GWU Statistics Student Seminar

Speaker 1: Fengyu Zhao, GWU

Title: A Bayesian Design of Oncology Dose Ranging Study with a Binary Outcome and Empirical Comparisons

Abstract: One of the key goals of dose-ranging study in phase II is to determine the optimal dose that will be used to go into the next phase III trial. The optimal dose should be high enough to demonstrate efficacy and be low enough in toxicity in the target population. There are a few strategies and operating characteristics to determine the optimal dose. We consider some key statistical questions in designing such a dose ranging study: 1. how many doses to be selected in dose ranging study? 2. Do we use a historical ORR or a randomized control? 3. How to make multiplicity adjustment in multiple pairwise comparison? In this presentation, we will present some simulation results to compare the efficiency (i.e., correct dose selection) of Bayesian and Frequentist hypothesis testing in above scenarios. Researchers could also derive the dose selection rate in their own settings based on our built-in R function.

Speaker 2: Yi Zhang, GWU

Title: Independence-Encouraging Subsampling for Robust Nonparametric Regression

Abstract: In the information era, big data of huge sample sizes are prevalent in many fields. Big data is a blessing since it may reveal interesting patterns small data cannot offer, but meanwhile it poses challenges for analysis, computation, and storage, especially when one aims to fit a nonparametric model. To address this problem, we propose an optimal subsampling method to select a small but informative subset of data, on which a good estimation performance can be expected. The proposed method not only alleviates the computation and storage burden but also provides other benefits, e.g., facilitating backfitting for nonparametric additive modeling. Its efficacy is demonstrated numerically.