Exploring the origin and conceptual framework of the EQ VAS

From the targeted search, a total of 27 papers were selected to inform this report [1, 4, 13‐15, 17‐38]. The search and selection flowchart and the summary table of the data extraction are detailed in Appendix 2 and 3. Much of the information on the origin of the EQ VAS were found from the grey literature search as well as EuroQol published materials (books).

As already introduced, VAS were already popular tools for eliciting values for health states (valuation) from individuals since the 1970’s [6, 7, 39]. From 1987 onwards, a group of researchers, later known as the EuroQol group, held regular meetings to develop a generic instrument to describe and value health, which led to the later establishment of the EQ-5D. During the early phases of EQ-5D development in 1987–90, a descriptive system to classify health was created (the current version includes five dimensions) and VAS was proposed as the preferred valuation method (other valuation methods were still in their infancy, such as time trade-off) [40]. In a book detailing the early meetings of the (EuroQol) research group, Brooks noted that the reason for choosing VAS over other valuation methods “was its relatively simplicity for scaling and quick way of getting valuations” [41]. Devlin and Parkin et al. recounted the reason for adopting VAS as the standard method: “self-completion questionnaires were seen as the only practical means of obtaining large population level valuation data sets and the VAS was the most suited to such a survey instrument” [42].

Grey literature (meeting notes) and publications documented the evaluation of a physical format of a VAS suitable for valuation. Various empirical studies tested VAS presentations. For instance, Sintonen tested vertical vs horizontal VAS, and vertical VAS with numerical labels and hash marks (much like a thermometer) were preferred (as illustrated by Fig. 1a–c) [43]. In parallel, a research team at the University of York tested the length and orientation of VAS and no difference across formats was found (as illustrated by Fig. 2d, e) [44]. Many research groups, including those in Rotterdam, Brunel and London, contributed to testing VAS design to find the most suitable format(s). By 1990 the final format of the VAS was fixed and, until current day, has not substantively changed: a single 20 cm vertical line with numerical indicators at 10-point intervals and hash marks at each millimeter with the top (100) labeled as “best imaginable health” and bottom (0) labeled as “worst imaginable health” [41], commonly known as a (feeling) thermometer or EuroQol VAS. As the valuation task in postal surveys, the EuroQol VAS was arranged in the middle of a letter-sized page with descriptions of hypothetical health states arranged on either side (reproduced in Fig. 3). Respondents were asked to draw a line from each health state to a position on the EuroOol VAS to indicate its numeric score.

Prior to 1994, the EuroQol instrument comprised of the five items of the EQ-5D profile (page 2), own health rating using the EuroQol VAS (page 3), followed by the valuation exercise using EuroQol VAS to elicit values for hypothetical health states. Pages 2 and 3 served as a “warm-up” task to familiarize respondents the EuroQol health states and the EuroQol VAS tool to facilitate the valuation task. The own health item (page 3) asked respondents to draw a line from a response box to the EuroQol VAS line to rate their own general health that day (shown in Fig. 4). The box-enclosed text matched the format used for the valuation task. In early applications of the instrument, pages 2 and 3 were sometimes used “off label” as a stand-alone instrument to collect health status information in clinical studies and trials. In 1994, pages 2 and 3 were endorsed by the EuroQol group as an independent questionnaire (the EQ-5D), and page 3 became known as the “EQ VAS” [41].

After nearly 2 decades of using EQ VAS as a self-reported health measure, a major revision occurred in 2009 due to several reporting issues. The modifications included asking the respondent to write a number into the box directly instead of drawing a line, as well as improving the text instruction. See Fig. 5 for an example of the current version of EQ VAS [45]. For a detailed report on the development of EQ-5D instrument, please refer to Brooks’s 2013 book “The EuroQol Group after 25 years” [41].

The search for conceptual framework was mainly based on the citation search (see Appendix 2 and 3 for an overview of included literature). We found a dearth of literature addressing the conceptual framework of EQ VAS as a measure of self-rated general health. A set of included studies (n = 6) addressed the conceptual framework and measurement theories of general VAS scales not specific to the EQ VAS [1, 26‐30]. One study by Karimi and colleagues established a 4-stage framework of how individuals value health states using EQ-5D, but the paper did not include the EQ VAS [19].

