EQ-5D Utility Value Transport

Transport reported EQ-5D health utility values between national value sets using copula-based population re-weighting.

What is this tool?

This application transports (maps) EQ-5D utility values reported under one country's value set to equivalent values under a different country's value set. It is designed for health economists, HTA analysts, and researchers who need to adapt published utility data for use in a different national context.

When is this needed? Clinical trials and observational studies typically report EQ-5D utility scores using a single national value set (tariff). When the results are used in a cost-effectiveness analysis for a different country, the reported utility must be re-expressed on that country's preference scale. Individual patient-level data are usually not available, so transport must work from summary statistics alone (mean and optionally SD).

Quick Start
  1. Go to the Transport tab.
  2. Enter the reported mean utility (and optionally the SD).
  3. Select the source value set used in the original study.
  4. Select one or more target value sets — results appear automatically and update on any input change.

The EQ-5D Instrument

The EQ-5D is a standardised multi-attribute utility instrument developed by the EuroQol Group. It comprises five dimensions — Mobility (MO), Self-Care (SC), Usual Activities (UA), Pain/Discomfort (PD), and Anxiety/Depression (AD) — each scored on a 5-level ordinal scale (EQ-5D-5L: 1 = no problems, 5 = extreme problems).

A value set (tariff) converts a health state profile (e.g., 1-1-2-3-1) into a single utility index. The index is anchored at 1 (full health) and 0 (dead); negative values are possible for states considered worse than dead. Value sets differ between countries because they encode different societal preferences.

This tool supports 83 national value sets : 43 for EQ-5D-5L and 40 EQ-5D-3L crosswalk sets.

Statistical Method

1. Reference Population Model

A Gaussian copula model is fitted to a reference dataset (Janssen 2012 crosswalk study, n = 3,806). This captures the joint distribution of EQ-5D health state profiles in a general population. The model has three components:

  1. Empirical marginals: The CDF \(F_d\) for each ordinal dimension (5 levels), capturing the frequency distribution of each dimension independently.
  2. Polychoric/polyserial correlations: The copula correlation matrix \(\Sigma\) is estimated via polycor::hetcor() — polychoric correlations for ordinal×ordinal pairs and polyserial correlations for ordinal×continuous (VAS). This estimates the latent Gaussian correlations appropriate for ordinal data (~0.6 sec, vs ~24 sec for pseudo-likelihood).
  3. Grid probabilities: 3,125 state probabilities computed via Sobol quasi-random sampling ( N = 250,000 points via randtoolbox::sobol ). Sobol points are transformed through the inverse-normal CDF, correlated via \(\text{chol}(\Sigma)\), mapped back through the normal CDF, and binned into discrete cells via findInterval() .

The Gaussian copula takes the form:

$$\mathcal{C}_{\Sigma}(u_1,\ldots,u_5) = \Phi_5\!\left(\Phi^{-1}(u_1),\ldots,\Phi^{-1}(u_5);\,\Sigma\right)$$

where \(u_d = F_d(x_d)\) is the empirical CDF of dimension \(d\), \(\Phi^{-1}\) is the standard normal quantile, and \(\Sigma\) is the \(5 \times 5\) copula correlation matrix estimated via polychoric correlations.

2. Kernel Re-weighting

Transport works by re-weighting the reference population so that its weighted mean utility matches the reported target mean. Each profile \(i\) receives a kernel weight based on an anchor variable \(A_i\) (a value-set-neutral severity measure):

$$w_i \propto P(\mathbf{x}_i) \cdot K(A_i;\, \boldsymbol{\theta})$$

The default anchor is the mean dimension level \(\bar{d}_i = (\text{MO}_i + \text{SC}_i + \text{UA}_i + \text{PD}_i + \text{AD}_i) / 5\), which is entirely independent of any value set. This achieved the lowest mean absolute error (MAE = 0.009) in validation.

3. Kernel Functions

1-parameter mode (mean only): When only a mean is reported, a lognormal kernel is used with a fixed population-calibrated shape (\(\sigma_{\log} = 1.136\)): $$K(\bar{d}_i;\, \mu) = f_{\text{LN}}(\bar{d}_i - 0.9;\, \mu, 1.136)$$ The location parameter \(\mu\) is optimised to match the reported mean.

2-parameter mode (mean + SD): When a SD is also available, both location and shape are jointly optimised (L-BFGS-B with multi-start) to match both reported moments. If the lognormal kernel cannot achieve the target SD (ceiling ≈ 0.27), the system automatically switches to a Beta kernel :

$$K(\bar{d}_i;\, \alpha, \beta) = f_{\text{Beta}}\!\left( \frac{\bar{d}_i - 1}{4};\, \alpha, \beta\right)$$

The Beta kernel supports U-shaped weight functions (\(\alpha < 1, \beta < 1\)) that push mass to the extremes, reaching SD values up to ≈ 0.45.

