Country-Related Differential Item Functioning in the EQ-5D-5L: Insights From the EuroQol Data Assessment of Population Health Needs and Instrument Evaluation (EQ-DAPHNIE)
Author(s)
Jiabi Wen, MSc1, Ademola J. Itiola, MPH, MSc1, Jeffrey A Johnson, PhD1, Mathieu M.F. Janssen, PhD2, Paula Lorgelly, BSc, PhD3, Tolulope Sajobi, PhD4, Fatima Al Sayah, PhD1.
1University of Alberta, Edmonton, AB, Canada, 2Maths in Health, Klimmen, Netherlands, 3University of Auckland, Auckland, New Zealand, 4University of Calgary, Calgary, AB, Canada.
1University of Alberta, Edmonton, AB, Canada, 2Maths in Health, Klimmen, Netherlands, 3University of Auckland, Auckland, New Zealand, 4University of Calgary, Calgary, AB, Canada.
OBJECTIVES: International health surveys such as EQ-DAPHNIE support cross-country comparisons of population health. For such comparisons to be valid, participants across countries should interpret health status measures consistently. Violations lead to differential item functioning (DIF): individuals with the same underlying health respond differently by country. This study examined the presence and magnitude of DIF in EQ-5D-5L dimensions using EQ-DAPHNIE data.
METHODS: EQ-DAPHNIE collected online cross-sectional data from 15 countries, targeting 4,500 respondents per country. Proxies for underlying health (trait) included EQ-5D-5L level sum score (LSS), EQ-VAS, EQ-HWB-9 LSS, and PROMIS-10 physical and mental health t-scores. Both uniform (constant across trait) and non-uniform (varying by trait) DIF were explored. DIF was assessed in three stages: 1) among English-speaking countries, 2) between English- and non-English-speaking countries, and 3) among non-English-speaking countries. Ordinal logistic regression models were used with EQ-5D-5L dimensions as outcomes, and trait, country, and their interaction as predictors. Meaningful DIF was defined as a two‐degree‐of‐freedom chi‐square test of country and interaction terms with p≤0.01 and a ≥3.5% point increase in pseudo-R² (Effect size: 3.5-7.0%, moderate; >7.0%, large).
RESULTS: Of 68,416 respondents, 67,178 (98%) had complete EQ-5D-5L data. No DIF was found among English-speaking countries (Australia, Canada, New Zealand, UK, US) or in comparisons with the ten non-English-speaking countries. Among non-English-speaking countries, no DIF was detected within Spanish-speaking (Spain, Argentina, Chile, Mexico) or Germanic (Germany, Netherlands) countries. However, moderate DIF (ΔR²=3.5-6.5%) appeared in four dimensions (excluding pain/discomfort) in comparisons between countries with different languages. Self-care dimension exhibited the most DIF: China consistently reported more problems than Japan, France, and Brazil.
CONCLUSIONS: DIF was not detected among English-speaking or between English- and non-English-speaking countries, supporting comparability across these groups. However, moderate DIF emerged among non-English-speaking countries, which may be due to linguistic adaptations or cultural differences. Potential bias should be considered when comparing EQ-5D-5L across countries.
METHODS: EQ-DAPHNIE collected online cross-sectional data from 15 countries, targeting 4,500 respondents per country. Proxies for underlying health (trait) included EQ-5D-5L level sum score (LSS), EQ-VAS, EQ-HWB-9 LSS, and PROMIS-10 physical and mental health t-scores. Both uniform (constant across trait) and non-uniform (varying by trait) DIF were explored. DIF was assessed in three stages: 1) among English-speaking countries, 2) between English- and non-English-speaking countries, and 3) among non-English-speaking countries. Ordinal logistic regression models were used with EQ-5D-5L dimensions as outcomes, and trait, country, and their interaction as predictors. Meaningful DIF was defined as a two‐degree‐of‐freedom chi‐square test of country and interaction terms with p≤0.01 and a ≥3.5% point increase in pseudo-R² (Effect size: 3.5-7.0%, moderate; >7.0%, large).
RESULTS: Of 68,416 respondents, 67,178 (98%) had complete EQ-5D-5L data. No DIF was found among English-speaking countries (Australia, Canada, New Zealand, UK, US) or in comparisons with the ten non-English-speaking countries. Among non-English-speaking countries, no DIF was detected within Spanish-speaking (Spain, Argentina, Chile, Mexico) or Germanic (Germany, Netherlands) countries. However, moderate DIF (ΔR²=3.5-6.5%) appeared in four dimensions (excluding pain/discomfort) in comparisons between countries with different languages. Self-care dimension exhibited the most DIF: China consistently reported more problems than Japan, France, and Brazil.
CONCLUSIONS: DIF was not detected among English-speaking or between English- and non-English-speaking countries, supporting comparability across these groups. However, moderate DIF emerged among non-English-speaking countries, which may be due to linguistic adaptations or cultural differences. Potential bias should be considered when comparing EQ-5D-5L across countries.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR64
Topic
Methodological & Statistical Research, Patient-Centered Research
Topic Subcategory
PRO & Related Methods
Disease
No Additional Disease & Conditions/Specialized Treatment Areas