Title: Survival analysis in two-stage randomized clinical designs using mixture distributions
Authors: Giovanna Ranzato - University of Padova (Italy) [presenting]
Giuliana Cortese - University of Padua (Italy)
Abstract: In many clinical designs, patients are treated by different combinations of therapies. In a two-stage design specifically, they are randomized to a primary therapy and eventually to a secondary therapy, depending on disease remission and patients' consent. Since the aim is to achieve the largest overall clinical benefit, the total effect of different combinations of first-stage and second-stage treatments on a survival outcome is of great interest. We propose a parametric estimator of the combined survival function for two-stage treatment strategies, using mixture distributions to model the possibly right-censored survival time. Observations are allowed to be censored in both stages; the first-stage duration is also modeled. The proposed parametric approach is particularly useful to investigate possible dissimilarities across strategies since parameters related to the first stage or second stage outcomes can be easily tested under a well-known likelihood framework. Simulation studies show a good performance of our estimator and an application to a two-stage randomized study on leukemia patients reveals that the procedure is easy to implement in practice.