Development and preliminary testing of the new five-level version of EQ-5D (EQ-5D-5L)
- First Online:
- 9.7k Downloads
This article introduces the new 5-level EQ-5D (EQ-5D-5L) health status measure.
EQ-5D currently measures health using three levels of severity in five dimensions. A EuroQol Group task force was established to find ways of improving the instrument’s sensitivity and reducing ceiling effects by increasing the number of severity levels. The study was performed in the United Kingdom and Spain. Severity labels for 5 levels in each dimension were identified using response scaling. Focus groups were used to investigate the face and content validity of the new versions, including hypothetical health states generated from those versions.
Selecting labels at approximately the 25th, 50th, and 75th centiles produced two alternative 5-level versions. Focus group work showed a slight preference for the wording ‘slight-moderate-severe’ problems, with anchors of ‘no problems’ and ‘unable to do’ in the EQ-5D functional dimensions. Similar wording was used in the Pain/Discomfort and Anxiety/Depression dimensions. Hypothetical health states were well understood though participants stressed the need for the internal coherence of health states.
A 5-level version of the EQ-5D has been developed by the EuroQol Group. Further testing is required to determine whether the new version improves sensitivity and reduces ceiling effects.
KeywordsHealth-related quality of life EQ-5D Development 5 level
The EQ-5D is a generic instrument for describing and valuing health. It is based on a descriptive system that defines health in terms of 5 dimensions: Mobility, Self-Care, Usual Activities, Pain/Discomfort, and Anxiety/Depression . Each dimension has 3 response categories corresponding to no problems, some problems, and extreme problems. The instrument is designed for self-completion, and respondents also rate their overall health on the day of the interview on a 0–100 hash-marked, vertical visual analogue scale (EQ-VAS). The EQ-5D has been widely tested and used in both general population and patient samples and has been translated into over 130 different language versions [www.euroqol.org].
The EQ-5D was designed to measure decrements in health. Substantial use of the instrument has shown that it can suffer from ceiling effects, particularly when used in general population surveys but also in some patient population settings [2, 3, 4, 5, 6, 7, 8]. As a result, there may be issues regarding its ability to measure small changes in health, especially in patients with milder conditions. In light of these possible limitations, and stimulated by demand from the clinical field, the EuroQol Group decided to explore ways of improving the EQ-5D’s measurement properties.
In 2005, a task force was established within the EuroQol Group to investigate methods to improve the instrument’s sensitivity to small and medium health changes and to reduce ceiling effects. Initial discussions focused both on expanding the descriptive system by adding additional dimensions and also on expanding the number of levels of severity in each dimension . The task force decided that there should be no change in the number of dimensions for a new version of EQ-5D. Twenty-five years’ experience of using the EQ-5D has provided evidence that the original choice of dimensions was a reasonable one, though there are some areas in which the range of dimensions included may not be optimal [10, 11]. Moreover, the EuroQol Group has considerable experience with the measurement and valuation of health using the current five dimension model and retaining that model would allow for an easier transition from the existing EQ-5D to a new version.
In terms of the number of levels per dimension, previously published studies by EuroQol Group members showed that prototype 5-level versions of EQ-5D could significantly increase reliability and sensitivity (discriminatory power) while maintaining feasibility and potentially reducing ceiling effects [12, 13, 14, 15]. The choice of a five-level descriptive system is also supported by substantial psychometric literature [16, 17, 18]. It was therefore decided that the new version of the EQ-5D should include five levels of severity in each of the existing five EQ-5D dimensions and that the new version would therefore be called the EQ-5D-5L. The existing EQ-5D will be renamed the EQ-5D-3L, which is how it will be referred to in the rest of this paper.
The objectives of the current study were to select severity labels for the EQ-5D-5L and to test the face and content validity of the resulting instrument. The study was performed simultaneously in the United Kingdom and Spain.
The EuroQol Group task force recommended that English and Spanish versions be developed in parallel, where they could also serve as root languages for further translations and adaptations of the expanded version.
The study consisted of two phases. In the first phase, carried out from June to November 2007, a pool of potential labels for the new levels was identified and provisional labels for the 5-level version were chosen from that pool after a response scaling task carried out in face-to-face interviews with convenience samples of lay respondents. In the second phase, carried out from May to July 2008, face and content validity of two alternative 5-level systems were tested in focus group sessions with healthy participants and those with chronic illness. The second phase was also used to test the face validity of a series of health states based on the 5-level versions. Different groups of respondents were used in the two phases of the study.
