Direct Elicitation

Direct Elicitation of Health State Utilities

In direct elicitation the utility for a health state comes from a person's own answer to a valuation task, such as how many years of life they would give up, or what risk of death they would accept, to avoid living in that state. The person may value their own health or a written description, and the answer sits on the scale where full health is 1 and dead is 0. The approach, also called direct valuation or direct utility assessment, is the alternative to indirect valuation, in which patients complete a questionnaire such as the EQ-5D and a published value set turns their answers into utilities. The choice between the two routes changes the size of QALY gains in a cost-utility analysis, and NICE places direct methods low in its hierarchy of preferred evidence. Value sets such as the UK EQ-5D-5L set are themselves built by direct elicitation with a sample of the public, so the two routes are linked stages of one system.

Direct and indirect routes to a utility value

Arnold and colleagues, comparing the two routes in the BMJ, describe direct methods as mapping preferences onto the utility scale by a trade-off, standard gamble or time trade-off, or by a visual analogue scale, which they note is less favoured. Indirect utility assessment instead passes through a generic quality of life questionnaire whose responses are converted to utilities by a tariff, or value set, obtained earlier from a general population sample that valued many states with a trade-off method.

Two different designs sit under the direct heading. In own-health valuation, patients value the state they are in, which NICE DSU Technical Support Document 8 calls "direct" valuation. In vignette studies, respondents who may never have had the condition value a bespoke description, or vignette, written by the researcher. TSD 11 notes that own-health valuation avoids the poor coverage and insensitivity that can affect descriptive systems, because the patient is experiencing the state.

The valuation tasks and their arithmetic

The time trade-off compares a longer period in the impaired state with a shorter period in full health. Janssen and colleagues, introducing the composite time trade-off used for EQ-5D-5L valuation, give the value as the time in full health judged equal to a given time in the impaired state, divided by that time, which is conventionally 10 years in EQ-5D studies.

$$u_{TTO} = \frac{x}{t}$$

where $u_{TTO}$ is the value of the state, $x$ is the number of years in full health at which the respondent is indifferent, and $t$ is the number of years in the impaired state.

That form cannot give values below zero. The composite time trade-off keeps the conventional task for states better than dead and, when a respondent trades away all 10 years, switches to a lead-time task in which 10 years of full health come before the 10 years in the state. Janssen and colleagues give the transformation for that stage.

$$u_{WTD} = \frac{r-10}{10}$$

where $u_{WTD}$ is the value of a state judged worse than dead, and $r$ is the number of years in full health, between 0 and 10, at which the respondent is indifferent to 10 years of full health followed by 10 years in the state. A response of $r = 0$ gives the floor of the task, a value of minus 1.

The standard gamble offers the state for certain or a gamble that gives full health with probability $p$ and immediate death otherwise. With full health set at 1 and death at 0, the canonical result under expected utility, as given by Drummond and colleagues, is that the value of the state equals the probability of full health at which the respondent is indifferent between the certain state and the gamble.

A visual analogue scale rating involves no trade-off, and TSD 8 states that values from choice-based methods are preferred to those that are not, giving the visual analogue scale as the example. Ordinal tasks such as a discrete choice experiment raise a different problem, since TSD 11 notes that most ordinal methods yield a latent variable that does not lie on the full health and dead scale.

Whose preferences are elicited

TSD 11 states that the source of values has important implications for the value obtained. It reports that patients valuing states they are in have consistently given higher values than respondents asked to imagine them, with some evidence that patients value mental health states lower than the general population does. The 2020 DSU report by Rowen and colleagues adds that patient values can incorporate adaptation and response shift, for example when full health is reinterpreted as full health for one's age.

The tasks also make ethical and practical demands. TSD 11 describes obstacles to asking patients life and death questions, and Arnold and colleagues note that time trade-off and standard gamble interviews need a trained interviewer who explains the task without distressing or leading the respondent.

Vignettes have a separate weakness. TSD 11 observes that a small set of vignettes cannot capture the spread of experience among patients in one clinical state, which a descriptive system completed by patients does capture. The 2020 DSU report adds that the valuation closest to the reference case is a choice-based method such as the time trade-off with a representative general population sample.

