A Method of Nonparametric Survival Analysis for Doubly Censored Data.

This thesis presents a new model and method of analysis for survival time data which can be right and /or left censored. Most of the recent work in survival analysis has been concerned with association between certain covariates, X, and time S (GREATERTHEQ) O to the occurrence of a certain event. This leads to the problem of determining the conditional survival distribution, G(s(VBAR)x) = P(S > s(VBAR)X=x). In the models previously presented, X is treated as taking fixed values. Here, the covariates are treated as r and om variables. Nonparametric estimates are found for the survival distribution, G(s) = P(S>s). The estimates can be computed using a General E-M Algorithm (Dempster et al (1977)). This method generalizes easily to left or right censored and tied data, and offers an alternative to the widely accepted Cox proportional hazards model. This approach yields a broadly applicable methodology for nonparametric estimation of the survival distribution.Ph.D.BiostatisticsUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/158442/1/8125110.pd