Cancer Control Research
5R29CA075971-04
Betensky, Rebecca A.
STATISTICAL METHODS FOR ANALYSIS OF FAILURE TIME DATA
AbstractDESCRIPTION (Adapted from applicant's abstract: Studies of cancer and AIDS
typically record clinical events and laboratory measurements for
participating subjects at several time points. Frequently, the times to the
clinical events and their association are of primary interest. The
relationship between the serial laboratory measurements and the times to
clinical events is of interest also for the improvement of patient
management and for the identification of potential surrogate markers. These
data are usually plagued by problems of missingness and censoring due to
missed study visits or to the discrete time observation of continuous
processes. Interval censored data arise when exact event times are censored
within intervals that are unique to each individual, and thus are
overlapping. The broad goal of this proposal is the development of methods
for analysis of interval and right censored data that addresses medical
questions frequently posed by clinicians.
The first specific aim of this proposal is the estimation of the
distribution function for bivariate and univariate interval censored failure
time data. Both nonparamaetric and "loosely parametric" estimators will be
derived. The second aim is he analysis of failure time data with
accompanying right and interval censored intermediate event times. Smooth
estimates of the hazard functions for the terminal event before and after
the intermediate event, an estimate of the survivor function adjusted for
the intermediate event, and estimates of the latency distribution between
the times will be derived. The third aim is the development of methods for
testing for independence between bivariate interval censored data, assessing
the impact of covariates on multiple interval censored outcomes, and
comparing adjusted survivor curves. The fourth aim is the development of
methods for analysis of interval censored data that are derived from serial
laboratory measurements. These include a new parametric frailty model for
interval censored data, adjustment for the measurement error and biologic
variation of the underlying processes, and estimation of smooth hazard
functions for interval censored data in the presence of a time-varying
covariate.
To accomplish these aims, local likelihood estimation, multiple imputation,
estimating equations, approximations of first passage time distributions for
continuous stochastic processes, and convex optimization theory will be
used.
|