Aphasia-specific or generic outcomes? a comparison of two health-related quality of life instruments for economic evaluations of aphasia treatments

This secondary analysis used deidentified, repeated measures data from the Constraint Induced or Multi-Modal Personalised Aphasia Rehabilitation (COMPARE) randomised controlled trial. The COMPARE trial was approved by institutional (La Trobe University) and hospital (Gold Coast University Hospital and participating hospital sites) ethics committees [16]. Participants were eligible if they had aphasia resulting from a stroke occurring at least six months prior to consent (Western Aphasia Battery-Revised Aphasia Quotient (WAB-R AQ) scores of < 93.8), medical stability, pre-stroke English fluency, independent toileting and a significant other available for assessments [16]. Exclusion criteria included previous non-stroke neurological events, severe apraxia of speech or dysarthria, major clinical depression or other mental health condition, and uncorrected sensory or other serious medical condition potentially impacting or influencing study participation [17]. Analysis for the current study included 201 COMPARE participants who completed the EQ-5D-3L and SAQOL-39g instruments at baseline [16]. Participants with incomplete outcome data or lost to follow-up were excluded from the 12-week analysis as utility values and total mean scores required complete the EQ-5D-3L or SAQOL-39g questionnaire responses.

COMPARE participants were recruited through direct community advertising and 22 hospitals in Australia and New Zealand (July 2016 to March 2021). From 342 people screened, 249 met the eligibility criteria and 216 were randomised to one of three treatment groups based on aphasia severity (WAB-R AQ scores) and location: constraint induced aphasia therapy- plus (nn = 71, 33%), multi-modality aphasia therapy (n = 75, 35%) and usual care (n = 71, 32%). Post-randomisation, 15 participants withdrew (three dropouts, nine due to treatment dissatisfaction, three due to COVID-19 restrictions), resulting in 201 participants completing baseline data collection [16]. Detailed recruitment, and withdrawal information is published elsewhere [16‐18]. Reporting of this study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [19].

The EQ-5D-3L measures HRQOL across five dimensions (mobility, self-care, usual activities, pain/discomfort and anxiety/depression) with three response levels: 1 = no problems, 2 = some problems, or 3 = extreme problems [9]. Patient responses create a five-digit health state profile matched to established country-specific value sets, generating utility values from 0 (death) to 1 (perfect health) [7, 8]. Utility values are multiplied by the duration of time spent in that health state to estimate QALYs. The EQ-5D-3L includes a Visual Analogue Scale (VAS) where patients rate their overall health from 0 (worst imaginable) to 100 (best imaginable) [7].

The SAQOL-39g contains 39 items across physical (16 items), psychosocial (16 items) and communication (7 items) domains with five response options: 1 = couldn’t do it at all, 2 = a lot of trouble, 3 = some trouble, 4 = a little trouble and 5 = no trouble at all. Domain and overall mean scores (1–5) are calculated, with higher score indicating better HRQOL [15]. Scores of four or below may be indicative of compromised quality of life [3].

The two instruments differ substantially (Fig. 1). The EQ-5D-3L covers five dimensions aligning with physical, mental, and social health functioning [20], while the SAQOL-39g focuses on three domains which are affected by post-stroke aphasia, including communication. Recall periods for the two instruments differs (today or within the last week, respectively). Lower SAQOL-39g scores, EQ-5D-3L utility scores, and VAS scores indicate poorer health. However, for EQ-5D-3L item responses, the lowest score (1) indicates best perceived functioning, and the highest score (3) represents the worst perceived functioning for each domain [9, 15].

Data that support the findings of this study are held by third parties and are available upon reasonable request to the corresponding author of the COMPARE trial (author M.L.R).

Ethical approval for this project was obtained from the La Trobe Human Ethics Low Risk Committee (Approval number: HEC23119). Approval to use deidentified COMPARE trial data was obtained from the data custodian (author M.L.R, La Trobe University).

Data were pooled for this analysis. Baseline participant characteristics were summarised using descriptive statistics. All other analyses were repeated in baseline and 12-week follow-up data. An algorithm validated in an Australian general population was used to convert baseline and follow-up EQ-5D-3L domain scores to utility values, as reported elsewhere [21]. Less than 10% of outcome data were missing, eliminating the need for data imputation [18]. Demographic characteristics of participants with missing versus complete baseline EQ-5D-3L and SAQOL-39g outcome data compared using chi-square tests for categorical variables and t-tests for continuous variables. Participant withdrawal is outlined alongside the COMPARE study results [18]. SAQOL-39g domain scores were reversed in the Spearman correlation and exploratory factor analyses for consistency of severity direction between the two measures. All analyses were conducted using STATA/SE 18.0 [22].

The distribution of scores at baseline and 12-week follow up were analysed with descriptive statistics and visualised with scatterplots for the EQ-5D-3L and SAQOL-39g. Convergent validity between the EQ-5D-3L and SAQOL-39g scores was explored by use of Spearman’s rank correlation coefficient. Correlation r values were interpreted as weak (0.1–0.39), moderate (0.4–0.69), high (0.7–0.89) or very high (0.9-1.0) per established benchmarks [23]. The discriminative ability of the EQ-5D-3L utility values and VAS was explored using a Receiver Operator Characteristic (ROC) curve analysis to examine how well they could classify participants according to their SAQOL-39g scores. We dichotomised SAQOL-39g scores at < 4 versus ≤ 4 for the ROC analysis, representing the lower half of the response scale (1–4 ‘couldn’t do it at all’ to ‘some trouble’ vs. the upper range (5: ‘no trouble at all’), based on the instrument developers’ description of the scores [3]. This cutoff point distinguishes between participants reporting some degree of difficulty (scores ≤ 4) versus those reporting no difficulties (score of 5) in HRQOL. Area Under the Curve (AUC) results range from 0.5 to 1. Given the exploratory nature of the analysis and the comparison between two subjective HRQOL measures, AUC values were described qualitatively rather than categorised using established diagnostic benchmarks. An exploratory factor analysis was conducted to understand the underlying structure of the EQ-5D-3L and explore construct validity in a population of people with aphasia. The number of factors retained was selected with reference to the Kaiser criterion, where factors with eigenvalues greater than one were retained. Uniqueness values ranged from 0 to 1, values of ≥ 0.6 were interpreted as high; with 40% or less of variance explained by common factors [24].

