EQ-5D Explained: Questionnaire, Value Sets and Utility Values

Most cost-per-QALY figures an HTA committee reads rest on a short questionnaire. EQ-5D asks patients about five aspects of their health, and a set of public preference weights turns the answers into the utility values that drive quality-adjusted life years. Which version and which value set an analyst uses can change the result of an appraisal.

The EuroQol Group began in 1987, when 14 people met to share ideas on measuring health; an interdisciplinary team from five European countries developed the instrument. The EuroQol Research Foundation manages its copyright on behalf of the EuroQol Group Association.

A 2020 review of 34 national pharmacoeconomic guidelines found EQ-5D named in 29 (85%) and preferred in 15. QALY basics are covered in our guide to what a QALY is; this article covers the measure behind the QALY weights.

The EQ-5D questionnaire: five dimensions and the EQ VAS

The EQ-5D questionnaire has two parts: a descriptive system, in which the respondent ticks one level of problems on each of five dimensions, and the EQ VAS, on which they rate their own health today.

The EQ VAS is a vertical scale from 0 to 100, labelled in the EQ-5D-5L “the worst health you can imagine” and “the best health you can imagine”. It records the patient’s own judgement, which is useful in trials, but it is not a utility: it is not anchored to dead and does not reflect general population preferences.

EQ-5D-3L vs EQ-5D-5L: what changed

The EQ-5D-3L, introduced in 1990, offers three levels per dimension and describes 243 health states. The EQ-5D-5L, introduced in 2009, offers five levels and describes 3,125. The extra levels were added to improve sensitivity and reduce the ceiling effect, where many respondents report no problems at all.

How an EQ-5D health state profile becomes a utility value

Each completed descriptive system is coded as a five-digit profile, one digit per dimension in a fixed order: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. A value set then assigns that profile a health state utility.

So 11111 means no problems on any dimension. On the 5L, 21232 means slight problems walking about, no problems with self-care, slight problems with usual activities, moderate pain or discomfort and slight anxiety or depression.

A value set lists a value for every possible profile, based on general population preferences in one country. It is typically estimated with a regression model in which each dimension level subtracts a decrement from full health:

The new UK EQ-5D-5L value set works this way: its largest level-5 decrements are for extreme pain or discomfort (0.479) and extreme anxiety or depression (0.391), and 55555 scores 1 − 0.279 − 0.206 − 0.212 − 0.479 − 0.391 = −0.567. Dolan’s UK 3L model added a constant for any problem and an extra decrement when any dimension is at its most severe level.

Worked example: from EQ-5D profile to utility values and QALYs

Once the value set is fixed, the arithmetic is simple: subtract the decrements, then multiply by time. The decrements below are invented round numbers for teaching. They are not a real value set. Level 1 carries no decrement.

Take a hypothetical treatment for a chronic pain condition that does not change survival, over a two-year horizon without discounting:

1. Score standard care. Patients typically report 21232: 1 − (0.05 + 0 + 0.05 + 0.15 + 0.05) = 0.70.
2. Score the new treatment. Patients typically report 21221: 1 − (0.05 + 0 + 0.05 + 0.05 + 0) = 0.85.
3. Multiply by time. Standard care gives 0.70 × 2 = 1.40 QALYs; the new treatment gives 0.85 × 2 = 1.70 QALYs.
4. Take the difference. Incremental QALYs are 1.70 − 1.40 = 0.30.

If the new treatment cost £7,500 more per patient, the ratio would be £7,500 ÷ 0.30 = £25,000 per QALY gained, the incremental cost-effectiveness ratio (ICER). You can rerun these numbers in our free QALY calculator.

Real analyses apply utilities cycle by cycle, with discounting, in a Markov or partitioned survival model; see Module 8: Economic Modelling.

How an EQ-5D value set is made: time trade-off, DCE and EQ-VT

A value set comes from a valuation study: a representative general population sample values a subset of EQ-5D health states, and a statistical model predicts the rest. For the EQ-5D-5L, EuroQol’s standard protocol runs in its EuroQol Valuation Technology (EQ-VT) software and combines composite time trade-off with a discrete choice experiment.

1. Design. The default design has 86 health states in ten blocks of ten for time trade-off, and 196 pairs of states in 28 blocks of seven for the choice task.
2. Sample. At 100 observations per block, that means 1,000 respondents, each valuing one block of states and one block of pairs.
3. Interview. Computer-assisted personal interviews, with quality control that retrains interviewers whose early interviews look rushed; EuroQol researchers report that self-completion leads to significant data quality problems.
4. Model. Time trade-off data alone, or hybrid models adding the choice data, estimate a decrement for each dimension level and predict all 3,125 values.

Composite time trade-off (cTTO) uses a conventional time trade-off for states better than dead and a lead-time version for worse ones, in which ten years in full health precede ten years in the state; values run from −1 to 1 (the trade-off logic is explained in our QALY guide). In a discrete choice experiment (DCE), respondents pick the better of two states. The choices sit on a latent scale, so they are combined with time trade-off data in hybrid models rather than used alone.

The new UK value set used composite time trade-off only: 1,200 interviews, 91.8% by videoconference, covering 102 health states, with a steering group including NICE and independent quality control.

Why EQ-5D utility values can fall below zero

A negative utility means that, on average, the people who valued a state judged living in it to be worse than being dead. Because value sets are anchored at 0 for dead, such states take values below zero, down to −1 in studies using the EQ-VT protocol.

