Norwegian and Swedish value sets for the EORTC QLU-C10D utility instrument

We relied on the previous methodology of King et al.[22]. The elicitation of EORTC QLU-C10D utility values relies on the assessment of trade-offs in health states using DCEs. This has been reported for various countries [22‐31]. Here, we determined the EORTC QLU-C10D utility weights for Norway and Sweden. This cross-sectional study includes responses from a general public online research panel for the countries Norway and Sweden.

Assessments took place from 10 October to 29 November 2022 and were self-completed by participants from any place in the respective country. Participants were recruited via a third-party provider online panel (https://​surveyengine.​com/​) and received a small payment for completing the assessment. We used quota sampling (by age and gender) in both general population samples to achieve representativeness for these two variables in each country. The initial eligibility criteria were sex (male, female) and aged between 18 and 80 years. Panel members who opted into the survey completed the introduction and disclosure sections, and those who consented progressed to the age and sex questions. Those who were younger than 18 years, did not declare sex male or female, fell within an age and sex quota that was already filled, or had entered the survey via a mobile (cell) phone were deemed ineligible for further participation and did not receive the compensation.

All participants provided electronic consent to participate, and the study was approved by the Ethics Committee of the Medical University of Innsbruck (approval number 1079/2023).

Participants provided information on socio-demographic data and clinical data (chronic diseases) and completed the EORTC QLQ-C30, the DCE to establish Norwegian and Swedish utility weights, the EQ-5D-5L [32], and the Kessler psychological distress scale (K10 [33]). Their feedback on the applicability and difficulty in completing the DCEs was also assessed. We do not report the EQ-5D-5L or the K10 data in this paper. Figure 1a and b shows the survey components and the order in which they were presented.

The EORTC QLU-C10D is a utility-scoring algorithm for the EORTC QLQ-C30 that was developed by the EORTC and the Multi-Attribute Utility in Cancer (MAUCa) Consortium [17, 20, 22]. The EORTC QLQ-C30 is a 30-item cancer-specific questionnaire that assesses functional health (physical functioning, role functioning, social functioning, emotional functioning, cognitive functioning), symptom burden (fatigue, pain, nausea/vomiting, sleep disturbances, dyspnoea, appetite loss, constipation, diarrhoea, financial impact), and global health or quality of life [16]. For the EORTC QLU-C10D, 10 key domains from the EORTC QLQ-C30 were selected: physical functioning, role functioning, social functioning, emotional functioning, pain, fatigue, sleep disturbances, appetite loss, nausea, and bowel problems (which merges the constipation and diarrhoea scales from the EORTC QLQ-C30). Items are scored on a four-point Likert scale to assess the severity levels from “not at all” to “very much”. The EORTC QLU-C10D describes a total of 4¹⁰ = 1,048,576 different health states [17]. The classification system of the EORTC QLU-C10D is shown in Supplementary Table 1. The robustness of the EORTC QLU-C10D valuation methodology and its psychometric properties, such as test–retest reliability, have been demonstrated in recent studies [20, 34‐36].

Utilities were elicited using DCE methodology, and the DCEs were performed analogously to the methodology of King et al. [20, 22]. More detailed information on the DCE is given in Supplementary Material Fig. 1. Participants were randomly assigned to complete 16 of a total of 960 possible choice sets. For each choice set, participants were instructed to select the preferred scenario. Scenarios described unique health states representing different severity ratings across the 10 domains of the EORTC QLU-C10D. In addition, each scenario was assigned a specific survival time to be lived in this state of health (levels: 1, 2, 5, or 10 years; “You live in this state for X years, and then die”). To keep participant burden acceptable, the scenarios within a choice set differed in only five of the 11 aspects (10 QLU-10D domains and survival time). The differing aspects were highlighted to increase their visibility for participants [20].

Descriptive statistics are reported for the sample characteristics including demographic and clinical information. Chi-square tests were used to compare the sample characteristics of gender, age, and education with national census data [37, 38]. To calculate utility weights, we followed an established procedure [39] that was also used in previous studies on the EORTC QLU-C10D [22‐24, 26‐31]. The utility (U) of an option “j” in the choice set “s” for a respondent “i” can be described by the following formula:where TIME_isj is the survival time in option j and X′_isj is a set of dummies related to the levels of the corresponding health state. Errors (ε_isj) are assumed to be independent and identically normally distributed [22, 39]. Generalized estimation equation models with first-order autoregressive covariance structure were used for parameter estimation (α, β) to adjust for correlated observations within respondents [26]. Our previous studies showed that these models produce estimators that are comparable to conditional logit cluster sandwich estimators [29]. As in previous valuation studies, estimates were converted into utility decrements based on the ratio of health state parameters (β) and the time coefficient (α) to reflect the trade-off between HRQOL and survival. Finally, as in previous valuation studies, we ran an additional analysis to correct for non-monotonic levels in the EORTC QLU-C10D utility decrements; i.e., for domains in which the severity level did not show an increasing decrement in utility weight. To do this, we combined non-monotonic levels and calculated the model with a reduced number of estimates in β.

Analyses were performed using IBM SPSS Statistics (version 27.0; IBM Corp., Armonk, NY, USA).

