Mapping Disease-Specific Patient-Reported Outcome Measures (PROMs) to EQ-5D Utility Scores: A Systematic Review
Author(s)
Alessio Pignatelli, BS1, Davide Boccaccio, MS2.
1Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA, 2Politecnico di Torino, Torino, Italy.
1Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, USA, 2Politecnico di Torino, Torino, Italy.
OBJECTIVES: To identify existing techniques to map disease-specific PROMs to preference-based general utility values.
METHODS: EconLit, Embase, PubMed, Scopus were searched from database inception to April 28, 2025 for mapping studies between disease-specific PROMs and utility based on EQ-5D.
RESULTS: Of 780 identified abstracts, 576 duplicates were removed. Two blinded reviewers independently screened 204 abstracts, selecting 143 for full-text review, resulting in 132 articles included in the analysis, with oncology (35 studies), rheumatology (15) and orthopedics (12) being the most-represented fields. 96 studies employed direct mapping, while 4 response mapping and 32 both. Direct mapping techniques included Ordinary Least Squares (OLS), Censored Least Absolute Deviations (CLAD), Generalized Linear Models (GLM), Adjusted Limited Dependent Variable Mixture Model (ALDVMM), Beta regression, Tobit regression, and two-part models. Most common response mapping techniques were Ordinal or Multinomial Logistic Regression and Ordered Probit. Notably, less common techniques, like Multivariate Ordered Probit and Seemingly Unrelated Ordered Probit, were those that allowed for the modelling of correlation between dimensions of EQ-5D. Few articles used machine learning techniques, for either direct or response mapping. Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were the most common metrics for model selection. Nearly all studies validated mapping results either on a new set of data or through k-fold cross-validation, using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) as the primary metrics.
CONCLUSIONS: There is a need to map PROMs to preference-based general utility values for many diseases, and several mapping algorithms to EQ-5D have been developed. Multiple mapping techniques exist and there is no clear winner as the best-fit technique varied by study (even among studies on the same PROM). Comparatively less effort has so far been put into developing response mapping techniques that adequately capture the correlation between the EQ-5D dimensions.
METHODS: EconLit, Embase, PubMed, Scopus were searched from database inception to April 28, 2025 for mapping studies between disease-specific PROMs and utility based on EQ-5D.
RESULTS: Of 780 identified abstracts, 576 duplicates were removed. Two blinded reviewers independently screened 204 abstracts, selecting 143 for full-text review, resulting in 132 articles included in the analysis, with oncology (35 studies), rheumatology (15) and orthopedics (12) being the most-represented fields. 96 studies employed direct mapping, while 4 response mapping and 32 both. Direct mapping techniques included Ordinary Least Squares (OLS), Censored Least Absolute Deviations (CLAD), Generalized Linear Models (GLM), Adjusted Limited Dependent Variable Mixture Model (ALDVMM), Beta regression, Tobit regression, and two-part models. Most common response mapping techniques were Ordinal or Multinomial Logistic Regression and Ordered Probit. Notably, less common techniques, like Multivariate Ordered Probit and Seemingly Unrelated Ordered Probit, were those that allowed for the modelling of correlation between dimensions of EQ-5D. Few articles used machine learning techniques, for either direct or response mapping. Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were the most common metrics for model selection. Nearly all studies validated mapping results either on a new set of data or through k-fold cross-validation, using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) as the primary metrics.
CONCLUSIONS: There is a need to map PROMs to preference-based general utility values for many diseases, and several mapping algorithms to EQ-5D have been developed. Multiple mapping techniques exist and there is no clear winner as the best-fit technique varied by study (even among studies on the same PROM). Comparatively less effort has so far been put into developing response mapping techniques that adequately capture the correlation between the EQ-5D dimensions.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
SA64
Topic
Methodological & Statistical Research, Patient-Centered Research, Study Approaches
Topic Subcategory
Literature Review & Synthesis
Disease
Cardiovascular Disorders (including MI, Stroke, Circulatory), Musculoskeletal Disorders (Arthritis, Bone Disorders, Osteoporosis, Other Musculoskeletal), Neurological Disorders, Oncology, Urinary/Kidney Disorders