Feasibility of the EQ-5D in the elderly population: a systematic review of the literature

In the past decades, the demographic trend of an ageing population has become one of society’s central challenges. Especially the proportion of the elderly population, i.e. people being 65 years and above [1], is growing faster than any other age group [2]. At the same time, the increase in life expectancy induces a higher individual risk of contracting one or multiple diseases over the course of life, eventually resulting in a growing number of multi-morbid patients with chronic diseases [3]. Chronic illness and multi-morbidity are known to be associated with disability, declined functional status and diminished quality of life (QoL) [4], which is further linked to higher health care utilisation and increased costs of health care [5]. Given the natural limit of health care resources, it is necessary to evaluate the cost-effectiveness of health care interventions from a societal perspective to sustain the health service provision [6].

Health-related quality of life (HRQoL) is a central outcome for the benefit assessment of health and social care interventions often measured as patient reports to gather the patients’ subjective assessment of their health condition. The EQ-5D is a generic preference-based measure of HRQoL to operationalise quality-adjusted life years (QALYs) in economic evaluation [7]. The EQ-5D consists of two sections. The first is a descriptive system covering five dimensions: mobility (MO), self-care (SC), usual activities (UA), pain or discomfort (PD) and anxiety or depression (AD). In the EQ-5D-3L each dimension can be described by three severity levels (1—no problems; 2—moderate problems; 3—unable to), thus allowing to distinguish 243 unique health states (3^5). The second part is a visual analogue scale (EQ VAS)—a vertical thermometer—ranging from 0 (‘worst imaginable health’) to 100 (‘best imaginable health’) gauging the respondent’s subjectively rated health which might cover aspects different from those in the descriptive system [8, 9]. A later variant, the EQ-5D-5L, covers the same five dimensions, but allowing the respondent to choose from five response levels, thus describing 3125 unique health states [10]. For both versions of the EQ-5D the descriptive system can be scored using a tariff, which provides the preference weights for each health state allowing to calculate an index value on the 0–1 QALY scale, whereas the EQ VAS rating is commonly analysed independently from the responses to the descriptive system [11].

The EQ-5D (hereafter, used to refer to both the EQ-5D-3L and EQ-5D-5L) has been shown to be applicable in various health conditions and populations and is the most widely used instrument for use in economic evaluation [12‐14]. Literature suggest that the EQ-5D is also frequently used in the economic evaluation of interventions for the elderly population [15‐19]. However, as of yet there is no established gold standard as to how HRQoL in the elderly should ultimately be measured. Especially with regard to the elderly population further requirements to the appropriate assessment of HRQoL are made, since people of higher age may be different to their younger counterparts in the general population in terms of their physical or mental abilities, education or understanding of health [17, 20]. Common criteria to evaluate an instrument’s performance are its reliability, validity as well as its feasibility [21]. Generally, the EQ-5D’s measurement properties are well examined [22, 23] with several studies specifically confirming the instrument’s reliability and validity in the older population, however, leaving the feasibility property widely untouched and warranting further research [18, 24, 25]. In this sense, feasibility is concerned with the difficulty or ease of applying the measure in a population, which translates into how well the measure is regularly completed [21, 26, 27]. Beyond these descriptions there is no gold standard to the definition or operationalisation of feasibility with regard to HRQoL measures. A prior search across all age groups indicates that the feasibility of the EQ-5D is typically associated with the proportion of missing values [13, 28‐31], time required for completion and the appropriateness of the administration mode [26, 32].

Age-related decline may be an obstacle for elderly respondents when self-reporting their HRQoL in a survey [33, 34]. Therefore, offering interviewer support or administration can help to reduce the burden to respondents and thereby have a positive impact on item response [35, 36]. There is recent qualitative evidence from older (60+ years) hip fracture patients suggesting that interviewer support was needed and had a positive effect on the instruments’ completion [37]. Even though the EQ-5D is a short measure there appears to be some demand for interviewer support, however, it is unclear how commonly this is applied when collecting EQ-5D data in the elderly.

Despite the steadily growing population of those aged 65 years and above, to date little is known about the feasibility of the EQ-5D in the elderly. Therefore, the aims of this literature review were as follows: (i) to assess the feasibility properties of the EQ-5D in the elderly and (ii) to examine the role of interviewer support in collecting EQ-5D data in samples of the elderly population.

