SF-6Dv2

SF-6Dv2 in Health Utility Measurement and Cost-Utility Analysis

SF-6Dv2 is the second version of the SF-6D, the scoring system that turns answers to the SF-36 health survey into utility values for quality-adjusted life years. Some clinical trials and cohort studies collect the SF-36 version 2 without the EQ-5D, and a health economist who needs utilities from those studies often has the SF-6D or the SF-6Dv2 as the most direct route. The revision matters for results because the UK value set for SF-6Dv2 reaches a worst state of −0.574, while the original UK SF-6D values sit between 0.301 and 1, so the same patients can show larger utility differences and different QALY gains depending on the version used. Here the name means the revised classification system together with its national value sets, and the original SF-6D appears only as a point of comparison.

Why the original SF-6D was revised

Brazier and colleagues describe the original SF-6D as one of the most widely used generic measures of health for calculating QALYs, and they set out to build a new classification system from the SF-36 version 2 (SF-36v2). The development team used factor analysis to establish dimensionality, Rasch analysis to understand how individual items performed, and tests of differential item functioning. The Rasch analyses were repeated in several subsets of two large cross-sectional datasets, one of recently discharged hospital patients and one of online patient samples. On the basis of this psychometric evidence, cross-cultural considerations and how easily the states could be valued, the six-dimension structure of the SF-6D was kept.

The changes fell within the dimensions. Physical functioning received a simpler classification with clearer separation between levels, role limitations gained a more detailed five-level description, vitality was described with negative wording, and the pain dimension moved to describing pain severity rather than pain interference. Brazier and colleagues judged that the new system describes more distinct levels of health and gives clearer wording for health state valuation.

When Mulhern and colleagues later valued the new system for the UK, they gave the narrow range of the existing SF-6D value set as one of the problems the new value set was meant to address. A narrow range compresses utility differences between mild and severe states, which in turn compresses QALY gains in a cost-utility analysis.

The six dimensions and how an SF-6Dv2 state is coded

An SF-6Dv2 health state is a six-digit code, one digit per dimension, built from a person's SF-36v2 answers. Each digit is the level reached on that dimension, with 1 the best level. The value set then attaches a utility to every possible code.

The dimensions are physical functioning, role limitations, social functioning, pain, mental health and vitality, in that order. Pain has six levels and the other five dimensions have five, which is why the best state is coded 111111 and the worst state is coded 555655 in the valuation studies. The number of distinct health states follows directly.

$$N = 5^{5} \times 6 = 3125 \times 6 = 18750$$

where $N$ is the number of health states the SF-6Dv2 can describe, the first factor counts the combinations of the five dimensions with five levels and 6 is the number of pain levels.

How the UK value set was estimated with a discrete choice experiment

The UK value set did not rely on an interviewer-led method such as the time trade-off. Mulhern and colleagues used an online discrete choice experiment with a duration attribute, a design that lets choices between pairs of health states lived for different lengths of time be placed on the scale where full health is 1 and dead is 0.

In the UK study, an online sample of 3,014 respondents each completed 10 choice sets with duration from an efficient design of 300 sets (Design 1) and 2 further choice sets from a design of 60 that added immediate death as a third option (Design 2). In Design 2 respondents ranked the options fully by picking the best and the worst. Conditional logit regression estimated value set models with and without interaction terms, and latent class models examined differences in preferences between groups of respondents.

A common way to see how duration anchors the scale is the linear specification below. It is the general form of a DCE with duration model, given here to explain the method rather than as the exact UK model.

$$V_{ij} = \alpha T_{ij} + T_{ij} \sum_{k} \beta_{k} x_{ijk} + \varepsilon_{ij} \qquad U_{h} = 1 + \sum_{k} \frac{\beta_{k}}{\alpha} x_{hk}$$

where $V_{ij}$ is the value respondent $i$ places on alternative $j$, $T_{ij}$ is the duration of that alternative in years, $x_{ijk}$ is a dummy for the level of dimension $k$ that departs from level 1, $\alpha$ is the value of a year in full health, $\beta_{k}$ is the coefficient on that level interacted with duration, $\varepsilon_{ij}$ is the random error, and $U_{h}$ is the utility of health state $h$ on the scale where full health is 1 and dead is 0.

