The National Institute of Diabetes and Digestive and Kidney Disease (NIDDK) has a long history of sponsoring clinical studies of new therapies. This application proposes new designs and statistical methods for NIDDK clinical studies. The primary emphasis of this application is the development of new designs for clinical trials and for dose-response studies which will better benefit the individual participant in the study. These new designs are response-driven; that is, responses from previous subjects in the study are used to ensure a higher probability that future subjects will be assigned to treatments or dose levels """"""""performing"""""""" better thus far. Response-driven designs have a rich history in biostatistical research and are attractive to clinicians, but are not utilized very often due to limitations of existing statistical techniques. A large part of this application addresses remedies which will make response-driven designs feasible for more Phase I-III clinical trials of new therapies for diseases of interest to the NIDDK. Little previous methodological work has addressed the use of response- driven designs for dose-response or toxicity studies. This project presents a randomized class of designs based on the generalized Polya urn model which allows targeting of a particular quantile and description of the dose- responsive relationship while protecting patients from being assigned to highly toxic dose levels. This class of designs is essentially nonparametric and allows a wide variety of design choices. Such designs will be of interest for pre-testing stages of new therapies for the obvious ethical benefits, but are also applicable to combined efficacy/toxicity responses, often of interest in Phase II trials. Problems with estimation and inference and issues in the practical implementation of the design are examined. Comparison with other existing methodology is proposed. Other response-driven designs are proposed, including a fully sequential design for dose-response studies, a design for titration or dose ranging experiments, a design appropriate for Phase III clinical trials with multiple treatment arms, and a design appropriate for survival trials. Each of these designs is motivated by realistic study scenarios, and ethical benefits are weighed along with issues of practical implementation; potential criticisms are addressed, and efforts made to propose solutions. Most NIDDK trials employ group sequential monitoring techiques. Two other topics, a monograph on group sequential monitoring of data in clinical trials, and a design for sequential monitoring of three treatment arms in a clinical trial when a standard randomization design is used, are also developed.
