Methodological Development for Detection of Informative Censoring

 
   Lead Investigator:    Cristina Murray-Krezan
   Institution      :    University of New Mexico
   E-Mail           :    cmurray-krezan@salud.unm.edu
   Proposal ID      :    860



   Proposal Description:
   Aim: Develop and evaluate a statistical test for the presence of informative      censoring.                                                                        The primary goal of this proposed research project is to develop a statistical    test for the presence of informative censoring. In our approach, we will using    the methods of Carroll and Wu (1989), which takes advantage of intra-individual   correlations between longitudinal changes over time and later censoring status.   We will conduct simulation studies as an initial evaluation of the test, but we   also would like to evaluate the performance of the test using real data sets.     Because the test will rely on associations between longitudinal changes in        patient measures and ultimate censoring status, we therefore require data sets    which have longitudinally measured data as we look into the performance of the    test.                                                                             Our proposed work has the potential to have a large impact on the analysis of     survival data. Current survival analyses rely on the assumption that the          censoring process occurs at random. Further, deviations from this assumption      cannot be statistically evaluated in simple ways, and no approach to identify     the presence of informative censoring has yet been established. We expect to      overcome the challenges with detecting non-random censoring by relying on         subject characteristics collected longitudinally over time, and applying          approaches from the linear models arena to determine whether longitudinal         changes in patient characteristics can successfully differentiate between those   who dropped out of the study prematurely from those who completed the study.      Should we succeed in developing such a test, which we fully expect, it will       provide an important tool for researchers performing time-to-event studies        using longitudinal data, and mak