DESCRIPTION (Adapted from the Applicant's Abstract): When patient outcome in a clinical trial includes both adverse and efficacy events, it is ethically and scientifically desirable to account for both types of treatment effects. Current typical practice, however, is to formally design such clinical trials based on a single efficacy or time-to-event outcome, while including additional, often informal provisions for safety monitoring. This may misrepresent the scientific goals of the trial, including the power of the test for efficacy. Moreover, informal safety monitoring procedures often have very undesirable properties. The long-term objective of this research is to provide models and methods for the design, conduct and analysis of clinical trials involving multivariate outcomes that include adverse events. The aim is to provide a more realistic and reliable basis for treatment evaluation while also formalizing safety monitoring by making early stopping rules for adverse events explicit. While each of the three research projects described below was motivated by one or more clinical trials at M.D. Anderson Cancer Center, the proposed methods will be broadly applicable to similar clinical trials, both in oncology and in other areas involving adverse treatment effects. To facilitate application, portable computer software to implement the proposed methods will be developed and freely distributed. Three different but related clinical settings will be considered: (1) In the first setting, treatment effect is characterized by a two-dimensional (efficacy, safety) parameter. This parameter may arise from multinomial or bivariate discrete outcomes, including such events as 50 percent shrinkage of a solid tumor, complete remission of leukemia, toxicity, or death; or from a bivariate non-negative valued random variable, such as disease-free survival time and an index of treatment-related morbidity. Using the two-sample, one-sided test proposed by Thall and Cheng (1998) to quantify efficacy-safety trade-offs, the specific goals are to derive optimal and minimax two stage designs and more general group-sequential designs, and to extend the test by allowing smooth decision boundaries and two-sided alternatives. (2) In the second setting, each patient receives multiple treatment courses according to the common medical practice of repeating a treatment that is successful and otherwise switching to a different treatment. The research goals are (a) to extend the model and multi-course treatment evaluation and selection methods of Thall, Millikan and Sung (1998) from binary to trinary outcomes, thus incorporating adverse events, and (b) to extend the method to accommodate dose-finding. (3) The third setting deals with short-term treatment-related adverse events, "toxicity,'' that affect long-term survival. We propose to develop mixture models, estimates and group-sequential tests for survival time that use toxicity data together with the usual right censored event times. The aim is to provide a basis for safety monitoring in phase III trials that is part of the formal test, rather than an additional ad hoc procedure.