EQ-5D-3L Value Set

EQ-5D-3L Value Set in QALY Calculation and NICE Appraisal

The three-level EQ-5D asks about five dimensions of health (mobility, self-care, usual activities, pain or discomfort, and anxiety or depression), and each answer of level 1, 2 or 3 says nothing about value until a set of weights is applied. Dolan's model produced the UK set of those weights from time trade-off interviews with members of the general public, and since his 1997 paper that set has turned EQ-5D-3L answers in trials and registries into the utilities behind QALY gains in UK cost-utility analysis. Its scale runs below zero, so about a third of the states it scores count as worse than dead, a feature that Devlin and colleagues described as rather unique among national value sets. The UK set also became the target of crosswalks and mapping models that let five-level data be expressed in three-level values, until NICE moved adult reference-case analyses to a new UK five-level set from 27 August 2026.

What the UK time trade-off set was built from

The EQ-5D-3L describes 243 health states, the product of three levels on each of five dimensions (3 × 3 × 3 × 3 × 3 = 243). Dolan noted that direct valuation of every one of those states was virtually impossible, so a procedure was needed to interpolate values for all of them from direct valuations of a subset. That logic, value a designed subset and model the rest, is shared by every later EQ-5D value set.

In the UK study, direct valuations of 42 EQ-5D-3L states were elicited with the time trade-off method from a representative sample of the UK population. Dolan and Roberts later reported that this original tariff rested on valuations from 2,997 members of the UK general population. The respondents valued described states, not their own health, so the weights express the general public's preferences about hypothetical health.

The scored values sit on the usual utility scale. A National Institute for Health and Care Excellence (NICE) Decision Support Unit technical support document on mapping explains that utilities are bounded above at full health, valued at 1, and below by the worst state the instrument can describe. For the EQ-5D-3L the worst state is 33333, which has a value of -0.594 in the UK value set, so every UK 3L utility lies in the range -0.594 to 1. Devlin and colleagues observed that approximately a third of EQ-5D-3L states received negative values in the UK set, and that all other countries had fewer worse-than-dead states and higher minimum values.

Modelling a three-level value set with main effects and the N3 term

Forty-two observed states cannot score 243 states without a model, and the model's form is part of the value set. Dolan's preferred model explained valuations through the severity level on each dimension, an intercept for any move away from full health and a term for whether any dimension sits at its most severe level. The EuroQol Group's reference volume edited by Szende, Janssen and Cabases tabulates national time trade-off sets in terms of a dummy for level 2 and level 3 on each dimension, and many of the tabulated sets include an N3 dummy equal to 1 when any dimension is at level 3. Not every national set takes this form. Shaw, Johnson and Coons built the US time trade-off set on a D1 model, which uses ordinal terms for departures from perfect health and interaction effects in place of N3.

$$U(s) = 1-\alpha D(s)-\sum_{d=1}^{5} \beta_{d}(l_d)-\gamma N_3(s)$$

where $U(s)$ is the utility of health state $s$, $\alpha$ is the constant decrement, $D(s)$ equals 1 when state $s$ has a problem on any dimension and 0 for state 11111, $l_d$ is the level reported on dimension $d$, $\beta_{d}(l_d)$ is the decrement attached to that level on that dimension and is zero at level 1, $\gamma$ is the N3 decrement and $N_3(s)$ equals 1 when any dimension of $s$ is at level 3 and 0 otherwise.

Three features of this form shape every utility a UK 3L analysis reports. The constant means that any departure from 11111 loses value at once, however mild the problem. The N3 term applies once, so a state with one dimension at level 3 bears the same N3 penalty as a state with all five at level 3. Separate decrements for each dimension mean the set embodies the public's view of which problems matter most, so two patient groups with the same number of problems can receive quite different utilities.

The N3 form has also been tested against alternative specifications. Dolan and Roberts re-estimated a tariff from the same data using differences in value between the worst state 33333 and every other state, and reported a mean absolute difference between actual and predicted values of 0.03, against 0.039 for the original model. Whichever model is used, the coefficients in an analysis should be taken from the source publication.

Worked example of the N3 term in an illustrative value set

The decrements in this example are illustrative and are not the published UK coefficients. They follow the model above, with $\alpha = 0.08$, mobility level 2 at 0.07, mobility level 3 at 0.30, self-care level 2 at 0.10, usual activities level 2 at 0.04, pain level 2 at 0.12, pain level 3 at 0.38 and $\gamma = 0.27$. The states are written in the order mobility, self-care, usual activities, pain and anxiety.

