Title: Evaluating longterm survival with generalised and weighted pseudovalues in paediatric stem cell transplantation studies
Authors: Ulrike Poetschger - Childrens Cancer Research Institute (Austria) [presenting]
Harald Heinzl - Medical University of Vienna (Austria)
Martina Mittlboeck - Medical University of Vienna (Austria)
Abstract: Increasing the rate of long-term survivors is the common primary interest in paediatric oncology studies. In particular, the research question, of whether allogeneic stem-cell transplantation can improve long-term survival probabilities as compared to conventional chemotherapy, is statistically challenging. To avoid selection bias, the analysis has to be based on donor availability rather than actually performed stem-cell transplantation. Since donor search is ceased after a patient suffers an event, donor availability status is incompletely observed and a simple two-group comparison (genetic randomisation approach) is not possible anymore. Due to non-proportional hazards, the almost exclusively used Cox model with time-dependent covariates leads to ambiguous results and unreliable conclusions. Models based on generalised and weighted pseudo-values are novel statistical approaches to properly address this situation without relying on proportional hazards. The approaches allow the estimation and comparison of survival probabilities of patients with and without an available donor. Simulation studies demonstrate that the approaches are approximately unbiased, show satisfying confidence interval coverage, and are able to outperform other commonly used statistical techniques.