因故取消，本次演講，造成不便之處，敬請諒解！

國立東華大學應用數學系

專    題    演    講

主講人：Regina Liu

                                     Rutgers University, USA

                                    講  題：Confidence Distribution (CD) in distribution inference and its applications to combining inferences from independent binary experiments with rare events

時  間：102 年 12 月 20 日 (星期五)  15: 20-16:50

地  點：理工一館 A324 會議室

摘   要

A challenging but recurrent problem in meta-analysis of binary experiments (e.g. clinical trials) with rare events is the presence of zero total event studies, where both treatment and control arms observe zero events. In this situation, the conventional meta-analysis approaches either exclude such studies from the analysis, or apply artificial continuity corrections to zero events. Both practices, however, are known to have undesirable consequences in inference. We propose to combine confidence distributions (e.g., p-value functions ) associated with the 2 by 2 tables. This approach can incorporate all available data in the analysis without using artificial corrections for zero events. It can also capture the appreciable difference between the effects from large and small zero total event studies, distinguishing, for example, a zero total event study with 1000 cases and 1000 controls from that with 10 cases and 10 controls. We establish an explicit formula for determining the overall type I error rate from the tests associated with individual studies. This formula provides a direct evaluation of the performance of the proposed approach and it also helps devise adjustments to further improve the power of the overall inference. In addition to the aforementioned desirable small sample properties, our approach is also shown to be efficient in the large sample setting. Numerical examples using simulated and real data on the inference of odds ratio show that, in the setting of rare events, our approach is superior to the Mantel-Haenszel, Peto and classical conditional methods. It should be stressed that although the proposed exact meta- analysis approach is motivated and illustrated throughout the paper by 2 by 2 tables, the approach is applicable to general settings of making exact inference on the common parameter in a series of independent studies. Throughout, we will also discuss the role of CD in the general framework of distribution inference.

This is joint work with Dungang Liu (Yale University) and Minge Xie (Rutgers University).

※※※   歡      迎      參      加   ※※※se1021220

附檔：