1. A starting state with two level 2 problems. State 21121 has moderate mobility problems and moderate pain.

$$U(21121) = 1-(0.08 + 0.07 + 0.12) = 0.73$$

where 0.08 is the constant, 0.07 the mobility level 2 decrement and 0.12 the pain level 2 decrement.

2. Two further level 2 problems. State 22221 adds self-care and usual activities problems at level 2.

$$U(22221) = 1-(0.08 + 0.07 + 0.10 + 0.04 + 0.12) = 0.59$$

where 0.10 and 0.04 are the self-care and usual activities level 2 decrements and the other terms are as in step 1. The utility falls by 0.14.

3. One move to level 3. State 21131 differs from 21121 only in pain moving to level 3, which brings in the N3 decrement.

$$U(21131) = 1-(0.08 + 0.07 + 0.38 + 0.27) = 0.20$$

where 0.38 is the pain level 3 decrement and 0.27 is $\gamma$. The utility falls by 0.53, of which 0.27 comes from N3 and 0.26 from the larger pain decrement.

4. A second dimension at level 3. State 31131 also moves mobility to level 3, and N3 is not charged twice.

$$U(31131) = 1-(0.08 + 0.30 + 0.38 + 0.27) = -0.03$$

where 0.30 is the mobility level 3 decrement and the other terms are as in step 3. The state is valued as worse than dead.

5. QALY gains and cost per QALY. Treatment A keeps patients in 21121 who would otherwise move to 21131, and treatment B keeps patients in 21121 who would otherwise move to 22221, each for 2 years without discounting and each at an extra cost of GBP 10,600. Treatment A gains 0.53 × 2 = 1.06 QALYs and treatment B gains 0.14 × 2 = 0.28 QALYs.

$$\frac{10600}{1.06} = 10000$$

where 10,600 is the incremental cost in GBP and 1.06 is the QALY gain of treatment A, giving GBP 10,000 per QALY. Treatment B costs about GBP 37,857 per QALY.

A single change of one level on one dimension is worth almost four times as much as two added level 2 problems, because the step to level 3 crosses the N3 threshold. In a model whose health states differ by the presence of any extreme problem, much of the QALY difference between states can come from that one coefficient, so a large utility gap between two model states is worth tracing to N3 before it is read as a large difference in health.

Why the negative range matters in cost-utility analysis

A value set with a deep negative range gives treatments for severe conditions large utility gains, because moving a patient out of a worse-than-dead state can be worth more than one full QALY a year. Under the UK 3L set the gap between 33333 and full health is 1.594, wider than the gap between dead and full health. Devlin and colleagues contrasted this with the 2018 England EQ-5D-5L set, which gave 5.1 per cent of states values below zero and a minimum of -0.285. That 2018 England set is not the one NICE adopted. NICE's earlier position statement did not recommend it, and PMG51 instead adopts the UK EQ-5D-5L value set published by Rowen and colleagues in 2026.

The depth of the scale also feeds into appraisals. Navega Biz, Hernández Alava and Wailoo, assessing the switch of NICE technology appraisals to five-level values, found that EQ-5D values generally rose after mapping to the 5L set, with larger increases for more severe 3L states. Utility gains measured with the two sets are therefore not interchangeable, and the size of the difference depends on where in the severity range a treatment acts. The EQ-5D-5L value set page covers the five-level sets themselves.

Linking five-level responses to three-level values

Van Hout and colleagues set out to give the five-level EQ-5D-5L interim values by mapping it to the existing three-level value sets. They built that crosswalk from 3,691 respondents in six countries who completed both the 3L and the 5L. They compared four models, found similar fit across them and chose a nonparametric approach for its simplicity. For each dimension and level of the 5L, the approach gives the probabilities of a response at each 3L level, and the 3L value set is applied to those probabilities to give a value for each of the 3,125 states that five levels on five dimensions can describe.

Later work took a different route. Hernández Alava, Pudney and Wailoo surveyed a large UK sample online in 2020 with both versions in randomised order, receiving almost 50,000 responses, and estimated a single joint ordinal model of the ten EQ-5D responses. A copula allows the correlation between 3L and 5L answers to differ at the upper and lower extremes of health, and the preferred specification includes age and sex as well as the EQ-5D responses. The model maps in either direction and can start from descriptive responses, individual utility scores or summary statistics.

Both approaches are a form of mapping function. When they predict three-level utilities, those utilities inherit the UK 3L set's properties, including its negative range and the N3 step. The Hernández Alava model also runs in the reverse direction, so 3L trial data can be expressed in five-level terms when a decision maker asks for five-level values.

