The 45th Annual Meeting of the European Society for Blood and Marrow Transplantation: Statistical Symposium – Poster Session

Abstract

Background: Studies in the field of pediatric leukemia that aim to compare allogeneic stem-cell transplantation (SCT) with conventional chemotherapy are the motivation of this work. To avoid selection bias, the design of such studies is usually based on a so-called genetic randomization. This is a simple analysis based on donor availability status according to registered donors. However, as donor search is often ceased after a patient s event, donor availability status is incompletely observed, so that this simple comparison is not possible. Here, the waiting time to transplant, the anticipated non-proportional hazards (due to increased early toxicity and potentially decreased later disease recurrences with SCT) and the primary interest in long-term survival (i.e. cure) needs to be considered simultaneously. With nonproportional hazards, commonly used statistical methods like Cox-regression with time-dependent covariates and landmark analysis show several limitations (relying on the definition of landmark times, poor interpretability as hazard ratios instead of long-term survival probabilities are evaluated, low statistical power) and these limitations may lead to wrong conclusions. A novel statistical approach to overcome these shortcomings is presented here. Methods: The pseudo-value regression technique is a powerful approach for modeling the impact of baseline covariates on long-term survival in a non-proportional hazards situation. We generalized the original pseudo-value approach to allow an adjustment for waiting time to donor identification. Although donor availability is incompletely observed, our approach mimics a genetic randomization and unbiasedly estimates survival probabilities with and without a donor at pre-specified time-points. The statistical evaluation is done in the framework of a statistical model. Hence, it is possible to adjust the analysis for potential risk-factors. This is a further major benefit of the given approach, that is useful in many practical applications. A further advantage of the proposed approach is the possibility to study the impact of waiting time on survival. Real data from childhood leukaemia are used to illustrate the practical value of the method. A simulation study was performed to investigate the statistical properties of the proposed model. Results: Regardless of whether the proportional hazard assumption holds or not, the estimated parameters are unbiased. With non-proportional hazards our approach clearly outperforms commonly used methods, like Cox-regression with time-dependent covariates and landmark analysis, with respect to statistical power and interpretability of the results. Conclusions: The proposed approach provides an unbiased, powerful and previously not available statistical tool to directly address the primary interest in survival probabilities in this common but methodologically difficult situation. Clinical Trial Registry: None Disclosure: Nothing to declare 12 34 56 78 90 () ;,: