Validity of the EuroQoL (EQ-5D) Instrument in a Greek General Population

Introduction

There is an increasing interest in instruments used to measure health-related quality of life (HRQOL) in general population surveys, as well as across a variety of diseases and conditions. Although disease-specific instruments have an essential role in measuring disease-related aspects, generic ones are generally more capable of capturing the overall impact of a disease, as well as the benefits of treatment [1]. Generic instruments are classified as either health profiles that measure health across a number of dimensions interpreted separately, or as multiattribute health status systems from which a single index is computed. The latter, having a clear basis on expected utility theory, reflect preferences for health states, weighted either by patients or the general population, and are predominantly exploited in cost-utility analyzes [2,3].

The EuroQol EQ-5D is a two-part, generic, preference-based measure of HRQOL developed by a multidisciplinary consortium of investigators from five European countries [4]. The initial version of the instrument addressed six domains, but was later revised to include five: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression [5,6]. Each domain is divided into three severity levels, corresponding to no problem, some problem, and extreme problem. A dimension for which there are no problems is said to be at level 1, while a dimension for which there are extreme problems is said to be at level 3. Combining one level from each domain results in defining 243 different health states ranging from 11111 (perfect health) to 33333 (worst health), and it is possible to obtain social values for each, thereby generating a tariff of health state valuations [7,8]. The second part of the EQ-5D consists of a vertical 0 to 100 visual analog scale (VAS), with 0 and 100 representing the worst and best imaginable health states, respectively. The subject marks a point on the scale to reflect overall health on the day of the interview, hence providing a single global rating of self-perceived health.

The EuroQol Group has closely monitored the translation of the EQ-5D into most major languages, and the instrument is currently being used in many different countries in a variety of clinical areas and evaluation programs, as well as in health policy formulation [9]. The instrument has been translated into Greek as well, and initial evidence on its applicability and adaptability to the Greek environment has been provided [10]. The aim of this study was to cross-sectionally examine, in a large and representative sample of the Greek general population, the construct validity of the three components of the EQ-5D: the descriptive system, the VAS, and the utility index (derived from time trade-off valuations from a general UK population [7,8]). Construct validity will be addressed via the accumulation of evidence on “known groups” validity, convergent, and divergent validity. Moreover, the issues of concurrent validity, internal consistency reliability, and sensitivity will also be investigated. The results are expected to support equivalence of metrics and meaning [11] with the original English version, and increase confidence in using the instrument in Greek studies, thus contributing to cross-cultural HRQOL comparisons.

Methods

Survey

The instruments used to measure HRQOL in this study were the EQ-5D and the SF-36 (Greek version 1.0, standard 4-week recall). The SF-36 is a generic HRQOL assessment instrument that includes eight dimensions: physical functioning (PF), role physical (RP), bodily pain (BP), general health (GH), vitality (VT), social functioning (SF), role emotional (RE), and mental health (MH) [12], and reliability and validity of the Greek version have been previously demonstrated [13,14]. Each dimension is scored on a 0 to 100 scale, with 0 and 100 corresponding to worst and best HRQOL, respectively [15], and the eight dimensions can be summarized in two summary scores of physical and mental health [16].

Subjects reported information on socio-demographic variables such as sex, age, marital status, education, and employment, with the latter two serving as proxy-estimators of socioeconomic status, because information on income was not available. Self-reported clinical conditions were also recorded, as this technique has been previously reported as reliable [17,18]. This survey focused on diabetes, hypertension, heart problems, asthma, hip/knee problems, depression, sleeping disorders, and obesity (BMI > 30). Furthermore, utilization of health services such as past-month physician consultations and past-year hospital admissions were recorded, as they have also been shown to be associated with HRQOL [19,20].

Psychometric Properties

The sensitivity of the EQ-5D refers to the extent to which the full range of health states is defined, as well as to floor and ceiling effects. Internal consistency reliability was estimated via Cronbach's alpha coefficient. Construct validity was assessed by examining hypothesized relationships between sociodemographic and clinical variables and EQ-5D scores, which have been previously reported in the literature. Specifically, it was expected that females, older subjects, persons widowed or divorced, those with a lower educational status, and the unemployed would report poorer health [21–23]. It was also logical to expect that those reporting greater use of health services and existing clinical conditions would also score lower on all three components of the EQ-5D [24,25].

