Improving the Design of EQ-5D Value Set Studies for China and Beyond

This thesis details in 6 studies studying different aspects of EQ-5D use in China. The population norm study provided the first norm based on urban Chinese self-reported health status, which not only provided insight into HRQoL variations among subgroups but also served as a reference point to quantify disease burden/intervention effect etc. The subsequent methodological studies offered suggestions for improving the design of future valuation studies that may strengthen health technology assessment and cost-utility analysis in China. Since EQ-5D is the most widely-used instrument worldwide for this purpose it is thus a good candidate for such instrument in China.
Chapter 2 describes a descriptive exercise that was performed to report EQ-5D-5L norm scores in the urban Chinese population. Additional analysis was undertaken to test whether self-reported HRQoL varied between different demographic groups. It was found that HRQoL outcomes did indeed differ over age, gender, education level, health insurance status, employment status, and the residence of origin groups. In Chapter 3, individual level inconsistency in the TTO task was related to factors varied during the interview and it was found that instead of respondents, the interviewer was most vital in reducing individual level inconsistency. The results suggested that the valuation process may have been influenced by potential interviewer effects before the implementation of the Quality Control (QC) tool, which was implemented in later EQ-5D-5L TTO research.
Commencing in Chapter 4, possible designs to be used for the EQ-5D valuation study were systematically examined and compared. First, in Chapter 4, an EQ-5D-3L saturated dataset was used as a gold standard to compare two oft-mentioned but somewhat conflicting design principles in selecting health states for direct valuation: the commonness of health states (the prevalence) versus statistical efficiency of a design. By simulating the modelling process, it was found that the principle of statistical efficiency outweighed the principle of commonness in achieving sufficient prediction accuracy for non-valued states. This result suggested that the designs for the previous valuation studies were not optimal, and that future valuation studies could use a smaller design if the statistical efficiency of that design was guaranteed.
In Chapter 4, the principle of commonness as a proxy for implausibility in health states selection was examined. In Chapter 5, it was reported how university students valued all EQ-5D-5L states and judged the implausibility of each state. The results showed that respondents lacked agreement concerning which states were implausible. As there was no universal implausible state, the mean value of a state from respondents who thought it was implausible was compared with the counterpart from respondents who thought it was plausible. The results showed that values from implausible observations were lower, but still in agreement with values from plausible observations.
Learning from design selection experience with EQ-5D-3L, in Chapter 6 the aim was to test the current EQ-VT design and to identify a possible smaller design for EQ-5D-5L valuation studies. The good performance in using an orthogonal design was confirmed again with EQ-5D-5L data, i.e. an orthogonal design with 25 states performed equally as well as the EQ-VT design with 86 states in terms of prediction accuracy for all 3,125 states. In Chapter 7, the most efficient TTO data design (the orthogonal) was tested in comparison with the standard EQ-VT design and again was found favorable.
In this thesis, attempts were made to understand the possible effects of sample and design choices in previous Chinese valuation studies. The findings of this thesis can also be generalized to other countries’ EQ-5D studies or to valuation studies employing other instruments than EQ-5D.