CMStatistics 2021: Start Registration
View Submission - CMStatistics
Title: Randomizing relative treatment effects Authors:  Dennis Dobler - Vrije Universiteit Amsterdam (Netherlands) [presenting]
Abstract: The relative treatment effect $p=P(T_{11} > T_{22}) + 0.5 P(T_{11} = T_{22})$ is a single, meaningful number that quantifies whether Treatment 1 is superior to another Treatment 2 ($p > 0.5$) or not ($p <= 0.5$). The assumed data structure is that paired survival times $(T_{1i}, T_{2i}), i=1,...,n$ are available where the first member of a pair has received Treatment 1 and the second Treatment 2. Because survival times could be independently right-censored, estimation of $p$ is based on Kaplan-Meier estimators. Inference is achieved by means of a randomization procedure that randomly flips the treatment labels and thus artificially creates a situation in which $p=0.5$ holds. Central limit theorems are obtained with the help of the newly developed, more general Randomization Empirical Process theory. The methodology will be illustrated with an application to data from a study on diabetic retinopathy in which the eyes of a patient underwent different treatments to prevent blindness.