Title: On the Efficiency of Group Testing and Retesting for Estimating Rare Events

　　Speaker: Liu Aiyi

　　Time: June 26, 2018, 10:30 am

　　Location: Room 216, Main Building

 

　　Speaker Profile:

　　Dr. Aiyi Liu is a senior investigator and current Acting Chief in the Biostatistics and Bioinformatics Branch of the National Institute of Health and Human Development, USA. His research interests in statistical methods development center around sequential methodology and adaptive designs; robust methods for multidimensional outcomes, design, analysis; and methods for semicontinuous outcomes. Dr. Liu is an active member of several professional societies, including the American Statistical Association (ASA) and the International Chinese Statistical Association (ICSA). He is an elected fellow of ASA and is currently the president of ICSA. He is an adjunct professor in department of biostatistics, bioinformatics and biomathematics in Georgetown University. He has published over 160 papers in statistical research and is the recipient of a number of NIH and NICHD merit awards.

 

　　Introduction：

　　Group testing, which utilizes pooled rather than individual samples, has been widely used as a cost-effective strategy in screening for and estimating the prevalence of binary characteristics with small prevalence. While retesting is necessitated to identify infected subjects, it is not crucial to estimate the prevalence. However, one can expect statistical benefits from incorporating retesting in the estimation due to the use of additional information. Study in this context is scarce, particularly for tests with testing errors. In this paper we investigate the efficiency of group testing procedures including retesting in estimating the prevalence of a disease. We show that, as compared to the procedures with only initial group testing outcomes, when the sensitivity of the test is less than 1, retesting on subjects within either positive or negative groups can improve the efficiency of estimates; when the test is subject to no misclassification, performing retesting on the positive groups can still yield more efficient estimates. More generally, if comparing two group testing procedures with retesting, we conclude that the one with more positive groups retested always prevails over the other in terms of estimation efficiency.