General population norms for the EQ-5D-3 L in Norway: comparison of postal and web surveys

Health-related quality of life (HRQOL) has become an increasingly important outcome of health care and public health interventions. HRQOL can be assessed using different instruments. The EQ-5D-3 L instrument is a standardized HRQOL questionnaire, which was developed by the EuroQol Group to provide a simple, generic measure of health for clinical and economic appraisal [1]. The EQ-5D-3 L questionnaire is available in more than 160 translated versions [2], and is the most commonly used instrument worldwide for assessment of utilities for use in cost-utility analyses to appraise health care interventions [3]. The EQ-5D is widely used in clinical trials, observational studies, and other health surveys [4, 5].

To aid in the interpretation of psychological tests, scores are often compared with a normative group, i.e. a group of test-takers that are representative of the population for whom the test is intended. Norms provide a frame of reference to interpret an individual’s scores relative to scores of others, whether the norms are national or local [6]. National population norms for the EQ-5D by socio-demographic characteristics are available from many countries [7], including the UK [8], Sweden [9], Finland [10], Denmark [11], Italy [12], Poland [13], Switzerland [14], United States [15, 16], China [17], Japan [18], Sri Lanka [19] and more [7]. The population norms present distributions of the five item scores, EQ VAS scores, or EQ-5D Index values that are based on different value sets and methods, e.g. the European VAS value set algorithms, or country-specific VAS or time trade-off (TTO)-derived values [7].

Most national and regional EQ-5D population surveys have used face-to-face interviews or computer-assisted personal interviews [7]; however some studies used postal surveys [20‐23] or computer-assisted telephone interviews [12, 24, 25]. Some studies also pool data from different postal surveys, for example to establish population norms for the EQ-5D [11]. For another commonly used health status measure, the SF-36 version 1, data from telephone and face-to-face interviews were combined to represent US general population norms [26]. However, we are not aware of population norms from web-based surveys for the EQ-5D.

Several Norwegian studies have used the EQ-5D-3 L to examine HRQOL in many diseased populations, including people with epilepsy [27], COPD [28], stroke [29, 30], diabetes [31], rheumatoid arthritis [32], hip fractures [33‐35], HIV/AIDS [36], brain tumors [37], and multiple myeloma [38]. To date there are no available population norms for this instrument in Norway.

This study aimed to provide population norms for the EQ-5D-3 L according to age, and sex, using nationally representative samples of Norwegians aged 18 years and above. The study compared results from a postal survey and a web-based data collection. We hypothesized that there would be no difference between the postal and web-based surveys after adjustment for age, sex and education.

This is the first study to examine the Norwegian population norms for the EQ-5D instrument. The major finding was that the distribution of scores for the EQ-5D descriptive system and its derived utility scores were rather similar in a postal survey and a panel survey with electronic data collection. Therefore, we think it is justified to pool the values in a combined norm set.

In the two surveys, and hence in the pooled sample, there was a deterioration of health status with increasing age, with some minor exceptions, as shown in previous studies and in a catalogue of population norms for the EQ-5D index and the EQ-VAS [7]. We noted a dip in EQ-5D index scores in the 51–60 years group among males, or an increase in the 61–70 years group. Another study noted a similar dip in EQ-5D index and other HRQoL scores at ages 55–64, or a slight hump in HRQoL in the 65–74 age group, compared to a linear, downward trend associated with age group; however, this was not considered a systematic effect [16]. The difference between genders in the present study was consistent across the EQ-5D index and EQ-VAS, in line with previous studies [7].

As far as we know, no previous collection of EQ-5D-3 L population norms has been collected using a web sample. Therefore, the difference in results between web-based and postal administrations is unknown. In principle, using a web panel could be problematic for two reasons: First, the mode of administration could impact on how people respond. Second, a web panel is by definition self-selected, and the question is whether the sample is representative of the general population.

Addressing the first question, a review of papers studying the agreement between electronic and paper administration of patient-reported outcome measures, including EQ-5D-3 L, found that the two modes of administration had a high level of equivalence [40]. The review was mostly based on studies of diseased populations, however, there is no obvious reason why the general population should have a larger difference in responses based on mode of administration than patients. Increasingly, patients who participate in studies fill out the EQ-5D electronically, and the new standard for performing valuation studies involves the use of computer-assisted interviews [41]. In that respect, electronically collected population norms have ecological validity since the mode of administration is the same.

