TY - JOUR TI - Two-stage design for phase II cancer clinical trials with multiple endpoints DO - https://doi.org/doi:10.7282/T31R6TQM PY - 2017 AB - The main purpose of a single-arm phase II cancer trial of a new regimen is to determine whether it has sufficient anti-tumor activity against a specific type of tumor to warrant its further clinical development. Such a research question can be answered under the frame- work of hypothesis testing. With the advent of targeted therapies that prolong disease stabilization, cancer patients typically experience stable disease (SD) rather than tumor shrinkage. It has been shown that patients with SD also achieve clinical benefits. There- fore, when evaluating the anti-tumor activity of a new treatment, clinicians are interested not only in overall response rate (complete or partial response(s)), but also in other types of measurements indicating clinical benefit. Taking two primary efficacy endpoints as an example, if the new treatment can improve on either endpoint(s), it may be promising for further evaluation. Therefore, "OR" logical relationship between the two primary efficacy endpoints is used when specifying the alternative hypothesis. In phase II cancer clinical trials, two-stage designs rather than single-stage ones are widely used for its possibility of early termination for futility to protect cancer patients. Motivated by two real cancer clinical trials, we propose a single-arm two-stage phase II cancer clinical trial design with two dichotomous alternative primary efficacy endpoints. Because of unknown correlation between two endpoints at the design stage, minimax rule is used to determine the optimal design, which minimizes the maximum of the expected sample size among all possible correlations, subject to the type I and II error constraints. Optimal designs for a variety of design parameters as well as the corresponding operating characteristics are provided. In addition, the statistical inferences of the design are studied. The MLE point estimators as well as confidence regions for the true event rates for the two efficacy endpoints are derived. Three types of confidence regions are obtained by inverting likelihood based test statistics: Wald, Score, and Likelihood ratio statistics. Among the three, the likelihood ratio-type confidence region performs the best in terms of good coverage probability and comparable expected area, and thus is recommended for this two-endpoint two-stage design. KW - Public Health KW - Clinical trials LA - eng ER -