The ability of the EQ-5D to discriminate between different levels of health was determined by comparing mean SF-36 scale and summary scores for subjects reporting no problem, for a given EQ dimension as opposed to reporting any problem, and it was expected that the mean SF-36 scores would be higher in the first case [26]. Convergent and divergent validity were tested by examining relationships with the Greek SF-36, and it was expected that comparable dimensions, e.g., EQ-5D mobility and SF-36 physical functioning, would correlate better, compared with less comparable dimensions, such as EQ-5D anxiety/depression and SF-36 physical functioning. Contrarily, the EQ VAS and index should correlate reasonably well with all SF-36 dimensions of health [27].

Statistical Analysis

Data were analyzed using SPSS version 13.0 (SPSS Inc., Chicago IL). The χ² (chi-square) and Mann–Whitney tests were used to examine differences in frequencies between categories and in average ranks, respectively, and the Kruskal–Wallis test for differences in average ranks between groups differing in key variables. The Spearman correlation coefficient was used to measure the association between EQ dimensions, EQ index and EQ VAS, and SF-36 subscale and summary scores, and correlations >0.50 were regarded as strong [28]. Odds ratios (95% confidence intervals) were used to estimate the association between reporting a problem on the EQ-5D descriptive system and the presence of a chronic disease or condition. Subgroup mean scores were adjusted for age and sex because these particular variables are potential confounders. For all tests, statistical significance was assumed for P-values < 0.05.

Results

Approximately one fourth of the health states generated by the EQ-5D descriptive system were reported by the respondents (60 out of 243). Forty-three percent of the sample rated their health state as 11111 (ceiling), implying no problems in any dimension, and only one person reported the most dysfunctional state 33333 (floor). Subjects reporting the best state were mostly male, younger, and of higher educational status (results not shown). Table 1, ranked in order of decreasing health according to the EQ index, shows the 20 most frequently reported health states, cumulatively accounting for 93.3% of the sample. It is evident that lower mean EQ index and VAS values correspond to poorer health states reported on the EQ-5D descriptive system. Individuals rating themselves as 11111, assigned a mean score of 86.70 to their health state on the EQ VAS, implying relatively high sensitivity, although it is also possible that minor health problems are not sufficiently recorded. Internal consistency reliability of the EQ-5D was good with Cronbach's alpha (α = 0.743), exceeding the recommended 0.70 standard for group-level comparisons [29].

Table 2 shows the percentage of individuals reporting problems in each of the EQ-5D dimensions and mean EQ index and VAS values according to socio-demographic and clinical variables. The mean EQ-5D index and EQ VAS scores (SD) were 0.80 (0.27) and 75.06 (20.49), respectively, whereas the Spearman correlation coefficient between the two was 0.62 (P < 0.001). Anxiety/depression was the most frequently reported problem (43.5% of the sample), and self-care was the least frequently declared problem (5.0%). Pain/discomfort was reported by 33.4% of the sample. Higher percentages of women reported problems on all the EQ-5D dimensions and rated their health as worse on the VAS (72.81 vs. 77.49 for men, P < 0.001). These differences are reflected in the assigned tariff values (0.76 vs. 0.84, respectively, P < 0.001).

Older respondents were more likely to report problems in all EQ-5D dimensions, and to rate their health lower on the VAS (Kruskal–Wallis, P < 0.001), although in the self-care and anxiety/depression dimensions, higher percentages were observed for the 55 to 64 group (and not ≥65). The proxy indicators of socioeconomic status (i.e., employment and more years of education) were, as expected, associated with the absence of problems in the EQ dimensions, and with higher scores in the EQ-index and EQ-VAS, implying that high socioeconomic status was positively related to HRQOL. The highest percentages of reporting a moderate or severe problem in the EQ-5D dimensions were observed in the group of retired respondents, and this is most likely because age, and perhaps clinical conditions, acting as covariates. Furthermore, being divorced/widowed (Kruskal–Wallis, P < 0.001), suffering from a clinical condition (Mann–Whitney, P < 0.001), last-month physician consultations (Mann–Whitney, P < 0.001), and past-year hospital admissions (Kruskal–Wallis, P < 0.001) all correlated positively with lower VAS and index values and with higher percentages of reported problems.