The favoured UK model added one interaction term that applies when any dimension is at its most severe level, and the value sets differed between Design 1 and Design 2. The resulting UK values run between −0.574 for the worst state and 1 for full health, so some SF-6Dv2 states are valued as worse than dead, which was not possible with the original UK SF-6D values. Mulhern and colleagues described the result as a provisional value set for use in resource allocation.

SF-6Dv2 value sets outside the UK

An SF-6Dv2 value set reflects the preferences of the population that was surveyed, so studies in each country produce their own country-specific value set. The Australian and Japanese studies followed an international protocol, while China combined the time trade-off with a DCE that included duration. The table lists a selection of published value sets, and others exist, including a Korean set and one for Tehran, Iran, both estimated with a DCE with duration.

The Australian study chose the Design 1 model with a moderating interaction as its value set and compared Australian and UK values. The Chinese study recruited respondents aged 18 to 90 in eight provinces and municipalities, and each respondent completed eight time trade-off tasks and ten DCE tasks with duration. In Quebec pain had the largest effect on values and vitality the smallest, the worst state 555655 was valued at −0.683, and the authors recommended a model combining both designs. In Japan physical functioning and pain had the largest influence on utility, and Japanese values tended to be lower than UK values for the same states.

SF-6Dv2 in NICE appraisals and other HTA decisions

The SF-6Dv2 sits outside the measure NICE prefers for its reference case, so its use in a NICE submission depends on the rules for non-EQ-5D data. The NICE technology appraisal and highly specialised technologies manual (PMG36) asks for health-related quality of life to be reported directly by patients and for the utility of changes to be based on public preferences using a choice-based method.

PMG36 sets out a hierarchy of preferred methods in figure 4.1 for when EQ-5D data are not available or not appropriate. In the manual text written before PMG51, mapping other quality of life measures to the EQ-5D was described as a departure from the reference case. If evidence shows that the EQ-5D is not appropriate, the first preference is another generic preference-based measure, a class that includes the SF-6Dv2. Section 4.3.10 sets out the empirical evidence needed to argue that the EQ-5D is inappropriate, starting with qualitative evidence on its lack of content validity.

NICE's interim methods statement on implementing the EQ-5D-5L value set (PMG51), published on 27 August 2026, takes precedence over the manual until the manual is updated, and NICE's FAQ says the value set will be used for new topics starting after the statement was published. The statement makes the EQ-5D-5L descriptive system and the new UK value set NICE's preferred approach for adults, and says that when a relevant study has no EQ-5D-5L data and a non-EQ-5D measure is used instead, those data should be mapped to utility values based on the UK EQ-5D-5L value set. For a trial that collected only the SF-36v2, an analyst preparing a NICE reference case would therefore map SF-36v2 or SF-6Dv2 data to the EQ-5D-5L with a mapping function. Reporting SF-6Dv2 utilities alongside the mapped values is an analyst's choice for showing how much the measure matters, not a NICE requirement.

Outside England, the Australian valuation team stated that its value set allows the SF-6Dv2 to be used in health technology assessment decision making in Australia. That is the authors' statement about the availability of local values, not a requirement set by an Australian HTA body.

Illustrative worked example comparing SF-6Dv2 and SF-6D utilities

All figures here are illustrative and undiscounted. A trial collects the SF-36v2 in two arms over three years and both the original SF-6D and the SF-6Dv2 UK values are applied to the same responses. A new treatment costs GBP 9,000 more than usual care over the period and does not change survival.

1. Utility gain per year. With the SF-6Dv2, mean utility is 0.62 in the treatment arm and 0.50 with usual care, a gain of 0.62 − 0.50 = 0.12 a year. With the original SF-6D the same patients score 0.71 and 0.65, a gain of 0.71 − 0.65 = 0.06 a year.

2. QALY gain over three years. The SF-6Dv2 gives 0.12 × 3 = 0.36 QALYs and the original SF-6D gives 0.06 × 3 = 0.18 QALYs.

3. Incremental cost per QALY.

$$\text{ICER}_{v2} = \frac{9000}{0.36} = 25000 \qquad \text{ICER}_{v1} = \frac{9000}{0.18} = 50000$$

where $\text{ICER}_{v2}$ and $\text{ICER}_{v1}$ are the incremental cost-effectiveness ratios in GBP per QALY using the SF-6Dv2 and the original SF-6D UK values, 9000 is the incremental cost in GBP, and 0.36 and 0.18 are the QALY gains from step 2.

