Primary Care Research Unit (PCRU), Department of General Practice, University of Melbourne, Australia.
Health Econ. 2010 Jun;19(6):683-96. doi: 10.1002/hec.1505.
An important issue in the measurement of health status concerns the extent to which an instrument displays lack of sensitivity to changes in health status at the extremes of the distribution, known as floor and ceiling effects. Previous studies use relatively simple methods that focus on the mean of the distribution to examine these effects. The aim of this paper is to determine whether quantile regression using longitudinal data improves our understanding of the relationship between quality of life instruments. The study uses EQ-5D and SF-36 (converted to SF-6D values) instruments with both baseline and follow-up data. Relative to ordinary least least-squares (OLS), a first difference model shows much lower association between the measures, suggesting that OLS methods may lead to biased estimates of the association, due to unobservable patient characteristics. The novel finding, revealed by quantile regression, is that the strength of association between the instruments is different across different parts of the health distribution, and is dependent on whether health improves or deteriorates. The results suggest that choosing one instrument at the expense of another is difficult without good prior information surrounding the expected magnitude and direction of health improvement related to a health-care intervention.
在健康状况测量中，一个重要问题涉及到测量工具在分布极值处（称为地板效应和天花板效应）对健康状况变化的敏感度缺乏程度。先前的研究使用相对简单的方法，侧重于分布的平均值来检查这些效应。本文的目的是确定使用纵向数据的分位数回归是否可以增进我们对生活质量工具之间关系的理解。该研究使用了具有基线和随访数据的 EQ-5D 和 SF-36（转换为 SF-6D 值）工具。与普通最小二乘法（OLS）相比，一阶差分模型显示出测量之间的关联度要低得多，这表明 OLS 方法可能会由于不可观测的患者特征而导致关联的估计值存在偏差。分位数回归揭示的新发现是，这些工具之间的关联强度在健康分布的不同部分有所不同，并且取决于健康状况是改善还是恶化。结果表明，如果没有与医疗保健干预相关的健康改善幅度和方向的良好先验信息，选择一种工具而放弃另一种工具是困难的。
