Background
The increasing number of older adults who are diagnosed with dementia has far-reaching implications for health service delivery and expenditures [
1]. Economic evaluations are performed more often to assist decision-makers in setting priorities, especially with regard to resource allocation [
2]. A central component of economic evaluations in health care is the use of preference-based instruments (also called value-based instruments) to measure changes in Health-Related Quality of Life (HRQoL). Preference-based measures, such as the EQ-5D [
3,
4], the SF-6D [
5] and the HUI [
6], are standardized multi-dimensional health state classifications [
7]. For each of these instruments, health states have been valued using techniques such as standard gamble (SG) or time trade-off (TTO) [
8]. These valuations were used for each instrument to generate a scoring algorithm of which a single utility score for each health state can be deduced.
The EQ-5D is commonly used to measure HRQoL and has been shown to be responsive, internally consistent and reliable in the normal population and other patient groups [
9,
10] as well as in patients with dementia [
11,
12]. However, concern has been raised that it may ignore elements of HRQoL of specific relevance to the elderly such as vision and hearing [
1] and in particular cognition [
1,
13‐
17]. It is known that cognitive problems have an impact on personality, mood, behavior and global functioning [
18], which are domains covered by the EQ-5D, but cognition might also be regarded as a separate dimension.
In response to the concern that the EQ-5D ignores cognition, the EQ-5D has been extended with a cognitive dimension (EQ-5D+C)[
15]. In this study of Krabbe et al. (which was an adapted Dutch replication of the Global Burden of Disease study commissioned by the World Bank) [
19], valuations (by means of a rating scale) elicited from EQ-5D+C descriptions were compared empirically with parallel EQ-5D descriptions in Dutch faculty members (i.e. scientific staff members and management members of the Department of Public Health, the Department of Clinical Epidemiology and Biostatics and the Institute of Social Medicine).
The EQ-5D+C generated different values compared with the EQ-5D. Whereas the content validity of the EQ-5D improved by adding cognition, both versions evoked equally reliable values. Based on these results, the authors emphasized the importance of considering the inclusion of a cognitive dimension. Furthermore, the EQ-5D+C was used to describe the health status of the Dutch population and to investigate sociodemographic differences [
14].
In this study, the content validity also improved through the addition of the cognitive dimension, while the reliability remained unaltered. It was concluded by the authors that the EQ-5D+C is an efficient tool for establishing the health status in the community. Another way to examine if the EQ-5D should contain a cognitive dimension is to investigate the performance of the EQ-5D and the EQ-5D+C in a population with cognitive impairments. The aim of this explorative study was to compare the performance of the EQ-5D and the EQ-5D+C by assessing their construct validity and responsiveness in patients aged 55 and older with cognitive impairments.
Discussion
The aim of this explorative study was to compare the performance of the EQ-5D and the EQ-5D+C by assessing their construct validity and responsiveness in patients aged 55 and older with cognitive impairments.
Based on our results it can be concluded that the construct validity of the EQ-5D and the EQ-5D+C is comparable in our study population, except for the VAS
5D. Results regarding construct validity of the EQ-5D are in line with the recent findings of Jönssen et al. [
17]. Contrary to our expectations, correlations between the cognitive dimension and the MMSE were almost similar to the correlations between the self-care and the usual activities dimensions and the MMSE. The presence of more and stronger correlations of both the EQ-5D and EQ-5D+C with the MMSE at the 12 month follow up measurement was possibly due to the fact that the dispersion of the scores using these instruments increased with time. Three studies also showed that cognitive function was positively related to HRQoL in cardiac rehabilitation patients [
34], in patients with progressive supranuclear palsy [
35] and in patients with hypertension [
36]. Another study [
37] failed to find a relationship between HRQoL and cognition in patients with dementia.
With regard to responsiveness, the EQ-5D performed slightly better than the EQ-5D+C, which is also in line with the findings of Jönssen [
17]. An important finding, again contrary to our expectations, is that changes in the MMSE corresponded better with changes in the self-care dimension and the usual activities dimension than with changes in the cognitive dimension.
However, no judgments were made about the strength of the correlations, which would provide us with a stricter criterion regarding the performance of the EQ-5D and EQ-5D+C. In the literature, different classifications were found [
38‐
41] a clear gold standard being absent. We therefore ignored the classifications and merely described our results. However, it is possible to compare our results with other studies. Our results were in line with correlations between the EQ-5D and clinical measures found in other studies involving diseases such as progressive supranuclear palsy (PSP) [
35], rheumatoid arthritis (RA) [
39] and stroke [
42].
The majority of authors [
39,
41,
42] and others) considered a Spearman's correlation of > 0.50 to be strong, a correlation of 0.30/0.35 – 0.50 to be moderate and a correlation < 0.30/0.35 to be weak. Using these classifications in our study, it can be concluded that both versions performed well with respect to construct validity, as indicated by strong correlations with the MMSE. Regarding responsiveness, it can be concluded that the EQ-5D performed moderately, whereas the EQ-5D+C did less well as indicated by weak correlations with the MMSE. When the more stringent classification of Landis and Koch [
38] is used (i.e. < 0.00 poor; 0.00–0.20 slight; 0.21–0.40 fair; 0.41–0.60 moderate; 0.61–0.80 substantial and 0.81–1.00 almost perfect), it can be concluded that the EQ-5D and the EQ-5D+C performed moderately with regard to the construct validity.