The choice of version halves the cost per QALY in this example, because the wider SF-6Dv2 scale records a larger utility difference between the arms. The direction is consistent with the comparison by McDool and colleagues, who found that the new valuation substantially changed the size of absolute utility differences between groups, which could change QALY results. The example does not show which version is closer to the truth. The comparison shows why an analysis should state the version and value set, and why a sensitivity analysis using the other version is informative when a decision is close.

How SF-6Dv2 differs from SF-6D, SF-36, EQ-5D and value sets

The SF-6Dv2 and its neighbours describe different layers of the same measurement chain. Keeping them apart avoids reporting a utility under the wrong name.

The SF-36 is the questionnaire that patients complete, and it produces a profile of eight domains rather than a single utility. The original SF-6D is the first classification system and UK value set built from SF-36 items, and the SF-6Dv2 is the revised classification system built from the SF-36v2, with its own value sets. A value set is the set of weights that turns any SF-6Dv2 code into a utility, and a country-specific value set is one estimated from a single country's public. The EQ-5D is a separate generic preference-based measure with five dimensions and its own descriptive system completed directly by patients, and in England its EQ-5D-5L version is the preferred measure for adults. Health-related quality of life is the construct all of these instruments try to measure.

Limitations and common misreadings

Most errors with the SF-6Dv2 arise when one version or one country's values are treated as interchangeable with another. The following misreadings recur, and most of them change the utilities that enter a QALY calculation.

• Mixing versions. McDool and colleagues compared UK values in 7,932 people from the Multi-Instrument Comparison project and found SF-6Dv2 utilities generally lower than SF-6D utilities, with larger standard deviations. Pooling or comparing values across the two versions mixes scales.
• Treating the floor as fixed. The original UK SF-6D cannot fall below 0.301, while the SF-6Dv2 UK values reach −0.574. Studies that assumed no state worse than dead with the SF-6D cannot carry that assumption across.
• Assuming better validity. In the same comparison the SF-6Dv2 separated disease and severity groups better than the SF-6D in some groups but not all, so a revised descriptive system does not guarantee better performance in every condition.
• Ignoring the model choice. The UK value sets differed between Design 1 and Design 2, and the published value set is one modelling choice among several, so the model used should be reported.
• Borrowing another country's values. Japanese values tended to be lower than UK values for the same states, so a value set from one country does not stand in for another population's preferences.
• Overstating NICE acceptance. A UK value set based on public choices does not make the SF-6Dv2 the NICE reference-case measure, which for adults in new topics is now the EQ-5D-5L under PMG51.

Sources

• Ameri H, Poder TG. Valuing SF-6Dv2 using a discrete choice experiment in a general population in Quebec, Canada. International Journal of Health Policy and Management. 2024;13:8404. doi:10.34172/ijhpm.8404
• Brazier JE, Mulhern BJ, Bjorner JB, Gandek B, Rowen D, Alonso J, Vilagut G, Ware JE; SF-6Dv2 International Project Group. Developing a new version of the SF-6D health state classification system from the SF-36v2: SF-6Dv2. Medical Care. 2020;58(6):557-565. doi:10.1097/MLR.0000000000001325
• McDool E, Mukuria C, Brazier J. A comparison of the SF-6Dv2 and SF-6D UK utility values in a mixed patient and healthy population. PharmacoEconomics. 2021;39(8):929-940. doi:10.1007/s40273-021-01033-6
• Mulhern B, Bansback N, Norman R, Brazier J; SF-6Dv2 International Project Group. Valuing the SF-6Dv2 classification system in the United Kingdom using a discrete-choice experiment with duration. Medical Care. 2020;58(6):566-573. doi:10.1097/MLR.0000000000001324
• Mulhern BJ, Norman R, Brazier J. Valuing SF-6Dv2 in Australia using an international protocol. PharmacoEconomics. 2021;39:1151-1162. doi:10.1007/s40273-021-01043-4
• 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. https://www.nice.org.uk/process/pmg36
• Shiroiwa T, et al. Valuation survey for SF-6Dv2 in Japan based on the international protocol. Quality of Life Research. 2025;34(2):445-455. doi:10.1007/s11136-024-03830-w
• Wu J, Xie S, He X, Chen G, Bai G, Feng D, Hu M, Jiang J, Wang X, Wu H, Wu Q, Brazier JE. Valuation of SF-6Dv2 health states in China using time trade-off and discrete-choice experiment with a duration dimension. PharmacoEconomics. 2021;39(5):521-535. doi:10.1007/s40273-020-00997-1