With respect to the clinical conditions reported by the subjects, all of them (except for heart problem) were significantly associated with increased odds of reporting a problem in at least one EQ-5D dimension, after adjusting for age and sex (Table 3). In particular, sleeping disorders and obesity were the ones with the greatest impact on mobility, usual activities, and pain/discomfort. Subjects with hip/knee problems were 3.5 times more likely to note a problem in the EQ dimension pain/discomfort, and those suffering from depression were 6.5 times more likely to report a problem in the EQ-5D dimension anxiety/depression, clearly marking the expected strong association between the two. Having diabetes resulted in significantly higher odds of reporting problems in the EQ-5D mobility and usual activities dimensions, whereas hypertension meant higher odds of reporting impairments in the two aforementioned dimensions and in pain/discomfort as well. Finally, asthma was linked to a 4.2 times higher probability of reporting problems in everyday activities. The existence of these clinical conditions also resulted in decrements in age and sex-adjusted EQ index and EQ VAS scores, with sleeping disorders generating the most noteworthy differences, namely 0.16 (P < 0.001) and 13.2 (P < 0.01), respectively. By contrary, hip/knee problems produced the smallest EQ index and VAS decrease after adjustment, implying that this condition is age-related.

The discriminatory ability of the EQ-5D was tested by comparing SF-36 scale and summary scores for subjects reporting no problems, for each of the EQ-5D dimensions, with subjects reporting any problems (Table 4). As expected, scores were significantly different (P < 0.001 for all, except EQ self-care and SF-36 MCS, where P < 0.05). The Spearman rank correlation coefficients (Table 5) between SF-36 scale and summary scores and EQ-5D dimensions, index, and VAS were high between dimensions expected to correlate well. For example, ρ = −0.66 between SF-36 RP and EQ-5D usual activities, ρ = −0.65 between SF-36 BP and EQ-5D pain discomfort, and ρ = −0.65 between SF-36 PF and EQ-5D mobility, just to mention a few, implying high convergent validity. On the other hand, dimensions expected to be less comparable demonstrated divergent validity by correlating weakly. In particular, ρ = −0.29 between SF-36 MH and EQ-5D mobility, ρ = −0.28 between SF-36 RP and EQ-5D anxiety/depression, and ρ = −0.23 between SF-36 SF and EQ-5D self-care. Finally, EQ index and VAS scores correlated fairly well, as hypothesized, with all SF-36 scale and summary scores.

Discussion

The adaptability of the EQ-5D to the Greek environment and the basic measurement properties of the instrument have been previously demonstrated [10]. This study reports on the first application in a large and representative sample of the Greek population, and on the first extensive evaluation of its basic psychometric properties. Initial evidence has been provided on construct and concurrent validity, reliability, and sensitivity of the instrument, supported by self-reported data on sociodemographic and clinical characteristics. This implies that the EQ-5D is potentially suitable for inclusion in large-scale health surveys in Greece, for providing HRQOL data to be used for calculating quality adjusted life years (QALYs) in cost-utility studies, and for cross-cultural quality of life comparisons.

The EQ-5D multiattribute system is limited to three response levels in each dimension, and this could impose a significant ceiling effect on health status assessment among relatively healthy people. In this study, 43% of the sample did not report any problems on the EQ-5D (health profile 11111), with a wide range of SF-36 scores observed for this particular group. Although central tendency of health status in a general population may be expected to be in the direction of good health, the discriminative ability of the EQ-5D (compared to the SF-36) is limited, and this has been recognized in several studies [23,24,26,30], and may limit the application of the EQ-5D in general population surveys. Available evidence from comparisons of SF-36 and EQ-5D in disease populations suggests that this is not as much of a limitation [31–33].

