The study methodology, including preference elicitation methods, health state selection, and sampling, has been described in detail elsewhere ( 25 , 26 ), so this section only provides a brief overview of the original methods.
EQ-5D-Y-3L
EQ-5D-Y-3L consists of a descriptive system and a visual analog scale (EQ VAS). The descriptive system assesses health-related quality of life across five dimensions: mobility (walking), caring for myself (washing or dressing), daily activities (going to school, hobbies, sports, play, doing things with family or friends), pain or discomfort, worry, sadness or unhappiness. Each dimension has three response levels. Level 1 represents ‘no problems’, ‘no pain or discomfort’, ‘not worried, not sad or unhappy’. Level 2 indicates ‘a bit of trouble’, ‘a bit of pain or discomfort’ and ‘a bit worried, sad or unhappy’. Level 3 means ‘having a lot of problems’, ‘having a lot of pain or discomfort’, and ‘being very worried, sad or unhappy’. The respondent’s health status profile is represented as a five-digit string, with each number corresponding to a severity level for that dimension. For example, ‘12323’ indicates no problems walking, some problems with washing or dressing, many problems with daily activities, some pain or discomfort, very worried, sad or unhappy. Each dimension has three levels, so 243(35) unique health conditions. The Level Sum Score (LSS) can be calculated by adding up the five-digit health states, with possible scores ranging from 5 (full health: 11111) to 15 (worst health: 33333). Next, the EQ VAS captures self-rated health on a vertical scale from 0 to 100. Here, 0 and 100 represent ‘worst imaginable health’ and ‘best imaginable health’ respectively. Existing evidence on the measurement properties of the EQ-5D-Y-3L supports its applicability (27). The official Hungarian version of the EQ-5D-Y-3L was used in this study.
Data Description
A secondary analysis was performed using data collected in the Hungarian EQ-5D-Y-3L evaluation study, which received ethical approval from the Research Ethics Committee of Corvinus University of Budapest (KRH/31/2021) (25). The cTTO task was performed using EQ-VT (v2.1) software, which included both a traditional 10-year TTO assessment for states better than death and a lead-time TTO variant for states worse than death (i.e., 10 years in perfect health and 10 years in EQ-5D-Y-3L states). Interviews were conducted by four graduate students with prior experience in Hungarian EQ-5D-3L and EQ-5D-5L parallel assessment studies (28). All interviewers received standardized training in assessment methods, EQ-VT protocol, and quality control procedures. Each interviewer completed 50 interviews for the EQ-5D-Y-3L assessment study.
Each recruited respondent valued two health states (e.g. being in a wheelchair), three real EQ-5D-Y-3L health states (21112, 32323, 13311), and 10 ‘real’ EQ-5D-Y-3L states from the perspective of a 10-year-old child (exact wording: ‘I consider your view of a 10-year-old child’). Respondents rated 10 health states in random order: 3 mild (11112, 11121, 21111), 2 moderate (22223, 22232), 4 severe (31133, 32223, 33233, 33323), and worst health (33333). After completing the assessment task, respondents were provided with a ranked list of 10 health conditions based on their responses (the ‘feedback module’). This module allows respondents to flag health status assessments that they believe do not reflect their preferences, even if the responses appear consistent. In the final phase of the interview, respondents ranked another four EQ-5D-Y-3L health states as important, this time from their own (adult) perspective. These conditions were randomly selected from the same set of 10 health conditions that had previously been assessed from the child’s perspective. No ‘feedback module’ was used to assess the adult perspective.
Overall, 200 Hungarian adults, representative of the general population in terms of age and gender, completed the cTTO task. In addition to the assessment activities, respondents completed questionnaires covering sociodemographic characteristics (e.g., education, civil status, number and age of children, region of residence), health status (e.g., chronic diseases and self-rated general health), and self-complete versions of the EQ-5D-3L and EQ-5D-5L. They also used a five-point Likert scale (from ‘strongly disagree’ to ‘strongly agree’) to rate their agreement with the statement ‘Children’s lives are more valuable than adults’.
Statistical analysis
Each respondent values 10 health conditions from a child’s perspective and 4 health conditions from an adult’s perspective, allowing direct pairwise comparisons of the 4 health conditions valued from both perspectives. Prior to analysis, observations flagged in the EQ-VT ‘feedback module’ were excluded to ensure data quality. The distribution of cTTO values for child and adult perspectives were first visualized separately using histograms. To ensure that the observed patterns were not due to random variation, we compared the clustering of extreme values (-1.0 and 1.0) between perspectives using McNemar’s test. The distribution of paired differences (i.e., child minus adult values) was examined using medians and deciles, and a Wilcoxon signed rank test was performed to assess whether the median was significantly different from zero. To illustrate the differences between the two perspectives, a Bland-Altman plot was generated, plotting the difference (y-axis) relative to the mean value of each matched pair (x-axis) as color-coded dots with LSS. Differences in means between perspectives were assessed using Student’s t-test. Mean cTTO values were first compared based on individual health status profile and then by severity category (e.g. mild, moderate, severe/worst) to improve statistical power. When the number of observations was sufficient, we further performed subgroup analyzes only for values better than death and values worse than death. The classification was based on the child’s perspective values. For example, if a state was rated as better than dead from a child’s perspective but worse than dead from an adult’s perspective, that state was classified as better than dead. cTTO values exactly equal to 0 were classified as better-than-dead or worse-than-dead, depending on the type of assessment task the respondent completed (traditional TTO in the former, lead-time TTO in the latter).
Four multivariate linear regression models were estimated to explore predictors of differences in cTTO values between perspectives. A random intercept model was used to account for repeated observations per respondent, and robust standard errors were applied to account for heteroskedasticity. In the first two models, cTTO values were regressed separately for child and adult perspectives to examine perspective-specific associations using all responses for each perspective (i.e., not limited to matching observations). Predictors included LSS of valued health status (with values ranging from 6 to 15, centered around subtracting 6 to simplify interpretation), age, gender, education, region of residence, number of children, and the view that a child’s life is more valuable than an adult’s life (recorded as a binary variable due to limited variability; responses were dichotomized as agree versus disagree or neutral). Individual-level covariates were selected based on previous literature (19,20,21,22). For subsequent models, the analysis was restricted to health states that were valuable from both perspectives. In the third model, cTTO values were combined into a single regression and a dummy variable for perspective was included as an additional predictor (coded 1 for child perspective and 0 for adult perspective). This specification allows us to test the overall effect of perspective while controlling for the same set of respondent characteristics. In the fourth model, we estimated the difference in cTTO values using the same predictors (as in the first two models), using the difference between child and adult values for each matched health condition as the dependent variable. Analyzes were performed using Stata/MP 18 (StataCorp LLC, 2023) and Bland-Altman plots were generated using the ‘ggplot2’ package in RStudio 2024.12.1 + 563 (Posit Software, PBC). Statistical significance was set as follows: blood < 0.05.
