基于对应分析的分类变量一致性评价量化指标的构建

The Structure of Quantitative Index for Categorical Variables' Agreement Evaluation based on Correspondence Analysis

目的将对应分析(correspondence analysis,CA)应用于临床诊断试验的一致性评价,构建基于CA的一致性评价量化指标,提高其在一致性评价应用中的客观性。方法根据CA的具体算法构造出距离矩阵,参考kappa系数的原理,利用SAS9.1构造一致性评价量化指标(分别简称CBD和CCBD);通过不同情形下的模拟数据和实例对量化指标进行可靠性和实用性考核。结果两个待评价方法时,CBD和kappa均与符合率存在较高的正相关;多个待评价方法时,CBD或CCBD分别与kappa或fleiss kappa成正相关。对于存在kappa悖论的资料,CBD表现出相对于kappa较高的稳定性。此外,CBD或CCBD与其相应的对应分析图得出的结论亦吻合。结论基于CA所构建的一致性评价量化指标是客观可靠的,对于临床科研中的一致性评价具有重要的理论和实际意义。

Objective Correspondence analysis（ CA） can be applied to the consistency evaluation of clinical diagnosis trials. It is expected to build quantitative indexes for agreement evaluation system with correspondence analysis to improve its objectivity. Methods First,the distance matrix is constructed according to the specific algorithm of CA. Then,referring to the principle of kappa,the quantitative indexes（ CBD and CCBD） were built. The reliability and utility of the quantitative indexes were tested with the simulative data under different situation and some examples. Results For two evaluated methods,positive correlations exist among kappa,CBD and concordance rate. In the case with more than 3 evaluated methods,CBD and CCBD are proportional to kappa and fleiss kappa coefficient respectively. As for the data in which kappa paradox exists,the stability of CBD is better compared with kappa. In addition,the conclusions drawn by the correspondence analysis diagrams are the same as that of CBD or CCBD. Conclusion The quantitative indexes of consistency evaluation built by CA is objective and reliable,and it is of important significance both in theory and in practice to the consistency evaluation in clinical research.