Direct elicitation in the NICE reference case

In the NICE technology appraisal manual, PMG36, section 4.3 sets the reference-case methods for measuring and valuing health effects. Section 4.3.1 asks for changes in health-related quality of life to be reported directly from patients, with their utility based on public preferences using a choice-based method, and section 4.3.4 specifies a representative sample of the UK population. Section 4.3.5 states that different methods produce different utility values, and section 4.10.3 asks for a table giving each utility value, its source and the method used, naming standard gamble as one example.

The interim methods statement PMG51, published on 27 August 2026, makes the EQ-5D-5L descriptive system and the UK value set by Rowen and colleagues NICE's preferred approach for adults. The value set must be used for appraisal topics whose invitation to participate is issued after publication, while ongoing topics continue to map 5L data to 3L utility values. PMG51 gives new wording for table 4.1 and sections 4.3.6, 4.3.8, 4.3.9 and 4.3.12 of PMG36, adds a new section 4.3.10, deletes section 4.3.16, and says the updated manuals will follow later in the year.

The PMG51 revision of figure 4.1 shows where direct elicitation sits. When evidence shows that the EQ-5D is not appropriate, the alternatives in order of preference are another generic preference-based measure, a condition-specific preference-based measure, vignettes developed to DSU best practice and valued by a general population sample with a technique such as the time trade-off, and then direct valuation of own health. PMG51 states that the evidence needed to show the EQ-5D is not the most appropriate measure has not changed, so the current PMG36 sections 4.3.10 and 4.3.11 still apply.

Direct elicitation also stands behind the preferred route. Rowen and colleagues built the UK EQ-5D-5L value set from 1,200 composite time trade-off interviews covering 102 health states with a representative sample of UK adults, modelled with a random effects Tobit model.

Worked example with two vignette states in a rare disease model

All numbers in this example are illustrative. A model of a rare condition needs utilities for a controlled and an uncontrolled state, evidence has shown the EQ-5D to be inappropriate, and no other preference-based measure fits. Two vignettes are valued by five members of the general public using the composite time trade-off with $t = 10$ years.

1. Value the controlled vignette. The indifference points are 9.0, 8.5, 8.0, 9.5 and 7.5 years in full health, giving 0.90, 0.85, 0.80, 0.95 and 0.75. The sum is 4.25 and the mean is 4.25 / 5 = 0.85.

2. Value the uncontrolled vignette. Three respondents give 5.0, 4.0 and 3.0 years, or 0.50, 0.40 and 0.30. Two trade away all 10 years and in the lead-time stage are indifferent at 8 and 6 years, so their values are (8 − 10) / 10 = −0.20 and (6 − 10) / 10 = −0.40. The sum is 0.50 + 0.40 + 0.30 − 0.20 − 0.40 = 0.60, and the mean is 0.60 / 5 = 0.12.

3. Find the utility gain. Moving a patient from the uncontrolled to the controlled state is worth 0.85 − 0.12 = 0.73 per year.

4. Convert to QALYs. If treatment keeps a patient controlled for 2 years who would otherwise be uncontrolled, the undiscounted gain is 0.73 × 2 = 1.46 QALYs.

5. Test a common error. If the two worse-than-dead answers were recorded as 0, the uncontrolled mean would be 1.20 / 5 = 0.24, the gain 0.85 − 0.24 = 0.61 per year and the QALY gain 0.61 × 2 = 1.22. The error removes 0.24 QALYs, about 16% of the gain, since 0.24 / 1.46 = 0.164.

6. Interpretation. The values reflect a small public sample judging two descriptions, so they carry wide uncertainty and belong in a sensitivity analysis. PMG36 section 4.10.3 would also require the elicitation method to be tabulated with each value.

Assumptions that move the values

Each task embeds assumptions. Janssen and colleagues note that the time trade-off assumes constant proportional trade-off, so the share of time traded does not depend on the time horizon, and that the lead-time stage adds additive independence between periods. Attema and colleagues report that neglecting time preference causes TTO scores to be underestimated, and that titration gives values 0.10 to 0.15 higher than a ping-pong search. The standard gamble rests on expected utility, so attitude to risk enters the value along with the view of the state.

