STATISTICAL METHODS FOR LONGITUDINAL STUDIES IN CANCER
Principal Investigator: James Rochon, Ph.D.
In this research, we proposed to address the issue of adjusting for confounders observed post-randomization in randomized clinical trials. Examples of these confounders included patient "compliance" measured through pill counts and other biochemical markers, the occurrence of co-morbid events, the use of concomitant and "rescue" medications, withdrawal from the assigned therapy, and so on. Since the confounder is observed following randomization, it can be considered as an outcome measure and analyzed accordingly. Previous research has begun from this premise and demonstrated how to adjust inferences on the primary endpoint for the influence of these confounding variables. In this study, we proposed to extend previously developed methodologies in several important directions:
(1) development of a methodology to perform inference on a primary response variable adjusting for a survival confounding measure in repeated measures experiments;
(2) development of a methodology in longitudinal data to adjust for informative censoring and other missing value problems;
(3) perform a simulation study of the proposed estimation and hypothesis testing procedures;
(4) development of procedures for analyzing bivariate repeated measure data; and,
(5) development of a general methodology for performing sample size calculations for discrete and continuous repeated measures studies.
Grant from NIH/NCI, 1-R01-CA70286- 01, 1996-1999.)