American Statistical Association
The now established approach to inference in survival problems, is heavily based on measure theoretic concepts and somewhat obscure convergence theorems such as the Rebolledo central limit result for stochastic integrals with respect to local square integrable martingales. For most biostatisticians, and for most applied statisticians, let alone clinical researchers with limited statistical training, the whole theory is inaccessible. It is only accessible to specialists who continue to specialize leaving everyone else to study other problems.
The purpose of this talk is to describe an alternative approach to inference based only on very well known classical results. The approach is very simple - based on what I call an asymptotic card trick - and is accessible to anyone with an elementary training in statistics. The implications however are not elementary and a very extensive range of results becomes immediately available to us. All of the well-known tests, such as the log-rank test, stratified log-rank test and weighted log-rank tests are straightforward. More complex problems such as time dependent covariates, time-dependent regression effects, crossing survival curves and multi-state models are also very straightforward. More powerful tests than those usually used are suggested for situations in which hazards may or may not be proportional and useful graphical techniques help underline inference.
|Date:||Wednesday, November 30, 2011|
|Time:||4:00 - 5:00 P.M.|
Memorial Sloan-Kettering Cancer Center
Department of Epidemiology and Biostatistics
307 East 63rd Street
(between First and Second Avenues)
3rd Floor Conference Room
New York, New York
Note: To gain access to the building, please follow the directions by the telephone in the foyer.