Health inequalities among population groups of various socioeconomic statuses (SES) are an important challenge for public health and social policy, both at the national and international levels (1). Existing health inequalities among countries largely have a socio-economic rather than a medical background (2, 3). Health inequalities are a societal concern; apart from the unfairness of the unequal share in good health per se, it is known that such inequalities impact economic and societal development (4).

Health inequality can be defined as the difference in health or health status between defined population groups (5). The uneven distribution of health determinants may be unjust or unfair and avoidable (5, 6). Regardless of the general level of development or prosperity, groups with a lower SES (measured by income, education, and/or occupation) have worse health outcomes, i.e., higher mortality rates and higher disease prevalence rates (7).

The difficulties of touching upon the causes and the resilience for change have resulted in a growing interest in the measurement and analysis of inequalities in health (8–10). So far, the focus has been mainly on the definition of pathways, the techniques to quantify inequalities, and the appropriate manner in which this topic can be communicated (1, 11, 12). Surprisingly, little attention has been paid to the health measure used as an indicator. In the analysis of clinical trials, health care performance measurement, and burden of disease research, the role of self-reported outcomes of patient/person on the health status has grown to the extent that this type of health information is regarded as critical (13–15). In health inequality analysis, however, simple dichotomous measures like mortality or disease indicators, and a single item 5- or 7-point subjective health scale are generally the norm (2, 7, 10, 16). Such simple measures facilitate the computation of inequalities, but also have some disadvantages. First, they give little information on the impact at the person level (general information only) and the contributing role of factors that require some specification on how they work. Second, the use of single-item self-report measures implies a smaller signal to noise ratio and a larger dependency on the homogeneous use of the response scale; consequently, the analytical power is then theoretically decreased (17–20).

Multi-item measures with a summary or utility outcome often outperform single-item measures (21). Therefore, these measures might be valuable in health inequality analyses as well. An example is the widely available and commonly used multidimensional health questionnaire, the EuroQol 5-Dimension (EQ-5D) (22). This self-report instrument is available in many languages, has been validated for many diseases and conditions, and has been used in health inequality assessment (23–26). The EQ-5D includes a descriptive system that consists of five dimensions, which include mobility, self-care, usual activities, pain/discomfort, and anxiety/depression, and the responses can be converted into a utility index or level sum score (LSS) (22, 27), thereby enabling burden of disease and cost-effectiveness estimations which are increasingly used for resource allocation and evaluation of care (28, 29). The EQ-5D also includes a visual analog scale (EQ VAS), which assesses a person's current (subjective) health status on a 0–100 scale. In the descriptive system of the EQ-5D, the extent of problems on each dimension is assessed, whereas in the EQ VAS, the extent of problems is translated into a score or rating of health status.

Theoretically, the multi-item EuroQol 5-Dimension 5-Level (EQ-5D-5L) might outperform the EQ VAS (or other single-item measures) in health inequality analyses, but this may be counterbalanced by the higher refinement of the EQ VAS scale (101 units). Therefore, the aim of this study was to explore the potential additive value of the EQ-5D-5L as an outcome measure in education-related health inequality analyses, relative to the EQ VAS, in a large three-country dataset (Italy, the Netherlands, and the UK). The secondary aim was to study the extent to which the EQ-5D-5L and EQ VAS scores are related to SES (in particular education) and to explore the role of chronic morbidity and other factors, which could play a mediating role.

This study explored the potential of the EQ-5D-5L LSS as an outcome measure in the health inequality analyses relative to the EQ VAS. As hypothesized, our study showed that the EQ-5D-5L LSS tended to better discriminate between participants with a different level of education compared to the EQ VAS, both in the general sample as well as in subgroups of participants with a specific chronic health condition. Furthermore, the present study showed that the health outcomes differed among the three countries studied, with the worst outcomes for participants from the UK and best outcomes for participants from Italy. Participants with a low level of education had worst outcomes and those with a high level of education had the best outcomes in both the UK and the Netherlands, with larger differences among subgroups in the UK. Health inequalities were thus found to be highest in the UK and smallest in Italy, confirming our hypothesis. Multivariable analyses showed generally stronger predictive relations in the UK and with the EQ-5D-5L LSS. The presence of chronic health conditions and being unable to work were strong independent predictors, canceling out the educational effects (apart from a small effect in the UK).

Results from our study are in line with earlier studies that showed that the EQ-5D reflects health inequalities: low SES (defined as low educational level/low status/unskilled manual group/low income/low social class) is associated with a lower EQ outcome (25, 34–40). Some studies used the EQ VAS as an outcome (25, 37, 38), while some used the separate EQ-5D domains as an outcome (25, 34, 37–39), and some others used the EQ utility score as an outcome (35, 36, 39, 40). Use of the separate domains might not be very useful as to disperse populations by the level of inequality, health differences are often summarized into one single value (2, 7, 10, 16). An earlier study used both the EQ-5D utility score and EQ VAS as outcomes and showed that the EQ-5D utility score better discriminated between ethnicity groups than the EQ VAS (41), which is in line with our results. However, in our study, in the UK participants with memory problems and in the subgroups of participants without a chronic health condition, the EQ VAS was discriminative between participants with a different level of education whereas the EQ-5D was not. Potential causes could be the mediation effect of the inability to work and the presence of chronic health conditions, and/or the fact that EQ VAS and EQ-5D scores could assess slightly different outcomes, which was also shown in previous work (42). It was reported that EQ VAS scores could be predicted from EQ-5D scores, but there was the contribution from other factors also, including psychological status, age, education, and distress (42). The EQ-5D-5L LSS thus seems to better discriminate between subgroups, which is valuable in health inequality analysis. However, the use of an instrument that includes more than one question also includes some disadvantages, namely the extra time needed to complete the instrument and the corresponding higher burden for those completing it, which might result in a potentially lower response rate. It is important to consider whether the improved discriminative power outweighs these disadvantages.

