings should be considered. Value of information assumes practice follows the decision, and Report 1 points out that implementation is often imperfect.
• More than one jurisdiction. The population also depends on the perspective and on whether the information might be generalisable to other jurisdictions, and Report 1 lists methods for a global value of information among priority areas for research.

Setting the effective time horizon

Report 1 describes future incidence and prevalence as estimable with reasonable accuracy, but the effective time horizon over which the decision stays relevant as less straightforward, citing factors such as time to patent expiry, the launch of an in-class substitute, new diagnostic or screening tests and expected price changes. It notes that published studies tend to use one year and then 10 to 20 years without clear justification, that VOI estimates can be highly sensitive to the horizon, and recommends justifying the horizon and testing alternatives in scenario analyses.

Philips, Claxton and Palmer argued in 2008 that a finite horizon is a proxy for the complex and uncertain process of future change in technologies, prices and information. When they modelled those changes explicitly, the value of information for the whole decision could rise or fall over time, but the value for the parameters that current research can evaluate tended to decline. The methods report still lists the appropriate horizon for research decisions among areas for future research.

Worked example with a prevalent cohort and ten years of incidence (illustrative)

All figures here are illustrative. A probabilistic analysis of a new treatment for a chronic condition gives a per-person EVPI of 0.05 QALYs at GBP 25,000 per QALY. Some 4,000 people already have the condition and await a treatment choice now, and 1,500 are newly diagnosed each year. The decision is judged relevant for ten years, periods 0 to 9, discounted at 3.5% a year.

1. Discount factors. For periods 0 to 4 they are 1, 0.96618, 0.93351, 0.90194 and 0.87144, and for periods 5 to 9 they are 0.84197, 0.81350, 0.78599, 0.75941 and 0.73373. All ten sum to 8.6077, matching the closed form.
2. Discounted beneficiary population. The incident cohorts contribute 1,500 x 8.6077 = 12,911.6 patients. The prevalent cohort adds 4,000 undiscounted, since it is treated in period 0, giving 16,911.6 in all.
3. Population EVPI. This is 0.05 x 16,911.6 = 845.6 QALYs, or 845.6 x 25,000 = GBP 21.14 million.
4. Without discounting. Counting 4,000 + 1,500 x 10 = 19,000 patients gives 0.05 x 19,000 = 950 QALYs, or GBP 23.75 million, about 12% too high.
5. Shorter horizon. If a competing product is expected after five years, periods 0 to 4 only, the incident sum is 1,500 x 4.6731 = 7,009.6, the population is 11,009.6 and population EVPI is 0.05 x 11,009.6 = 550.5 QALYs, or GBP 13.76 million.
6. Waiting for the study to report. With the ten-year horizon but a trial taking three years, patients in periods 0 to 2, including the prevalent cohort, are treated on current evidence. Only periods 3 to 9 benefit, with factors summing to 8.6077 minus 2.8997, or 5.7080. The population is 1,500 x 5.7080 = 8,562.0 and population EVPI is 0.05 x 8,562.0 = 428.1 QALYs, or GBP 10.70 million.

Suppose the proposed trial would cost GBP 12 million. On the base case and the five-year horizon population EVPI exceeds that cost, so research is potentially worthwhile and the analysis would go on to EVPPI and EVSI. Once the three-year wait is allowed for, population EVPI falls below the trial cost, which is enough to show that this trial cannot be worth funding for this decision. The same per-patient EVPI supports opposite conclusions depending only on population and timing assumptions.

In the similar example in Report 1, a per-person EVPI of 2 units of net health benefit, 100 beneficiaries a year and a five-year horizon give a discounted population EVPI of 943 units, about USD 47 million at USD 50,000 per QALY.

Population EVPI as a screen for research and access decisions

Report 1 sets out the decision rule. A population EVPI below the estimated cost of research is a sufficient condition for establishing that research is not of value, and the value of information process should stop. A population EVPI above the cost is a necessary but not sufficient condition for research to be potentially worthwhile, because perfect information cannot be bought with a finite study.

The 2026 report on uncertainty from the Health Economics Methods Advisory (HEMA) argues for population-level results because the ultimate impact of decisions is at the population level. It proposes setting the expected incremental net health benefit of immediate access, scaled to current patients, against the population value of further research, and where the population impact of uncertainty is high compared with that of immediate access, decision makers may need to consider linking funding to further research. In the example, a per-person incremental net health benefit of 0.08 QALYs for the 4,000 current patients would give 0.08 x 4,000 = 320 QALYs from immediate access, well below the 845.6 QALYs at stake from uncertainty, the situation in which HEMA suggests decision makers consider linking funding to further research. Its recommendations ask that the consequences of uncertainty be reported and, at a minimum, include the population EVPI.

PMG36 sets no formal value of information requirement, though it provides for managed access, which it describes as designed to allow further evidence generation in the NHS (section 6.4.11).

Population EVPI beside EVPPI, EVSI and ENBS

The same scaling applies across the value of information family. The methods report states that population EVPPI is an expected upper bound on the value of research that would remove uncertainty about a subset of parameters, and that the difference between population EVSI and the cost of data collection is the expected net benefit of sampling, the net value of a particular study.

These comparisons only make sense when the population and horizon are the same across measures. The reporting delay in step 6 belongs in population EVSI and ENBS as well, since it depends on the study design. Population EVPI comes first in the task force sequence. It is the widest bound and, as the methods report notes, needs no sampling beyond the probabilistic analysis.

Limitations and common misreadings

Population EVPI inherits the limits of the per-person EVPI, which values only the uncertainty represented in the model. Multiplying by a large population magnifies any error in the per-person value, including the error from leaving structural uncertainty unquantified.

Common misreadings include the following.

• Population EVPI as a research budget. It is an expected upper bound for one decision, so it can rule research out but cannot show that a given study is worth its cost.
• Patients multiplied by years. Leaving out discounting, or adding the prevalent cohort in every period, overstates the population.
• A conventional horizon. One-year and 10 to 20 year horizons lack clear justification, as Report 1 notes, so the horizon needs a stated reason and a scenario analysis.
• Ignoring the reporting delay. Patients treated before a study reports gain nothing from it, and in the example a three-year wait roughly halves the population.
• Perfect implementation. Low uptake reduces the value actually realised.

Report 1 also notes that the assumption of a risk-neutral decision maker has been challenged, even for a population-level decision maker.

Sources

• 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. Sections 4.2.16, 4.5.1, 6.3.2 and 6.4.11. https://www.nice.org.uk/process/pmg36
• Philips Z, Claxton K, Palmer S. The half-life of truth: what are appropriate time horizons for research decisions? Medical Decision Making. 2008;28(3):287-299. https://doi.org/10.1177/0272989X07312724
• 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