Background Some care home residents are unable to complete the measures of health utility needed for cost-utility analyses. Proxy measures are an alternative source of data; however, the level of agreement in the EQ-5D between residents and care home staff and the impact of clustering on this agreement has not been investigated previously. Using data nested in a large clinical trial, we aimed to (1) identify inter-rater agreement levels for the reporting of EQ-5D by care home residents and carers, (2) investigate whether proxies have a greater tendency to choose ‘some problems’ at the domain level and (3) investigate factors influencing agreement.
Methods Kappa statistics were used to examine the level of agreement between proxy and self-completed responses at the domain level and intraclass correlation coefficients (ICCs) were used for EQ-5D index scores, visual analogue scales and total quality-adjusted life-years (QALYs). Importantly, the calculation of the Kappa and ICC relies on the assumption of independence of the observations, so the impact that clustering might have upon our results was also examined. The percentage of residents and proxies choosing each level of each domain was compared to examine whether responders have a consistent preference for one of the levels. Finally, a mixed model was used to consider the extent to which the difference in EQ-5D-S and EQ-5D-P was affected by other variables at baseline.
Results We found slight to fair agreement at the domain level (cluster-adjusted Kappa 0.02–0.30) and fair agreement at the score level (cluster-adjusted ICC 0.30–0.32) and for QALYs (cluster-adjusted ICC 0.38). Proxies had a greater tendency to use the ‘some problems’ compared to self-report. The random effect mixed model showed that a higher GDS-15 (and likelihood of depression) and lower MMSE score (and lower cognitive impairment) were both associated with greater discrepancy between proxy and self-completed scores.
Conclusion Proxies appear to be a reliable source of data for QALY scores but may be less reliable if individual domains are considered. As expected, the presence of more severe depression or cognitive impairment in residents reduced their agreement with proxy scores. The frequent reporting of ‘some problems’ by proxies suggests that a 4-level measure would counter this tendency to choose the ‘middle’ option.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.