Comparison of our findings on differences among countries are in line with the existing literature and confirm our hypothesis (37). A large European comparison study found that the relative index of inequality was the highest in the UK, lower in the Netherlands, and the lowest in Italy, when these three countries were compared (7). The same pattern was seen in a study on socioeconomic inequalities in self-assessed health in 17 European countries (43). Suggested potential factors that might contribute to these smaller health inequalities in Italy include the Mediterranean diet and the relatively low number of female smokers (44, 45). Within our Italian sample, it was surprising that differences in SES did not seem to exist. Previous studies reported the same patterns of worse health outcomes being associated with a low level of education and better health outcomes with a high level of education (7, 43).

Our results indicate that having a chronic health condition does not explain all the differences that arise from the level of education. This confirms earlier studies that showed that a low SES is associated with worse health outcomes, even after controlling for having chronic diseases (25, 41, 46, 47), although the inability to work (the companion variable explaining the education effect) was not always identically covered. Furthermore, within samples with a specific disease, studies also reported that a low level of education is associated with worse outcomes (40, 48–51). However, our results were inconsistent after controlling for relevant demographic factors. Apart from an association between low level of education and worse EQ-5D-5L LSS in the Dutch sample and low level of education and worse EQ VAS in the UK sample, no other association between educational level and outcomes was observed. Thus, other factors seemed to be stronger associated with a worse health outcome, with a consistently strong negative impact of having a chronic health condition. This association was shown to increase with an increase in the number of chronic health conditions. In the subgroup analysis, it is noteworthy that for the UK, low and middle income had a borderline significant negative impact on tLSS for the diseased subgroup, which may be indicative of more pronounced income inequality effects on health in the UK. The subgroup analysis also showed how unable to work had a much larger impact on tLSS in the diseased group when compared to the healthy subgroup for all countries, as expected.

The present study included some strengths and limitations. Strengths included the large sample size and the invitation of persons from three countries that were representative of their population with respect to age, sex, and educational level. Also, some limitations should be considered. By the use of a web-based survey, we only included participants who had access to a computer and internet and who were able to read and understand the survey, which may have led to participation bias as persons with low health literacy (often those with a low SES) may not have been included. We tried to mitigate this by selecting participants with certain characteristics (e.g., age, sex, and educational level) from the existing large panel in order to increase the representativeness of the study sample for the adult population in the selected countries. Also, a detailed analysis of non-responders was not possible due to the system of recruitment used.

Second, we used the educational level as a proxy for SES rather than an indicator that also incorporates income level or financial security. However, educational level is a common and widely used proxy for SES in the field of social epidemiology. It is a characteristic that has strong associations with a multitude of factors linked to health such as risky health behavior, limited access to financial and social resources, and type of work and residency (52–54).

Third, the outcome measure in our study was the LSS rather than the utility scores calculated with value sets of the three countries. Value sets for the EQ-5D reflect the preferences of the population of a country for the EQ-5D-5L health states, and the utility scores for the same health state calculated with different value sets vary substantially (55). Currently, there is no EQ-5D-5L value set available for Italy. Therefore, in order to use the EQ-5D-5L utility scores as an outcome measure, a proxy value set should be applied to assess the EQ-5D-5L utilities for the Italian study sample. In our study, the choice of the alternative value to assess utility scores for the Italian respondents, the Dutch or the UK value set, is arbitrary. The use of one of these value sets would therefore mean that differences in the utility scores between the Italian and UK or Dutch respondents reflect the differences in the EQ-5D-5L profile only and not the differences in the preferences for health states between these countries.

The present study showed that in three different European countries, EQ-5D measures show the presence of education-dependent health inequalities, which are universally explained in regression analysis by, independently, the presence of chronic health conditions and the inability to work. In stratified analysis, the EQ-5D-5L LSS shows slightly better discrimination between participants with different levels of SES compared to the EQ VAS.

The data analyzed in this study is subject to the following licenses/restrictions: the datasets generated and/or analyzed during the current study are not publicly available due to privacy/ethical restrictions but are available from the corresponding author on reasonable request. Requests to access these datasets should be directed to Inge Spronk, i.spronk@erasmusmc.nl.

The studies involving human participants were reviewed and approved by Leids Universitair Centrum—Commissie Medische Ethiek. The participants provided their written informed consent to participate in this study.

IS conceptualized and designed the study, analyzed and interpreted data, drafted the initial manuscript, and reviewed and revised the manuscript. SP, EL, and MJ conceptualized and designed the study, interpreted data, and reviewed and critically revised the manuscript. JH and GB conceptualized and designed the study, analyzed and interpreted data, and reviewed and critically revised the manuscript. All authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.

This work was supported by the EuroQol Research Foundation (Grant No.: 20180630).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.