中文版EQ-5D-3L量表应用于慢性病人群测量特性的Rasch模型分析

Analysis of the Measurement Characteristics of Chinese Version of EQ-5D-3L Applied in Population with Chronic Disease:Rasch Model

目的应用Rasch项目反应理论评价中文版EQ-5D-3L量表在慢性病人群中的适用性。方法应用基于Rasch模型的RUMM2030软件,从总体拟合效果、内部一致性、标的性和单维性等方面,对量表在非慢性病和常见慢性病人群中的测量效果进行评价。结果除贫血组外,其他人群的项目残差和个体残差的均值都小于0,说明模型属于过度拟合; 15个人群中有2个拟合效能水平为"好",11个为"可接受",2个为"低",说明量表内部一致性在中等水平;个体位置均数均为正且偏度系数均为负,表明各人群都出现了程度不一的天花板效应;主成分分析的残差中约40%的信息来自HRQo L之外的其他潜在特质。结论中文版EQ-5D-3L量表在慢性病人群中的测量效果好于非慢性病人群。本研究的结果验证了EQ-5D-3L量表存在的两个不足;一是量表的单向性不足;二是项目的敏感度不够,尤其对健康状况较好者不够敏感。

Objective To evaluate the fitness of the Chinese Version of EQ-5D-3L to a Rasch item response model in populations with or without chronic diseases. Methods The Rasch model-based software named RUMM2030 was used to evaluate the validity of Chinses version of EQ-5D in non-chronic disease population and 13 types of chronic disease population. The goodness of fit, person separation index （PSI）, the targeting and the unidimensionality was assessed. Results Except for anemia, the item residual and individual residual in all the other chronic populations were small than 0, which indicated that the models were overfitted. For the level of the fitness of the 15 models, 2 of them were “Good”, 10 models were “Reasonable” and the other 2 models were “Low”, which meant that the discriminate ability of the questionnaire was too bad and not well and the internal consistency was at moderate level. All the individual locations were positive and the coefficients of skewness of them were negative, which indicated that the ceiling effects existed in all of the 15 populations. About 40% of the information in the residual of principle component analysis belonged to the latent trait which was not the latent trait of HRQoL. Conclusion The Chinese version of EQ-5D-3L was adopted well in populations with chronic diseases than those without chronic diseases. This research found two limitations of Chinese version of EQ-5D-3L. First, the questionnaire does not include enough contents of general health status because there are only five items in it and the unidimensionality of latent trait of HRQoL is not well. Second, the items in the questionnaire are not sensitive enough especially in health population.