In the lead-time task, a respondent who would give up all ten years in the state, and some of the lead time too, produces a negative value; giving up the entire lead time gives −1. In the invented value set above, 55555 would score 1 − 1.50 = −0.50.

The share of negative values depends heavily on the value set: 34.6% of states in the UK EQ-5D-3L value set, 5.1% in the earlier English 5L value set (Devlin and colleagues, 2018; minimum −0.285) and 14.8% in the new UK 5L value set. In models of severe disease, time in a worse-than-dead state subtracts QALYs, so a value set with more or deeper negative values makes preventing those states look more valuable.

NICE and EQ-5D-5L: the 2026 switch from the 3L value set

As of October 2026, NICE’s preferred approach for adults is the EQ-5D-5L descriptive system scored with the new UK EQ-5D-5L value set (Rowen and colleagues, 2026). NICE confirmed it in an interim methods statement, PMG51, published on 27 August 2026, replacing the UK EQ-5D-3L value set as the basis for reference-case utilities.

What the switch means in practice

• One value set. All adult utility values in reference-case analyses should come from the UK 5L value set.
• 3L data. The consultation proposed mapping EQ-5D-3L data onto the 5L value set with the Decision Support Unit (DSU) mapping function, which works in both directions; the DSU updated TSD 22 and its tools in 2026.
• Timing. The consultation proposed applying the change only to evaluations that start after the updated methods are published (for technology appraisals, at the invitation to participate), without reviewing existing 3L-based recommendations solely to switch to 5L.
• Severity. TSD 23 adds EQ-5D-5L general population utilities by age and sex for NICE’s severity modifier.

The switch shifts results. Re-running 39 NICE technology appraisals published from 2016 to 2024, Sheffield researchers found that incremental QALYs rose by 12.5% on average for survival-extending cancer treatments and fell by 37.5% for non-cancer treatments that only improve quality of life. NICE attributes the change to both the new instrument and newer valuation methods: respondents were less willing to trade life years to avoid poor health. See also our guide to NICE technology appraisals in 2026.

EQ-5D requirements at other HTA agencies

Agencies that rely on cost-utility analysis usually prefer a generic preference-based measure scored with national preferences. EQ-5D is the most common choice, but the rules differ in detail.

NICE notes that Canada, France, the Netherlands and Spain have adopted local EQ-5D-5L value sets. For QALYs in US decision-making, see health technology assessment in the United States; for Germany and France, see AMNOG and the G-BA early benefit assessment and how HAS evaluates medicines in France.

Mapping EQ-5D when it was not collected

Mapping predicts EQ-5D utilities from another measure, such as a disease-specific questionnaire, using a model estimated in a dataset that collected both. NICE treats it as a departure from the reference case, for use when EQ-5D data are unavailable. Mapping between the 3L and 5L is often called a crosswalk.

NICE DSU Technical Support Document (TSD) 22 (2023, updated September 2026), which replaced TSD 10, sets out good practice:

1. Check the overlap. Mapping is inappropriate if EQ-5D itself does not suit the condition.
2. Use the right data. Individual-level data with both measures at the same time points, covering the severity range in the model.
3. Avoid linear regression. It is biased at both ends of the health spectrum; mixture models suited to bounded, skewed data performed well.
4. Include age and check fit visually. Condition on age for adults, and plot predicted against observed means by clinical subgroup rather than relying on summary statistics.
5. Carry the uncertainty. Report the variance-covariance matrix so mapped utilities can enter probabilistic sensitivity analysis.

Form matters in 2026. Direct mapping predicts the index value and is tied to one value set; response mapping predicts the answer on each dimension, so it can be re-scored with a new value set. An ISPOR good practices task force report (2017) gives similar international guidance.

EQ-5D-Y for children and EQ-5D licensing basics

EQ-5D-Y is the youth version, with child-friendly wording such as “feeling worried, sad or unhappy”. Every EQ-5D version requires registration, and commercial users pay a licence fee.

EQ-5D-Y

EQ-5D-Y-3L dates from 2009 and EQ-5D-Y-5L from September 2024. EuroQol recommends self-completion from age 8 to 15, with proxy versions for parents or caregivers when self-report is not feasible. As of October 2026 value sets exist only for the Y-3L, under their own valuation protocol; EuroQol expects a limited number of Y-5L value sets in 2027. NICE recommends no specific measure for children.

Licensing

Using, reproducing or translating EQ-5D needs prior written consent from the EuroQol Office, obtained by registering. EuroQol states that use is free for non-commercial academic, educational and public health purposes once registration is approved; commercial users pay a fee based on the registration details, and amounts are not published there.

Common pitfalls with EQ-5D utility values

Many EQ-5D errors come from mixing instruments, value sets, scales or levels of aggregation:

• Treating the EQ VAS as a utility. It is a self-rating, not a preference-based value anchored to dead.
• Mixing or borrowing value sets. Use one value set per model and say which; NICE now expects the UK 5L value set throughout for adults.
• Mapping a mean instead of patients. TSD 22 prefers individual-level data, because mapping an average utility loses information.
• Ignoring ageing. Over long horizons, NICE expects utilities to decline with age, as in the general population, and not to exceed general population values.
• Hiding who answered. NICE prefers patients’ own reports, then carers rather than healthcare professionals; for children, report who completed the measure.
• Comparing across the switch. ICERs scored with the 3L value set are not directly comparable with 5L-based results.