Assignment errors and the valuation of EQ-5D health states-do responses mean what we think they mean?

Cost-utility analysis is used within the health technology assessment processes of many countries. For these analyses, patients typically indicate the health state that they are in based on a pre-defined descriptive classification. Each health state corresponds to a utility value; these values are obtained from members of the general public who are asked what they would be willing to give up to avoid spending time in that health state. If people struggle to do this, they might imagine (and so value) different states to those they are prompted with. Here, they assign a different meaning to the prompt than intended, which we designate as an ‘assignment error’. This paper formally defines these errors for the EQ-5D-3L and explores MVH dataset used to construct the UK EQ-5D-3L tariff. We modify the regressions used to form this tariff to include potential assignment errors and find that these errors are significant predictors in the regressions. For some states, there is evidence that over half the respondents answering valuation questions may make an assignment error. As these errors will affect some states more than others, they are a potential source of bias and hence distortion in resource allocation. The size of this distortion is explored in the UK context using the regressions identified by this paper, all of which suggest that less weight be given to curing moderate illness and more weight be given to curing more severe illness.