We also performed the following data quality checks. We determined the percentage of participants who always chose the same response option in each of the 16 scenarios of the DCE (always “A” or always “B”); smaller percentages of these cases indicated high data quality. To determine whether DCE results were affected by the speed of completing the survey, we removed the 10% fastest respondents and compared the utility decrements of the remaining 90% with those of the entire sample by determining the mean absolute difference (MAD) between the two sets of utility decrements.

An overview of the participant flows for Norway and Sweden are shown in Fig. 1a and b, respectively. A total of 22.499 persons were invited via email to participate in the survey. In Norway, 1696 persons entered the survey, and 307 were excluded because of quota sampling (i.e., they were not within the correct age or gender stratum). Of the remaining 1489 persons, 1019 (68.4%) completed the survey. In Sweden, 3563 persons started the survey, and 1517 were excluded for not being in the correct quota. Of the remaining 2046 persons, 1048 (51.2%) completed the survey.

The socio-demographic and clinical characteristics, and a comparison of these between the sample and census data for the general population are shown in Table 1. Because of quota sampling, the distributions of age and gender in the sample matched well with those in the population. Among both the Norwegian and Swedish participants, a significantly larger percentage of participants were highly educated in the sample than in the general population (details in Table 1). A smaller percentage of participants had chronic diseases in the Swedish sample compared with the general population, but this was not observed for the Norwegian sample.

Both in Norway and Sweden, most participants considered the DCE as clear or very clear (Norway 59.0%, Sweden 64.0%). Only a small percentage regarded the DCE as unclear or very unclear (Norway 17.5%, Sweden 11.7%). More than one-third of the participants in Norway (37.3%) and almost half of those in Sweden (46.9%) considered the DCE difficult or very difficult, and a considerable percentage regarded it as easy or very easy (Norway 37.4%, Sweden 25.0%). For the response strategy, a large percentage of the participants considered most or all aspects of the response sets (Norway 49.3%, Sweden 43.4%), and about 25% considered only the highlighted aspects (Norway 25.5%, Sweden 25.3%). A small percentage concentrated on only a few aspects (Norway 14.4%, Sweden 19.3%) or used other strategies (Norway 10.7%, Sweden 12.1%) (see supplementary Table 2).

Descriptions of the findings of the DCE analysis are given in Table 2 and Figs. 2 and 3 for Norway and in Table 3 and Figs. 4 and 5 for Sweden.

In Norway, problems with non-monotonicity (non-increasing utility decrements with increasing level of severity) were observed for seven of the 10 domains (all domains except physical functioning, role functioning, nausea). However, none of the cases of non-monotonicity reached statistical significance. Both the uncorrected and monotonicity-adjusted utility weights are shown in Table 3 (uncorrected values on the left and corrected values on the right). In the reminder of this paper, we concentrated on the monotonicity-adjusted findings. The largest utility decrements were seen for the domain of physical functioning (decrement of − 0.263 for highest level “very much”), followed by pain (decrement − 0.205 for “very much”) and role functioning (− 0.139). Among the cancer-specific domains, nausea showed the largest utility decrement (− 0.124), but the other cancer-specific domains had small utility decrements. The lowest possible utility value with a severity level of “very much” in all domains amounted to − 0.129 and corresponded to a utility worse than death.

In Sweden, lack of monotonicity was encountered only in two domains: fatigue and sleep disorders. Again, the effects of non-monotonicity (differences between adjacent levels) were small and did not reach statistical significance. Non-monotonicity was adjusted for in the adjusted utility values (Table 3). As in Norway, the largest utility decrements were observed for physical functioning (− 0.207 for level “very much”), followed by pain (− 0.139), role functioning (− 0.133), and nausea (− 0.119). Emotional functioning also exhibited sizable utility decrements (− 0.115) (see Table 3). The lowest possible utility value for the health state with a severity of “very much” in all domains was − 0.099, which was slightly larger than the corresponding value for Norway.

Utility weights can be calculated by subtracting the decrements (we recommend to use corrected decrements) from Tables 2 and 3 from 1 (state of perfect health). For example, the utility score for the worst possible health state (severity level of “very much” for all domains) would be calculated as follows for Sweden (see also https://​qol.​eortc.​org/​eortc-qlu-c10d/​):

Extreme response patterns (always response option “A” or “B” in the DCE) were observed only in a few cases: in 1.2% of the participants from Norway and in 1.5% from Sweden. After removing the 10% fastest respondents from the sample, DCE results remained relatively stable. In Norway, the differences in utility decrements between the complete and reduced samples ranged from − 0.032 to 0.019, and the MAD was 0.011. In Sweden, the differences ranged from − 0.014 to 0.012, and the MAD was 0.006.

In Norway, the National Health Insurance is responsible for evaluating medicines issued under its scheme. Additionally, the Norwegian National System for Managed Introduction of New Health Technologies within the Specialist Health Care Services guides funding decisions for the implementation of new technologies and medicines for use in Norwegian hospitals. These entities rely on principles of priority setting, which entail the benefit criterion, resource criterion, and severity criterion. The benefit criterion is specified further, and the use of QALYs is mentioned explicitly for quantifying the benefits [1, 2, 40].