Participants in both phases were recruited to ensure a wide range of socio-demographic characteristics. For the response scaling phase, the UK participants were recruited via local newspaper advertisements, local community advertisements, and from an existing participant database. The Spanish participants were recruited from among parents from local schools and from patient associations. Patient focus groups included primarily individuals with arthritis, diabetes, or asthma. In all groups, adequate written and oral fluency in English or Spanish was required.
Written informed consent to participate was obtained from all participants in both phases of the study.
Phase 1: response scaling
Potential labels for the EQ-5D-5L were identified from a review of existing health-related quality-of-life instruments, a review of the literature on response scaling, hand searching of dictionaries and thesauruses, and informal interviews with native speakers of the target languages to establish how they described different severities of health problems. The same process was carried out in English and Spanish and, where possible, equivalent terms were sought in both languages. Labels included in the initial pool clearly had to fit with the lexical structure used in the EQ-5D-3L, such as ‘I have no problems doing my usual activities’ and ‘I have some problems doing my usual activities’.
In order to select labels from the pool for the new levels, an interviewer-administered response scaling exercise similar to those used in previous studies [14, 19, 20] was adopted to estimate the severity represented by each label. For this exercise, respondents were shown a rating scale in the form of a vertical, hash-marked, 40 cm visual analog scale (VAS) with end points of 0 and 100 to be used as a visual aid in grading label severity. For the Mobility, Self-Care and Usual Activities dimensions, the same set of labels was used. The interviewer placed a card labeled ‘No problems’, ‘No pain/discomfort’, or ‘No anxiety/depression’ as appropriate at the bottom of the scale (0) to act as the lower anchor and a card labeled ‘Unable to, ‘The worst pain or discomfort I can imagine’, ‘As anxious or depressed as I can imagine’ as the upper anchor (100). The respondent was then shown other labels from the pool singly in a quasi-random order and asked to assign a score between 0 and 100 to indicate label severity in relation to the lower and upper anchors.
The interviewer noted all scores, and when the respondent had rated all labels for a particular dimension, the interviewer laid them out in rank order alongside the VAS and asked the respondent to review the ranking and make any changes he or she thought necessary. If labels were reordered at this point, the respondent was asked to assign a new score to the relevant labels. Final scores assigned were recorded in an answer booklet. The scaling task was repeated for each dimension. Before finishing with the cards, the respondent was asked whether any of the labels sounded unusual, or should not be used in relation to a particular dimension.
Respondents rated labels for all five dimensions. The three functional dimensions (Mobility, Self-Care and Usual Activities) were always interspersed by the Pain/Discomfort and Anxiety/Depression dimensions, so that the respondent did not rate the same label types consecutively. Before rating the actual labels, respondents performed a practice task based on levels of overall health to get used to the study requirements. Data on age, level of education, main activity, and use of any current treatment for health problems, together with the existing EQ-5D-3L descriptive system and EQ-VAS, were collected after the response scaling task.
Before the main response scaling task, a pilot test was performed to test study procedures and materials. Based on the results of the pilot study, some labels were eliminated from the initial pool to achieve a more manageable number for the response scaling task. In particular, any labels using additional modifiers such as ‘very’ or ‘quite’ were eliminated as were any that were considered excessively colloquial or too high a level of language. After pilot testing, it was concluded that the feasible limit was about 10–12 labels per dimension for an individual respondent.
Responses to the scaling task were analyzed by calculating means and medians and the corresponding standard deviations and interquartile ranges (IQR). Labels to go forward for further testing were selected based on criteria that had been identified before data collection started. These included selecting labels close to or at the 25th, 50th, and 75th centiles on the VAS, ensuring consistency across dimensions and coherence with wording in the descriptive system. No quantitative comparison of label scores was carried out in deciding which labels to carry forward to the next stage; median scores were simply used as a guide to determine which labels fell closest to the 25th, 50th, and 75th centiles. Labels were also required to be in colloquial language. The choice of labels and their appropriateness was discussed by the task force at several meetings during the course of the study.
