Many international health technology assessment organisations, such as the National Institute of Health and Care Excellence (NICE) in England [
1], use economic evaluation, the comparative assessment of the costs and benefits of alternative interventions, to support resource allocation decisions in healthcare. The need for consistency and comparability in their recommendation decisions has resulted in the increased use of quality adjusted life years (QALYs) which combine length of life with health-related quality of life (HRQoL), measured using utilities, into a single metric. Utilities are generated using generic preference-based measures, which are applicable to any disease area, in contrast to condition-specific measures that have limited generalisability, and therefore risk commissioning decisions across disease areas being inconsistent [
2]. The EQ-5D-3L [
3] is one of the most widely used and preferred generic preference-based measures [
4] and is currently NICE’s recommended measure for economic evaluation [
1].
Generic preference-based measures are not always included in trials as condition-specific measures may be considered more informative. Reducing the number of measures also minimises the additional burden on patients. In mental health, there are a number of condition-specific HRQoL measures for common conditions such as depression and anxiety e.g., the Patient Health Questionnaire-9 (PHQ-9) [
5] and the Generalised Anxiety Disorder-7 (GAD-7) [
6]. These measures assess a patient’s mental health but are not designed to inform QALY estimation. Mapping between the condition-specific measure and a generic preference-based measure using regression analysis is one method for indirectly obtaining utilities. The mapping regression results can then be applied to other trials and settings where the preference-based measures are missing. NICE recommend that EQ-5D can be estimated from another measure using statistical mapping when EQ-5D is appropriate but not available in the relevant study [
1].
A recent review [
7] found only a limited number of published mapping studies using mental health measures, including the PHQ-9 and GAD-7 [
8]. Most of the mapping studies in the review, including those looking at mental health measures, used ordinary least squares (OLS) regression as the regression approach [
7]. There are limitations with using OLS as utilities are bounded, errors are not normally distributed, and measures such as EQ-5D-3L have trimodal distributions [
9]. Alternative and more flexible approaches have been developed to address these concerns [
10]. A more recent study employed equipercentile linking analysis to map from PHQ-9 to EQ-5D-3L [
11] but the study was criticised for not following the most recent guidelines on mapping [
12]. A different study provided mapping from the PHQ-9 and the GAD-7 to the EQ-5D-5L United States (US) utilities and the mapped EQ-5D-3L UK utilities using more appropriate mapping approaches [
13]. EQ-5D-5L is a newer version of the EQ-5D [
14] but the three-level version has been used in older trials and observational data where data may be drawn from to inform models. The EQ-5D-3L also continues to be recommended for use by NICE [
1]. The utilities generated from the two EQ-5D versions are not equivalent therefore where analysts want to generate utilities from the PHQ-9 and the GAD-7 that are comparable to the EQ-5D-3L, an appropriate mapping algorithm is required. The objective of this study was therefore to address this gap in the literature by generating mapping functions between two commonly used measures of mental health, the PHQ-9 and GAD-7, and the EQ-5D-3L.