EQ-5D data are often summarised by an EQ-5D index. The distribution for the original version of the index, the EQ-5D-3L, often shows two distinct groups in patient populations. This is a result of both the distribution of ill health and how the index is constructed (Parkin et al., 2016). To date, there is little evidence available about the distribution of the EQ-5D-5L index. This project aims to:
- explore whether or not the EQ-5D-5L index distribution also demonstrates this two-group distribution
- test the extent to which clustering of EQ-5D-5L profile data drives any observed clustering of the EQ-5D-5L index, and the extent to which clusters result from the characteristics of the value sets used to create the index;
- discuss the implications of our results for statistical analysis of EQ-5D-5L index data.
Data from Cambridgeshire Community Services NHS’s electronic patient records data warehouse were analysed. EQ-5D-5L profiles before treatment were obtained for 30,284 patients across three patient groups: community rehabilitation services (N=6,919); musculoskeletal therapy services (N=19,999); and nursing services (N=3,366).
The EQ-5D-5L index is calculated using both a ‘mapped’ (crosswalk) value set (MVS) and the English value set (EVS). We examined the distribution of 1,730 of the 3,125 profiles described by the EQ-5D-5L to check for clustering of the EQ-5D-5L index. The k-means cluster method and the Calinski–Harabasz pseudo-F index stopping rule were used to search for the clusters in the index. We examined the impact on the results of using different initial values in the clustering analysis.
Clustering within the EQ-5D-5L index distribution is suggested by both clustering methods, within each of the three patient groups and all patients together. For the all patients’ data, we found two robust clusters for the MVS-based index, compared to three robust clusters for the EVS-based index. The EQ-5D-5L profile data alone do not obviously drive these index clusters.
The results highlight the importance of undertaking careful exploratory data analysis for health related quality of life measures such as the EQ-5D, to ensure that statistical testing takes account of clustering and other features of the data distribution.
To access the full paper click here.
This project was funded by a research grant from the EuroQol Research Foundation. The research was also supported by the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Health Research and Care East of England at Cambridgeshire and Peterborough NHS Foundation Trust.
For more information about OHE’s programme of EuroQol-related research, please contact Nancy Devlin.
Posted in EQ-5D and PROMs, Research | Tagged Research Papers