Beyond the Mean: Do Distribution-Based Metrics Offer Greater Insight Into Predictors of Societal Health Preferences?: A Secondary Analysis of 42 EQ-5D-5L Value Sets
Author(s)
Annushiah Vasan Thakumar, BSc, PhD1, Ling Jie Cheng, PhD, MPH, BSN (Hons), RN2.
1School of Pharmacy, Taylor's University, Subang Jaya, Malaysia, 2National Perinatal Epidemiology Unit, Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom.
1School of Pharmacy, Taylor's University, Subang Jaya, Malaysia, 2National Perinatal Epidemiology Unit, Nuffield Department of Population Health, University of Oxford, Oxford, United Kingdom.
OBJECTIVES: Relying on central tendency measures to summarise preference-based instrument data may mask underlying distributions of societal preferences, which are often left-skewed due to a concentration of high utility values. This study assessed whether distribution-based indicators offer additional explanatory power in predicting value set variability.
METHODS: We analysed 42 published general population-based EQ-5D-5L valuation studies. Each value set was summarised with the following seven summary indicators used as dependent variables: mean, median, skewness, interquartile range (IQR), 75th percentile (P75), 25th percentile (P25), and the value assigned to the 'pits' state (55555). Independent variables included publication year, global mean age, gender ratio, sample size, CREATE reporting quality score, WHO region, EQ-VT version (1, 2, or other), and valuation technique (composite time trade-off vs hybrid). We applied univariate and multivariable linear regression models to explore predictors for each indicator. Model performance was evaluated using R², root mean square error (RMSE) and mean absolute error (MAE).
RESULTS: These studies (2016-2025) were conducted in Europe (38%), Asia/Western Pacific (29%), Africa/Middle East (19%), and the Americas (14%). EQ-VT version 2 and the hybrid approach were most frequently used (67%). Region and EQ-VT version consistent predictors across indicators. Compared to Europe, value sets from Asia/Western Pacific reported significantly lower mean (β = -0.102, 95%CI [-0.199,-0.004]), median (β = -0.110, 95%CI [-0.207,-0.013]), and P75 (β = -0.089, 95%CI [-0.167,-0.010]) utilities. Compared to EQ-VT version 2, version 1 consistently predicted higher mean, median, and P25 but was not associated with P75. IQR demonstrated the strongest explanatory performance (R² = 48.8%, RMSE = 0.069, MAE = 0.050), with EQ-VT version 1 being the sole predictor (β = -0.114, 95%CI [-0.187,-0.042]).
CONCLUSIONS: Summary metric choice influences observed predictors, highlighting the limitations of relying solely on central tendency in EQ-5D-5L analyses. The strong explanatory performance of IQR highlights the importance of accounting for value dispersion.
METHODS: We analysed 42 published general population-based EQ-5D-5L valuation studies. Each value set was summarised with the following seven summary indicators used as dependent variables: mean, median, skewness, interquartile range (IQR), 75th percentile (P75), 25th percentile (P25), and the value assigned to the 'pits' state (55555). Independent variables included publication year, global mean age, gender ratio, sample size, CREATE reporting quality score, WHO region, EQ-VT version (1, 2, or other), and valuation technique (composite time trade-off vs hybrid). We applied univariate and multivariable linear regression models to explore predictors for each indicator. Model performance was evaluated using R², root mean square error (RMSE) and mean absolute error (MAE).
RESULTS: These studies (2016-2025) were conducted in Europe (38%), Asia/Western Pacific (29%), Africa/Middle East (19%), and the Americas (14%). EQ-VT version 2 and the hybrid approach were most frequently used (67%). Region and EQ-VT version consistent predictors across indicators. Compared to Europe, value sets from Asia/Western Pacific reported significantly lower mean (β = -0.102, 95%CI [-0.199,-0.004]), median (β = -0.110, 95%CI [-0.207,-0.013]), and P75 (β = -0.089, 95%CI [-0.167,-0.010]) utilities. Compared to EQ-VT version 2, version 1 consistently predicted higher mean, median, and P25 but was not associated with P75. IQR demonstrated the strongest explanatory performance (R² = 48.8%, RMSE = 0.069, MAE = 0.050), with EQ-VT version 1 being the sole predictor (β = -0.114, 95%CI [-0.187,-0.042]).
CONCLUSIONS: Summary metric choice influences observed predictors, highlighting the limitations of relying solely on central tendency in EQ-5D-5L analyses. The strong explanatory performance of IQR highlights the importance of accounting for value dispersion.
Conference/Value in Health Info
2025-11, ISPOR Europe 2025, Glasgow, Scotland
Value in Health, Volume 28, Issue S2
Code
MSR46
Topic
Methodological & Statistical Research, Real World Data & Information Systems
Topic Subcategory
PRO & Related Methods
Disease
No Additional Disease & Conditions/Specialized Treatment Areas