Sensitivity analysis of EQ-5D-3L index scores in terms of discriminative and known-groups validity in SLE: introducing Adequate Health State

Abstract

Objectives

To investigate the ability of different EuroQol 5-Dimensions 3-Levels (EQ-5D-3L) index scores to discriminate between verum drug and placebo (discriminant validity) as well as between responders and non-responders (known-groups validity) in the SLE patient population of two phase III clinical trials of belimumab.

Methods

Data from the BLISS-52 (NCT00424476) and BLISS-76 (NCT00410384) trials (N = 1684), which both showed superiority of belimumab to placebo, were utilized. Responders were defined as SLE Responder Index 4 (SRI-4) achievers at week 52. The Pearson’s χ² and Mann–Whitney U tests were used for comparisons, and logistic regression analysis was used for adjustments for confounders and assessment of independence.

Results

While full health state (FHS; EQ-5D index score 1) showed the best ability to discriminate between belimumab and placebo [adjusted odds ratio (OR) 1.47; 95% CI 1.11, 1.96; P = 0.008] and between SRI-4 responders and non-responders (adjusted OR 3.47; 95% CI 1.29, 10.98; P = 0.020), the discriminative ability of EQ-5D index scores 0.800 or more reached statistical significance for both discriminant validity (adjusted OR 1.29; 95% CI 1.02, 1.63; P = 0.036) and known-groups validity (adjusted OR 3.08; 95% CI 1.16, 9.69; P = 0.034).

Conclusion

Overall, higher EQ-5D index scores were associated with increasing ability to discriminate between belimumab and placebo, and between responders and non-responders. EQ-5D index scores less stringent than FHS may be clinically relevant health-related quality of life goals of treatment in patients with SLE, introducing the concept of EQ-5D adequate health state when FHS is not achievable.

Introduction

SLE is a chronic autoimmune disease that can cause inflammation in any organ or tissue in the human body, and mainly affects women of childbearing age [1]. Despite significant advances in the understanding of SLE pathogenesis, patients with SLE still experience considerably poorer health-related quality of life (HRQoL) compared with healthy individuals and compared with patients with other common chronic diseases [2, 3]. Since HRQoL was established as one of four core outcomes for SLE clinical trials during the OMERACT IV consensus conference, along with disease activity, medication side-effects and organ damage [4], a shift towards an increasing endorsement of the patient perspective and patient-reported outcomes has been witnessed in SLE research. This is especially important in light of the well-reported discordance in the perception of disease activity and health status, as well as concern priorities, between patients with SLE and treating physicians [5].

The EuroQol 5-Dimensions (EQ-5D) questionnaire [6] is a commonly used generic instrument for assessment of HRQoL that has displayed good psychometric properties for SLE patients regarding validity and reliability [7]. Its short and simple format makes it easy for the patient to fill in, as well as for clinicians and researchers to follow and interpret, as illustrated by its high survey completion rate and degree of patient acceptability [7, 8]. Patient responses can be converted into a single index score, which is valued against the preferences of the general population and ranges from <0–1. An index score of 1, also known as full health state (FHS), corresponds to optimal self-perceived health.

In a previous study, we showed that FHS displayed a strong ability to discriminate between drug and placebo as well as between responders and non-responders in the BLISS-52 [9] and BLISS-76 [10] clinical trials of belimumab [11]. In the pooled BLISS study population, 23.0% of the study participants reached FHS at week 52. This frequency was only 48.7% in a US general population–based reference sample, matched for age and sex. For comparison, a study by Zrubka et al. [12] which involved patients with chronic diseases, including autoimmune diseases such as RA and SSc, demonstrated an FHS frequency of 20.7%. In a Swiss study [13] examining multimorbid patients, i.e. having at least three chronic conditions, only 13.6% reported FHS. Hence, FHS can be difficult to achieve, especially in severely ill populations, and may thus be considered a stringent outcome. Although there have been studies on lower EQ-5D index scores and their discriminative ability in disease cohorts other than SLE [14–16], data are lacking for SLE patient populations.

The aim of this study was to investigate whether EQ-5D-3L index scores below the maximum value of 1.00 have the ability to discriminate between verum drug (belimumab) and placebo (discriminant validity), as well as between responders and non-responders (known-groups validity) in the SLE patient population of two large phase III clinical trials of belimumab.

