主 题：Stochastic Processes and Prediction of Eventual Outcomes of Clinical Trials in Pharmaceutical R&D
主讲人：Ming T. Tan 教授
主持人：林华珍教授
时 间：2015年3月6日星期五下午3:00-4:00
地 点：通博楼B座212学术会议室
主办单位：统计学院 统计研究中心 科研处
主讲人简介：
Ming T. Tan教授 Georgetown University 生物统计系主任。是Biometrics 、Statistics in Medicine、Drug Design, Development and Therapy 等期刊副主编。Ming T. Tan教授在多个美国卫生研究院（NIH）、科学审查委员会、数据与安全监测委员会和FDA咨询委员会任职。他是美国统计协会 Fellow ，国际统计学会Elected Member。发表论文共计169篇，出版专著7本。
内容提要：
The share value of a pharmaceutical company depends greatly on clinical trials on developmental drugs. This research is motivated from clinical studies to test whether one treatment is better than a standard one. The data are monitored during the course of the trial. Statistical tests are often performed sequentially in multi-stage adaptive to data from patients at earlier stages. It is of both clinical interest, and financial interest and statistical interest to predict the eventual outcome of the clinical trial based on emerging data, which with repeated analysis form a stochastic process. The talk will focus on the case with multiple endpoints as very limited work has been done in multi-stage statistical design for multidimensional outcomes, partly due to the mathematical complexity of the problem. We construct a rank based test statistic and show that the stochastic process formed by sequentially computing this test statistic converges to a Brownian motion measured at finite information time points. Upper bound of the difference in covariance structure between the proposed stochastic process and the Brownian motion is provided for accessing the finite sample performance of the Brownian motion approximation. By combining this stochastic process with existing multi-stage sequential procedures, we can obtain a multi-stage test for comparing multidimensional outcomes of different data types. We show through simulation that such multi-stage design can preserve both type I error and statistical power well even for studies of moderate sample size, and apply the methods to two randomized clinical trials and further show its application to early stage cancer immunotherapy trial where multiple cell surface parameters are needed to characterize treatment effect. This work is in collaboration with Peng Huang, Johns Hopkins University, Baltimore, Maryland, USA.