The EQ-5D dimensions, index, and VAS were able to distinguish between groups of respondents in the expected manner (known-groups validity) on the basis of sex, age, socioeconomic status, self-reported health problems, and health services utilization (as a proxy of HRQOL), providing evidence of its construct validity. Furthermore, scores on the Eqindex—derived by the time trade-off from a UK population—were higher than scores on the VAS, and this has been observed in many previous studies [23,34,35]. Convergent and divergent validity of the EQ-5D descriptive system was supported by confirmation of expected relationships with the SF-36 scale and summary scores. Additionally, reporting a problem with a given EQ dimension was generally associated with lower SF-36 scores.

Health conditions, which are known to be reliable when self-reported, had an effect on EQ dimensions, index, and VAS. In particular, subjects suffering from diabetes, hypertension, heart problems, asthma, hip/knee problems, depression, sleeping disorders, or obesity were more likely to report a problem in the EQ dimensions and lower VAS values, after adjustment for age and sex. Similar results have been reported in other general population studies and support EQ-5D construct validity [24,25,27]. A proportion of subjects had more than one of the above-mentioned conditions, but most multimorbid groups were small, and it was chosen to study each condition individually to obtain adequate sample sizes. These health conditions generated expected differences in EQ index scores as well. This offers support to previous assertions that valuations for a standard set of EuroQol health states are broadly similar in different countries, suggesting cross-cultural applicability [36,37]. Nevertheless, it should be noted that other EQ-5D valuation studies have reported preferences differing from those of the UK population [38], and hence future efforts could be focused on obtaining a Greek tariff as well.

All SF-36 scale scores differ significantly between subjects grouped by their responses to the EQ-5D. Differences in SF-36 scores, between comparable dimensions of the EQ-5D and the SF-36, were expectedly larger than between noncomparable ones, implying good concurrent validity of the EQ-5D. With respect to correlations between SF-36 scales and EQ-5D dimensions, the only unexpected result was the correlation between SF-36 mental health and EQ-5D anxiety/depression, which, although significant, was weaker than what might have been expected (ρ = −0.43). Furthermore, the fact that EQ-5D response categories are very limited, in comparison with the SF-36, is an acknowledged problem, for which the proposed solution is to use a separate disease-specific instrument, alongside the generic EQ-5D, if more information about health domains is required [39]. For use in economic evaluations, however, the observed concurrent validity of the EQ-5D is satisfactory, especially in view of the highly significant positive correlation between the EQ-5D index and the SF-36 scores.

This study has some limitations that should be taken into account. Although cross-sectional construct validity, as well as internal consistency reliability and sensitivity of the Greek EuroQol EQ-5D have been fairly demonstrated, other issues such as test–retest reliability, longitudinal construct validity, and responsiveness have not been addressed. This is particularly important because health status changes over time, and the instrument should be able to detect these changes, particularly those of clinical importance. Furthermore, as the literature on the use of preference-based instruments in Greece is extremely scarce, future studies are required to compare the EQ-5D with other measures such as the SF-6D, derived from the SF-36 [40]. Furthermore, clinical conditions were self-reported, and it has been shown that the reliance on such data may result in biased estimates of the prevalence of some conditions [41]. Moreover, significant differences have been found between self-reported and measured anthropometric indexes in young Greek study participants [42] that might lead to erroneous estimations of obesity. Regarding hypertension, 39% of Greek hypertensives are unaware of their health problem [43].

In conclusion, the EQ-5D descriptive system, the EQ VAS, and the UK population-based index have demonstrated construct validity in a large and representative sample of the Greek general population. The EQ components have been shown to discriminate well between subgroups differing in sociodemographic and health-related characteristics. Correlations between the EQ-5D and the validated Greek SF-36 provided evidence of convergent and divergent validity. The relatively large ceiling effect, implying perhaps moderate sensitivity, has to be taken into consideration when administering the EQ-5D to large samples from the general population. Nevertheless, this potential disadvantage is balanced by the instrument's brevity, ease of use, and the existence of tariffs for use in cost-utility analyzes.

Source of financial support: None.