Methods

Study design and population

This study was a post hoc analysis performed on data from the BLISS-52 (NCT00424476) and BLISS-76 (NCT00410384) clinical trials, two multicentre, double-blinded, placebo-controlled trials with similar design and endpoints. The trials included 865 and 819 participants, respectively. For inclusion, patients had to (i) fulfil the revised ACR classification criteria for SLE [17], (ii) be ≥18 years of age, (iii) have a Safety of Estrogens Lupus Erythematosus National Assessment– SLEDAI (SELENA-SLEDAI) [18] score of ≥6, (iv) show two positive ANA or anti-dsDNA test results of which one or more needed to be sampled at screening, and (v) be on stable non-biological standard therapy for ≥30 days, which could comprise antimalarial agents, glucocorticoids and conventional immunosuppressants.

The study participants were randomized to intravenously receive placebo, belimumab 1 mg/kg or belimumab 10 mg/kg at regular intervals. The similarities in study design enabled pooling of the data prior to analysis. EQ-5D data were used to indicate patients’ HRQoL at baseline, and weeks 4, 8, 12, 24 and 52 in both trials [19]. In the analysis evaluating EQ-5D-3L utility index scores for their ability to discriminate between drug and placebo, we considered patients who received the licenced dose of intravenous belimumab (10 mg/kg; N = 563) and patients who received placebo (N = 562). For the comparisons between responders and non-responders, we considered the entire trial populations (N = 1684). The datasets were made available by GlaxoSmithKline through the Clinical Study Data Request (CSDR) portal.

EQ-5D-3L

The EQ-5D questionnaire consists of two parts, i.e. a visual analogue scale which is used for assessing overall health, and a descriptive system that comprises five dimensions of health, i.e. mobility, self-care, usual activities, pain or discomfort, and anxiety or depression. In the three-level version (EQ-5D-3L), there are three levels of severity for each dimension, i.e. ‘no health problems’ (level 1), ‘moderate health problems’ (level 2) and ‘extreme health problem’ (level 3). Combining the reported levels from each of the dimensions yields a unique health state, with a total of 3⁵ = 243 possible health states. Each health state is referred to as a 5-digit code. For example, 11111 indicates no problems in any of the five dimensions, whereas 33333 indicates the worst possible state. The health state may also be converted into a single utility index score, which is derived by applying a formula that attaches values to each of the levels in each dimension. The compilation of values for all possible EQ-5D states is called a value set, most of which are based on the preferences of the general population in a certain country, ensuring that the score is reflective of that specific society. As in early reports from the BLISS studies, the US value set [20] was used in this post hoc analysis. The index value can range from <0 (with 0 being a health state experience equivalent to death, and negative values being valued worse than death) to 1 (considered optimal health, also known as FHS). Thus, an index score of 1 is intended to reflect the desired perception of health status [6].

Outcome measures

The SLEDAI 2000 (SLEDAI-2K) [21] was used to assess SLE disease activity, and the SLICC/ACR Damage Index (SDI) [22] was used to assess organ damage. Responders were defined as achievers of the SLE responder index 4 (SRI-4) criteria [23] at week 52 from treatment initiation. The SRI-4 at week 52 was the primary efficacy endpoint in the BLISS-52 and BLISS-76 trials, with its criteria being derived from a preceding phase II trial of belimumab [24], which was also used as the outcome of response in the present study.

Statistical analysis

Data are presented as numbers (percentage), or the mean (s.d.) in the case of normal distributions or the median [interquartile range (IQR)] in the case of non-normal distributions. The Pearson’s chi-squared (χ²) test was used for comparisons of unrelated binary variables and the Mann–Whitney U test was used for comparisons of unrelated continuous variables. Logistic regression analysis was employed for different EQ-5D-3L utility index scores as the primary exposure and the licenced dose of belimumab, i.e. intravenous infusions of 10 mg/kg (reference: placebo) as the outcome. Furthermore, the relationship between EQ-5D-3L utility index scores and SRI-4 response at week 52 (reference: non-response) was investigated using logistic regression analysis; for adjustments, multivariable models included age, sex, ancestry, any organ damage accrued by the end of follow-up (SDI score >0 at week 52) and exposure to belimumab.

P-values <0.05 were considered statistically significant. The IBM SPSS software version 28 (IBM Corp., Armonk, NY, USA) and R software version 4.1.0 (R Foundation for Statistical Computing, Vienna, Austria) were used for statistical analyses and data management.

Patient involvement

A patient research partner (Y.E.) was involved in the study concept and design, interpretation of data and editing of the manuscript.