In Sweden, two main entities are responsible for health economic evaluation. Firstly, 20 directly elected county councils are responsible for providing health services to the population and setting preferences about the types of services to be offered. Secondly, the Dental and Pharmaceutical Benefits Agency is responsible for drug appraisals of the prescription and disposal of medicines. Sweden relies on cost-effectiveness analysis in health economic appraisal, but this analysis is subordinate to the other two guiding principles of human value and need and solidarity when making resource-allocation decisions [1, 4]. In general, the use of CUA and QALYs is recommended by the Dental and Pharmaceutical Benefits Agency [41].

The importance of CUA and QALYs for prioritization and evaluation in health care is emphasized in governmental documents and guidelines, such as those mentioned above, in both Norway and Sweden. Although generic utility measures such as the EQ-5D and the SF-6D are used in cancer care, our current study offers national utility weights to be used with the cancer-specific EORTC QLU-C10D. The EORTC QLU-C10D is based on the widely used EORTC QLQ-C30 questionnaire [16], which has been used in academic and pharmaceutical research for decades, and data from clinical trials that included the EORTC QLQ-C30 can be used to generate utility scores (even retrospectively) that can be used to calculate QALYs. The Norwegian and Swedish findings for studies using the EORTC QLU-C10D can be compared to the growing body of cancer-specific utility weights for this questionnaire available for different countries [23, 24, 26‐31, 42].

The finding of the largest utility decrement for physical functioning from both Norway and Sweden has also been observed in other European countries such as Denmark [30], Germany [26], and Austria [31]. Overall, pain was the domain with the second-largest decrements and the largest for the symptom domains in the EORTC QLU-C10D questionnaire. This pattern has also been observed in other countries such as Denmark [30], Austria [31], Germany [26], Spain [29], France [27], and the UK [24]. The worst attainable health status (i.e., having the largest impairment in all domains) was − 0.129 for Norway and − 0.099 for Sweden; these values indicate that, in both countries, this health status would be rated as worse than being dead. In general, the utility decrements for the EORTC QLU-C10D in Norway and Sweden and the rank order of domains were similar to those in other European countries.

Normally, the Norwegian and Swedish peoples are presumed to be similar, and this appeared to be so for the variables examined in the present study. However, a few differences were identified in our study. The domain of emotional functioning had a larger utility decrement in Sweden than in Norway, and this difference suggests that emotional outcomes are more valued or considered to have a stronger impact in the Swedish population. We do not presume this to be a difference based on wording, as the translations of “depressed” are similar in Norwegian and Swedish. Similarly, a large impact of emotional function was identified in Denmark [30]. In international comparisons, only Denmark [30], France [27], the UK [24], and Australia [42] have larger utility decrements for emotional functioning than Sweden.

Pain was the second-largest utility decrement in both countries and the symptom domains with the largest decrement. The utility decrement for pain was larger in Norway than in Sweden, and the Norwegian decrement was similar to those reported in Dutch [28], French [27], and German [26] studies. This pattern was the opposite for fatigue in that a smaller utility decrement was observed in the Norwegian population than in the Swedish population. Other countries report utility decrements for the fatigue domain in between these two [26, 28, 29, 31].

In terms of monotonicity [38] of utility decrements, we found more problems with non-monotonicity for Norway (seven domains) than for Sweden (two domains). These may have resulted from the translation of the four response levels (“not at all”, “a little”, “quite a bit”, “very much”), which differs somewhat between the Norwegian and Swedish versions of the EORTC QLQ-C30 and EORTC QLU-C10D, with the wording of the Norwegian response options being a bit vaguer than the Swedish ones.

The participants from both countries included in the present study had the same age and gender distributions as the general populations in these countries [37, 38]. However, the Swedish participants had fewer chronic diseases compared with the general population [38], but this was not the case for the Norwegian participants [37]. The abovementioned differences may have influenced the external validity of the results, and the external validity may be better for the Norwegian than the Swedish findings.

A strength of this study is that we included 1019 Norwegian and 1048 Swedish participants who were representative of their respective general populations with regard to age and gender. The standardised methodology used in the present study is reliable [17], which enables sound comparisons between countries. One limitation of this study is that the participants are not representative of the Norwegian and Swedish population in terms of education level because the participants were more highly educated than the general population. A significantly smaller percentage of the Swedish participants had chronic diseases compared with the general population. Another limitation is that non-monotonicities were encountered; i.e., an increasing EORTC QLU-C10D domain level did not coincide with increasing utility decrements in some domains. In general, differences between adjusted and non-adjusted utility decrements were quite small (see Tables 2 and 3). These differences could be the result of respondents having problems differentiating between some adjacent levels in the DCE for some domains. However, none of these non-monotonicities reached statistical significance. Imposed constraints were used to remove non-monotonicities in the final analyses, as in all previous EORTC QLU-C10D valuations [22‐24, 26‐31].

Finally, we only had access to respondents who completed the whole survey. Hence we are not able to compare complete responders and partial responders, as for example, done previously [43], where only minor differences between responders and partial responders where found.