Our aim was to explore the responsiveness of the EQ-5D in patients receiving treatment for depression and anxiety. This was done by comparing change in the EQ-5D to change in the disorder-specific measures BDI-II and BAI. We hypothesised that the EQ-5D should show magnitude of change as the BDI-II and BAI during treatment. The ES was large (d > .80) for all measures, ranging from Cohen’s d 1.07–1.84. For the SRM, which accounts for variability in treatment response by dividing change scores by the standard deviation of change scores, the BDI-II, the EQ-5D value and the EQ VAS all showed large magnitude of change. The BAI showed moderate magnitude of change on the SRM when accounting for its higher correlation between baseline and end of treatment scores. Furthermore, the EQ-5D value change scores showed moderate correlation with the BDI-II and the BAI change scores. The hypothesis that the EQ-5D should show similar magnitude of change as the condition-specific measures during treatment thus seems confirmed.
We then examined if the EQ-5D utility could correctly classify patients deemed as “Recovered” according to the condition-specific measures. Results from the ROC analyses indicate that this was the case: AUROC were from fair to good on all analyses when distinguishing “Recovered” patients from “Improved” or “Unchanged”. This was true for the total sample (AUROC .72 – .82), for patients with depression (AUROC .75 and .80), and for patients with anxiety (AUROC .83 and .84). In a similarly consistent pattern, the EQ-5D showed poor ability to distinguish between “Improved” and “Unchanged” patients for the total sample, for depression, and for anxiety, (AUROC .52 – .61). The ability of the EQ-5D to consistently identify recovered patients indicates that our second hypothesis was confirmed. We also calculated Youden’s index, as this may be informative for clinicians and serve as a reference for future research. For recovery from both depression and anxiety, cut-off point as defined by highest combined sensitivity and specificity was an EQ-5D utility ≥ .768 at end of treatment.
Data on the responsiveness of the five-level version of the EQ-5D in mental health is limited, though cross-sectional measures have indicated moderate to good correlation with condition-specific measures [10]. Comparing to the three-level version, one study found moderate responsiveness to anxiety disorders. Similar to the present study, patients were classified as having either “more”, “constant”, or “less anxiety” according to the BAI. T-tests showed significant differences in change scores for the EQ-5D value and the EQ VAS. However, that study found that the SRM were moderate to small, and ES were large for the EQ-5D value only when patients were deteriorated [17].
Reviews of the literature on the three-level version have indicated reasonable responsiveness in depression and anxiety [9], suggesting that the five-level may have similar properties. One recent study compared the responsiveness of the three-level and five-level versions of Anxiety/depression dimension for mental health patients. Although the five-level version was found to be more responsive, both showed limited ability to capture changes in mental health [18]. The Anxiety/depression dimension did show significant change from baseline to end of treatment in the present study. Future research may determine if it is useful as a standalone tool.
A previous cross-sectional study did find that the EQ-5D value could screen for depression and anxiety in patients with type 2 diabetes [19]. In the present study, the EQ-5D value showed similar performance in a longitudinal design in patients with depression and anxiety as primary diagnoses. That the EQ-5D value may perform better than the Anxiety / depression dimension alone is perhaps reasonable, as it may better capture the wide-ranging impact of depression and anxiety on health and quality of life [4, 5].
The EQ-5D is increasingly used when evaluating health status in surveys and clinical trials [6], and decision-making bodies recommend its use in evaluating health technologies [7, 44]. Demonstrating its validity in diverse patient groups is therefore essential for assisting sound decision-making when allocating health care resources. In this study, the EQ-5D showed good responsiveness to change for patients with depression and anxiety. This suggests that the EQ-5D can be a valid and useful tool for evaluating impact of disease and benefit of treatment for these patients, for instance through estimating QALYs. It also suggests that the EQ-5D can be useful in clinical setting, such as evaluating treatment outcomes.
Strengths and limitations
The main strength of the study is adding to a limited evidence-base concerning the responsiveness of the five-level version of the EQ-5D in patients with depression and anxiety. The study included a fairly large clinical sample who were assessed and diagnosed by clinical psychologists before entering treatment. We can thus be reasonably certain of the clinical characteristics of the sample. The study took part in a national health service clinic, suggesting that these patients are somewhat representative of clinical populations with depression and anxiety in Norway. The patients saw substantial treatment gains as reflected by the large ES and SRM, which gave an opportunity for evaluating the ability of the EQ-5D to identify recovered patients.
Several limitations to the study have to be considered. The study only included patients who completed treatment, and treatment gains were large. The study could therefore not evaluate the ability of the EQ-5D to detect smaller changes, that still may be of importance to patients. A related limitation is that the large rate of recovered patients in the study meant that “Unchanged” patients formed a small subgroup. The findings concerning the unchanged patients should be treated with caution. We also lack adequate data to determine if the EQ-5D would be equally responsive to deterioration as improvement during treatment. The study also lacked data on comorbidity.
The current study uses the UK value set for converting to EQ-5D value scores, as there is currently no Norwegian value set available. Choice of value sets has shown to influence the estimation of QALYs, which suggests that it would be useful to replicate the present findings when a Norwegian value set is available [13].
As new measures of health status become available, such as the Recovering Quality of Life (ReQoL) or capability-oriented instruments, it will be important to compare and contrast these against the EQ-5D to judge which instrument is best suited for patients with depression and anxiety [45, 46]. There is evidence that a wide range of outcomes that are important to patients with mental health problems are not adequately captured by commonly used instruments [5, 8]. Further research is needed to assess whether the EQ-5D could reflect key changes in a wider range of outcomes [5], or if other instruments or bolt-on dimensions may be better for capturing psycho-social factors of importance to patients [47].