Categoria: Seminari e Convegni
Stato: Corrente

Survival Data Analysis for Cancer Data

2-6 ottobre 2017 - AULA BUZANO presso il Dipartimento di Scienze Matematiche "G. L. Lagrange" - Politecnico di Torino

Survival Data Analysis for Cancer Data

Prof. Matthieu Resche-Rigon, Prof. Sylvie Chevret
ECSTRA Team, Inserm, University of Paris Diderot

1- Introduction to survival data (2h courses, 3 hours tutorials)
We start with the definition of survival (failure time) data, and the description of their
specificities. A brief review of probability distributions for survival data is given. We then
develop the concept of likelihood for model parameter estimation in presence of right
censoring, then that of nonparametric estimation. Kaplan-Meier estimator of survival function
is presented. Last, Wald, likelihood ratio, and score tests for drawing statistical inference are
presented, and formulae for sample size computations are derived.
2- Regression models for survival data (2h courses, 3 hours tutorials)
We introduce the concept of partial likelihood function for right-censored data, and discuss
the main regression models for hazard function. We discuss the inference and model
checking for fitting such models. Composite endpoints used in Cancer (event-free survival,
progression free survival, etc.) are presented and discussed.
3- Introduction of the competing risks framework (2h courses, 3 hours tutorials)
We describe the setting of competing risks, and describe the probability functions for such
data. Nonparametric estimator of probability distribution is derived as well as statistical
testing and regression models. Strategy of modeling is discussed.
4- Multistate modeling for multivariate survival data (2h courses, 3 hours tutorials)
We consider recurrent survival data and multiple survival data, with examples from Cancer
settings. We discuss the modeling strategies according to the type of data, comparing
marginal models with conditional models.
5- Modeling clustered survival data (2h courses, 3 hours tutorials)
We describe modeling approaches for clustered survival data: stratified models, fixed and
random effects modeling. We describe permutation tests for cluster effect on a covariate,
such as treatment-by-cluster interaction in clinical trials settings.

© Politecnico di Torino - Credits