Ethics

All participating sites have obtained ethics committee or institutional review board approvals and patient recruitment and follow-up was conducted in accordance with the principles of the Declaration of Helsinki (NCT00424476 and NCT00410384). All study participants provided written informed consent prior to enrolment in the BLISS-52 and BLISS-76 trials. Access to the datasets was granted by GlaxoSmithKline through the CSDR consortium. The present study was approved by the Swedish Ethical Review Authority (2019-05498).

Results

Study population

Demographic and clinical characteristics, including the EQ-5D-3L index scores for the pooled BLISS population, are presented in Table 1. The total number of patients was 1684, comprising patients who received belimumab (N = 1122) and patients who received placebo (N = 562). Demographic characteristics were balanced between the treatment arms. Mean age at baseline was 37.8 years and most patients were female (94.1%), while the most common ancestry was white/Caucasian (47.4%). The proportion of patients with an SDI score >0 was 41.9% at baseline and 44.0% at week 52. The median EQ-5D-3L index score was 0.778 at baseline for both the patients receiving standard therapy plus belimumab and those receiving standard therapy alone. In the pooled population, 45.1% achieved SRI-4 at week 52, denoting responders, with a larger proportion achieving SRI-4 in the belimumab arm compared with the placebo arm (48.4% vs 38.6%; P < 0.001).

Discriminative ability

At week 52, there was a statistically significant difference in the EQ-5D-3L index score between patients receiving belimumab and patients receiving placebo (P = 0.023), as seen in Table 1. Patients receiving belimumab had a higher median index score (0.816; IQR 0.708–0.860) compared with patients receiving placebo (0.800; IQR 0.708–0.843).

Verum drug vs placebo

The discriminative ability of different EQ-5D-3L index scores to separate between verum drug (belimumab 10 mg/kg) and placebo at week 52 is delineated in Fig. 1. FHS had the best discriminative ability in the pooled BLISS study population. However, lower scores than FHS also showed ability to separate belimumab from placebo. As illustrated in Fig. 1, the discriminative ability increased with an increasing EQ-5D-3L index score, with the nine highest EQ-5D-3L index scores among those reported by the patients in the BLISS trials reaching a statistically significant discriminative ability. Detailed results from this analysis are presented in Table 2.

FHS yielded the highest odds ratio (OR) for separating verum drug from placebo (OR 1.471; 95% CI 1.11, 1.96; P = 0.008). The OR generally increased with an increasing index score. The lowest cut-off that was statistically significant in the pooled BLISS study population was 0.816 (OR 1.287; 95% CI 1.017, 1.630; P = 0.036).

Responders vs non-responders

Fig. 2 shows the predicted probability of SRI-4 given a certain EQ-5D-3L index score, based on analysis adjusted for confounders (Table 3). As shown in Fig. 2, patients who reported FHS had the highest probability of having achieved SRI-4 response at week 52, whilst patients with an EQ-5D-3L index score of –0.109 had the lowest probability. Notably, the probability of achieving SRI-4 increased in a linear fashion with an increasing EQ-5D-3L index score.

Results from logistic regression analysis are detailed in Table 3. EQ-5D-3L index scores at week 52 (adjusted OR 1.019; 95% CI 1.01, 10.3; P < 0.001) and belimumab use (adjusted OR 1.448; 95% CI 1.17, 1.79; P = 0.001) were associated with attainment of SRI-4. Patients of Asian ancestry were less likely to achieve SRI-4 response compared with patients of other ancestries (adjusted OR 0.655; 95% CI 0.50, 0.86; P = 0.003), whilst patients of Indigenous American ancestry were more likely to achieve SRI-4 response compared with patients of other ancestries (adjusted OR 1.416; 95% CI 1.10, 1.83; P = 0.008). Age, sex and organ damage accrued by week 52 showed no significant association with SRI-4 response.

Finally, EQ-5D-3L index scores were grouped into categories to facilitate interpretation (Table 4). As seen in Table 4, the lower the EQ-5D-3L index score category, the lower was its discriminative ability. EQ-5D-3L index scores of 0.8 and above favoured the ability of the index to predict SRI-4 response in both unadjusted and adjusted models, with the comparator being EQ-5D-3L index scores below 0.8 (adjusted OR for EQ-5D-3L index scores 0.8–0.9: 3.081; 95% CI 1.16, 9.69; P = 0.034; adjusted OR for EQ-5D-3L FHS: 3.472; 95% CI 1.29, 10.98; P = 0.020). Also in this analysis, patients of Asian ancestry were less likely to achieve SRI-4 response compared with patients of other ancestries (adjusted OR 0.664; 95% CI 0.50, 0.87; P = 0.004), and patients of Indigenous American ancestry were more likely to achieve SRI-4 response compared with patients of other ancestries (adjusted OR 1.429; 95% CI 1.11, 1.85; P = 0.006), while age, sex and organ damage at week 52 yielded no significant association.

