|
Abstract:
|
In medical research we often need to compare the reliabilities of two measurement techniques by which physicians examine the patient's disease status. It is straightforward to calculate a reliability (or agreement) score such as intraclass correlation coefficient (ICC) for each technique. However, the statistical inference for the difference of the ICCs related to the two techniques has not been fully developed. In JSM 2015, we presented a likelihood ratio test based on a linear mixed-effects model for testing the difference in the ICC between two measurement techniques (Abstract # 317064). In this presentation we propose three bootstrap methods for estimating a confidence interval (CI) for the ICCs' difference: non-parametric basic method, bias corrected and accelerated method, and parametric method that resample both random effects and residuals. We will examine and compare the properties of the CIs through the simulations. A study for immunohistochemical evaluation of MYC protein expression in diffuse large B-cell lymphoma (DLBCL) patients will be illustrated as an example of real data.
|
