Study design and sample
The study used data from the Norwegian EQ-5D-5L valuation study. Data collection started in November 2019, but stopped in March 2020 due to the COVID-19 pandemic, at which point 542 interviews were completed. The study intentionally oversampled selected groups typically hard to reach, including ethnic minorities, those with lower socio-economic status and parents of young children [
22].
Respondents were invited to the study via randomly sampled locations within different geographic areas in Norway and location type strata aimed at reaching different respondent groups. Contact persons at each location assisted, where feasible, to meet quotas according to gender and age. Child day-care facilities and primary schools were sampled to increase the number of respondents with young children.
Interviews and questionnaires
Data were collected by one-to-one pc-assisted interviews, following standard EQ-VT protocol version 2.1 [
9], and guided by a trained interviewer. See the original study protocol for more details on training and use of valuation technology [
22]. Interviews were completed at sampled locations, for example libraries, schools, workplaces or recreational centres. Where possible, interviews were completed in separate rooms. Standard EuroQol quality controls (QC) were assessed throughout data collection [
23], with flags related to time spent on the task, the introduction to lead-time TTO, and inconsistent valuations of the worst possible health state. Protocol compliance was found to be excellent, with few interviews flagged for poor data quality.
Interviews were conducted using the EQ-PVT, a portable version of the EQ-VT software developed by EuroQol. The EQ-PVT provides a similar visual presentation of the TTO tasks as the EQ-VT software, presented as two horizontal scales indicating number of years in Life A (a life in full health) and Life B (a life in the health state to be valued). The respondent values each health state by choosing between Life A and Life B in an iterative process until the respondent perceives the two lives to be of about the same value.
Following EQ-VT protocol, composite time trade-off (cTTO) was administered [
8,
24]. The cTTO is a modified version of the TTO, where lead-time TTO is used for the valuation of states identified as worse than being dead (WTD). When states are judged to be WTD, the respondent is offered an additional 10 years in full health lead-time in Life B, a total of 20 years (10 years in full health, followed by 10 years in the health state to be valued), as an alternative to 10 or fewer years in full health in Life A.
Interviewers followed a standardised interview guide with scripted introduction and recommended responses, introducing all parts of the cTTO task, including the concept of WTD, and how to give such values using the cTTO. Respondents practiced by valuing three practice states before completing 10 cTTO valuations.
The interviewer guided the respondent through the entire interview and answered questions respondents had throughout. The interviewer was instructed to not comment on seemingly illogical responses, but to encourage respondents to think aloud and carefully consider each health state presented. After completing all TTO tasks, respondents were asked to review responses and flag any they deemed inconsistent in a feedback module, without comment from the interviewer.
In addition to the valuation tasks, each respondent defined their own health with the EQ-5D-5L and VAS, and completed a paper questionnaire with items describing their background. The standard EuroQol visual analogue scale (EQ VAS) from 0 to 100 was used, with 100 representing best imaginable health and 0 worst imaginable health. Respondents defined their own health state prior to completing valuation tasks, and to conclude completed the rest of the questionnaire, including questions about significant others.
Information on significant others was collected from questionnaire items where respondents indicated how many children under 18 years of age they had responsibility for, as well as their marital/partner status. The items were formulated as “Do you have responsibility for children under the age of 18?”, where respondents indicated the number of children for whom they were responsible, and “What is your marital status?”, with the response categories “Single”, “Married”, “Cohabiting”, “Divorced/separated”, “Widowed”. Responses for these items were recoded to “with children under 18 years of age” if they stated that they had responsibility for at least one child under 18 years of age, and with a partner if they indicated that they were either married or co-habiting.
Statistical analysis
Descriptive statistics summarized respondent characteristics. Linear regression models assessed the association of the main analysis variables with use of the feedback module and QC flags.
Each respondent provided ten individual TTO valuations, all of which were included in the primary analyses, irrespective of flagging in the feedback module. To account for the nested nature of the data, a mixed model with a random intercept at the respondent level was used to estimate the effects of having significant others on willingness to trade. We used disutility (= 1-utility) in the analyses. Values elicited using the cTTO procedure are left-censored at − 1. Correspondingly, disutility values were handled as being right-censored at 2, i.e. a Tobit model. The effect of age on elicited values was explored prior to final modelling using descriptive methods and loess regression (Supplementary Fig. 1).
We tested five different models. Model 1 included only having children as a significant other, as well as age, sex, and higher education. Model 2 was similar to Model 1, but with a dummy variable for having a partner instead of children as significant other. Model 3 included both variables for having children and having a partner, and Model 4 included an interaction between the two. Model 5 included only a dummy variable for any significant other, indicating either children or a partner, in addition to age, sex and higher education, as in previous models. We defined dummy variables coded 1 for: individuals with children < 18 years (CHILD); individuals living with a partner or married (PART); individuals with either children or a partner (SIGNIF); female respondents (FEM); individuals with higher education (EDU). Age in years was included as a continuous variable (AGE). For more flexible modelling and to account for the non-linear relationship of age and disutility, we made use of natural splines; a form of flexible interpolation by use of a pre-defined set of polynomials. In the equation, ns represents a function for cubic (3-knot) natural splines. Knots were placed at the quartiles of age in the data, giving four estimates in total. Final number of knots was determined by the Akaike information criterion. The five models were defined as following:
\({\text{Model 1: }}\,disutility \sim \alpha + \beta_{{ns\left( {AGE} \right)}} + \beta_{EDU} + \beta_{FEM} + \beta_{CHILD } + b_{0id}\).
\({\text{Model 2: }}\,disutility \sim \alpha + \beta_{{ns\left( {AGE} \right)}} + \beta_{EDU} + \beta_{FEM} + \beta_{PART } + b_{0id}\).
\({\text{Model 3: }}\,disutility \sim \alpha + \beta_{{ns\left( {AGE} \right)}} + \beta_{EDU} + \beta_{FEM} + \beta_{CHILD} + \beta_{PART } + b_{0id}\).
\({\text{Model 4: }}\,disutility \sim \alpha + \beta_{{ns\left( {AGE} \right)}} + \beta_{EDU} + \beta_{FEM} + \beta_{CHILD} + \beta_{PART } + \beta_{CHILD:PART} + b_{0id}\).
\({\text{Model 5: }}\,disutility \sim \alpha + \beta_{{ns\left( {AGE} \right)}} + \beta_{EDU} + \beta_{FEM} + \beta_{SIGNIF } + b_{0id}\).
Sensitivity analyses controlled for interviewer effects, being married versus co-habiting, the health state valued, respondent’s self-reported health (EQ VAS), and included respondents with missing values for number of children, and excluded responses flagged in the feedback module. We coded missing values for children as not having indicated any children under the age of 18. To control for the health state we performed two analyses, including dummy variables per level per dimension of the health state, as well the level sum score (representing the deviation from full health) as a measure of the health state’s general severity.
R version 3.6.2 was used for the statistical analyses [
25]. We chose a 5% significance level, using two-sided tests.
The Regional Committee for Medical and Research Ethics reviewed the study and stated that their approval was not required. The Norwegian Institute of Public Health approved the Data Protection Impact Assessment for the study 30th September 2019.