Discussion

The purpose of this study was to investigate whether EQ-5D-3L index scores below the maximum value of 1.00 have ability to discriminate between verum drug and placebo as well as between responders and non-responders in the SLE patient population of two successful phase III clinical trials of belimumab, which both met their primary endpoint, i.e. superiority of belimumab over placebo in terms of proportions of SRI-4 responders at week 52. Through logistic regression analysis, EQ-5D-3L FHS was found to possess the best discriminative ability at week 52. Moreover, increasing EQ-5D-3L index scores were associated with a gradually greater discriminative ability. These associations remained significant after adjustments for confounding factors in multivariable logistic regression analysis. In the analysis for discrimination between drug and placebo, as well as in the analysis between responders and non-responders, EQ-5D-3L index scores above ∼0.800 showed discriminative ability that reached statistically significant separation, both before and after adjustment for confounders.

In a previous study by Lindblom et al. [11], FHS showed a robust ability in distinguishing belimumab from placebo as well as responders from non-responders in the same patient population, at multiple times during the study period, following similar methodology. Herein, EQ-5D-3L FHS displayed the best discriminative ability among EQ-5D-3L utility index scores. There was a linear increase of the predictive ability of gradually increasing EQ-5D-3L index scores, corresponding to an overall better self-perception of HRQoL, to discriminate between treatment arms and between clinical responders and non-responders. This is well in line with previous studies demonstrating that biological agents including belimumab contribute to HRQoL improvements in SLE patients [19, 25], with responders showing overall greater improvements compared with non-responders [26]. Thereby the results of the present investigation lend additional support for the benefits of belimumab in the context of patient-reported HRQoL [27, 28].

Change in EQ-5D-3L was not found to discriminate between belimumab and placebo in a study of the same patient population by Strand et al. [19]. In contrast to our study, Strand et al. investigated mean changes of EQ-5D index scores from baseline to week 52 rather than attainment of a specific health state as in the present study (i.e. EQ-5D FHS and gradually less stringent health states) at a specific timepoint (e.g. at week 52), using an analysis of covariance model for comparison of the treatment arms. The different approaches may explain the contrasting, yet complementary, results, introducing the concept of using EQ-5D health states rather than score changes as HRQoL goals of treatments in clinical trial design, following the concept of states of clinical activity as main treatment targets, i.e. remission or low disease activity.

We found EQ-5D-3L index scores above a cut-off of ∼0.800 to be statistically significant in discriminating between drug and placebo, as well as between clinical responders and non-responders. Whilst this cut-off is specific for the cohort of the present study and cannot be extrapolated for use in any SLE population, other studies also point towards the clinical relevance of index scores lower than FHS. For instance, a study from Argentina by Machado Escobar et al. involving 147 SLE patients found that an EQ-5D index scores 0.739 or above defined a good quality of life [29]. Machado Escobar et al. used latent class analysis, values of Bayesian Information Criterion as optimal criterion, probability calculations and receiver operating characteristic curve analysis to establish an optimal cut-off [29]. A study by Cooper et al. involving chronic arthritis patients compared EQ-5D value sets from Sweden and the UK [16]. This was done by establishing the EQ-5D cut-off values beyond which patients were satisfied with their current condition based on their level of pain and level of general function. The optimal cut-off was set at 0.69 for the EQ-5D British value set and 0.81 for the Swedish value set. The values were determined using receiver operating characteristic curve analysis.

While FHS denotes the desirable health status as perceived by the patients, lower cut-offs that are less stringent may designate an ‘adequate health state’. This is of relevance as FHS may not always be achievable, as shown in the study by Lindblom et al. [11] in which the FHS frequency post-treatment was 23.0%. In the same study, disease activity and organ damage scores were found to be independently associated with lower frequencies of FHS, while in a subsequent study, experience of FHS was found to be associated with reduced accrual of organ damage [30]. Thereby, ‘adequate health state’ may be a clinically relevant HRQoL treatment goal whenever FHS is not achievable. This follows the treat-to-target (T2T) principle, according to which remission is desirable, but whenever remission is not achievable, low disease activity should be the target [31]. Based on our findings, adequate health state could allow for ‘some problems’, i.e. a level 2 response in up to one dimension, but does not allow for ‘extreme problems’, i.e. a level 3 response in any of the dimensions. While further study in other cohorts is needed for validation purposes, the introduction of the concept of adequate health state with the above simplified definition could substantiate a patient-reported outcome measure (PROM)-based constituent in T2T approaches. Such concepts may inform current efforts aiming at incorporation of PROMs in treatment evaluation in a systematic fashion. In this regard, it is worth noting that a global academic–industry partnership was recently formed with the goal to develop a novel, patient-centred, clinician-reported outcome measure for SLE randomized clinical trials, the Treatment Response Measure for SLE (TRM-SLE), accounting for drug efficacy in HRQoL aspects [32].