Two qualitative studies offer insight into what the EQ VAS measures [13, 14]. Tan and colleagues examined the cultural appropriateness of EQ VAS in three Asian countries (China, Japan and Singapore) [13], using open-ended questions (e.g. What does “best imaginable health” mean to you?) to understand participants’ interpretation of the EQ VAS. Based on 144 face-to-face interviews, they found that interpretations of “the best imaginable health” varied amongst participants and can be categorized into 5 themes: physical health, mental well-being, social relationships, medical conditions and treatment, and health promotion knowledge and behaviors. Some participants considered the “best imaginable health” to be unachievable. Interpretations of “the worse imaginable health” also varied and can be categorized into three themes: “death,” “disease,” and “disability.” In another qualitative study, Ernstsson and colleagues investigated how patients with type-1 diabetes think and reason when reporting and valuing their own current health, using EQ-5D-5L, EQ VAS, and an open-ended TTO question [14]. The face-to face interviews also found that respondents had difficulties defining, imagining, and/or relating to the “best imaginable health” label of EQ VAS. Most respondents used the best imaginable health as a reference point and related/ compared their health to it.

Amongst the empirical studies focused on EQ VAS, the majority (n = 7) investigated the difference in or associations with social-demographic, economic, behavior, health condition or environmental variables. Age, education, income, employment, drinking, and exercise were common factors which demonstrated their impact on EQ VAS scores.

Amongst studies of general VAS, one publication [30] noted that VAS, as a global scale, cannot be explained by additive measures, which echoes the more recently proposed emergent model of single-item self-rated health measures [46]. Another argue that earlier development and uses of VAS scales were intended to evaluate within-subject change but more current applications assess between-subject differences [28]. One paper [26] expressed concerns using VAS scales for utility estimations.

Many of the papers excluded during screening addressed measurement properties of the EQ VAS including construct and/or known-group validity in various health conditions. Although not directly relevant to the framework of the EQ VAS, it is important to note that these studies demonstrated satisfactory to strong validity. As found by a systematic review of the psychometric properties of the EQ VAS by Cheng et al.—using the COSMIN framework to assess the validity, test–retest and responsiveness—overall construct validity has been demonstrated across disease groups but less consistently in Asian populations [15]. The authors of the review hypothesized that this finding could be due to the greater variability in the interpretation of the EQ VAS amongst Asian respondents as observed by Tan and colleagues [13]. Rest-retest and responsiveness of EQ VAS were two areas with less satisfactory performance [15].

Many general VAS publications cautioned about end-point descriptors, with the earliest published paper we identified advising not to use extreme language [1]. The end-point labels of the EQ VAS with the top defined as “the best imaginable health” and the bottom as “the worst imaginable health” have been used since the EQ VAS was used as a warm-up task for valuation. Qualitative studies found that these labels are difficult to imagine or relate to, or even be considered as unachievable (for the best imaginable health) [13, 14], leading to avoidance of the end-points of the EQ VAS. Likely this issue may explain the phenomena that among respondents reporting no problems on the EQ-5D descriptive system, the average EQ VAS scores are further away from its full score of 100 [17, 47]. This finding has often been interpreted as the EQ VAS covering more aspects of health than the 5-dimensions of the description system [48], but could also be due to end-point aversion bias. Furthermore, “the best imaginable health” and “the worst imaginable health” capture two different concepts and applying both to a single continuum is a bipolar design. Wewers and Lowe [28] cautioned that bipolar designs are less effective than unipolar designs, where only one concept is applied to the continuum (e.g. “no pain” and “extreme pain”). Ernstsson and colleagues found some respondents to be less attentive to the “worse imaginable health” endpoint than the “best imaginable health” endpoint [14]. The use and impact of end-point label of the EQ VAS warrant further empirical research.

EQ VAS includes hash marks (scale marks) to indicate integers from 0 to 100 with larger marks at deciles. One of its earliest “thermometer-like scale” applications for valuations was the study by Torrance (1976) where a category scaling method—a 100 mm “desirability line”, assuming 101 equal-interval categories (also known as equal-interval scaling)—was used [6]. A similar form of the design was borrowed by EuroQol group. In comparison to a traditional or general VAS (a straight line without hash marks or numbers, Fig. 1a and b), researchers can easily ascertain the numerical value indicated by the respondent without the need of a ruler to measure the location of the mark on the scale line. However, Freyd in a 1923 publication advised against using breaks and division lines in rating scales [1]. It is interesting to note that Tan et al.’s qualitative study found that some respondents reported the hash marks of the EQ VAS to be “too granular”: they recommended that removing hash marks may make the scale easier to understand [13]. However, it reminds unexplored whether this finding can be generalized. The impact of EQ VAS scale marks and whether they lead to an optimal response scale design is in need of further research.