Phase 2: testing the face and content validity of alternative 5-level versions
The results of the response scaling task led to an intermediary result of two, rather than one, alternative 5-level versions in both UK English and Spanish (for an explanation, see Results). The second part of the study aimed to assess the ease of use, comprehension, interpretation, and acceptability of these two versions and to use these results to decide on a final, definitive version for validation work. A further aim of this part of the study was to evaluate the face validity of some hypothetical health states generated by the 5-level descriptive systems. To this purpose, the two alternative versions were tested in 8 focus groups in each country (total of 16 groups); four of these were composed of healthy participants and four under treatment for a health condition.
Groups were led by an experienced moderator, and sessions were audio-recorded and transcribed for analysis. A previously prepared script was followed in all groups. All participants in each group first completed either Alternative 1 or Alternative 2 of the EQ-5D-5L (depending on the group they were assigned to), followed by the EQ-VAS. Participants were then asked to review their answers and what they had thought about while they completed the survey. Further questions were used to probe their reactions to the questionnaire in more detail, particularly their reactions to the severity labels used. Participants then provided socio-demographic information before being asked to complete the complementary Alternative 2 or Alternative 1, again on their own, after which there was further group discussion on their reactions. At the end, participants were asked their preferences for the alternative descriptive systems. The order of administration of versions 1 and 2 was alternated between the groups to control for possible ordering effects, and groups were assigned randomly to the different orders.
Examples of two of the health states tested in the phase 2 focus groups
Health state 1
Slight (mild) problems in walking about
No problems washing or dressing myself
Unable to do my usual activities
Slight pain or discomfort
Not anxious or depressed
Health state 2
Severe problems in walking about
Moderate problems washing or dressing myself
Slight problems doing my usual activities
Severe pain or discomfort
Extremely anxious or depressed
The focus groups were run using a structured ‘script’ or guide, so the analysis was based initially on grouping and contrasting participant statements relating to each of the specific issues addressed. Thematic content analysis  was used to explore issues in more depth and to examine the transcripts for other, non-scripted statements and expressions.
Sample characteristics of participants in the response scaling task with UK and Spain general population figures for comparison
UK (n = 40)d
Spain (n = 37)*
UK general populatione
Spain general population
Low (no schooling or only primary)
Middle (left school 16–18 yrs)
High (university or similar)
In paid employment
Looking for work
Looking after home/family
EQ-VAS score, mean (SD)
N (%) of respondents in EQ-5D state 11111
Comparison of median (IQR) scores for mobility, self-care, and usual activities labels, UK and Spain (Spanish labels in parenthesis)
Comparison of median (IQR) scores for pain/discomfort and anxiety/depression labels, UK and Spain
Based on this decision, two alternative 5-level versions were identified: in the case of the functional dimensions, the UK alternatives tested were ‘No problems-Minor problems-Moderate problems-Major problems-Unable to’ and ‘No problems-Slight problems-Moderate problems-Severe problems-Unable to’. In the Pain/Discomfort and Anxiety/Depression dimensions, alternative labels tested were ‘mild’ and ‘slight’ as the second level, and ‘severe pain’ or ‘a lot of pain’ and ‘severely’ or ‘very’ anxious or depressed at the 4th level. A similar process for label selection was followed in Spain.
Sample characteristics of respondents in the focus groups; healthy participants and patient groups, UK and Spain
Healthy N = 15
Patientsa N = 15
Healthy N = 18
Patients N = 19
Women, N (%)
Years, mean (SD)
Educational level, N (%)
Further education after leaving school
Main activity, N (%)
In both Spain and the UK, participants generally found both of the two alternative versions easy to understand and complete, giving comments such as “Questions are well-formulated and specific”. With reference to the new severity labels, participants commented that “they are very clear points, and there is no doubt that you go from less to more in each dimension” and that “all different levels seem to be covered”. Some Spanish respondents thought that it might be difficult to distinguish between some of the labels, particularly at the lower end of the scale. However, the results of the response scaling exercise and comments regarding the type of problems reflected by each of the labels suggested that most respondents were perfectly able to distinguish between the different labels used.
