Treatment Comparison with Survival and Non-survival Primary Endpoints
Embargo until
2015-05-01
Date
2014-04-25
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Johns Hopkins University
Abstract
Multiple biomarkers (surrogate endpoints) are often used to predict the failure event, as well as to evaluate the effect of treatment. In biomarker researches, Behrens-Fisher problem is a nonparametric two sample comparison problem, which is vital due to its generality. In this thesis, we proposed a new testing method for Behrens-Fisher problem. Our method dealt with multiple primary endpoints and focused on the global effect of a treatment rather than effect for every single endpoints, which separated us from most of the current studies. We reviewed existing methods dealing with multiple endpoints Behrens-Fisher Problem with complete data, and introduced basic characteristics and testing methods for survival data. In light of the limited ability to process censoring data with current methods, we created this new method combining information of non-survival markers and survival time to improve accuracy of the evaluation. Inspired by the idea of rank sum test or U statistics, we built an adjusted U-statistic to perform the hypothesis test for general two sample comparison problems. This test offered a reasonable approach to evaluate treatment effect and inform clinical decision-making when the length of follow-up is available and the importance of the primary endpoints could be of equal-weight. In addition, the result of simulation indicated that our new test would have satisfactory performance under different real-world scenarios.
Description
Keywords
Multiple Endpoints, Survival Analysis, Behrens-Fisher Problem, Global Treatment Effect, U-statistics.