4. Optimisation

Mean-only: Brent's method on a scalar objective: $$\hat{\mu} = \arg\min_\mu \left( \bar{m} - \sum_i w_i(\mu)\, U_{\text{src},i}\right)^2$$

Mean + SD: L-BFGS-B minimises the combined squared deviation: $$\min_{\boldsymbol{\theta}} \left[ \left(\bar{m} - \sum_i w_i\, U_{\text{src},i}\right)^2 + \left(s - \text{wSD}(U_{\text{src}}, \mathbf{w})\right)^2\right]$$ where \(\text{wSD}\) uses exact population variance via \(\sqrt{\mathrm{E}[U^2] - (\mathrm{E}[U])^2}\).

5. Transport

Once optimal weights are found, the transported mean for any target value set \(T\) is simply the weighted mean:

$$\hat{\mu}_T = \sum_i w_i^* \, U_{T,i}$$

Because all 3,125 profiles are pre-scored under every value set, transporting to multiple targets is instant once the weights are determined.

Validation

The method was validated against the Multi-Instrument Comparison (MIC) study, which provides individual patient-level EQ-5D data across 9 disease groups scored under 10 national value sets (810 transport pairs).

Configuration Mean MAE Max |Error|
Mean level, 1-param (mean only) 0.0086 0.1007
Utility anchor, 1-param 0.0102 0.0946
Mean level, 2-param (mean + SD) 0.0093 0.0949
Utility anchor, 2-param 0.0108 0.0962

MAE = mean absolute error in utility units (polychoric copula, Sobol QMC grid). The mean-level anchor with 1-parameter mode achieves the best mean accuracy. 2-parameter mode improves SD estimation at a modest cost in mean MAE.

Anchor Variables

Three anchor variables are supported, each offering a different trade-off:

Mean level
Arithmetic mean of the five dimension scores \(\bar{d} \in [1, 5]\). Entirely value-set-neutral. Recommended (best MAE).
Utility
Source value set utility score. Directly targets the reported scale, but inherits value-set-specific nonlinearities.
Consensus
Mean across all available value sets. Reduces idiosyncratic scale effects; performance between utility and mean-level anchors.

Cite this tool

Rand, K., Estévez-Carrillo, A. “A method for transporting reported mean EQ-5D utilities to different value sets”, EuroQol Plenary Meeting 2026, Lisbon, Portugal.

References

  1. Janssen MF, et al. (2013). Measurement properties of the EQ-5D-5L compared to the EQ-5D-3L across eight patient groups. Quality of Life Research 22(7):1717–1727.
  2. Devlin NJ, et al. (2018). Valuing health-related quality of life: An EQ-5D-5L value set for England. Health Economics 27(1):7–22.
  3. Richardson J, et al. (2014). The Multi-Instrument Comparison (MIC) project. PharmacoEconomics 32(12):1157–1172.
  4. Van Hout B, et al. (2012). Interim scoring for the EQ-5D-5L. Value in Health 15(5):708–715.
  5. EuroQol Group (1990). EuroQol — a new facility for the measurement of health-related quality of life. Health Policy 16(3):199–208.

Distributional data

Create, inspect, download and upload eqdistr reference distributions — the polychoric correlations and dimension marginals used as the source of a transport.

Create from a CSV of individual profiles

Upload a CSV with 5 columns mo, sc, ua, pd, ad (ordinal levels 1–5), one row per respondent (header optional). Only the distilled summary is kept — the individual data is not stored.

mo,sc,ua,pd,ad
1,1,1,2,1
2,1,1,3,2
1,1,2,2,1
...

Upload an existing eqdistr file

Accepted formats: .rds , .json , .yaml / .yml / .txt .

Distribution preview

Marginal level distribution
Polychoric correlations

Download

R object (.rds), JSON (.json), or human-readable (.yaml).

R object (.rds) JSON (.json) Readable (.yaml)

Reference

Rand, K., Estévez-Carrillo, A. “A method for transporting reported mean EQ-5D utilities to different value sets”, EuroQol Plenary Meeting 2026, Lisbon, Portugal.

Reference

Rand, K., Estévez-Carrillo, A. “A method for transporting reported mean EQ-5D utilities to different value sets”, EuroQol Plenary Meeting 2026, Lisbon, Portugal.

Transported Utility Means Across Anchors

One panel per selected anchor variable.


Comparison Table

Select methods and click 'Compare Methods'.

Reference

Rand, K., Estévez-Carrillo, A. “A method for transporting reported mean EQ-5D utilities to different value sets”, EuroQol Plenary Meeting 2026, Lisbon, Portugal.