Over the past three decades the EQ-5D has been applied in an extensive list of populations and settings and amongst those elderly populations and patients were frequently examined, too. Even though the EQ-5D’s feasibility properties were confirmed and found to be unproblematic for the overall general population by two recent major reviews [22, 23], feasibility of the EQ-5D is not systematically explored for the elderly population and warrants further examination [18, 49]. Therefore, this review summarised the available information on the feasibility properties of both the EQ-5D-3L and EQ-5D-5L in the elderly population. In the light of the reviewed studies, the terminology around this measurement property and which aspects of feasibility are commonly reported are inconsistently used. Overall, missing values of approximately up to 10% on the descriptive system and completion rates of around 90% seem to be ballpark figures in elderly populations. On the other hand, completion of the EQ VAS seems to pose a higher burden to older respondents, since missing rates are generally higher. Also of interest is the high prevalence of interviewer-assisted or even interviewer-based administration of the EQ-5D in the older population.

This review identified several different synonyms for feasibility such as ‘completion’, ‘practicality’ or ‘acceptability’. However, all labels aim to describe the applicability of the measure to the target population in a similar manner, viz. how conveniently and successfully the measure can be completed. Yet, to the best of our knowledge, none of the leading QoL organisations such as the EuroQol group, the international society for quality of life research (ISOQOL), World Health Organization (WHO) or the European Organisation for Research and Treatment of Cancer (EORTC) Quality of Life Group provide a definition of feasibility. To aid the standardisation of reporting this measurement property and the usability of available evidence for researchers, we suggest summarising these labels under the term ‘feasibility’.

Regardless of the label used in individual studies, there was little agreement on how feasibility was operationalised and therefore included studies examined feasibility in terms of missing values, completion rate, time required to complete the EQ-5D or—more broadly—in qualitative statements referring to the completion of the instrument. However, the amount of available information on each of these parameters varied greatly. One of the more frequently reported outcomes were missing values. Overall, the proportion of missing values was mostly below 7% on the five dimensions with few exceptions, e.g. in a study by Michalowsky et al. [50] including cognitively impaired respondents; but generally missing values did not exceed 11%. Where detailed information on the dimension-level was available, those suggested that dimensions were equally affected with no particular one sticking out. Two recent systematic reviews provide a reference value of 5% missing values for both the EQ-5D-3L and EQ-5D-5L [22, 23]. This upper value largely seems to hold for the included studies with only a marginal excess effect in missing values in the elderly population, which was already described elsewhere [30, 51‐53]. Even though the proportion of missing values seems to increase slightly with higher age, the EQ-5D compared more favourably in terms of missing values when compared to other instruments that are commonly applied with the elderly such as the SF-36 [20, 40], the AQoL [43] or ICECAP-O [54].

A related concept is the completion rate of the descriptive system. Findings from this review suggest that the proportion of incomplete responses is commonly less than 10%, i.e. more than 90% of index values are computable. Considerably lower completion rates were reported by Hickson and Frost [47], arguing that patients were too ill to complete the questionnaire. Also, van Laar et al. [55] report an equally low completion rate of 83.3%, however, it remains unclear to what extent the study design or the disease negatively interfered with completion rates. Again, comparing our findings with the benchmark completion rates of more than 93.4% (3L) and 96% (5L) provided by Buchholz et al. [22], we find that the share of computable index values seems to slightly decrease in the elderly population. This is congruent with findings from Luthy et al. [41] who found that completion rates for the EQ-5D-3L were negatively associated with increasing age. However, this finding is not surprising given that missing values were also slightly more prevalent in the elderly, which in turn lead to incomplete health state information. Based on the reviewed evidence on missing values and completion rates of the descriptive system, we would like to argue that a magnitude of approximately less than 10% missing values and about 90% of computable EQ-5D index values demonstrate reasonable feasibility of the EQ-5D in the elderly population.

A further component that was found to be more frequently missing was the EQ VAS. The range of missing EQ VAS ratings was 2.3–25.3% and differs considerably, but generally exceeds the proportion of missing values in the descriptive system. There is evidence that the concept of the EQ VAS is more difficult to comprehend than the descriptive system and causes the most problems [43, 46‐48], which was found in older populations before [26]. Then again, higher proportions of missing EQ VAS ratings [32] and execution problems [56] were also reported for the younger populations, yet to a lesser extent.

