Examining the psychometric properties of a split version of the EQ-5D-5L anxiety/depression dimension in patients with anxiety and/or depression

This study used a cross-sectional study design to recruit participants from Amanuel Mental Specialised Hospital (AMSH), the only specialised psychiatric hospital serving people from all regions of Ethiopia. Respondents with anxiety and/or depression who visited the hospital for outpatient services were eligible to participate in this study. After completing their outpatient visit, research nurses who had access to patients' records (patient case notes) invited eligible patients to an interview, which was done in a separate room where clinicians or other hospital staff were not involved. The sample size for the current study was determined using a single proportion formula [12]. A proportional number of participants (around one-third of each) with anxiety (149), depression (159) and with both anxiety and depression (154) were recruited on a consecutive basis until the target sample was reached.

Five research nurses collected data between June and October 2021. The interviewer-administered approach was used to collect data, given that self-report is associated with some challenges due to low literacy levels among the Ethiopian population. The Amharic version of the EQ-5D-5L was used, which was translated as part of the Ethiopian valuation study using the standardized approach recommended by the EuroQol group [13, 14]. The EQ-5D-5L has been previously used in other studies in Ethiopia [15‐17]. Validated and commonly used tools to measure anxiety and depression, the GAD-7 and the PHQ-9 scales, were used to assess the severity of anxiety and depression. The severity of depression was classified based on PHQ-9 scores (1 to 4 points = minimal, 5 to 9 points = mild, 10 to 14 points = moderate, 15 to 19 points = moderately severe and 20 to 27 points = severe) [11]. The GAD-7 scores calculated from seven questions were used to classify the level of anxiety (0 to 4 minimal, 5 to 9 mild, 10 to 14 moderate and 15 to 21 severe) [10]. During the data collection, respondents were asked to complete the EQ-5D-5L with amended split version of the anxiety and depression subdimensions in a random order after the main descriptive system, followed by the PHQ-9 and the GAD-7. Two survey versions were used: one with standard EQ-5D-5L questions followed by split anxiety and depression subdimensions in sequential order and a second version where the split anxiety and depression subdimensions were placed between P/D and A/D questions of the standard EQ-5D-5L. After completing the EQ-5D-5L questions, participants were asked to select whether they preferred the composite EQ-5D-5L A/D question or the split questions. They were then asked an open-ended question to describe the reasons for their preferences. The data collection tool also included questions related to respondents’ socio-demographic characteristics and clinical characteristics. The type of diagnosis (including types of anxiety and depression) was based on the medical diagnosis by clinicians (psychiatrists) at the AMSH, while the severity of anxiety or depression was based on the self-report from the respondents. In all questions, except the clinical characteristics, Amharic, the official working language of the country, was used to ensure comprehensibility of the questions by the patients.

The data were entered into Excel and then imported into STATA version 14, where most statistical analyses were carried out. The informativity analysis were done in Excel. A sankey diagram was generated using R-Studio to show the relationship between EQ-5D-5L response levels and the severity of anxiety or depression. Descriptive statistics were used to describe the study population by demographic and clinical characteristics, including frequency and percentage. The proportions of ‘no problems’ reports were compared using the chi-square test. We have used the more common convention of a p value of less than 0.05 to denote statistical significance. Agreement between rating for composite and split dimension was compared using percent agreement and Cohen’s Kappa (k) statistics, with scores interpreted as k: ≤ 0 = poor, 0.01–0.20 = slight, 0.21–0.40 = fair, 0.41–0.60 = moderate, 0.61–0.80 = substantial and 0.81–1 = almost perfect [18]. Convergent validity was assessed by examining the correlation (Spearman’s rank correlation coefficients) between all EQ-5D-5L dimensions, the split dimensions and measures of anxiety severity and depression severity (i.e. GAD-7 and PHQ-9). Correlation coefficient was interpreted as weak (≤ 0.4), moderate (0.4–0.7) and strong (> 0.7) [19]. To examine the known-group validity, one-way ANOVA was used to compare the mean scores of the EQ-5D-5L A/D and the split dimensions. Scheffe’s post hoc comparisons and effect size (Cohen's d statistic) were used to show the difference in mean values between all possible pairs of groups. Effect sizes of less than 0.2 are considered as small, around 0.5 as moderate and greater than 0.8 as large [20]. The Shannon index (H') and Shannon Evenness index (J’) were used to demonstrate the power of informativity of the composite and split dimensions. The H’ value represents absolute information captured by possible categories.