EQ-5D Utility Value Transport: Methods and Validation

1. Background

The EQ-5D is a standardised multi-attribute utility instrument developed by the EuroQol Group. It comprises five dimensions - Mobility (MO), Self-Care (SC), Usual Activities (UA), Pain/Discomfort (PD), and Anxiety/Depression (AD) - each scored on a 5-level ordinal scale (EQ-5D-5L), plus a visual analogue scale (EQ VAS, 0-100).

Country-specific value sets (tariffs) convert EQ-5D profiles into a utility index anchored at 0 (dead) and 1 (full health); negative values are possible for states worse than death. Value sets differ substantially between countries because they reflect national societal preferences.

Why transport is needed: Clinical studies often report utilities referenced to a single national value set. When results are used in a different country's health technology assessment, the reported utilities must be adapted to the target country's value set. Because individual patient profiles are usually unavailable, only reported summary statistics (mean, possibly SD) can serve as inputs.

2. Statistical Method

2.1 Latent-Normal Reference Distribution

A reference dataset (Janssen 2012 crosswalk study, n = 3,806) is summarised by the empirical marginal level distributions of the five EQ-5D dimensions and their 5x5 polychoric correlation matrix (estimated with polycor::hetcor; theoretically matched to ordinal data and far faster than pseudo-likelihood). Health states are modelled as discretisations of a latent multivariate normal:

$$\mathcal{C}_{\Sigma}(u_1,\ldots,u_5) = \Phi_5\!\left(\Phi^{-1}(u_1),\ldots,\Phi^{-1}(u_5);\,\Sigma\right)$$

Expected probabilities for all 3125 EQ-5D-5L states are obtained deterministically from this distribution (Sobol quasi-random integration or exact rectangle probabilities) - no respondent-level resampling.

2.2 Re-weighting via Anchor Variable

Let \(A_i\) denote the anchor variable for profile \(i\). Weights \(w_i \ge 0\) are optimised so that the weighted mean of source utility values equals the reported aggregate mean \(\bar{m}\). The optimal weights are then applied to target value set utilities to obtain the transported estimate.

2.3 Anchor Variables (retained)
  • Utility anchor: Anchor on the source value-set utility score. Gaussian kernel weight: $$w_i \propto \exp\!\left(-\tfrac{(U_{\text{src},i}-\hat{\mu})^2}{2\hat{\sigma}^2}\right)$$
  • Consensus anchor: Anchor on a country-neutral utility score (mean across included value sets). Same kernel form; avoids value-set-specific scale effects.
  • Mean level anchor: Anchor on the mean EQ-5D dimension score \(\bar{d}_i = (\text{MO}+\text{SC}+\text{UA}+\text{PD}+\text{AD})/5\in[1,5]\). Entirely value-set-neutral. Optimal fixed bandwidth \(\hat{\sigma}=0.70\) (empirically calibrated). Achieves best mean MAE overall (0.0103 on MIC benchmark).
2.4 Optimisation

Mean-only transport (1 parameter, Brent method): $$\hat{\mu} = \arg\min_\mu \left(\bar{m} - \textstyle\sum_i w_i(\mu)\,U_{\text{src},i}\right)^2$$

With reported SD available, 2-parameter optimisation (L-BFGS-B) jointly minimises squared deviations from both reported moments \((\bar{m},\,s)\): $$\min_{\mu,\sigma}\left[\left(\bar{m}-\textstyle\sum_i w_i\,U_{\text{src},i}\right)^2 + \left(s - \text{wSD}(U_{\text{src}},w)\right)^2\right]$$

3. SD-Fitting Extension (2-Parameter Mode)

When a reported SD is available, 2-parameter optimisation jointly fits both mean and SD. Evaluation on the MIC benchmark (9 diseases, 10 value sets, 810 transport pairs) compares 1-parameter (mean only) and 2-parameter modes.

MIC Benchmark Results
Configuration Mean MAE Max |Error| Mean SD err Max SD err
Gaussian kernel + utility (1-param) 0.0120 0.0946 0.0261 0.0942
Gaussian kernel + consensus (1-param) 0.0120 0.0934 0.0258 0.0899
Gaussian kernel + mean level (1-param) 0.0103 0.1007 0.0455 0.1207
Gaussian kernel + utility (2-param) 0.0142 0.0962 0.0137 0.0574
Gaussian kernel + consensus (2-param) 0.0136 0.0958 0.0146 0.0687
Gaussian kernel + mean level (2-param) 0.0133 0.0949 0.0138 0.0616
Key Findings
  • Mean level (1-param) achieves the best mean MAE (0.0103) when only mean utility is reported. The value-set-neutral anchor avoids idiosyncratic scale effects of specific national tariffs.
  • 2-parameter mode dramatically improves SD accuracy (47-70% reduction in mean SD error vs 1-param) at a modest cost in mean MAE (+0.002-0.003).
  • When SD is reported, all 2-param anchors achieve comparable SD accuracy (~0.014 mean SD error), with mean-level having the lowest maximum errors.
  • Recommended strategy: use mean-level anchor (1-param) when only mean is available; use 2-param mode with any anchor when SD is also reported.