NICE's use of the three-level set and the 2026 change

NICE sets the value set its appraisals use through its reference case, and in 2026 that choice moved away from the three-level set for adults. NICE's interim methods statement PMG51, published on 27 August 2026, replaced the earlier position statement on the EQ-5D-5L value set for England. The overview page of the NICE technology appraisal and highly specialised technologies manual (PMG36) states that from 27 August 2026 the EQ-5D-5L descriptive system and the UK value set should be used for all adult reference-case analyses, and the interim statement takes precedence over PMG36 until the manual is revised later in 2026.

The three-level set therefore stops being the default source of adult reference-case utilities for NICE, while remaining the scoring rule behind a large body of published utilities and past appraisals. Analysts reusing older UK utilities need to know which set produced them before combining them with new five-level values in one model.

How a three-level value set differs from nearby concepts

A value set is any collection of weights that scores an instrument's states, and a country-specific value set is one elicited from one country's public. An EQ-5D-3L value set is narrower again, tied to the 243-state three-level descriptive system, and the UK time trade-off set is one member of that family. The EQ-5D-3L questionnaire records responses and carries no values until a set is applied.

A crosswalk applies a 3L value set to 5L responses through a statistical link, whereas a five-level value set values 5L states directly through new health state valuation work. A mapping function predicts EQ-5D utilities from a different instrument or from clinical measures, and still needs a value set at the end of the chain. The EQ visual analogue scale, a respondent's rating of their own health, is not a value set either, since it does not attach public preference weights to described states.

Common misreadings of three-level utilities

The floor of -0.594 belongs to the UK time trade-off set and not to the EQ-5D-3L as an instrument. Other national sets score 33333 differently, and Devlin and colleagues reported in 2018 that every other country's set had fewer worse-than-dead states and a higher minimum, so a negative mean utility in a UK analysis is partly a property of the UK weights.

Another misreading treats N3 as a decrement for each dimension at level 3. N3 is a single indicator, charged once whatever the number of extreme problems, so a hand calculation that subtracts it repeatedly will understate utilities for the most severe states.

A third error mixes values across sets. A health state utility value scored with the UK 3L set, one mapped through the van Hout crosswalk and one scored with a five-level set can differ for the same patient, and NICE's 2026 change makes the source of each utility something to report alongside the value itself.

Sources

• Devlin NJ, Shah KK, Feng Y, Mulhern B, van Hout B. Valuing health-related quality of life: an EQ-5D-5L value set for England. Health Economics. 2018;27(1):7-22. doi:10.1002/hec.3564
• Dolan P. Modeling valuations for EuroQol health states. Medical Care. 1997;35(11):1095-1108. doi:10.1097/00005650-199711000-00002
• Dolan P, Roberts J. Modelling valuations for EQ-5D health states: an alternative model using differences in valuations. Medical Care. 2002;40(5):442-446. https://pubmed.ncbi.nlm.nih.gov/11961478/
• Hernández Alava M, Pudney S, Wailoo A. Estimating the relationship between EQ-5D-5L and EQ-5D-3L: results from a UK population study. PharmacoEconomics. 2023;41(2):199-207. doi:10.1007/s40273-022-01218-7
• Wailoo A, Hernández Alava M, Pudney S. NICE DSU Technical Support Document 22: mapping to estimate health state utilities. NICE Decision Support Unit, University of Sheffield; June 2023 (DSU list shows a 2026 update). https://sheffield.ac.uk/media/118671/download?attachment=
• National Institute for Health and Care Excellence. Interim methods statement: implementing the EQ-5D-5L value set (PMG51). Published 27 August 2026. 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. Overview and update information. https://www.nice.org.uk/process/pmg36
• Navega Biz A, Hernández Alava M, Wailoo A. Switching from EQ-5D-3L to EQ-5D-5L in England: the impact in NICE technology appraisals. Value in Health. 2026;29(5):794-801. https://pubmed.ncbi.nlm.nih.gov/41895683/
• Shaw JW, Johnson JA, Coons SJ. US valuation of the EQ-5D health states: development and testing of the D1 valuation model. Medical Care. 2005;43(3):203-220. doi:10.1097/00005650-200503000-00003
• Szende A, Janssen B, Cabases J, editors. Self-reported population health: an international perspective based on EQ-5D. Dordrecht: Springer; 2014. Table 2.2. doi:10.1007/978-94-007-7596-1
• van Hout B, Janssen MF, Feng YS, Kohlmann T, Busschbach J, Golicki D, Lloyd A, Scalone L, Kind P, Pickard AS. Interim scoring for the EQ-5D-5L: mapping the EQ-5D-5L to EQ-5D-3L value sets. Value in Health. 2012;15(5):708-715. doi:10.1016/j.jval.2012.02.008