比较成对有序观测数据的一种新方法

A New Method for Testing Non-Inferiority between Two Diagnostic Procedures in Paired-Sample Ordinal Data

在生物医学统计研究中,如果诊断试验过程比传统诊断过程更加易于操作,需要的费用少,那么在采用试验过程之前,就非常有必要去评价诊断试验的诊断精度是否不低于传统诊断过程的诊断精度.考虑多于两类的有序变量诊断结果,并且两种诊断结果构成成对有序观测数据,提出了比较这种观测数据的一种新方法,给出了诊断试验的诊断精度不低于传统诊断过程诊断精度的标准及定义.基于大样本理论,构造了假设检验问题,给出了F ieller-type检验统计量.为验证样本量有限时,检验方法是否有效,进行了随机模拟,取得了满意的结果.

A new method for evaluating equivalence between two diagnostic procedures is proposed in sample ordinal data. The standard of non-inferiority is defined. On the basis of large sample theory, the paired score evaluation procedure for detecting non-inferiority is developed. Monte Carle simulation is conducted to evaluate the finite sample performance of this test procedure. The simulation results show that the proposed asymptotic test procedure performs well in most of situations.