Participants were predominately male (69%), aged between 55 and 70 years, with a median two and a half years since their stroke (Table 1). The majority of the cohort (94%) had mild or moderate aphasia (Table 1). At baseline, 201 participants completed both the EQ-5D-3L and SAQOL-39g (Table 2). Participants with missing baseline data were more likely to be aged over 70 years and have a self-reported nervous system disorder. No other participant characteristics differed significantly between groups.

Participants reported minor problems with quality of life on the EQ-5D-3L utility scores (mean 0.79, SD 0.20), with just over a quarter reporting perfect health at baseline (utility value = 1, 26.4%, n = 53). SAQOL-39g scores (range: 3.0-4.09) indicated participants experienced minor problems across all domains of HRQOL, however, no participants reported ‘no trouble at all’ across all three domains at baseline (Table 2). Ceiling effects were observed on the EQ-5D-3L: At baseline, more than 50% of responders reported ‘no problems’ across all domains of HRQOL except the usual activities domain (45%; Table 2). This pattern was consistent at 12-week follow up, although the proportion of responders indicating ‘no problems’ on usual activities rose to 50%. On the SAQOL-39g, most responses were spread between the three middle categories (a little trouble, some trouble, a lot of trouble), with ‘no trouble at all’ being the least selected response category at baseline (7%) and 12 weeks (10%). Scatterplots demonstrated a weak positive association between the EQ-5D-3L and SAQOL-39g domain scores, with broadly dispersed data at both baseline and 12-week follow up. This pattern indicates a weak relationship between the two measures at both time points (Fig. 2).

The communication domain of the SAQOL-39g demonstrated a consistently weak relationship with domain, index, and VAS scores on the EQ-5D-3L at both time points. The SAQOL-39g physical domain correlated moderately, except for the anxiety/depression domain which was weak. The SAQOL-39g psychosocial domain correlated weakly to the EQ-5D-3L domains, other than anxiety and depression and the utility values, which were moderate. The SAQOL-39g total mean score was moderately correlated to the EQ-5D-3L at baseline (r = 0.63) and 12-week follow up (r = 0.60). Overall, correlations were not found to be strong enough to enable the values on one measure to predict the values on the other (Table 3).

ROC curves were used to explore how well the EQ-5D-3L utility values and VAS could classify people with aphasia according to their self-rated HRQOL on the SAQOL-39g (Fig. 3). The sizes of the AUC were calculated as exploratory analyses. Results suggested the EQ-5D-3L demonstrated reasonable classification performance when distinguishing participants with lower HRQOL (AUC = 0.80 at baseline, 0.78 at 12 week follow up), as defined by total mean scores of four or below on the SAQOL-39g. The VAS showed moderate classification performance at baseline (0.78), and more limited performance at 12-week follow up (0.69; Supplementary Fig. S1).

Factor analysis identified two distinct factors with eigenvalues above 1 at both time points. At baseline, factor 1 explained 45% of variance (eigenvalue 3.16), and factor 2 explained 19% of variance (eigenvalue 1.33). Factor 1 was characterised by loadings from the SAQOL-39g physical domain in addition to the EQ-5D-3L mobility, self-care, and usual activities domains. Factor 2 included the SAQOL-39g psychosocial and communication domains, and the EQ-5D-3L pain/discomfort and anxiety/depression domains (Table 4). At 12 weeks, factor 1 explained 27% of variance (eigenvalue 1.89), and factor 2 explained 20% (eigenvalue 1.39). The factor structure was relatively stable at the two time points. Collectively, the instruments demonstrated some overlapping constructs (physical functioning and psychosocial well-being), confirming the construct validity of both measures. However, each instrument captured a unique domain of HRQOL: the EQ-5D-3L uniquely measured pain and discomfort (uniqueness 0.66), while the SAQOL-39g uniquely measured communication (uniqueness 0.73). These findings indicate that communication-related HRQOL, as captured by the SAQOL-39g, would not be adequately assessed using the EQ-5D-3L alone in people with aphasia.

This study provides novel evidence comparing the structure and performance between a generic and an aphasia-specific HRQOL instrument. The standardised collection of data within a large-scale randomised control trial is a further strength. Although these findings are important for guiding the future direction for economic evaluations of aphasia treatment, it was a preliminary exploration of an established dataset. Therefore, patient perceptions of EQ-5D-3L acceptability and comprehensibility were not investigated. The study population comprised primarily participants with mild to moderate chronic aphasia after stroke, which may present limitations in generalising our findings to people with more severe aphasia Additional analyses comparing responses between participants with mild and severe aphasia were not possible with the small sample of people with severe aphasia (n = 6).The algorithm used to convert EQ-5D-3L domain scores to utility values was validated in an Australian-representative sample. Algorithms converting EQ-5D scores to utility values have significant influence on whether the resulting utility values are meaningful. Research has demonstrated substantial cross-country variation in EQ-5D-3L utility values, depending on the valuation method used. Therefore, while our use of an Australian algorithm was appropriate for this study population, the resulting utility values may not be generalisable to other countries with different health state preferences [49]. Results from patients with aphasia after an acute stroke, or converted using a different valuation algorithm, may have also produced different results.