As this study was a post hoc analysis of previously conducted trials, there are some limitations connected to its design. The BLISS trial protocols excluded patients with severe active LN, CNS vasculitis or neuropsychiatric SLE; therefore, the results cannot be extrapolated to these SLE patient populations. In addition, data on some potential confounders were unavailable, examples being comorbidities such as FM, as well as socioeconomic status, diagnosis of depression or data on daily physical activity. Moreover, as the threshold values determined in this study can only be considered specific for this cohort, the results should be interpreted with caution and further analysis of clinically relevant EQ-5D index score targets in future studies may elucidate how optimal cut-offs differ depending on the population characteristics and the value sets used. In this regard, application of country-specific value sets may yield divergent results and could form the scope of a future investigation. Altogether, such aspects lessen the generalizability of our results.

Strengths of this study included the large cohort, with patients from a broad variety of countries, as well as the low to minimal degree of data missingness. Considering that SLE is a rare disease in many countries, the size of the BLISS cohort makes it unique and highly valuable for post hoc studies. This also contributed to sufficient power in the statistical analyses and allowed for assessment of and adjustment for several variables with confounding potentiality, including organ damage, demographic, laboratory and clinical parameters, as well as concomitant treatments.

Conclusions

EQ-5D-3L FHS had the best ability to discriminate between verum drug (belimumab) and placebo, as well as between responders and non-responders at week 52, in the patient population of the BLISS-52 and BLISS-76 clinical trials of belimumab in SLE. We found an association between increasing EQ-5D-3L index scores and improving discriminative ability. Our findings suggest that EQ-5D cut-offs less stringent than FHS might prove useful in separating verum drug from placebo in SLE trials of drugs with similar properties and effect magnitude as that of belimumab. An EQ-5D ‘adequate health state’ defined as some problems in no more than one dimension and no response of extreme problems may constitute a pragmatic HRQoL goal in SLE patients whenever EQ-5D ‘full health state’ cannot be achieved, resembling the T2T principles for remission and low disease activity in SLE. Overall, our findings provide additional support for the usefulness of EQ-5D as a PROM in SLE, and the integration of PROMs in the global evaluation of patients with SLE along with the clinical assessments.

Data availability

The datasets used and analysed during the current study can be made available through the CSDR consortium.

Contribution statement

Conception and design of the work: I.P. Data management: N.H., A.G., J.L., D.G., I.P. Statistical analysis and interpretation of data: N.H., A.G., J.L., S.E., Y.E., D.G., E.H., M.R., I.P. Patient research partner: Y.E. Critical revision of the manuscript for important intellectual content: all authors. All authors reviewed and approved the final version of the manuscript prior to submission and agree to be accountable for all aspects of the work.

Funding

This work was supported by the GlaxoSmithKline Investigator-Sponsored Studies (ISS) programme and grants from the Swedish Rheumatism Association (R-969696), King Gustaf V’s 80-year Foundation (FAI-2020-0741), Swedish Society of Medicine (SLS-974449), Nyckelfonden (OLL-974804); Professor Nanna Svartz Foundation (2021-00436), Ulla and Roland Gustafsson Foundation (2021-26), Region Stockholm (FoUI-955483) and Karolinska Institutet. The funders had no role in the design of the study, in the collection, analyses or interpretation of data, in the writing of the manuscript or in the decision to publish the results.

Disclosure statement: J.L., E.H., M.R. and I.P. have received research funding from the EuroQol Research Foundation. I.P. has additionally received research funding and/or honoraria from Amgen, AstraZeneca, Aurinia Pharmaceuticals, Elli Lilly and Company, Gilead Sciences, GlaxoSmithKline, Janssen Pharmaceuticals, Novartis, Otsuka Pharmaceutical and F. Hoffmann-La Roche AG.

Acknowledgements

The authors would like to thank GlaxoSmithKline (Uxbridge, UK) for sharing the data from the BLISS-52 (NCT00424476) and BLISS-76 (NCT00410384) trials with the Clinical Study Data Request (CSDR) consortium, as well as all participating patients.

References