NICE DSU and EEPRU Release Mapping Tools Ahead of EQ-5D-5L Value Set Adoption

The NICE Decision Support Unit (DSU) and the NIHR Policy Research Unit in Economic Evaluation (EEPRU) have released updated Stata, Excel and R commands for mapping between EQ-5D-3L and EQ-5D-5L. These give health technology assessment (HTA) analysts a way to convert existing three-level utility data before the United Kingdom adopts a new EQ-5D-5L value set. In two recorded presentations from July 2026, the DSU team explained the background to the mapping model and how to use the software.

The trigger is familiar to UK HTA practitioners. NICE has recommended the EQ-5D-3L value set since its 2022 methods manual update, but a new UK EQ-5D-5L value set published by Donna Rowen and colleagues earlier in 2026 changes the situation. Professor Allan Wailoo, Director of the NICE DSU, sets out what this means in practice.

“NICE has indicated that it will adopt this value set at some point in 2026, and when they do, that means that analysts that have 3L data will need to map that to the new 5L.”

Ad

Wailoo explains that the joint DSU-EEPRU work provides a way to perform that mapping in either direction, although the 3L-to-5L route is the one most likely to be required in future. He asks analysts who find the work useful to cite the underlying PharmacoEconomics paper, Hernández Alava, Pudney and Wailoo’s “Estimating the Relationship Between EQ-5D-5L and EQ-5D-3L: Results from a UK Population Study” (2023; 41:199-207).

A bidirectional, age- and sex-specific model

Professor Mónica Hernández-Alava described the statistical machinery behind the mapping. The model differs from the usual disease-specific mapping exercises in two ways: it is designed to work in both directions within a single model, and it models the responses people give to the instruments rather than the value sets themselves.

“We wanted the same model to be able to go from 3L to 5L but also from 5L to 3L.”

Modelling responses rather than value sets means the same model can be re-used with different national value sets without re-estimation. Hernández-Alava presents this as a deliberate design choice. The resulting model is a joint ten-equation system covering the five dimensions of the 3L and the five dimensions of the 5L, conditional on age and sex. Because there is no unique relationship between the two instruments, both age and sex must be entered as covariates.

The team kept the model flexible. Any assumptions rejected by the data would leave the model misspecified and biased, so three features appear that are rarely seen in mapping models: copulas, which allow flexible dependency patterns between the 3L and 5L in every dimension; mixtures of normal distributions, which can approximate almost any shape; and a latent factor to account for the fact that the same person answers all ten questions.

“Our goal here was to estimate a model that was as flexible as possible to avoid imposing any unnecessary restrictions.”

The underpinning data came from a large UK online panel collected in April 2020 through the polling company OnePoll, with randomised instrument ordering and web-based interviews. A pilot experiment in the UK Household Longitudinal Study’s innovation panel had earlier established that question ordering matters and therefore matters for the mapping, so the main study was designed to randomise the sequence of the two instruments.

Stata, Excel and R

Solutions are available in all three platforms. In Stata, the eq5dmap command can be installed directly from within the program by typing search eq5dmap and following the on-screen instructions. The Excel workbook, the latest version labelled “EQ-5D map Jan 26”, maps either a single value or multiple values at a time and requires macros to be enabled. For R users, the download bundles the mapping functions, a demonstration script and an example dataset; the R implementation builds on earlier scripts.

All three produce identical results across the worked examples, which the presenters confirm deliver the same mapped values whether computed in Stata, Excel or R. The commands on the DSU website were updated in April 2026 to align with the Rowen et al. value set.

What this means for analysts

The practical upshot is that the move to EQ-5D-5L, once NICE completes its adoption, will require analysts with legacy 3L data to convert it. The DSU and EEPRU have now provided a validated, consistent and freely available route to do so. The age- and sex-specific nature of the model means practitioners must record patient age and sex alongside their utility inputs, and the bandwidth parameter must be set correctly when only summary scores are available.

For those mapping children under 16, the DSU notes the data cover adults only, so analysts must make their own choice, for example setting age to 16, with the associated caveats. Full instructions, example data and the commands themselves are available on the NICE DSU website, alongside the two instructional recordings.

Source: NICE Decision Support Unit, University of Sheffield, “Mapping between EQ-5D-5L and EQ-5D-3L” (recorded presentations, July 2026).