How direct and indirect values compare

Direct and indirect methods applied to the same patients tend to disagree in one direction. Arnold and colleagues found 32 studies giving 83 comparisons for adult health states. Mean utilities were 0.81 for the standard gamble and 0.77 for the time trade-off, against 0.59 for the EQ-5D, 0.63 for the SF-6D, 0.75 for HUI-2 and 0.68 for HUI-3. In one study of intermittent claudication, the same respondents gave an EQ-5D utility of 0.57 but direct values of 0.82 by time trade-off and 0.91 by standard gamble.

The authors concluded that indirect methods tend to give lower values, sometimes substantially, which could relatively favour treatments that are not lifesaving. Possible reasons they give include the few categories of a generic questionnaire and the gap between the public who supply tariffs and patients who have adapted to their condition.

Limits and common misreadings

Direct values are easy to misuse once they leave their study. Five errors follow from the points made earlier.

1. Treating a rating as a utility. Visual analogue scale scores involve no trade-off, and NICE asks for a choice-based method.
2. Pooling values from different tasks. PMG36 section 4.3.5 warns that methods give different values, so mixing them across model states needs to be stated and tested.
3. Presenting patient values as public values. Own-health valuations can embed adaptation and sit last in the figure 4.1 list for when the EQ-5D is not appropriate.
4. Assuming indirect valuation avoids elicitation. The UK EQ-5D-5L value set is a model of public composite time trade-off responses.
5. Recoding worse-than-dead answers. Flooring negative responses at zero inflates the value of severe states, as the worked example shows.

The wider process of turning preferences into a health state utility value is covered under health state valuation.

Sources

• Arnold D, Girling A, Stevens A, Lilford R. Comparison of direct and indirect methods of estimating health state utilities for resource allocation: review and empirical analysis. BMJ. 2009;339:b2688. https://doi.org/10.1136/bmj.b2688
• Attema AE, Edelaar-Peeters Y, Versteegh MM, Stolk EA. Time trade-off: one methodology, different methods. European Journal of Health Economics. 2013;14(Suppl 1):S53-S64. https://doi.org/10.1007/s10198-013-0508-x
• Brazier JE, Longworth L. NICE DSU Technical Support Document 8: An introduction to the measurement and valuation of health for NICE submissions. Decision Support Unit; August 2011. https://doi.org/10.15131/shef.data.33738301
• Brazier JE, Rowen D. NICE DSU Technical Support Document 11: Alternatives to EQ-5D for generating health state utility values. Decision Support Unit; March 2011. https://doi.org/10.15131/shef.data.33744721
• Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. 4th ed. Oxford University Press; 2015.
• Janssen BMF, Oppe M, Versteegh MM, Stolk EA. Introducing the composite time trade-off: a test of feasibility and face validity. European Journal of Health Economics. 2013;14(Suppl 1):S5-S13. https://doi.org/10.1007/s10198-013-0503-2
• National Institute for Health and Care Excellence. Interim methods statement: implementing the EQ-5D-5L value set (PMG51). Published 27 August 2026. Sections 1.7 to 1.18, appendix A and changes to figure 4.1. https://www.nice.org.uk/process/pmg51
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 31 January 2022, last updated 13 October 2026. Sections 4.3.1, 4.3.4, 4.3.5 and 4.10.3. https://www.nice.org.uk/process/pmg36
• Rowen D, Brazier J, Wong R, Wailoo A. Measuring and valuing health-related quality of life when sufficient EQ-5D data is not available. NICE Decision Support Unit report; 31 July 2020. https://www.sheffield.ac.uk/nice-dsu/methods-development/measuring-health-related-quality-life
• Rowen D, Mukuria C, Bray N, et al. A United Kingdom value set for the EQ-5D-5L. Value in Health. 2026;29(5):858-869. https://doi.org/10.1016/j.jval.2026.03.008