Regarding responsiveness, fair correlations were found between changes in the EQ-5D and EQ-5D+C and changes in the MMSE. The relatively low responsiveness of the EQ-5D in this study could be due to the, on average, small changes in cognition in a year, or to a ceiling effect because there are only three levels for each dimension of the EQ-5D. Patients' health may improve or decline but not enough to go up or down one level. Instruments that have a greater number of possible responses may be more responsive. Furthermore, it is possible that adaptation to illness on the part of the proxy leads to a lack of responsiveness, especially with a chronic condition such as dementia [
39]. It should also be noted that a lack of clarity exists with regard to the definition and adequate approach for evaluating responsiveness. Some authors argued that there is no need for an additional concept like responsiveness, since it can be viewed as either longitudinal validity or magnitude of the treatment effect [
32,
43,
44]. The definition and approach used in this study has also been referred to as longitudinal validity [
32].
There are several limitations to this study that need to be recognized. An important limitation of this study concerns our study design. The origin of this study, the MEDICIE trial, was designed to compare the effects of a multidisciplinary diagnostic observation centre for psychogeriatric patients (DOC-PG) with care as usual on HRQoL, mental and physical health, and the costs and use of health care facilities by patients with psychogeriatric problems. Therefore, studying the usefulness of the EQ-5D+C in this patient population was framed in this RCT. The EQ-5D was administered first, that is the five dimensions followed by the VAS5D. Subsequently, the proxies were asked to answer the sixth dimension concerning cognitive functioning, whereupon the VAS5D+C was valued. It would have been better to administer the EQ-5D+C completely as well in order to make valid comparisons between the 2 versions. However, considering the explorative nature of this study, we did not want to burden the participants of the MEDICIE trial by administering a similar questionnaire twice.
Second, regarding the assessment of the EQ-5D+C, the proxies may have focused their attention on the cognitive dimension when scoring the VAS
5D+C, even though they had been instructed to rate the VAS
5D+C again based on the overall health. This effect is called a framing effect, which suggests that how something is presented (the 'frame') influences the choices people make [
45]. Hence, it is possible that the higher correlations of the VAS
5D+C with the MMSE are due to a framing effect. However, according to Parkin et al. [
46], the framing bias also exists when assessing the EQ-5D, meaning that values of the VAS
5D are affected by end-state descriptors (last named dimensions).
Another possible limitation is the use of proxies to complete the questionnaires. Previous research indicated that there is generally fairly good proxy-patient agreement for observable items such as mobility, self care and usual activities, but poor agreement for non-observable items such as pain and affect [
16]. Others have found agreement to be poor for the domains most affected by dementia (self-care and usual activities) [
17]. In the light of the longitudinal nature of our study, the complex health problems of our study population and their progressive global deterioration of intellect and personality, the method of proxy rating had been chosen. It is generally acknowledged that in the later stages of dementia proxy measures are required since patients are no longer capable of making an adequate evaluation of their HRQoL [
12,
17]. Furthermore, the use of proxy reports throughout the course of a longitudinal study, rather than substituting them only when the person with dementia becomes unable to report his or her HRQoL, reduces bias over time [
47]. The overall picture of previous research is that rating by proxy is a valid alternative for assessing HRQoL in the presence of dementia [
17,
47‐
49], although it is possible that the scores in the EQ-5D and the EQ-5D+C were biased because of perceived caregiver burden [
50].
A final limitation also concerns the design of our study. Comparisons between the EQ-5D and the EQ-5D+C were merely based on the dimensions and VAS-scores of both versions and not on the utility scores since these are not available for the EQ-5D+C. It should be noted that an algorithm has been developed for EQ-5D+C health states, based on Dutch disability weights [
51,
52]. In the Dutch disability weights study, a comprehensive set of disease-specific disability weights for 175 disease stages associated with 52 disease categories was obtained [
53,
54]. Based on these disability weights, an EQ-5D+C regression model was fitted. However, the origins of the EQ-5D+C disability weights and the EQ-5D utility scores differ significantly. First, the algorithm is based on valuations of health experts instead of valuations of the general public. Second, EQ-5D+C health states were valued by means of the person trade-off (PTO) method, whereas EQ-5D health states were valued by means of the time trade-off (TTO) method [
27]. PTO differs from TTO in that subjects are required to trade-off person years lived healthy against person years lived with some defined disability, thus making choices in the context of a decision involving other people rather than themselves. Whether the PTO technique is able to reflect actual preferences is still under debate [
55,
56]. Finally, besides the EQ-5D+C health state description, subjects were given specific information with respect to the disease, which differs from the EQ-5D valuation procedure [
27]. Therefore, in our opinion, no valid comparison of EQ-5D+C disability weights with EQ-5D utility scores can be made. In order to develop a new scoring algorithm of which utility scores for the EQ-5D+C can be deduced, a valuation procedure similar to the one used for the EQ-5D should be applied. Presenting EQ-5D+C health states to members of the general population should reduce the framing effect described earlier, as the cognitive dimension will then be 'just' one of the six dimensions in the health states. Furthermore, although in the descriptive part of the EQ-5D a proxy effect may still be present, by using a utility score based on valuations of the general population, possible proxy effects are expected to decrease.
Competing interests
The author(s) declare that they have no competing interests.
Authors' contributions
FV was the principal investigator of the larger trial and is guarantor. FV, AK, CD, and JS designed this trial. CW and DW carried out the outcome measures. AK and CW were responsible for statistical analysis and interpretation of the study, and all authors contributed to interpretation. CW drafted the manuscript, and all authors critically revised it for scientific content and approved the final version.