The Shannon index (Hʹ) was defined by

With pi being the proportion of respondents with one specific health profile (e.g. 11,223), and c being the total number of theoretically possible health profiles (i.e. 3125 for the EQ-5D-5L, and 15,625 for the EQ-4D-5L + anxiety + depression). If a profile is not observed, its contribution is zero to this summation. For example, when comparing the EQ-5D-5L A/D composite (five dimensions) version to the split EQ-6D-5L version (6 dimensions), we expect the H ’ of the split to be higher to reflect more discriminatory performance. The higher the H’ value, the more information is captured by the system, and maximum H' (H'max) is observed in homogenous distribution. The J’ value represents the extent to which information is evenly distributed (relative informativity) across the categories and estimated by dividing H’ by H’max [21]. The free text describing the reasons for patient preferences towards composite versus split questions was analysed using content analysis, which is a qualitative data analysis method that is used to identify common patterns and themes within the data [22]. One author (YB) read and re-read the text comments, generated codes, and organised those into categories.

Most participants reported that they had no problems on MO (78.8%), SC (73.6%) and UA (57.4%). Less than half of the respondents reported they had no problems on the composite P/D (43.9%) or A/D (30.5%). The distribution of responses across the EQ-5D-5L and the split anxiety and depression subdimensions is presented in Appendix 2 in the Supplementary File. In the split version, 26% and 23.8% reported no problems on the anxiety and depression subdimensions, respectively, while 13.2% reported no problems on both the anxiety and depression subdimensions. The proportion of no problem reports on the composite and split questions was significantly different [χ2(1, N = 924) = 40.55, p < 0.001]. Cross-tabulation analyses by clinical diagnoses in Table 2 showed that the proportion reporting no problems on the composite A/D dimension was 30.9% in patients with anxiety, 52.2% in patients with depression and 7.8% in patients with both anxiety and depression. Conversely, the proportion of no problems on both anxiety and depression subdimension was 12.8% in respondents with anxiety, 24.5% in respondents with depression and 1.9% in respondents with both anxiety and depression. These proportions were significantly different for respondents with anxiety [χ2(1, N = 298) = 14.34, p < 0.001], depression [χ2(1, N = 318) = 25.75, p < 0.001] or for respondents with both anxiety and depression [χ2(1, N = 308) = 5.68, p = 0.017]. The details of the chi-square test are presented in Appendix 3 in the Supplementary File.

In comparison to the anxiety subdimension (r = 0.58), the depression subdimension had a higher correlation(r = 0.65) with the EQ-5D-5L A/D, although the anxiety and depression subdimensions had lower correlations (r = 0.48) with each other. Furthermore, the depression subdimension had a higher correlation with PHQ-9 (r = 0.53) and GAD-7 (r = 0.33) than the composite EQ-5D-5L A/D dimension (r = 0.36 and r = 0.28, respectively), indicating minimally better convergent validity (Table 3).

