Exposició GRASS
Ponents: Llorenç Badiella // Klaus Langohr - Dimecres 27 de novembre 2013
Dimecres 27 de novembre
De 10:00 a 12:00
* Exposició Llorenç Badiella: Corbes ROC
Títol: From ROC curves to Kaplan-Meier: P(X > Y) as an effect size measure for censored data.
Resum:
In order to test non-parametrically that two samples are drawn from the same survival function in the presence of censored data, several approaches are commonly used: Gehan's generalized Wilcoxon test, logrank test, Peto and Peto's test. These tests do not provide a clinical measure of the magnitude of the differences between groups. On the other hand, Efron (1967) proposed an unbiased estimation of the Mann-Whitney parameter P(X < Y) under random censorship. This statistic, commonly used in the context of diagnostic tools, can be easily interpreted from a clinical point of view and used for assessing differences between survival curves, as an effect size measure. We propose a method to estimate its variance and obtain confidence intervals based on inverse probability of censorship weighted (IPCW) U-statistics for right-censored data. A simulation study under different censorship patterns is conducted to evaluate the accuracy of the confidence intervals and the performances of the aforementioned tests.
Key words: Two-sample Non-parametric tests for censored data, Mann-Whitney parameter, Inverse probability of censorship weighting.
* Exposició Klaus Langohr: Tema 1 del llibre "CAUSAL INFERENCE".
Link del llibre: http://www.hsph.harvard.edu/miguel-hernan/causal-inference-book/