Assessing health state utilities for people with myalgic encephalomyelitis/chronic fatigue syndrome in Australia using the EQ-5D-5L, AQoL-8D and EQ-5D-5L-psychosocial instruments

Myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) is a chronic, multisystemic disease with complex manifestations [1, 2]. Its clinical manifestations are heterogenous in nature with an identifiable pattern; however, it generally presents with disabling extreme post-exertional malaise (PEM) [2]. PEM is the pathological inability of the body system to generate adequate energy on demand: a massive and prolonged energy deficit which results in the inability to perform normal and routine activities [2]. PEM usually arises in concert with other disabling and prolonged systemic manifestations such as unrefreshing sleep, orthostatic intolerance and cognitive impairment [3, 4]. The chronic nature of ME/CFS affects the work productivity and health-related quality of life (HRQoL) of people living with the condition over time [5‐8]. Furthermore, ME/CFS affects more persons from the age groups most likely to be in the workforce [5, 6, 9].

Health state utilities (HSU) are metrics that measure the strength of preference for a particular health state, represented as a number between 0 and 1, where 0 is anchored to death (or health states equivalent to being dead) and 1 corresponds to ideal health [10‐12]. Health states worse than death are possible, represented by negative HSU [13]. HSU are used in cost-utility analyses (CUA) for Health Technology Assessments [11]. There are several approaches to deriving HSU, with multi-attribute utility instruments (MAUI) commonly used [11]. As well as being an input metric to CUA, HSU have also been shown to be independent predictors of patient outcomes, including all-cause mortality and development of complications [14]. Moreover, clinicians have found that measuring HRQoL and HSU is beneficial to patients regarding clinical assessment, relationships, communication, and management [15].

Just three studies assessing HSU for people living with ME/CFS have been published, originating from the UK [16, 17] and Denmark [8]. Overall, mean HSU were consistently low, ranging between 0.36 and 0.56 using the EQ-5D-3L [8, 16, 17] instrument. However, to the best of our knowledge there are no studies that assess HSU against disease severity classifications for ME/CFS, hence assessing cost-utilities for ME/CFS remains challenging.

Some studies have used generic HRQoL measures that are not MAUI (and therefore do not measure HSU) such as the SF-36 [7] or World Health Organization Quality of Life questionnaire (WHOQOL-BREF) [18] to reveal diminished HRQoL for people living with ME/CFS. An Australian study examined the impact of sociodemographic and patient symptoms using the SF-36 and reported significantly lower scores across all domains compared to the general population [7]. Similarly, another Australian study explored HRQoL (physical functioning and psychological distress) for people living with ME/CFS and reported profound effects on physical functioning in addition to increased manifestations of psychological distress [19]. Another much older Australian study (1995) used the Sickness Impact Profile and patient interviews to demonstrate that ME/CFS had a significant impact on the quality of life of people with ME/CFS, especially on their social functioning [20]. A multi-country study reported similarly reduced HRQoL for people living with ME/CFS [21].

The primary aim of this study was to investigate HSU for people with ME/CFS. In addition, we aimed to identify a MAUI with a propensity for the physical and/or psychosocial domains of health and that is preferentially sensitive for Australians with ME/CFS.

A national survey using convenience sampling was conducted between August and December 2021 to assess the human and economic impacts of ME/CFS. To inform the development of a tailored questionnaire, we conducted focus groups with people living with ME/CFS. Based on these, we included questions about costs incurred due to ME/CFS: costs related to electronic and digital equipment and ‘apps’, childcare, everyday living (e.g., cleaning, gardening services), specialised home modification or renovations, and special dietary requirements. Furthermore, we worked closely with the Patient Advisory Group to ensure the survey captured relevant information and was delivered in such a way as to reduce cognitive burden on participants. The online survey was hosted using the Qualtrics platform in modules with an estimated completion time presented for each. This was done so participants could pace themselves and complete modules during separate sessions. The survey included questions on disability and fatigue severity using the de Paul Symptom Questionnaire-Short Form (DSQ-SF) [22], and on co-morbidities (including diabetes, cardiovascular disease and mental health and classified as 0, 1, 2 and 3 or more co-morbidities). HSU were assessed using the AQoL-8D, EQ-5D-5L and EQ-5D-5L-Psychosocial instruments. To minimize bias that may arise from asking repetitive questions on HRQoL, the EQ-5D-5L and AQoL-8D instruments were delivered to participants in random order, with 50% completing the EQ-5D-5L first and 50% completing the AQoL-8D first. Ethics approval was granted by the University of Tasmania’s Health and Medical Research Ethics Committee (H0018473).