The Sankey diagram in Fig. 1 shows that a greater proportion of respondents who indicated moderate anxiety or depression in symptom scores reported ‘slight problems’ on the composite and split subdimensions. In contrast, many respondents who reported mild anxiety or depression based on symptom scores selected ‘no problems’ and ‘slight problems’ on the EQ-5D’s composite and split dimensions. To further examine this relationship, a one-way ANOVA compared the mean scores by symptom severity of anxiety and depression. Composite EQ-5D-5L A/D and split subdimensions distinguished participants based on the diagnoses and illness severity (i.e. clinical diagnosis and self-reported symptom scores) with significant differences in the mean score (p < 0.001). Scheffe's post hoc analyses revealed that both split subdimensions and composite dimension significantly discriminated groups by anxiety severity (p < 0.001); however, no significant difference was reported between respondents with moderate and severe anxiety. Furthermore, the largest effect size (0.81) was reported for the depression subdimension, followed by the anxiety subdimension (0.72) and the EQ-5D-5L A/D dimension (0.71). Furthermore, the depression subdimension showed a large effect size (1.63) between depression severity groups (severe vs mild depression), followed by the EQ-5D-5L A/D (0.97) and anxiety subdimensions (0.78). Overall, both the split and composite dimensions demonstrated good known-group validity, with moderate to large effect sizes, especially between groups with mild and severe conditions (Table 4).

Adding an anxiety or depression subdimension to a modified EQ-5D-5L tool increased absolute and relative informativity, as evidenced by the increase in both Shannon indices. EQ-4D-5L + anxiety (H’ = 5.4; J’ = 0.47) and EQ-4D-5L + depression (H’ = 5.31; J’ = 0.46) result in slightly better informativity than EQ-5D-5L (H’ = 5.19; J’ = 0.45). Compared to standard EQ-5D-5L, adding two split dimensions instead of the composite dimension increased both Shannon indices (H’ = 6.39; J’ = 0.46) (Table 5).

The content analysis summarized why respondents preferred the split anxiety and depression subdimensions over the composite EQ-5D-5L A/D. Among the 462 individuals, 163 (35.3%) preferred the composite EQ-5D-5L A/D, whereas 299 (64.7%) preferred the split subdimensions. Respondents provided more than one justification for their preference.

Based on the pattern of occurrence, the reasons were classified into three main categories that described

Respondents who preferred the composite EQ-5D-5L A/D question stated that the question was clear (38%), easy to respond to (5%), or straightforward (3%). In addition, respondents considered the composite EQ-5D-5L A/D question to be short and better than the two subdimensions to describe (9%) or explain (3%) their health problems.

Illustrative quote

Both feelings are well represented in the combined question and are easily understandable. The question that combines anxiety and depression describes me well.

Those who preferred the two subdimensions said that they were easier to answer (5%), while 4% of them claimed better clarity and understanding as the reasons for their preference for the split questions.

Illustrative quote

"Split questions are clearer and simple; therefore, I understand them easily and they also express how I feel today."

Some respondents preferred the combined questions because of their health conditions. Participants who preferred the composite questions in this category claimed they had both conditions (anxiety and depression). The question that described both conditions in combination was easier to answer than the split subdimension (15%). This group of participants claimed to have anxiety and depression symptoms therefore they preferred the questions that blended these two feelings.

Illustrative quote.

I have both feelings, and it is better to respond in a combined format.

Those, on the other hand, who preferred the split questions felt that the two subdimensions described their health condition in a better and more accurate way (31%). Respondents stated that they felt comfortable when being asked about anxiety or depression (3%) separately. Respondents who preferred the split subdimension, generally, seemed to experience either anxiety, depression but not both, which is why they found the split versions easier to complete.

Illustrative quote

The split anxiety question tells a lot about my health and explains it very well, but depression is not my concern.

Some participants preferred the composite or split question because of cognitive and interview burden. It was difficult for respondents to differentiate between the two conditions, and they were confused and some reported feeling anxious to respond to the split questions. Respondents claimed they were worried about answering the split questions and preferred a combined question as it avoids talking. The split question posed an additional burden of thinking long and hard about it (3%). Some respondents said that the two health conditions have something in common and cannot differentiate the two feelings; therefore, they preferred the composite question, which was easy to comprehend and respond to.

Illustrative quote

When it is decomposed, it is a thought-provoking question; it takes time."