Generally, missing data on both components the EQ VAS and the descriptive system diminish the available sample size for analyses and, if item nonresponse occurs systematically, this may result in biased results [57]. The decision on how to handle missing data in the analysis should be guided and justified based on the mechanism of missing data rather than the proportion of missing values per se [21, 39, 58‐60]. Generally, preventing missing data before they occur is more efficient than an analytical remedy [61]. In this sense, a re-occurring topic in included studies is the high proportion of older respondents requiring assistance to complete the EQ-5D or even an interviewer-based approach [40, 46, 48]. Several issues, such as asking for explanations and clarifications due to limited item interpretation [40, 46, 48], were identified that may warrant interviewer support when collecting EQ-5D data in the elderly [44]. Similar findings were presented from a qualitative study with younger hip fracture patients (60+ years) by Rohr et al. [37] arguing that without interviewer support missing values would be significantly higher. At least three of the included studies acknowledged that interviewer support was needed, but did not further assess the amount of support that was required by the participants [47, 62, 63]. With regard to this, Coast et al. [48] report age-dependent probabilities of requiring interviewer assistance, further suggesting an increased need of assistance with higher age. On the other hand, several studies report good feasibility properties in the older population relying on self-completed measures without assistance [20, 42, 64], whereby the EQ-5D appears to be well applicable as a self-report measure in the elderly. Even though interviewer support was frequently provided in included studies, the effect of interviewer assistance on preventing missing values on the descriptive system and EQ VAS or increased completion rates cannot be quantified based on the available evidence for two reasons. First, included studies did not compare feasibility aspects between assisted and non-assisted respondent sub-groups and, secondly, studies were too heterogeneous in their outcomes and study characteristics to identify factors that facilitate adequate feasibility. Furthermore, adopting an interview approach is resource intensive and may introduce additional bias or measurement error, if implemented improperly and, thus, may offset potential benefits [65]. The controversial evidence and the lack of studies exploring the effect of an interviewer approach express the need for further research on this topic. Future qualitative research as done by Rohr et al. [37] and van Leeuwen et al. [44] may help to better understand the role of interviewers in the data collection process. Alternatively, a cognitive de-briefing of interviewers in quantitative studies may shed a light on the heterogeneity of the elderly and help exploring sub-groups that might benefit from interviewer assistance.

From our review, we found only two of the included studies applied the EQ-5D-5L in the elderly population and analysed aspects of the instrument’s feasibility, which limits the generalisability of our findings beyond these two studies for the EQ-5D-5L. However, we would expect the EQ-5D-5L’s feasibility to be comparable or potentially even better than those of the EQ-5D-3L, since the EQ-5D-5L kept the brevity. Further, enabling respondents to better map their health onto the descriptive system in more detailed distinctions might potentially increase engagement to the task [66]. But then again, this may have also increased the cognitive burden hampering instrument completion and data quality [67‐69], however, not to an extent which is believed to overburden respondents [66]. Since none of the included studies comparatively examined both the EQ-5D-3L and EQ-5D-5L in the elderly population, this gap warrants further research.

Some limitations of this study should be considered. In accordance with the underlying age definition for the elderly population, this study only included publications, which focussed on study populations exclusively being 65 years and above. Due to this strict age-related eligibility criterion, some studies were excluded where the majority of respondents may have met the inclusion criteria, whilst only a minor share of respondents was not eligible. Also, included studies exclusively sampled respondents from western and developed countries, which does not allow any generalisation on the feasibility in elderly populations in Asia, Africa or South America, where the older population may be assumed to be different due to their cultural believes or lower literacy in rural regions. Furthermore, the heterogeneity of the target population, which may include healthy and independently living respondents, but also frail or cognitively impaired participants, also limits the generalisability of our findings, especially with regard to the EQ-5D-5L, which was only included in two studies. Moreover, despite searching several databases, we might have missed relevant publications due to an inconsistent use of terminology in addressing feasibility and more importantly due to different search engines for the databases, which may have partially prevented the identification of studies addressing feasibility as a secondary or even only as a descriptive analysis. As a potential remedy to this limitation, we searched all reference lists of included articles to mitigate this risk.

This study aimed to assess the EQ-5D’s feasibility properties when used in the elderly population. Our findings suggest that missing values or comprehension problems—especially with the EQ VAS—are slightly more prevalent in older age groups compared to the younger general population. However, these aspects are well within an acceptable range and still considerably lower than in other measures such as SF-36 or ICECAP-O, which are frequently used in the elderly. Furthermore, older respondents seem to have a higher propensity of requiring some degree of assistance or even an interviewer-based approach. Overall, evidence from the reviewed literature indicates that the EQ-5D-3L has good feasibility properties and, hence, is highly applicable in older respondents. Moreover, further research is needed to explore feasibility properties of the EQ-5D-5L in older respondents too, whilst examining the proportion of missing values, completion rate and completion time considering the role of any interviewer support in the data collection process.