Eligibility criteria for our participants included a self-reported diagnosis of ME/CFS and aged 18 years or older and living in Australia. Participants were recruited through newspaper advertisements, newsletters, and social media posts of Emerge Australia (a national organisation providing education, advocacy, research, and support services).

Table 1 shows the characteristics of the three MAUI. The EQ‐5D-5L asks participants to indicate whether they have problems on a five-level scale for each of the five dimensions of health: mobility, self‐care, usual activities, pain/discomfort, and anxiety/depression. The EQ-5D-5L was developed to address the limited sensitivity (lack of descriptive richness and serious ceiling effects) of its predecessor the EQ-5D-3L [23] and describes 3125 health states. The algorithmic range for most of the instrument’s country-specific value sets describes HSU ranging from < 0 to 1.0[13].

The AQoL-8D was originally developed to achieve sensitivity not only in health states affected by physical disorders, but also in those affected by mental disorders [24]. This instrument contains 35 items in eight dimensions and was derived using psychometric methods for achieving content validity. Three of the dimensions (independent living, pain, senses) load to a physical super-dimension; the other five (mental health, happiness, coping, relationships, and self-worth) load to a mental super-dimension. The size of the instrument means that it can define billions of health states [24].

The new EQ-5D-5L-Psychosocial was developed by Chen and Olsen in 2020 [25] and was externally validated for a large cohort with multiple sclerosis (MS) [26]. It was developed to address the psychosocial deficiencies of the EQ-5D-5L by including four additional bolt-on dimensions of vitality, relationships, sleep and community connectedness, adopted from the AQoL-8D (Table 1) [25]. The developmental phase of this new instrument revealed that vitality was the most important dimension with regard to HRQoL [25]. Given the dominant position of the EQ-5D-5L in applied studies, the developers suggested that identifying a set of bolt-on dimensions that captured the psychosocial aspects of health would serve as a realistic alternative for developing a completely new extended generic preference-based measure [25]. The scoring algorithm was developed from a mapping analysis that mapped responses to nine items (five EQ-5D-5L and four bolt-on items) onto the AQoL-8D utilities.

Australian population norms for the AQoL-8D and EQ-5D-5L were sourced from literature to correspond to the mean age of our study population i.e. norm for the AQoL-8D is 0.77 [27] and for the EQ-5D-5L is 0.89 utility points [28].

Table 1 shows the characteristics of the DSQ-SF [29]. It consists of 14 questions (with two 5-point Likert scales for each question) and has been validated against the Canadian Consensus Criteria [2]. Disability severity was calculated using the DSQ-SF. We adopted a validated methodology used in a previous work on multiple sclerosis to map from the Patient Determined Disease Steps (9 questions regarding gait) to the Expanded Disability Status Scale [30]. Following the logic of this disability severity classification, we calculated disability severity for each individual who entered the study by assigning values of no disability (0), mild disability (1), moderate disability (2) and severe disability (3) to the five Likert responses (with increasing disability) to the 28 DSQ-SF questions with 0 = 1; 1 = 2; 2 = 3; and 3 = 4 and 5 for each question and then calculating the average across the 14 questions. The final disability severity score used the cut points of < 0–0.5 (no disability); 0.6–1.5 (mild disability); 1.6–2.5 (moderate disability); and > 2.5 (severe disability). Fatigue severity was classified by selecting seven questions directly related to fatigue symptoms such as fatigue, tiredness after minimal exercise and unrefreshing sleep, and followed the same method to select fatigue severity questions (Supplementary Table 1).

