Correspondence on "How can we estimate QALYs based on PHQ-9 scores? Equipercentile linking analysis of PHQ-9 and EQ-5D" by Furukawa et al.

Furukawa et al posed the question: how can we estimate qualityadjusted life years (QALYs) based on Patient Health Questionnaire-9 (PHQ-9) scores? They recommend equipercentile linking analysis between the depression severity PHQ-9 and preferencebased EQ5D threelevel version (EQ5D3L; UK value set), the latter used to estimate utility data for QALYs. Furukawa et al refer to the process of ‘crosswalking’, whereby the practice of fitting a statistical model to health utility data has been referred to as ‘mapping’ and 'crosswalking’. Furukawa et al reference two mappingrelated papers (their references 7 and 9); however, their analysis seems to have missed rigorous mapping methodology and previous studies which have used these mapping processes, alongside other conceptual considerations when wanting to ‘crosswalk’/‘map’ from a nonpreferencebased (often conditionspecific) measure such as the PHQ-9 to the preferencebased EQ5D3L. Clear guidance for mapping has been set out by Wailoo et al. A case for equipercentile linking for mapping has been made based on suggested limitations of the more commonly used regression methods; the case for regression is described by Alava et al. A systematic review of mapping studies published in 2019 states: ‘There were 180 papers with 233 mapping functions in total [identified]...The last 10 years has seen a substantial increase in the number of mapping studies and some evidence of advancement in methods with [...] greater reporting of predictive ability of mapping functions’. From this review, the majority of mapping functions were generated to obtain EQ5D3L/EQ5D fivelevel version (EQ5D5L)/childfriendly EQ5D version (EQ5DY) scores (n=147) among other preferencebased measure scores; eg, ShortForm SixDimension (SF6D, n=45). Furukawa et al reference one study, which was also identified by Mukuria et al, which maps from the PHQ-9 to the SF6D (not EQ5D3L), which concluded that: ‘mapping from mental health conditionspecific measures, such as the widely used PHQ-9, GAD [(Generalized Anxiety Disorder)] and HADS [(Hospital Anxiety and Depression Scale)], may not be an appropriate approach to generating EQ5D and SF6D scores as these measures focus on specific symptoms and not on the wider impact of mental health conditions’ (their reference 7). Furukawa et al is mapping and therefore existing rigorous mapping methods should be used and compared with the suggested equipercentile linking analysis. We recommend not using the suggested conversion table by Furukawa et al until further conceptual and statistical analyses have been conducted, including reporting of performance statistics to allow method performance to be judged and compared against existing mapping studies in the empirical literature. We make this recommendation on the basis that Furukawa et al currently provides no reported performance statistics or comparisons to suggest the potential predictive ability of using the conversion table; therefore there is no way to judge to what extent the conversion table could lead to biased, inaccurate, and imprecise QALY estimations which could lead to suboptimal decisionmaking.