The alternative versions were not equally attractive and in both countries participants tended to prefer version 2, which used ‘slight’, ‘moderate’, and ‘severe’ for the central levels in the Mobility, Self-Care, and Usual Activities dimensions, as opposed to ‘minor’, ‘moderate’, and ‘major’ problems. The latter were generally considered less colloquial. A typical comment was that you might use them “talking to a doctor or something…but I can’t imagine saying to a friend or family having minor problems walking about”. ‘Slight’ and ‘severe’ were described by one participant as being “common language” which “would trigger a response without having to think about it a lot”. A smaller number of participants did prefer ‘minor’ and ‘major’, suggesting that it was ‘more modern language’; other participants suggested that there was very little difference between the alternative sets of labels.
In the Pain/Discomfort and Anxiety/Depression dimensions, participant preferences regarding labels were not so clear. In both the UK and Spanish versions, therefore, it was decided to maintain the same scaling as in the functional dimensions (‘slight’, ‘moderate’, ‘severe’).
Participants’ comments regarding the way they interpreted the severity labels showed that the labels functioned well at the intended level of measurement. For example, to describe ‘slight problems’ in the self-care dimension a patient suggested “Maybe if you have a pulled muscle in your back and it’s difficult to wash your hair.” When referring to ‘moderate problems’ with mobility, participants explained that “even though I have to use a crutch to get around, I can still get up on my own, and I can get around” and “I have moderate problems with walking because of my knee. …it describes it very well…neither a lot nor a little.” On the other hand, to describe ‘severe problems’ in this dimension, examples included people who experienced great pain when walking due to arthritis or a herniated disc.
Testing of health states based on new labeling systems
Participants found it relatively easy to understand the health states, whichever version was used. In fact, comments focused more on health state content, and particularly on what they saw as contradictions or a lack of realism in the health states, rather than on the way the health states were worded. For example, one respondent said that for her “washing and dressing are everyday activities and are therefore covered by usual activities [so the two dimensions should not be separate]”. On the other hand, the labels used were not an impediment to understanding the health states, and participants were in general easily able to distinguish between health states. Both alternative versions of the 5-level descriptive system appeared to work equally well in this sense, though when asked explicitly the majority of participants preferred the ‘slight-moderate-severe’ alternative in the first three dimensions.
This paper reports the process and results of developing a new 5-level version of the EQ-5D in UK English and Spanish for Spain. By using response scaling and focus groups, it was possible to develop 5-level versions in UK English and Spanish that have demonstrated initial content and face validity. The results of the response scaling exercise suggest that the labels selected are well-distributed across the health continuum and that their distribution was similar in the two countries.
Although 5-level versions of the EQ-5D have previously been developed and applied, they were experimental versions prepared and tested by individual group members or research teams [12, 13, 14, 15]. The UK English and Spanish versions reported here are the first to be produced as a result of an official EuroQol Group initiative, and they should be considered the definitive versions, dependent on further testing of validity, reliability, and sensitivity to change. The opportunity was also taken to harmonize wording within the instrument by, for example, rewording the poor health extreme of the Mobility dimension as ‘Unable to walk about’ instead of ‘Confined to bed’.
As regards the decision to use 5 levels in each dimension, this issue was discussed at length as it was also possible to use different numbers of levels across domains (in fact, the first version of the EQ was a six-domain instrument, with 3 domains having 3, the others having 2 levels ). Two lines of argument resulted in the choice of a uniform five-level instrument. First of all, there seemed to be no natural or obvious argument to apply different levels: all domains of the current EQ refer to ‘uncountable’ entities, where the full range must be referred to by general grading terms. These can be based on frequency or intensity of dysfunction/disability, but the principle is the same for all EQ domains. Likewise, we had no a priori preference for trying to discriminate more (or less) on a given domain than on others. Second, there are obvious practicalities in having an equalized system. Self-report of own health status (for description) and trade-off tasks (for valuation) are arguably easier to explain and understand: using a dissimilar number of levels may lead to questions about ‘missing’ levels. Consistency in choice of labels across dimensions (using ‘slight’, ‘moderate’, ‘severe’ wherever possible) should simplify operational aspects of using the questionnaire by facilitating respondent interpretation, aiding the construction of health states and simplifying the translation process. We are aware that terms such as ‘slight’, ‘moderate’, and ‘severe’ can be open to intra- and inter-cultural variability in interpretation and, for that reason, have modified the translation procedure for the 5L version in order to test respondent interpretations of these terms more thoroughly.
