Professor of Biostatistics
Dr. Gray has worked primarily on problems relating to the analysis of censored data and clinical trials. Some areas of current research focus are developing smoothing techniques for censored failure time data, methods for design and analysis of studies with competing risks endpoints, and methods for multivariate failure time data.
In examining the effect of potential prognostic factors for cancer patients, and in other regression problems with failure time endpoints, methods for exploring relationships are needed as an aid in model building. Methods of inference which require minimal assumptions for validity are also needed. In complete samples smoothing techniques and nonparametric estimation have been extensively studied for these purposes. Dr. Gray has worked on extending some of these techniques, especially spline and kernel estimation methods, to censored data problems. In applications to breast cancer clinical trials, these methods have made it possible to identify nonproportionality in the effects of ER status and tumor necrosis, explore appropriate modeling for variables such as number of positive nodes, tumor size, age, and percent cells in S phase, and identify an interaction in the effects of number of positive nodes and tumor size, among other effects.
In clinical trials studying therapies for chronic diseases in the elderly, it will often happen that some patients will fail from unrelated causes prior to the failure of the disease under study. For situations like this, Dr. Gray has recently studied methods for modeling the probability of failing from particular causes over time, which follows earlier work on testing for equality of these probability functions among treatment groups. Another area of interest is the appropriate definition of endpoints for such trials.
Dr. Gray’s work on methods for multivariate failure time data has focused on clustered data problems, which could arise in multi-center clinical trials if patients from the same center are more likely to have similar outcomes than patients from different centers. He has developed methods for testing for the presence of clustering, and for estimating regression parameters for clustered data. He is also currently working on methods for exploring the degree of within cluster association and estimating association parameters, and on methods for improving the efficiency of generalized estimating equations for marginal regression analysis.
Ph.D., 1983, Oregon State University