Title: Relative treatment effects in two dependent samples: An alternative to logrank or sign tests
Authors: Dennis Dobler - Vrije Universiteit Amsterdam (Netherlands) [presenting]
Abstract: Relative treatment effects quantify the probability that someone who received Treatment A survives longer than someone who received Treatment B. We call Treatment A superior if that probability exceeds 50\%. The underlying survival data are assumed to be independently right-censored. We will discuss different possible definitions of relative treatment effects in paired data, depending on which estimation is based on Kaplan-Meier or Aalen-Johansen estimators. Two big advantages over existing tests are that inference procedures based on relative treatment effects are still useful in the case of crossing hazards and they are based on statistics that allow for easy interpretations. A randomization method and the bootstrap are used to produce reliable inference procedures. We will apply the methods to datasets in the context of diabetes (time to blindness).