EQ-5D Index Clustering: Causes and Implications for Health Care

In a paper presented at the January 2013 Health Economics Study Group meeting, David Parkin of the NHS and OHE’s Nancy Devlin and Yan Feng examined what may cause the two-cluster shape of EQ-5D indexes. This is important because these indexes are used widely in other analyses that affect health care resource decisions.

In a paper presented at the January 2013 Health Economics Study Group meeting, David Parkin of the NHS and OHE’s Nancy Devlin and Yan Feng examined what may cause the two-cluster shape of EQ-5D indexes. This is important because these indexes are used widely in other analyses that affect health care resource decisions.

The EQ-5D is used widely in economic analyses, population health surveys and, more recently, for routine assessment of patients’ health — for example, in the NHS Patient Reported Outcome Measures (PROMs) programme. The EQ-5D instrument comprises two elements that ask the patient to report health. The first is the EQ-5D self-classifier, where respondents create a “health profile” by ticking boxes to indicate which of three levels of problems (none, some or extreme) they have on each of five dimensions (mobility, self care, usual activities, pain and discomfort, and anxiety and depression). The second is the EQ-VAS, where respondents rate their overall health on a visual analogue scale from 0 (worst health imaginable) to 100 (best health imaginable). Both provide valuable data and both are used in evaluations of health services and decisions about health care allocation.

The most common way of analysing data from the EQ-5D is to use index values that summarise the profile data. Such indexes are used widely, serving as the bases for a range of other important statistical analyses and modelling. Their reliability and appropriate use, then, both are crucial.

The focus of this paper is on the tendency of EQ-5D index values to cluster into two groups. It has not been clear whether this phenomenon in fact represents two distinct patient groups or is an artefact of the methods used to weight the profile data to produce indexes. To gain insight into this issue, the authors analysed data from the English NHS PROMs programme (hip and knee replacements, and varicose vein and hernia repairs) and from a study of two chronic conditions (asthma and angina). The distribution of EQ-5D index values was compared with distributions from data that are not weighted; the ungrouped profile data were used to determine whether and how their characteristics might affect the distribution. The EQ-5D index values also were compared to distributions both from condition-specific indexes for the same kinds of patients and from the EQ-VAS.

The authors conclude that the underlying profile data and the weighting system each affect the observed two-cluster distribution. The EQ-5D classification system captures differences between patients with the same condition on dimensions that are observed mainly at level 2 or 3. The weights commonly used to calculate the index exacerbate this by placing greater weight on level 3 observations, creating a noticeable gap in index values between the groups.

As the authors note, the analyses in the paper highlight the important point that any index “in effect obscures useful information about health states and may even produce misleading information”. They recommend that all the underlying data for all health indexes, not just the EQ-5D, be well understood and that health indexes be used with care.

Parkin, D., Devlin, N. and Feng, Y. (2013) What determines the shape of an EQ-5D index distribution? Paper presented at the Health Economists Study Group meeting. Exeter, UK. 9-11 January 2013.

For additional information, please contact Nancy Devlin.

For an overview of OHE’s extensive work on patient-reported outcomes measures, including the EQ-5D and the EQ-VAS, click here.