American Statistical Association
Many biomedical research studies involve following individuals in time to observe certain types of events, generically referred to as failures. While methods for the analysis of univariate failure time data, including Kaplan-Meier curves, censored data rank tests, and Cox regression, are well developed, the same cannot be said of bivariate or higher dimensional failure time data analysis methods. Even the basic topic of nonparametric estimation of the bivariate survivor function is not fully resolved, with available methods either incorporating negative ‘mass’ assignments, or having poor moderate sample size efficiency properties. In this talk, the bivariate survivor function nonparametric maximum likelihood estimation problem will be redefined in a manner that avoids uniqueness problems, by temporarily setting aside certain doubly-censored observations and imposing local independence at failure time ‘grid’ points where empirical information is lacking. The doubly-censored observations are then incorporated in the NPMLE estimation using self-consistency. On-going work to extend these likelihood-based approaches to the regression analysis of multivariate failure time data will also be briefly mentioned.
|Date:||Wednesday, October 30, 2013|
|Time:||11:00 A.M. - 12:00 P.M.|
Rockefeller Research Laboratories
430 East 67th Street
New York, New York