What Determines the Shape of an EQ-5D Index Distribution?

In a new OHE Research Paper, OHE’s Nancy Devlin and Yan Feng join David Parkin from King’s College London in analysing characteristics of the EQ-5D indexing process that may obscure useful, and important, information about health states. Because the EQ-5D is one of the most used health-related quality of life instruments worldwide, it is crucial that the characteristics of the data be well understood.

In a new OHE Research Paper, OHE’s Nancy Devlin and Yan Feng join David Parkin from King’s College London in analysing characteristics of the EQ-5D indexing process that may obscure useful, and important, information about health states.

Because the EQ-5D is one of the most used health-related quality of life instruments worldwide, it is crucial that the characteristics of the data be well understood. The EQ-5D is used to ask patients or others to self-report their health in terms of five dimensions and three levels of problems (‘no’, ‘some’ or ‘extreme’). That information from patients is then summarised by a single number, anchored at 0 (dead) and 1 (full health) to show how ‘good’ or ‘bad’ each state is. The authors of this paper note that distributions of EQ-5D index values in patient and general populations have an interesting characteristic: they typically divide into two distinct groups. This might reflect the actual distribution of ill health, but it also might be an artefact of how the EQ-5D index is constructed.

In this OHE Research paper, the authors examine the determinants of the shape of EQ-5D distributions, particularly the origins of the ‘two groups’ distribution. They analyse data from elective English NHS surgery patients (hip and knee replacements, and varicose vein and hernia repairs) and a study of primary care patients with chronic disease (asthma and angina). The distributions of EQ-5D index values are compared to distributions based on data that have not been weighted; with the distributions that arise when different countries’ weights are used to summarise patients’ data; and with condition-specific indexes for the same patients.

The research shows that a very small number of profiles account for most of the observed data: just twelve EQ-5D health state profiles account for 86% of all the health states reported by patients. The explanation for the ‘two groups’ characteristic seems to lie both in the grouping of profiles and in the nature of the weights applied to them. Examining the group with higher index values, the most commonly observed health states all had ‘some’ problems with mobility, usual activities and pain, and had either ‘no’ or ‘some’ problems with self care and anxiety/depression. In the group with lower index values, the most commonly observed health states all had ‘extreme’ problems with pain/discomfort and some also had the worst level of usual activities. The weights commonly used to calculate the index place more weight on level 3 health problems, creating a noticeable gap in index values across the two groups.

The authors also show that the ‘two-groups’ characteristic of EQ-5D index distributions is not uniquely associated with the use of the UK value set. Applying different value sets produces different distributions, but the two-group distribution arises in most cases. However, the difference between the distributions in each case serves as a reminder that, for any given set of patients’ EQ-5D data, which value set is used to summarise them will have an important bearing on the results. The authors point out that concentrating on the EQ-5D index may obscure useful information about health states and possibly produce misleading information. They emphasise the importance of exploratory analysis of EQ-5D data — both to improve analyses of EQ-5D data for comparison and inference purposes, and to help develop more accurate mapping across different health measures. Any health status index based on weighted profiles, the authors note, will benefit from this analytical approach.

Download Parkin, D., Devlin, N. and Feng, Y., 2014. What determines the shape of an EQ-5D distribution? Research Paper 14/04. London: Office of Health Economics.

For further information, please contact Professor Nancy Devlin. For an overview of OHE’s extensive activities in patient-reported outcomes measures, please click here.