Summary statistics describing the baseline sociodemographic and clinical characteristics (Table 2) of participants were generated and presented as mean and standard deviation (SD) or median and interquartile range (IQR) (Supplementary Table 3) for continuous variables, and frequency counts and percentages for categorical variables (Table 3). Completion rates for all three MAUI were assessed and summarized. HSU were described using frequency distributions and summary statistics (mean, 95% CI and range). HSU were generated using the Australian tariffs [13, 25, 31] noting that there can be 10 missing patient-reported responses for the AQoL-8D, and no missing values for the EQ-5D instruments. An Australian value set for the EQ-5D-5L has not been published to-date; therefore we used an Australian algorithm based on a discrete choice experiment [13]. The algorithm for the EQ-5D-5L-Psychosocial (using Australian data) was adopted [25]. Histograms were used to visualize the frequency distribution of the individual HSU in each MAUI (Supplementary Fig. 2).

We visualized pairwise combination of the three instruments and examined the sensitivity of each by comparing the means and differences of HSU using Bland–Altman’s analysis (Fig. 2a–c). This method was used to show the agreement or bias between two paired quantitative measures [32]. Spearman’s correlation coefficient (Supplementary Table 4) was used to assess pairwise relationships between the instruments [10]. Spearman’s correlation coefficient is a non-parametric statistical measure used to test the strength and the direction of association between two ranked variables [33].

We examined floor and ceiling effects for the three MAUI (Table 4) Ceiling effects were assessed as HSU = 1.0 (full health). Floor effects were assessed as the worst patient-reported HSU for both the AQoL-8D and EQ-5D-5L-Psychosocial, and < 0 for the EQ-5D-5L (given that an HSU less than 0 is a health state deemed to be worse than death). We cross-matched against the individual HSU generated by the alternate instruments for individuals with ceiling and floor effects and then investigated the summary statistics for these HSU.

We investigated summary HSU for each disability and fatigue severity classification. To validate our method, we expected that as disease and fatigue severity increased, mean and median HSU would decrease. We also investigated the proportions of responses for the four bolt-on dimensions of the EQ-5D-5L-Psychosocial across the five levels of dimensions (Table 5).

Univariable linear regression models were used to examine the association between mean HSU of the sociodemographic variables and disease classifications for the three MAUI (Table 6). A p-value of < 0.05 was assumed as statistically significant. STATA (version 17, Stata Corp), R Package and Microsoft Excel were used for statistical analyses.

Overall, 201 participants attempted to complete the AQoL-8D and EQ-5D-5L, with 198 providing sufficient responses to calculate HSU. Non-completers and completers were similar across sociodemographic variables. Table 2 describes participants’ characteristics. The mean (SD) age was 48.7 years (14.3) with four-fifths being female (n = 158, 79.8%). Using the Australian Bureau of Statistics’ Statistical Geography Standard [34], over half resided in major cities (58.9%) and one-quarter in inner regional areas (28.1%). Almost half (43.9%) of the participants were either married/defacto whilst a third were single. Almost half (46.5%) said they were too unwell to work whilst one-quarter (26.7%) worked on a part-time basis. Of the participants with a source of income, two-fifths (40.5%) earned between $AUD400 and $AUD799 per week. The majority of participants reported no comorbid health conditions (64.7%), and one-quarter had at least one comorbidity (24.2%). For disability and fatigue severity, 69.2% reported moderate disability severity and 59.6% reported moderate fatigue severity respectively.

Of the 198 participants for whom we could generate an HSU using the EQ-5D-5L, we observed between 1 and 4 missing responses each in questions for the AQoL-8D relating to confidence (4), family relationship (4), mobility (4), cope with problems (4), help around house (6) and close relationships (3) respectively-(Supplementary Table 2). These missing values did not affect the utility values generated for the AQoL-8D.

Table 3 and Supplementary Table 3 summarizes HSU for the three MAUI. Mean HSU for the three instruments were similar: EQ-5D-5L 0.46 (95% CI 0.42–0.50); AQoL-8D 0.44 (95% CI 0.42–0.45); and EQ-5D-5L Psychosocial 0.44 (95% CI 0.42–0.46). Notably, these HSU were substantially reduced (almost half) compared to the general population norms (i.e. EQ-5D-5L: 0.89 and AQoL-8D: 0.77) [27, 28]. The median (IQR) HSU were similar for the EQ-5D-5L Psychosocial and AQoL-8D but higher for the EQ-5D-5L.

