Assessing the proxy response bias of EQ-5D-3 L in general population: a study based on a large-scale representative household health survey using propensity score matching.

ABSTRACT

BACKGROUND: Proxy respondent-someone who assists the intended respondent or responds on their behalf-are widely applied in the measurement of health-related quality of life (HRQL). However, proxies may not provide the same responses as the intended respondents, which may bias the findings. OBJECTIVES: To determine whether the use of proxies is related to socio-demographic characteristics of the intended respondent, and to assess the possible proxy response bias of Chinese version of EQ-5D-3 L in general population. METHODS: A cross-sectional study based on a provincially representative sample from 2013 National Health Service Survey (NHSS) in Shaanxi, China was performed. HRQL was measured by Chinese version of EQ-5D-3 L. Propensity score matching (PSM) was used to get matched pairs of self-reports and proxy-reports. Before and after PSM, univariate logistic and linear models including the indicator of proxy response as the only independent variable, were employed to assess the possible proxy response bias of the dimensional and overall health status of EQ-5D-3 L respectively. RESULTS: 19.9% of the responses involved a proxy. Before PSM, the proxy-report group was younger in age and reported less unhealthy lifestyle, lower prevalence of disease, and less hospitalization than the self-report group. After PSM, it showed that the proxy-report group was statistically more likely to report health problem on each dimension of EQ-5D-3 L, with odds ratios larger than one comparing with self-report group. The means of EQ-5D-3 L index and EQ VAS of proxy-report group were 0.022 and 0.834 lower than self-report group. CONCLUSIONS: Significantly negative proxy response bias was found in Chinese EQ-5D-3 L in general population, and the magnitude of the bias was larger in physical dimensions than psychological dimensions after using PSM to control confounders.