McHorney et al. noted that in a randomized trial on the effects of data collection method on the summary scores of a different health status instrument, the SF-36, respondents to telephone interviews tended to score higher on the MCS than those in a postal survey, although there was no effect on the PCS [42]. Furthermore, telephone-administered US general population norms for the SF-36v2 in 2005–2006 [43] were higher than for norms collected by mail administration in 1998 [44] which was attributed to the effect of telephone administration.

Response rates to epidemiological surveys in general [45] and in the Nordic countries have declined over the last decades [46]. The consequences of the use of web surveys have been debated and challenges raised [47]. Often, results obtained with electronic versions are generally similar to paper versions, in terms of outcomes and psychometric properties, although the response rates to internet surveys tend to be lower and the respondents less representative than in mail surveys [48, 49]. However, established offline norms for established offline tests may still not be appropriate for use with online versions, as evidenced from psychometric tests in different settings [50]. Moreover, Buchanan states that when comparing online scores with normative data, “normative data gathered online must be used” [50]. This may apply for HRQoL as well.

Self-selection bias is a general problem when conducting surveys; people who agree to participate in surveys are systematically different from people who decline [51]. However, one could assume that web panels consist of individuals who are the most inclined to participate in studies. To counter this effect, survey companies are able to put extra effort into recruiting individuals with typical non-responder characteristics to their panels, for instance individuals from the youngest and oldest age groups. As an additional step in studies where representativeness is important, the companies may over-sample groups which have low response rates.

The pooled postal and web sample in this study was comparable to the Norwegian general population regarding gender and education, although the proportion aged 19–40 years was somewhat lower and the proportions 41–60 and > 60 years of age were somewhat higher, than in the general population in 2010 [52]. We did not weight the sample to better approximate the general population. Some studies of HRQoL have pointed at various method for adjusting web-based panel results to better approximate the scores of the general population, such as post-stratification adjustment [53‐55]. One of these studies reported web-based norms for two HRQoL instruments, the AQoL-6D and AQoL-8D, using post-stratification weights to address the effects of self-selection in the web-based survey [55].

Internet coverage is important when assessing the potential bias in using a web panel, since in some cases only selected parts of a population – the wealthy and highly educated, has had access to the relevant technology. In Norway, using computers connected to the internet is free in public libraries. Further, there is a high coverage rate of internet in Norway, as in 2010 90% of the Norwegian households had access to the internet, and 97% in 2016 [56].

Recently, the Norwegian Medicines Agency has published draft guidelines for pharmaco-economic analyses [57]. The agency recommends using the EQ-5D for estimating utilities, and suggests using Swedish population norms [9, 23] as a reference, because Norwegian norms are unavailable. The means and SDs for the population values in the various age strata in the present study were very similar to the Swedish norms collected in 1998 [9]. The norms in both the Swedish and the present study were based on the same EQ-5D-3 L tariff from the UK [39], although the Swedish study assigned a value of 0 to states rated as worse than death [9], in contrast to the present study and the original UK tariff. There are also other important methodological differences. For example, the Swedish norms were based on regional data only (from Stockholm County) from a postal survey, and it had three postal reminders and a telephone follow-up of non-respondents resulting in a higher response rate than in the present study. Furthermore, the Swedish study presented data separately for those > 80 years of age [9].

Some limitations of this study should be noted. We have discussed some of the short-comings and problems with a web-based panel, where it is difficult to evaluate the non-response and representativeness as traditionally done in postal surveys. The sample size was sufficient to provide normative values according to sex and age groups in 10-year intervals, although the number of respondents above > 70 years was too small for splitting this into several age groups. This may be important, as health status might be expected to deteriorate rapidly with increasing age > 70 years.

Our observations of equivalence between general population normative values collected in a web panel aimed to be representative of the general population and those from a postal survey, in which the survey was sent to random individuals drawn from the Population Register of Statistics Norway, supports the use of web samples when collecting EQ-5D population norms. Because of the similarity of the scores, we have pooled the scores to a larger data set, which may be used as general population norms for the EQ-5D-3 L in Norway. This may be useful for example in assessment of health care interventions or in pharmaco-economic analyses.