Mean HSU were higher in males compared to females across the three instruments and were observed to be highest amongst males for the EQ-5D-5L at 0.53 (95% CI 0.44–0.62). Participants in the < 45-year age category had lower mean HSU compared to the older age category (46–84 years) across all three instruments. The EQ-5D-5L recorded the highest mean HSU of 0.51 (0.45–0.57) for the older age category.

Regarding the disability and fatigue severity classifications, mean HSU for all three instruments showed the expected inverse relationship: as disability severity and fatigue severity increased, the concomitant HSU decreased. Ideal health (1.00 for the EQ-5D-5L) and near-ideal health (0.99: AQoL-8D and 0.94: EQ-5D-5L-Psychosocial) for participants with no disability was observed. In addition, substantially diminished mean HSU for the severe disability severity classification of EQ-5D-5L (0.02), AQoL-8D (0.30) and EQ-5D-5L-Psychosocial (0.25) were reported. The substantially lower HSU generated from the EQ-5D-5L was due to its broader algorithmic range as noted in Table 1. Figure 1 shows the trend of mean HSU for disability severity and fatigue severity categories across the three MAUI.

Figure 2a–c show Bland Altman’s plots with positive mean differences between EQ-5D-5L and AQoL-8D = 0.03 (95% CI − 0.53 to 0.47); EQ-5D-5L and EQ-5D-5L-Pychosocial = 0.02 (95% CI − 0.72 to 0.68); and AQoL-8D and EQ-5D-5L-Psychosocial = 0.01 (95% CI − 0.40 to 0.42). The Spearman’s correlation coefficient for the three instruments was strong and positively correlated for the EQ-5D-5L-Psychosocial and the AQoL-8D (ρ 0.83; p < 0.0001) and the least positive correlation for the AQoL-8D and EQ-5D-5L (ρ 0.65; p < 0.0001). However, we note that a high correlation does not imply close agreement and is blind to the possibility of systematic bias [10, 32, 35]. To verify these results, we analysed Bland–Altman plots. Visualisation of the plots in Fig. 2a–c also revealed that there was systematic variation between the EQ-5D-5L and EQ-5D-5L-Psychosocial and the AQoL-8D and EQ-5D-5L. On the other hand, there was no evidence of systematic variation between the AQoL-8D and EQ-5D-5L-Psychosocial suggesting pairwise agreement for these two instruments, validating our results from the correlation analysis.

Table 4 shows the floor and ceiling effects for each instrument. For the EQ-5D-5L, a ceiling effect (i.e., participants reporting full/perfect health) was observed for 2.5% (n = 5) of participants, and a floor effect was observed for 13.6% (n = 27). In contrast, no floor or ceiling effects were observed for the EQ-5D-5L-Psychosocial or AQoL-8D instruments although both showed two high—but not perfect—utility valuations (EQ-5D-5L Psychosocial = 0.91 and 0.98); AQoL-8D (0.95 and 0.99).

To investigate the floor effect for the EQ-5D-5L in more detail, Table 5 shows a comparison of the responses to the bolt-on questions for the EQ-5D-5L-Psychosocial for the participants who reported a floor effect (n = 27) and those who did not (n = 171). We observed that 89% of participants in the n = 27 group always lacked energy whilst only 50% of the participants in n = 171 group felt the same way.

Table 6 summarises the univariable regression models assessing the associations between mean HSU and clinical variables across the three instruments. The strongest effects were associated with disability severity and fatigue severity. Firstly, compared to no disability, moderate and severe disability were associated with substantial decreases in HSU across all three instruments (p < 0.01). Mild disability was associated with decreased HSU with the AQoL-8D and EQ-5D-5L psychosocial instruments respectively (p < 0.01), but not for the EQ-5D-5L.

Effects were also observed for age, employment, and marital status. For age, small associations between older ages (45–84 years) and higher HSU were observed for the EQ-5D-5L alone and with the psychosocial questions (p < 0.01 and p = 0.02 respectively). Unsurprisingly, compared to the employment category “too unwell to work”, being employed full- or part-time was associated with improved HSU for all three instruments. In addition, being retired was also associated with higher HSU (p < 0.01 for all three instruments). With regard to relationship status, being married was associated with higher HSU for each instrument (p ≤ 0.01).