Some methodological issues with the analysis of preference-based EQ-5D index score

Abstract

The EQ-5D index is widely used to assess the preference-based health status. In this paper, we examine the analytical issues of regression models for the US preferencebased EQ-5D index score. We propose a two-part approach to model the special features of the EQ-5D index. The first part is a logistic model for the probability of reaching the maximum score 1.0. The second part is a model for the rest of the scores that are less than 1.0, which can be a least squares regression with robust standard errors for the conditional mean, or a quantile regression for conditional quantiles such as the median. We show that the two-part model has some desirable features that are not available in the previously published regression methods for the EQ-5D index. We illustrate the proposed approach with data from the Medical Expenditure Panel Survey. The proposed method may be used for other utility or health related quality of life scores of similar features.

Estimating an EQ-5D Value Set for Malaysia Using Time Trade-Off and Visual Analogue Scale Methods

“…Although the N3 and D1 models using VAS valuations were virtually identical, the N3 model was marginally the better model because it had fewer large AEs (AEs Ͼ 0.1) between the predicted and actual health state values. Furthermore, the N3 VAS valuation model minimizes the gap between health state 11111 and the next best health state [13], as shown in Table 5 and Appendix 5 in Supplemental Materials found at doi:10.1016/j.jval.2011.11.024. Therefore, value sets derived from the VAS valuation would appear to perform better in the Malaysian setting.…”

Tails from the Peak District: Adjusted Limited Dependent Variable Mixture Models of EQ-5D Questionnaire Health State Utility Values

“…However, it is unclear that a two-class model is capable of overcoming this challenge unless either the underlying distribution for the classes is itself suitable for censored data or if one class has a zero variance, in which case the model is equivalent to the two-part model. Similarly, Li and Fu [11] applied a twopart model to US EQ-5D questionnaire data. Specifically, they ex- plored how the second part of the model, for those individuals who do not score 1, may be approached in a number of ways.…”

Marginal differences in health-related quality of life of diabetic patients with and without macrovascular comorbid conditions in the United States

“…The US preference-based EQ-5D index score is not normally distributed [26]-its distribution has been noted as semi-continuous with spikes at certain values. Based on the most recent recommendations [27], we employed a two-part model with the first part a logistic regression and the second part a least square regression with robust standard errors to analyze the EQ-5D index score. The distribution of EQ-VAS is somewhat similar to that of EQ-5D index but is continuous.…”