Results on the response scaling task were substantially similar in the UK and Spain, and scores assigned to labels generally varied only minimally across dimensions. For example, response scaling scores for ‘moderate’ always fell between 40 and 50 regardless of both country and dimension. These results suggest some robustness in the response scaling scores.
Likewise, although most of the labels used in the EQ-5D-5L can be considered colloquial, comments from some focus group participants indicated that certain terms (particularly ‘moderate’) sounded unusual in this context. On the other hand, the consistency of the results obtained with the response scaling exercise suggests that respondents did not in fact have difficulty in understanding the level of problems referred to. It was also difficult to find any other suitable term approaching the central point on the severity continuum. Quantitative testing of the new EQ-5D-5L will provide additional evidence on the appropriateness of the labels selected.
This study provided an opportunity to test the comprehensibility and face validity of health states derived from the EQ-5D-5L. Again, participants had little difficulty understanding the level of problems that the new labels were intended to describe; the majority of comments instead referred to what participants in both countries considered to be unlikely or self-contradictory health states. For example, some respondents thought that ‘having no problems with washing or dressing’ would sit uncomfortably with ‘being unable to walk’ in the same health state. However, this is more an issue of the relationship among the attributes than the within attribute level descriptions and as such is likely to be pertinent for both the 3 and 5 level versions of the EQ-5D. To take this type of comment into account, the EuroQol Group has been discussing the use of such plausibility testing and cognitive debriefing prior to future valuation studies for the 5-level version.
Spanish and English were chosen as the two languages for the initial development of the EQ-5D-5L because they are two of the most widely spoken languages worldwide and because they can, to a certain extent, act as root languages for translation into a number of other languages. French and Chinese versions of the EQ-5D-5L have also recently been developed using a similar methodology.
One limitation of the current study was that, for practical reasons, the test–retest reliability of the response scaling scores was not assessed. It was also observed that none of the major instruments had undertaken test–retest in this type of scaling exercise [19, 20]. A further limitation of the study may have been the response scaling method used, in which labels were initially rated independently with only the VAS anchors providing context. Although respondents later had the opportunity to redress any ratings they saw as inconsistent when labels were ranked based on the ratings they had supplied, values may have differed if labels had been rated initially in the context of other labels, e.g., using a paired-choice exercise. Nevertheless, the findings were quite consistent across dimensions and countries, with good face validity in the focus groups, suggesting that the final ordering was acceptable to respondents. A further limitation was that we used convenience samples for the response scaling exercise which were not representative of the national populations and the sample sizes used were quite small, though in line with similar studies [20, 23]. These issues may limit the generalizability of the findings. Finally, in both the response scaling exercise and the focus groups, the proportion of participants with tertiary level education was high. This may have led to more consistent results and greater acceptance of wording than would have been the case if the sample had included more respondents at lower educational levels. Future studies of this type should aim to include more balanced, representative samples.
The next step in development will be to field test the EQ-5D-5L and EQ-5D-3L together in general population and clinical samples to evaluate the psychometric properties (sensitivity, validity, and reliability) of the EQ-5D-5L and to compare them with the EQ-5D-3L. Further work is also required to determine the degree of cross-cultural equivalence of the severity labels. For this, properly constructed samples using equal probability of selection methods are required which are sufficiently large to investigate the issues raised in this paper. It will also be necessary to develop value sets for the EQ-5D-5L based on new, large-scale valuation exercises. Preparation for these valuation exercises is on-going.
In conclusion, official versions of the new EQ-5D-5L now exist in UK English and Spanish for Spain, and translations have already been produced for use in a further 25 countries. The UK English and Spanish for Spain versions have shown initial content and face validity, though further psychometric testing is required not only of validity and reliability but also sensitivity to change of the EQ-5D-5L, which is a necessary prerequisite for the development of a valuation set for the EQ-5D-5L. It is expected that the EQ-5D-5L will have better discriminative capacity and sensitivity to change than the EQ-5D-3L as well as smaller ceiling effects.
The study was financed by grants from the EuroQol Foundation.
Conflict of interest
All of the authors are current members of the EuroQol Group and the current study was financially supported by the EuroQol Group.
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.