Please activate JavaScript!
Please install Adobe Flash Player, click here for download

ISCB2014_abstract_book

62 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsWednesday, 27th August 2014 • 9:00-10:48 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust was tested with likelihood-ratio-tests. Both models were found to be flex- ible enough to reproduce the main features of the data structure and led to almost identical interpretation. References: a Estève et al, Stat-Med 1990;9:529-538 b Remontet et al, Stat-Med 2007;26:2214-2228 c Mahboubi et al, Stat-Med 2011;30:1351-1365   C33.2 An excess hazard model adjusting for lack of additional life table variables C Touraine1,2 , N Grafféo1,2 , R Giorgi1,2,3 1 INSERM, UMR912 (SESSTIM), Marseille, France, 2 Aix Marseille University, UMR_S912, IRD, Marseille, France, 3 APHM, Hôpital Timone, BIOSTIC, Marseille, France   Excess hazard model is commonly used in population-based cancer stud- ies to estimate the real impact on the excess mortality of prognostic fac- tors that influence overall mortality. In this model, the mortality observed is usually decomposed into the sum of the overall mortality and the excess mortality due to cancer. Overall mortality is obtained from population life tables stratified by sociodemographic variables (typically age, sex, calen- dar year). However, some additional variables are known to impact overall mortality. They could have a potential effect on excess mortality and are often absent in life tables (for example, ethnicity or deprivation). It has been shown that the use of a life table that lacks stratification by such a variable can lead to a biased estimate of its effect and of the other covari- ate effects on excess mortality. In this work, we propose an excess hazard model that adjusts for addition- al variables in order to reduce this bias.We extended a model proposed by Cheuvart and Ryan for grouped data in a clinical framework. In our model, the overall mortality is allowed to differ from the one from life tables by a scale parameter assuming a proportional effect. Estimates are obtained from individual data using a maximum likelihood approach. A likelihood ratio test allows testing the significance of the scale parameter.The perfor- mance of the model was evaluated by simulations considering different scenarios.The interest of the model is illustrated using a population-based dataset on colon cancer with life tables stratified or not by ethnicity. C33.3 Generalization of a log-rank type test to compare net survival distributions N Grafféo1,2 , F Castell3 , A Belot4,5,6 , R Giorgi1,2,7 1 INSERM, UMR912 (SESSTIM), Marseille, France, 2 Aix Marseille Université, UMR_S912, IRD, Marseille, France, 3 Aix Marseille Univ, CNRS, Centrale Marseille, I2M, UMR 7373, Marseille, France, 4 Hospices Civils de Lyon, Service de Biostatistique, Lyon, France, 5 Université Lyon 1, UMR 5558 Laboratoire Biostatistique-Santé, Villeurbanne, France, 6 Institut de Veille Sanitaire, DMCT, Saint- Maurice, France, 7 APHM, Hôpital Timone, BIOSTIC, Marseille, France   Net survival is the survival that would be observed, in a hypothetical world, if the disease under study were the only possible cause of death. In cancer research, by removing the effect of death from causes other than cancer, net survival allows us to compare cancer survival between differ- ent groups. Pohar-Perme et al. proposed a non-parametric consistent es- timator of net survival. However, to the best of our knowledge, there is no statistical test for the comparison of Pohar-Perme net survival functions for more than 2 groups. Our purpose is to build a generalized log-rank-type test for comparing net survival functions of several groups. Following the approach used in our previous work in the context of two groups, we expressed the log-rank type test in the counting process framework. As done in the Pohar-Perme estimator, we introduced the inverse probability weighting procedure in the counting and the at risk processes. Covariance matrix of our test statistic was obtained thanks to the martingale theory. We proved the as- ymptotic distribution of our test statistic under the null. Simulation studies were performed to evaluate the performance of our test in terms of type I error and power. We generated survival times depending on age, sex, and a covariate X defining the groups to compare. Different effects of X were considered to obtain similar groups or not regarding net survival. Results obtained under different scenarios show that our log-rank type test per- forms well in terms of type I error and power.   C33.4 Additive relative survival multistate semi-Markov model F Gillaizeau1,2,3 , E Dantan1 , M Giral2,3 , Y Foucher1,2 1 Université de Nantes, Nantes, France, 2 Centre Hospitalier Universitaire de Nantes, Nantes, France, 3 INSERM CR1064 ITUN, Nantes, France   Medical researchers are often interested to investigate the relationship be- tween explicative variables and times-to-events like disease progression or death. Such multiple times-to-events can be studied using multistate models. For chronic diseases, it may be relevant to consider semi-Markov multistate models because the transition intensities between two clinical states more likely depend on the time already spent in the current state than on the chronological time. When the cause of death for a patient is unavailable or not totally attributable to the disease, it is not possible to specifically study the associations with the excess mortality related to the disease. Relative survival allows an estimate of the net survival in the hypothetical situation where the disease under study would be the only possible cause of death. We propose here a new semi-Markov additive relative survival (SMRS) model that combines the multistate and the relative survival approaches. Using simulated data, we highlight the effectiveness of the SMRS model whose results tend to those obtained if the different causes of death are known. Regardless the parameter considered, absolute biases were lower than 0.04, and coverage rates greater than 92% (proportion of samples in which the 95% confidence intervals includes the theoretical value). The usefulness of the SMRS model is illustrated for a cohort of kidney trans- plant recipients. We have developed a package in R for the analysis of semi-Markov additive relative survival models. C33.5 Diagnostic tools for model building in net survival: use and comparison of two methods to test the proportional hazards assumption C Danieli1,2 , N Bossard1,2 , L Roche1,2 , A Belot1,2,3 , Z Uhry1,2 , L Remontet1,2 1 Hospices Civils de Lyon, Service de Biostatistique, Pierre-Bénite, France, 2 Université Lyon 1, CNRS, UMR5558, LBBE, Villeurbanne, France, 3 Institut de Veille Sanitaire, Saint-Maurice, France   Net survival is the most relevant indicator to compare cancer survivals between countries or periods. It can be obtained using the Pohar-Perme estimator or an excess mortality hazard model including the demographic variables that define the expected mortality (usually age, sex and year of diagnostic). The latter solution involves a complex model-building strate- gy that requires diagnostic tools. The main assumptions to check concern the baseline distribution, the link function, the functional form and the proportional effect of the covariate of interest. We focus on the only two methods developed to check the assumption of proportionality within the net survival context: the Stare method based on the partial residuals (similar to Schoenfeld residuals) and the Cortese method, extended from Lin method and adapted to the semi-parametric excess hazard model based on partial score processes.

Pages Overview