Principal Investigator: John, M. Lachin, Sc.D.

rs. Lachin and Bautista , in collaboration with Dr. K.K. Gordon Lan, former faculty member of the Center, are undertaking research into various statistical methods related to the analysis of clinical trials in cancer, and to trials in general. The principal focus of the research is to develop methods for the analysis of event count data, the analysis of longitudinal data, the properties of group sequential procedures, especially with application to these areas. Other matters of general interest such as rank analyses with informative censoring and the intention-to-treat principle have also been addressed.

or the analysis of count data, that is common in the assessment of adverse events in clinical trials, methods have been developed that allow the estimation of the mean rate, and the variance of the mixing distribution, for an over-dispersed Poisson process without the need to specify the form of the mixing distribution. We also extend these results to Poisson regression models with distribution-free estimates of the over- dispersion variance component. We have also shown that a process of these statistics, computed sequentially, can be characterized as Brownian motion.

or the analysis of longitudinal data, we have likewise relaxed the usual assumption of a random effect that is normally distributed and have described the properties of such analyses when computed sequentially over time. For distribution-free multivariate analyses of repeated measures, we have also described the joint distribution of a sequence of K-df chi-square tests.

e have described the power of group sequential procedures in general and the factors that affect the power of such sequential tests. We have also shown that if the boundary is crossed but the trial is not terminated, then the previously spent type I error may be retrieved with negligible effect on the final type I error of the sequential test and with minimal effect on power. Grant from NCI 5-R01-CA55098 1997-2000.