Current topics on treatment comparison for survival endpoints with non-proportional hazards
Description
TitleCurrent topics on treatment comparison for survival endpoints with non-proportional hazards
Date Created2021
Other Date2021-10 (degree)
Extent1 online resource (xi, 93 pages) : illustrations
DescriptionThe proportional hazards (PH) assumption is the key assumption which may need to be examined in each survival analysis. It is the underlying assumption for the Cox proportional hazards and the log-rank test achieves its maximum efficiency when the PH assumption is satisfied. However, when the PH assumption is violated, we may not be able to provide a valid clinical interpretation for the hazard ratio of treatment effect, and the log-rank test may no longer achieve its desired level of power. Many alternative tests and measures have been developed for the non-proportional hazards situation. Four of them, Weighted log-rank test, Max-Combo Test (MX), Maximin Efficiency Robust Test (MERT) and The Restricted Mean Survival Time (RMST), are studied in this dissertation. Since previous papers only introduced MERT results by using uncensored case correlation matrix, MERT result with real censored correlation is discussed in this dissertation. The performance of the four non-proportional approaches, namely, Weighted log-rank test, MX, MERT, and RMST, is evaluated under the null and six typical proportional/non-proportional hazards conditions. The strength and weakness for each approach is explored in this dissertation. In addition, this proposal compares the test performance between Fleming-Harrington family G0,0, G1,0, G0,1, G1,1 and family G0,0, G2,0, G0,2, G2,2 for Weighted log-rank test, MX and MERT. The best non-proportional hazards test among those four tests and two families is explored in this dissertation. In clinical trials, two stage design is commonly used to dismiss ineffective treatment early at the end of the first stage, in order to avoid the high cost and long trial duration. Conditional power is the conditional probability of a significant result at the end of the trial given the data observed thus far. It is commonly used to evaluate the possibility of stopping the trial for futility. It also can be used to calculate the new sample size in the sample size re-estimation at the interim stage. The RMST is a robust measure of the survival time distribution. It provide a clinically interpretable summary and is widely used in the non-proportional hazards situation, since it does not rely on the PH assumption. This dissertation describes how to calculate the CP for the RMST endpoint at the interim stage for non-proportional hazards models. An approach for predicting survival curve at the final stage is developed in order to calculate the correlation in test statistics between the interim and final stages.
NotePh.D.
NoteIncludes bibliographical references
Genretheses
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.