4. Dropped Methods

The following weighting functions were evaluated on the MIC benchmark and removed from the active implementation due to consistently inferior performance relative to the Gaussian kernel.

6.1 Triangular Weighting

$$w_i = \max\!\left(0,\, 1 - \frac{|A_i - \hat{p}|}{h}\right)$$

One-parameter method with a fixed bandwidth \(h\). Linear falloff creates hard cutoffs at the boundary, producing unstable gradients and poor performance at distribution tails. Mean MAE approximately 40% higher than Gaussian kernel on the MIC benchmark.

6.2 Gradient Weighting

Linear gradient from anchor minimum to maximum, tilted toward the estimated location \(\hat{p}\). Cannot represent peaked or narrow distributions. Consistently worst-performing across all disease groups (mean MAE > 0.030 with utility anchor).

6.3 Mixed Weighting

Arithmetic average of triangular and gradient weights. Inherits limitations of both components. Marginal improvement over gradient alone but substantially worse than Gaussian kernel.

6.4 Beta Density Weighting

$$w_i \propto f_{\text{Beta}(\hat{\alpha},\hat{\beta})}\!\left(\tfrac{A_i}{100}\right)$$

Two-parameter method fitting beta-distributed weights to the VAS anchor. Requires \(A_i \in [0,100]\) (VAS only) and is not applicable to utility- or mean-level anchors. Performance comparable to Gaussian kernel on VAS anchor but not portable to utility-based anchors; removed for consistency.

6.5 VAS Anchor

The EQ VAS (0-100) was evaluated as an alternative anchor to utility scores. While VAS is value-set-neutral by design, its anchoring mechanism differs fundamentally from utility-based transport: VAS reflects patients' subjective rating rather than societal preference-weighted health states. On the MIC benchmark, VAS anchor produced mean MAE 10-15% higher than utility anchor. VAS support is retained in the package API but no longer recommended for routine use.

6.6 Power-Law / Exponential Severity Models

Six parametric severity models (power-law, ROC-based, IRT, beta-binomial, probit, mixture) were tested as alternative marginal distributions for quantile-mapped re-weighting. All models failed to improve on simple independence baseline (2-8x worse MAE) because they assume identical marginal distributions across all 5 EQ-5D dimensions, contradicting the copula's dimension-specific empirical marginals (e.g., Self-Care: 70% at level 1; Anxiety/Depression: 40% at level 1). This fundamental incompatibility rules out parametric severity functions as a viable extension.

5. R Package: eq5dtransport

available_value_sets()
83 national EQ-5D value sets (43 x 5L + 40 x crosswalk).
eqdistr(data, source)
Summarises a dataset into polychoric correlations + marginals (shareable, no raw data).
default_eqdistr()
The bundled Janssen 2012 reference distribution (the app's default).
expected_probabilities(distr, ...)
Expected probabilities for all 3125 EQ-5D-5L health states.
transport_eq5d(mean_val, sd_val, from_vset, to_vsets, model, ...)
Transports a reported utility mean (and optionally SD).
compare_transport(mean_val, from_vset, to_vsets, model, anchors, ...)
Benchmarks anchor combinations against each other.

6. References

  1. Janssen MF, et al. (2013). Measurement properties of the EQ-5D-5L compared to the EQ-5D-3L across eight patient groups. Quality of Life Research 22(7):1717-1727.
  2. Devlin NJ, et al. (2018). Valuing health-related quality of life: An EQ-5D-5L value set for England. Health Economics 27(1):7-22.
  3. EuroQol Group (1990). EuroQol - a new facility for the measurement of health-related quality of life. Health Policy 16(3):199-208.
  4. Van Hout B, et al. (2012). Interim scoring for the EQ-5D-5L. Value in Health 15(5):708-715.
  5. Richardson J, et al. (2014). The Multi-Instrument Comparison (MIC) project. PharmacoEconomics 32(12):1157-1172.

Reference

Rand, K., Estévez-Carrillo, A. “A method for transporting reported mean EQ-5D utilities to different value sets”, EuroQol Plenary Meeting 2026, Lisbon, Portugal.

eq5dtransport 0.10.0 · eq5dtransportshiny 0.16.0